|
{ |
|
"paper_id": "O07-1002", |
|
"header": { |
|
"generated_with": "S2ORC 1.0.0", |
|
"date_generated": "2023-01-19T08:08:06.949502Z" |
|
}, |
|
"title": "Bayesian Topic Mixture Model for Information Retrieval", |
|
"authors": [ |
|
{ |
|
"first": "\u5433\u5b5f\u6dde", |
|
"middle": [], |
|
"last": "\u8a31\u8ed2\u777f", |
|
"suffix": "", |
|
"affiliation": { |
|
"laboratory": "", |
|
"institution": "Cheng Kung University", |
|
"location": {} |
|
}, |
|
"email": "" |
|
}, |
|
{ |
|
"first": "\u7c21\u4ec1\u5b97", |
|
"middle": [], |
|
"last": "\u570b\uf9f7\u6210\u529f\u5927\u5b78\u8cc7\u8a0a\u5de5\u7a0b\u5b78\u7cfb", |
|
"suffix": "", |
|
"affiliation": { |
|
"laboratory": "", |
|
"institution": "Cheng Kung University", |
|
"location": {} |
|
}, |
|
"email": "" |
|
} |
|
], |
|
"year": "", |
|
"venue": null, |
|
"identifiers": {}, |
|
"abstract": "In studies of automatic text processing, it is popular to apply the probabilistic topic model to infer word correlation through latent topic variables. Probabilistic latent semantic analysis (PLSA) is corresponding to such model that each word in a document is seen as a sample from a mixture model where mixture components are modeled by multinomial distribution. Although PLSA model deals with the issue of multiple topics, each topic model is quite simple and the word burstiness phenomenon is not taken into account. In this study, we present a new Bayesian topic mixture model (BTMM) to overcome the burstiness problem inherent in multinomial distribution. Accordingly, we use the Dirichlet distribution for representation of topic information beyond document level. Conceptually, the documents in the same class are generated by the associated multinomial distribution. In the experiments on TREC text corpus, we show the results of average precision and model perplexity to demonstrate the superiority of using proposed BTMM method.", |
|
"pdf_parse": { |
|
"paper_id": "O07-1002", |
|
"_pdf_hash": "", |
|
"abstract": [ |
|
{ |
|
"text": "In studies of automatic text processing, it is popular to apply the probabilistic topic model to infer word correlation through latent topic variables. Probabilistic latent semantic analysis (PLSA) is corresponding to such model that each word in a document is seen as a sample from a mixture model where mixture components are modeled by multinomial distribution. Although PLSA model deals with the issue of multiple topics, each topic model is quite simple and the word burstiness phenomenon is not taken into account. In this study, we present a new Bayesian topic mixture model (BTMM) to overcome the burstiness problem inherent in multinomial distribution. Accordingly, we use the Dirichlet distribution for representation of topic information beyond document level. Conceptually, the documents in the same class are generated by the associated multinomial distribution. In the experiments on TREC text corpus, we show the results of average precision and model perplexity to demonstrate the superiority of using proposed BTMM method.", |
|
"cite_spans": [], |
|
"ref_spans": [], |
|
"eq_spans": [], |
|
"section": "Abstract", |
|
"sec_num": null |
|
} |
|
], |
|
"body_text": [ |
|
{ |
|
"text": "\u96a8\u8457\u8cc7\u8a0a\u5927\uf97e\u6c3e\uf922\uff0c\u5404\u7a2e\uf969\u4f4d\u6587\u4ef6(digital documents)\u7684\u907d\u589e\uff0c\u4f7f\u5f97\u8cc7\u8a0a\u6aa2\uf96a\u7cbe \u78ba\ufa01\u548c\u6587\u4ef6\u6a21\u578b\u7684\u5efa\uf9f7\u65e5\u986f\u91cd\u8981\u3002\u5728\u8cc7\u8a0a\u6aa2\uf96a\u548c\u6a5f\u5668\u5b78\u7fd2\u7814\u7a76\u4e0a\uff0c\u7d71\u8a08\u578b\u672c\u6587\u6a21\u578b (statistical text model)\u5df2\u9010\u6f38\u6210\u70ba\u4e00\u500b\u91cd\u8981\u7684\u8b70\u984c\u3002\u5c31\u8cc7\u8a0a\u6aa2\uf96a\u7684\u7814\u7a76\u8005\u800c\u8a00\uff0c\u5927\u591a\uf969\u5c07 \u6587\u4ef6\u8996\u70ba\u662f bag-of-word \u7684\u8868\u793a\u6cd5\uff0c\u5617\u8a66\u7528\u7d71\u8a08\u7684\u65b9\u6cd5\uff0c\u64f7\u53d6\u6587\u5b57\u7684\u7279\u5fb5\u4ee5\u5efa\u69cb\u8cc7\u8a0a\u6aa2 \uf96a\u7684\u6a21\u5f0f\uff0c\u6b64\uf9d0\u65b9\u6cd5\u4ea6\u7a31\u70ba\u5411\uf97e\u7a7a\u9593\u6a21\u578b [32] \u3002Bag-of-word \u7684\u7f3a\u9ede\u662f\uf967\u8003\u616e\u4eba\uf9d0\u8a9e\u8a00 \u7684\u540c\u7fa9\u5b57\u8a5e(synonym)\u4ee5\u53ca\u591a\u7fa9\u5b57\u8a5e(polysemy) \u3002\u518d\u8005\uff0c\u6b64\u65b9\u6cd5\u7684\u7a7a\u9593\u7dad\ufa01\u8868\u793a\u76f8\u7576 \u65bc\u5b57\u5178\u500b\uf969\u7684\u5927\u5c0f\u3002\u9019\u610f\u8b02\u6709\u8a31\u591a\u7684\uf96b\uf969\u5fc5\u9808\u88ab\u4f30\u8a08\uff0c\u5bb9\uf9e0\u5c0e\u81f4\u6548\u80fd\u7684\ufa09\u4f4e\u3002\u5728\u6587\u737b\u4e0a\uff0c \u5df2\u6709\u4e00\u4e9b\u6587\u4ef6\u8868\u793a\u6cd5\u88ab\u63d0\u51fa\u89e3\u6c7a bag-of-word \u65b9\u9762\u7684\u4e00\u4e9b\u554f\u984c\u3002\u9996\u5148\uff0c\u6f5b\u5728\u8a9e\u610f\u5206\u6790 (Latent Semantic Analysis, LSA) [10] \uff0c\u662f\u5c07\u6587\u4ef6\u4ee5\"\u5b57\u8a5e-\u6587\u4ef6\"\u77e9\u9663\u8868\u793a\u7684\u65b9\u6cd5\u3002\u900f\u904e\u5947 \uf962\u503c\u5206\u89e3(Singular Value Decomposition, SVD)\u5c07\u6587\u4ef6\u6295\u5c04\u5230\u4e00\u500b\u4f4e\u7dad\ufa01\u7684\u8a9e\u610f\u7a7a\u9593\uff0c\u4e26 \u5047\u8a2d\u6bcf\u4e00\u5947\uf962\u503c\u53ca\u5176\u5c0d\u61c9\u7684\u5947\uf962\u5411\uf97e(singular vector)\u4ee3\u8868\u5176\u6f5b\u5728\u4e3b\u984c\u6216\u6982\uf9a3\uff0c\u4e14\u6bcf\u4e00\u6587 \u4ef6\u53ef\u7531\u53f3\u5947\uf962\u77e9\u9663\u8f49\u7f6e\u7684\ufa08\u5411\uf97e\u8868\u793a\u3002\u5728\u8cc7\u8a0a\u6aa2\uf96a\u548c\u8a9e\u97f3\u8fa8\uf9fc\u4e0a\u5df2\u8b49\u660e\u662f\u6709\u50f9\u503c\u7684\u5206\u6790 \u5de5\u5177 [2] [3] [24] \u3002\u7b2c\u4e8c\uff0c\u6a5f\uf961\u6a21\u578b(Probabilistic Model)\u7684\u57fa\u672c\u5047\u5b9a\u70ba\u89c0\u6e2c\u8cc7\uf9be\u4e0b\u7684\u4e00\u500b\u751f \u6210\u6a21\u578b\uff0c\u6b64\u6a21\u578b\u53cd\u61c9\u8cc7\uf9be\u672c\u8eab\u7684\u67b6\u69cb\u3002\u76ee\u524d\uff0c\u5df2\u6709\u4e00\u4e9b\u6a5f\uf961\u6a21\u578b\u7684\u6280\u8853\u88ab\u5ee3\u6cdb\u5730\u4f7f\u7528\u3002 \uf9b5\u5982\uff0c\u6a5f\uf961\u6f5b\u5728\u8a9e\u610f\u5206\u6790(Probabilistic Latent Semantic Analysis) [16] [17] \u4ee5\u53ca Latent Dirichlet Allocation [6] \u3002 PLSA \u6a21 \u578b \u4f5c \u6cd5 \u662f \u64f7 \u53d6 \u8207 \u6587 \u4ef6 \u95dc \uf997 \u7684 \u610f \u5411 \u6a21 \u578b (Aspect model) [18] \u3002PLSA \u6a21\u578b\u6709\u5e7e\u9805\u7f3a\u9ede [6] \uff0c\u9996\u5148\uff0c\u662f\u6c92\u6709\u76f4\u63a5\u7684\u65b9\u6cd5\u5c07\u6a5f\uf961\u5206\u914d\u7d66\u5148\u524d\u672a \u51fa\u73fe(unseen)\u7684\u6587\u4ef6\u3002\u5176\u6b21\uff0c\uf96b\uf969\uf969\uf97e\u6703\u96a8\u8457\u6587\u4ef6\uf969\uf97e\u7dda\u6027\u64f4\u589e\u3002LDA [6] \u70ba\u4e00\u500b\u8f03\u5b8c\u6574 \u7684\u751f\u6210\u6a21\u578b\uff0c\u5176\u65b9\u6cd5\u662f\u5c07\u6bcf\u4e00\u7bc7\u6587\u4ef6\u7684\u6a5f\uf961\u8996\u70ba\u6f5b\u5728\u4e3b\u984c\u4e2d\u96a8\u6a5f\u5b57\u8a5e\u6a5f\uf961\u7684\u6df7\u5408\u6a21\u578b\uff0c \u9032\u800c\u6c42\u5f97\u8a72\u7bc7\u6587\u4ef6\u51fa\u73fe\u7684\u6a5f\uf961\u503c\u3002\u7136\u800c\uff0c\u5176\u8fd1\u4f3c\u63a8\uf941\u6f14\u7b97\u6cd5\u4e26\uf967\u5bb9\uf9e0\u5be6\u73fe\u3002\u518d\u8005\uff0c\u6587\u4ef6 \u4ee5 \u591a \u9805 \u5206 \u4f48 \u8868 \u793a \u6cd5 \uff0c \u7121 \u6cd5 \u6709 \u6548 \u53d6 \u5f97 \u5b57 \u8a5e \u5728 \u6587 \u4ef6 \u4e2d \u7684 \u7a81 \u767c \u73fe \u8c61 (burstiness phenomenon) [ ", |
|
"cite_spans": [ |
|
{ |
|
"start": 202, |
|
"end": 206, |
|
"text": "[32]", |
|
"ref_id": "BIBREF31" |
|
}, |
|
{ |
|
"start": 397, |
|
"end": 401, |
|
"text": "[10]", |
|
"ref_id": "BIBREF9" |
|
}, |
|
{ |
|
"start": 576, |
|
"end": 579, |
|
"text": "[2]", |
|
"ref_id": "BIBREF1" |
|
}, |
|
{ |
|
"start": 584, |
|
"end": 588, |
|
"text": "[24]", |
|
"ref_id": "BIBREF23" |
|
}, |
|
{ |
|
"start": 725, |
|
"end": 729, |
|
"text": "[16]", |
|
"ref_id": "BIBREF15" |
|
}, |
|
{ |
|
"start": 730, |
|
"end": 734, |
|
"text": "[17]", |
|
"ref_id": "BIBREF16" |
|
}, |
|
{ |
|
"start": 766, |
|
"end": 769, |
|
"text": "[6]", |
|
"ref_id": "BIBREF5" |
|
}, |
|
{ |
|
"start": 826, |
|
"end": 830, |
|
"text": "[18]", |
|
"ref_id": "BIBREF17" |
|
}, |
|
{ |
|
"start": 845, |
|
"end": 848, |
|
"text": "[6]", |
|
"ref_id": "BIBREF5" |
|
}, |
|
{ |
|
"start": 908, |
|
"end": 911, |
|
"text": "[6]", |
|
"ref_id": "BIBREF5" |
|
}, |
|
{ |
|
"start": 1073, |
|
"end": 1074, |
|
"text": "[", |
|
"ref_id": null |
|
} |
|
], |
|
"ref_spans": [], |
|
"eq_spans": [], |
|
"section": "\u4e00\u3001\u7dd2\uf941", |
|
"sec_num": null |
|
}, |
|
{ |
|
"text": "i w \u8868\u793a\u5b57\u5178\u4e2d\u7684\u5b57\u8a5e\u5728\u6587\u4ef6\u4e2d\u51fa\u73fe\u7684\u983b\uf961\u503c\uff0c\u800c\u5b57\u5178\u901a\u5e38\u7531\u6587\u4ef6\u96c6\u4e2d\u7684\u8a13\uf996\u96c6\u5408\u6240\u64f7 \u53d6\u5f97\u5230\u3002\u6574\u500b\u6587\u4ef6\u96c6\u53ef\u4ee5\u900f\u7531\u5b57\u8a5e\u6587\u4ef6\u77e9\u9663\uf92d\u8868\u793a\uff0c\u5982\u4e0b\u6240\u793a \u23a5 \u23a5 \u23a5 \u23a5 \u23a6 \u23a4 \u23a2 \u23a2 \u23a2 \u23a2 \u23a3 \u23a1 = nm n n m m w w w w w w w w w L M M M L L 2 1 2 22 21 1 12 11", |
|
"cite_spans": [], |
|
"ref_spans": [], |
|
"eq_spans": [], |
|
"section": "\u4e00\u3001\u7dd2\uf941", |
|
"sec_num": null |
|
}, |
|
{ |
|
"text": "(1) ", |
|
"cite_spans": [], |
|
"ref_spans": [], |
|
"eq_spans": [], |
|
"section": "A", |
|
"sec_num": null |
|
}, |
|
{ |
|
"text": "EQUATION", |
|
"cite_spans": [], |
|
"ref_spans": [], |
|
"eq_spans": [ |
|
{ |
|
"start": 0, |
|
"end": 8, |
|
"text": "EQUATION", |
|
"ref_id": "EQREF", |
|
"raw_str": "\u5176\u4e2d ij w \u8868\u793a\u5b57\u5178\u4e2d\u7684\u7b2c i \u5b57\u8a5e\u5728\u7b2c j \u7bc7\u6587\u4ef6\u4e2d\u51fa\u73fe\u7684\u983b\uf961\u503c\u3002\u5728\u4e0a\u8ff0\u8868\u793a\u6cd5\u4e2d\uff0c\u7f3a\u4e4f\u4efb \u4f55\u6709\u95dc\u5b57\u8a5e\u4e4b\u9593\u7684\u8a9e\u610f\u8a0a\u606f\u3002\u56e0\u6b64\uff0c\u6709\u5176\u4ed6\u5b78\u8005\u8003\u616e\u6b64\uf9d0\u76f8\u95dc\u8a0a\u606f\uf92d\u63cf\u8ff0\u6587\u4ef6\uff0c\u7a31\u70ba\u6f5b \u5728\u8a9e\u610f\u5206\u6790(Latent Semantic Analysis, LSA)[10]\u3002LSA \u57fa\u672c\u7684\u6982\uf9a3\u662f\u4ee5\u4f4e\u7dad\ufa01\u7684\u5171\u540c \u8a9e \u610f \u56e0 \u5b50 \u5448 \u73fe \u539f \u5148 \u6587 \u4ef6 \u548c \u5b57 \u8a5e \u4e4b \u9593 \u7684 \u95dc \uf997 \u3002 \uf9dd \u7528 \u5947 \uf962 \u503c \u5206 \u89e3 ( Singular Value Decomposition, SVD) \u627e\u51fa\u5b57\u8a5e\u5c0d\u61c9\u6587\u4ef6\u7684\u8a9e\u610f\u7d50\u69cb\uff0c\u53ef\u5c07\u9ad8\u7dad\ufa01\u7684\u77e9\u9663\u8cc7\uf9be\ufa09\u4f4e\u70ba r \u7dad \ufa01\u5927\u5c0f\u4e4b\u7279\u6027\u3002\u5176\u5947\uf962\u503c\u5206\u89e3\u4e4b\u67b6\u69cb\u793a\u610f\u5716\uff0c\u5982\u5716\u4e00\u6240\u793a\u3002 A U S T V x x 1 W m W 1 D n D 1 u M u 1 v n v word vectors words documents document vectors 0 0 \u2245 ( ) n \u00d7 m ( ) r m \u00d7 ( ) r r \u00d7 ( ) n r \u00d7 \u5716\u4e00\u3001\u5947\uf962\u503c\u5206\u89e3\u4e4b\u67b6\u69cb\u793a\u610f\u5716 (\u4e8c)\u3001\u6587\u4ef6\u6df7\u5408\u6a21\u578b\u4e4b\u63a2\u8a0e 1\u3001Mixture of Unigrams Mixture of Unigram (MU)\u6a21\u578b\u662f\u5c07 Unigram \u6a21\u578b\u7d93\u7531\uf9ea\u6563\u96a8\u6a5f\u4e3b\u984c\u8b8a\uf969\u800c\u64f4\u589e [31]\u3002\u5728\u6b64\u6df7\u5408\u6a21\u578b\u4e0b\uff0c\u6bcf\u4efd\u6587\u4ef6\u7d93\u7531\u6240\u9078\u64c7\u7684\u4e3b\u984c\u6240\u7522\u751f\uff0c\u63a5\u8457\uff0c\u5f9e\u4e3b\u984c\u76f8\u95dc\u7684\u591a\u9805 \u5f0f\u7368\uf9f7\u7522\u751f\u5b57\u8a5e\u3002\u5176\u6587\u4ef6\u7684\u6a5f\uf961\u8868\u793a\u5982\u4e0b \u220f \u2211 \u220f\u2211 \u220f \u2211 = = = = w z w z w z z w P z P z P z w P w P d P z P z w P w P ) | ( ) ( ) ( ) | ( ) ( ) ( ) ( ) | ( ) (", |
|
"eq_num": "(2)" |
|
} |
|
], |
|
"section": "A", |
|
"sec_num": null |
|
}, |
|
{ |
|
"text": "EQUATION", |
|
"cite_spans": [], |
|
"ref_spans": [], |
|
"eq_spans": [ |
|
{ |
|
"start": 0, |
|
"end": 8, |
|
"text": "EQUATION", |
|
"ref_id": "EQREF", |
|
"raw_str": "N M d z w ) (d P ) | ( d z P ) | ( z w P \u5716\u4e8c\u3001PLSA \u6a21\u578b\u793a\u610f\u5716 PLSA \u6a21\u578b\u4e3b\u8981\u7684\u7279\u5fb5\uff0c\u662f\u91dd\u5c0d\u5b57\u8a5e\u548c\u6587\u4ef6\u5171\u540c\u4e8b\u4ef6\u5c0b\u6c42\u4e00\u500b\u751f\u6210\u6a21\u578b[16][17]\u3002 \u672c\u6587\u8cc7\uf9be\u96c6\u662f\u7531\u5b57\u8a5e-\u6587\u4ef6\u5c0d ) , ( w d \u6240\u7d44\u6210\uff0c\u6587\u4ef6\u4ee5 } , , { 1 N d d K \u2208 d \u8868\u793a\uff0c\u5176\u500b\uf969\u70ba N ; \u53e6\u5916\uff0c\u5b57\u8a5e\u4ee5 } , , { 1 M w w K \u2208 w \u8868\u793a\uff0c\u5b57\u5178\u76f8\u7576\u65bc\u662f M \u500b\u5b57\u8a5e\u6240\u5f62\u6210\u4e4b\u96c6\u5408\u3002\u5047\u8a2d\u6bcf\u4e00 \u5b57\u8a5e\u5728\u7d66\u5b9a\u7684\u6587\u4ef6\u4e2d\u6f5b\u5728\u4e3b\u984c } , , { 1 K z z K \u2208 z \u4e0b\u7522\u751f\u3002\u5c07\u5b57\u8a5e-\u6587\u4ef6\u5c0d ) , ( w d \u5171\u540c\u51fa\u73fe (co-occurrence)\u7684\uf997\u5408\u6a5f\uf961\u4ee5\u5f0f(3)\u8868\u793a \u2211 \u2211 = = z z d z P z w P d P z d P z w P z P w d P ) | ( ) | ( ) ( ) | ( ) ( ) ( ) , (", |
|
"eq_num": "(3)" |
|
} |
|
], |
|
"section": "A", |
|
"sec_num": null |
|
}, |
|
{ |
|
"text": "EQUATION", |
|
"cite_spans": [], |
|
"ref_spans": [], |
|
"eq_spans": [ |
|
{ |
|
"start": 0, |
|
"end": 8, |
|
"text": "EQUATION", |
|
"ref_id": "EQREF", |
|
"raw_str": "\u5728 PLSA \u6a21\u578b\u4e2d\uff0c\u6587\u4ef6\u5247\u7d93\u7531 ) | ( z w P \u7684\u56e0\u5b50\u7684\u6df7\u5408\u63cf\u7e6a\u5176\u7279\u6027\u3002\u5c07 z \u8996\u70ba\u6f5b\u5728\u8b8a\uf969\uff0c \u53ef\u4ee5\u5bb9\uf9e0\u5730\u5c0d PLSA \u6a21\u578b\uf9dd\u7528 EM \u6f14\u7b97\u6cd5\uf92d\u5b78\u7fd2\uf96b\uf969\u3002\u6700\u5927\u5316\u5c0d\uf969\u76f8\u4f3c\ufa01\u53ef\u4ee5\u8868\u793a\u6210\uff1a \u2211\u2211 \u2211 \u2211\u2211 = = d w z d w z w P z d P z P w d n w d P w d n L ) | ( ) | ( ) ( ) , ( ) , ( log ) , ( PLSA (4) \u5176 ) , ( w d n \u8868\u793a\u5b57\u8a5e\u5728\u6587\u4ef6\u4e2d\u7684\uf969\uf97e\u3002\u5728 E-step \u4e2d\uff0c\uf9dd\u7528\u76ee\u524d\u4f30\u8a08\u7684\uf96b\uf969\uf92d\u8a08\u7b97\u6f5b\u5728\u8b8a \uf969\u7684\u4e8b\u5f8c\u6a5f\uf961\uff0c\u5176\u5f0f\u5b50\u5982\u4e0b \u2211 = z z w P z d P z P z w P z d P z P w d z P ) | ( ) | ( ) ( ) | ( ) | ( ) ( ) , | ( PLSA (5) \u5728 M-step \u4e2d\uff0c\uf9dd\u7528\u6f5b\u5728\u8b8a\uf969\u5728 E-step \u4e2d\u7684\u4f30\u6e2c\uff0c\u4f7f\u5f97\u89c0\u5bdf\u7684\uf997\u5408\u5c0d\uf969\u76f8\u4f3c\ufa01\u7684\u671f\u671b\u6700 \u5927\u5316\u3002\u5176\u6240\u6709\uf96b\uf969\u7684\uf901\u65b0\u5982\u4e0b \u2211\u2211 \u2211 = w d d w d z P w d n w d z P w d n z w P ) , | ( ) , ( ) , | ( ) , ( ) | ( PLSA (6) \u2211\u2211 \u2211 = d w w w d z P w d n w d z P w d n z d P ) , | ( ) , ( ) , | ( ) , ( ) | ( PLSA (7) \u2211\u2211 \u2211\u2211 = d w d w w d n w d z P w d n z P ) , ( ) , | ( ) , ( ) ( PLSA (8) PLSA \u5728 \u8cc7 \u8a0a \u6aa2 \uf96a \u4e2d \uff0c \u53ef \u4ee5 \u85c9 \u7531 \u4f4e \u7dad \u7684 \" \u6f5b \u5728 \" \u7a7a \u9593 \u4ee3 \u66ff \u539f \u59cb \u6587 \u4ef6 \u7684 \u8868 \u793a \u3002 \u5728 Hofmann[16][17]\uf9e8\uff0c\u4ee5 ) | ( d z P \u4f5c\u70ba\u5728\u4f4e\u7dad\u7a7a\u9593\u4e4b\u6587\u4ef6\u7684\u7d44\u6210\uff0c\u5c0d\u65bc\u672a\u770b\ufa0a(unseen)\u4e4b\u6587 \u4ef6\u6216\u67e5\u8a62\uf906\uff0c\u7d93\u7531\u6700\u5927\u5316\u5c0d\uf969\u76f8\u4f3c\ufa01\u548c\u56fa\u5b9a ) | ( z w P \u53ca\u8a08\u7b97\u800c\u5f97\u3002 3\u3001Latent Dirichlet Allocation \u8fd1\u5e7e\uf98e\uf92d\uff0cLatent Dirichlet Allocation (LDA)\u88ab\u63d0\u51fa\uf92d\u6a21\u7d44\u6587\u96c6\u7684\u6f5b\u5728\u4e3b\u984c[6]\u3002\u5728 \u5927\u8a5e\u5f59\u81ea\u52d5\u8a9e\u97f3\u8fa8\uf9fc\u7cfb\u7d71\u4e0b\u4f7f\u7528\u5728\u8a9e\u8a00\u6a21\u578b\u7684\u8abf\u6574[30][33]\uff0c\u4ee5\u53ca\u5176\u4ed6\u6a5f\u5668\u5b78\u7fd2\u61c9\u7528\u4e0a \u7686\u6709\uf967\u932f\u7684\u6210\u6548[4][5]\u3002LDA \u4e3b\u8981\u662f\u514b\u670d PLSA \u6a21\u578b\u4e2d\u4e0a\u8ff0\u7684\u7f3a\u9ede\uff0c\u6bd4\u8f03 LDA \u8207 PLSA \u6a21\u578b\u76f8\uf962\u4e4b\u8655\uff0c\u5728\u65bc LDA \u5c07\u6bcf\u4e00\u7bc7\u6587\u4ef6\u7684\u6a5f\uf961\u90fd\u8996\u70ba\u6f5b\u5728\u4e3b\u984c\u4e2d\u96a8\u6a5f\u5b57\u8a5e\u6a5f\uf961\u7684\u6df7\u5408 \u6a21\u578b\uff0c\u85c9\u6b64\u53d6\u5f97\u8a72\u7bc7\u6587\u4ef6\u51fa\u73fe\u7684\u6a5f\uf961\u503c\u3002LDA \u6a21\u578b\u4f7f\u7528\u96a8\u6a5f\u8b8a\uf969\u03b8 \uf92d\u4ee3\u66ff PLSA \u6a21\u578b\u4e2d ) | ( d z P \uf96b\uf969\u3002\u03b8 \u548c z \u6709\u76f8\u540c\u7684\u7dad\ufa01\uff0c\u8868\u793a\u6587\u4ef6\u4e2d\u4e3b\u984c\u7684\u6df7\u5408\u3002\u03b8 \u5c0d\u6bcf\u4e00\u6587\u4ef6\u5f9e Dirichlet \u5206\u4f48\u53d6\u6a23\uff0c\u4ee3\u66ff\u4f30\u8a08\u6bcf\u4e00\u8a13\uf996\u6587\u4ef6\u7684\u6df7\u5408\u6a5f\uf961 ) | ( d z P \uff0c\u5c0d PLSA \u6a21\u578b\u800c\u8a00\uff0cLDA \u6240\u9700 \u8981\u7684\uf96b\uf969\uf97e\u8f03\u5c11\u3002\u5728 PLSA \u6a21\u578b\u4e2d\uff0c\u6709 K*N \u500b ) | ( d z P \uf96b\uf969\uff0c\u800c LDA \u6a21\u578b\uff0c\u5c0d\u6587\u4ef6\u7684 \u53d6\u6a23\uff0c\u03b8 \u53ea\u9700 K \u500b\uf96b\uf969\u3002 \u5728 LDA \u6a21\u578b\uf9e8\uff0c\u5047\u8a2d\u6587\u4ef6\u5f9e\u6f5b\u5728\u4e3b\u984c\u4e0a\u96a8\u6a5f\u6df7\u5408\u53d6\u6a23\uff0c\u900f\u904e\u5b57\u8a5e\u4e0a\u7684\u5206\u4f48\u63cf\u7e6a\u6bcf \u4e00\u4e3b\u984c\u7684\u7279\u6027\u3002\u5728\u6b64\u6a21\u578b\u4e2d\uff0c\u6587\u4ef6\u70ba\u89c0\u5bdf\u8b8a\uf969\uff0c\u8996\u70ba\u5b57\u8a5e\u7684\u96c6\u5408\uff0c } , , 1 { M K \u2208 d \uff0c\u6bcf\u4e00 \u5b57\u8a5e\u53d6\u6c7a\u65bc\u672a\u89c0\u5bdf\u8b8a\uf969(\u4e5f\u5c31\u662f topic) z\uff0c\u8868\u793a\u5728 } , , 1 { K K \u7684\u53ef\u80fd\u503c\uff0c\u4e26\u4e14 K \u8d85\uf96b\uf969 (hyperparameter) \u5fc5 \u9808 \u88ab \u6c7a \u5b9a \u3002 \u5728 \u6587 \u4ef6 \u7a7a \u9593 \uf9e8 \uff0c LDA \u6a21 \u578b \u5b58 \u5728 \u672a \u89c0 \u5bdf \u8b8a \uf969 \uff0c 0 ), , , ( 1 > = k K \u03b8 \u03b8 \u03b8 \u03b8 K \u4e14 1 = \u2211 k k \u03b8 \u3002\u5176\u6a21\u578b\u5982\u5716\u4e09\u6240\u793a\uff0c\u03b1 \u8868\u793a\u70ba\u4e3b\u984c\u6df7\u5408\u03b8 \u4e4b Dirichlet priori\uff0c\u800c\u5b57\u8a5e\u6a5f\uf961\u900f\u904e M K * \u77e9\u9663 \u03b2 \uf96b\uf969\u5316\uff0c\u5176\u4e2d ) | ( z w P = \u03b2 \u3002 z w M N theta alpha beta \u5716\u4e09\u3001LDA \u6a21\u578b\u793a\u610f\u5716 \u6587\u4ef6 d \u548c\u4e3b\u984c\u6df7\u5408\u03b8 \u7684\uf997\u5408\u5206\u4f48\u70ba \u220f \u2211 \u23a5 \u23a6 \u23a4 \u23a2 \u23a3 \u23a1 = w w d n z z P z w P P P ) , ( LDA ) | ( ) | ( ) | ( ) | , ( \u03b8 \u03b1 \u03b8 \u03b1 \u03b8 d (9) \u5176\u4e2d\uff0c ) , ( w d n \u8868\u793a\u5b57\u8a5e w \u5728\u6587\u4ef6 d \u4e2d\u51fa\u73fe\u7684\u500b\uf969\uff0c ) | ( \u03b1 \u03b8 P \u70ba\u03b8 \u7684 Dirichlet \u6a5f\uf961\u5206\u5e03\u3002 \u6211\u5011\u53ef\u4ee5\u5f97\u5230\u6587\u4ef6\u7684\u908a\u969b\u5206\u4f48 \u03b8 \u03b8 \u03b1 \u03b8 \u03b1 d z P z w P P P w d n w z ) , ( LDA ) | ( ) | ( ) | ( ) | ( \u222b \u220f \u2211 \u23a5 \u23a6 \u23a4 \u23a2 \u23a3 \u23a1 = d", |
|
"eq_num": "(10)" |
|
} |
|
], |
|
"section": "A", |
|
"sec_num": null |
|
}, |
|
{ |
|
"text": "EQUATION", |
|
"cite_spans": [], |
|
"ref_spans": [], |
|
"eq_spans": [ |
|
{ |
|
"start": 0, |
|
"end": 8, |
|
"text": "EQUATION", |
|
"ref_id": "EQREF", |
|
"raw_str": "\u5b9a \u7fa9 \u4e00 \u500b \u5206 \u4f48 ) , | , ( \u03c6 \u03b3 \u03b8 z q \u7684 \u8fd1 \u4f3c \u7fa4 \uff0c \u4e26 \u4e14 \u9078 \u64c7 Variational Parameters\u03b3 \u548c\u03c6 \u63a5\u8fd1\u771f\u5be6\u7684\uf969\u503c\u3002Variational \u5206\u4f48\u5b9a\u7fa9\u70ba \u220f = z z z q q z q ) | ( ) , ( ) , | , ( \u03c6 \u03b3 \u03b8 \u03c6 \u03b3 \u03b8 (11) \u5c0d\u65bc\u9019\u65b0\u6a21\u578b\uff0c\u53ef\u4ee5\u7d93\u7531 Variational Distribution \u548c True Posterior \u4e4b\u9593\u7684 KL Divergence \u6700\u5927\u5316\u5f97\u5230 ) , , | , ( \u03b2 \u03b1 \u03b8 d z P \u7684\u8fd1\u4f3c\uff0c )) , , | , ( || ) , | , ( ( min arg ) , ( ) , ( * * \u03b2 \u03b1 \u03b8 \u03c6 \u03b3 \u03b8 \u03c6 \u03b3 \u03c6 \u03b3 d z P z q D =", |
|
"eq_num": "(12)" |
|
} |
|
], |
|
"section": "A", |
|
"sec_num": null |
|
}, |
|
{ |
|
"text": "\uf96b\uf969\u4f30\u6e2c\u904e\u7a0b\uf9dd\u7528 variational EM\uff0c\u4f7f\u5f97\u5c0d\uf969\u76f8\u4f3c\ufa01\u6700\u4f4e\u754c\u9650(lower bound)\u6700\u5927\u5316\uff0c\u57fa\u65bc \u8fd1\u4f3c\u4e8b\u5f8c\u5206\u4f48 ", |
|
"cite_spans": [], |
|
"ref_spans": [], |
|
"eq_spans": [], |
|
"section": "A", |
|
"sec_num": null |
|
}, |
|
{ |
|
"text": "EQUATION", |
|
"cite_spans": [], |
|
"ref_spans": [], |
|
"eq_spans": [ |
|
{ |
|
"start": 0, |
|
"end": 8, |
|
"text": "EQUATION", |
|
"ref_id": "EQREF", |
|
"raw_str": ") | , ( d z P \u03b8 \u7684\u4e00\u7a2e\u8b8a\u5316\u5206\u4f48\uf92d\uf901\u65b0\uf96b\uf969\uff0c\u900f\u904e\u4e0b\uf99c\uf978\u500b\u6b65\u9a5f\u8fed\u4ee3\u904e\u7a0b\u3002\u5728 E-step \u4e2d\uff0c\u4f7f\u7528\u8b8a\u5316\u7684\u4e8b\u5f8c\u5206\u4f48\u8fd1\u4f3c\uff0c\u5c0d\u6bcf\u4efd\u6587\u4ef6\u627e\u5230\u591a\u8b8a\uf96b\uf969 } , { \u03c6 \u03b3 \u7684\u6700\u4f73\u5316\u503c\uff0c ]} | ) [log( exp{ \u03b3 \u03b8 \u03b2 \u03c6 E n \u221d (13) \u2211 + = n n \u03c6 \u03b1 \u03b3 (14) \u5728 M-step \uf9e8\uff0c\u4f7f\u5f97\u6709\u95dc\u6a21\u578b\uf96b\uf969\u5c0d\uf969\u76f8\u4f3c\ufa01\u6700\u5c0f\u754c\u9650\u6700\u5927\u5316\uff0c\u5c0d\u689d\u4ef6\u591a\u9805\uf96b\uf969\u7684\uf901\u65b0 \u53ef\u4ee5\u8868\u793a\u5982\u4e0b \u2211\u2211 \u221d d n dn dn w \u03c6 \u03b2", |
|
"eq_num": "(" |
|
} |
|
], |
|
"section": "A", |
|
"sec_num": null |
|
}, |
|
{ |
|
"text": "EQUATION", |
|
"cite_spans": [], |
|
"ref_spans": [], |
|
"eq_spans": [ |
|
{ |
|
"start": 0, |
|
"end": 8, |
|
"text": "EQUATION", |
|
"ref_id": "EQREF", |
|
"raw_str": "J K \u2208 h \u8868\u793a\uff0c\u4ee5\u53catopic\uff0c\u4ee5 } , , 1 { K K \u2208 z \u8868\u793a\u3002\u5176\uf96b\uf969 \u03c4 \u03c0 , \u548c \u03b2 \u500b \u5225 \u8868 \u793a theme \u7684 \u6df7 \u5408 \u7a0b \ufa01 \u6240 \u4f54 \u7684 \u6bd4 \uf9b5 ) ( j h P = \u3001 topic \u7d66 \u5b9a theme \u7684 \u6df7 \u5408 \u7a0b \ufa01 ) | ( j h z P = \u4ee5\u53ca\u6bcf\u4e00\u5b57\u8a5e\u7d66\u5b9a\u6bcf\u4e00\u4e3b\u984c\u7684\u6a5f\uf961\u503c ) | ( z w P \u3002 z w M N theme beta tau pi \u5716\u4e94\u3001TTMM \u793a\u610f\u5716 \u6bcf\u500b\u6587\u4ef6\u53ef\u4ee5\u8996\u70ba theme h \u7684\u6df7\u5408\uff0c\u8868\u793a\u70ba \u2211 \u220f \u2211 \u2211 \u23a5 \u23a6 \u23a4 \u23a2 \u23a3 \u23a1 = = = = = = j w w d n z j j h z P z w P j h P j h d P j h P P ) , ( ) | ( ) | ( ) ( ) | ( ) ( ) (d (16) \u5176\u4e2d\uff0c ) | ( j h d P = \u8868\u793a\u7d66\u5b9a\u4e00\u500b\u4e3b\u984c j h = \uff0c\u5176\u6587\u4ef6\u7684\u751f\u6210\u6a5f\uf961\uff0c\u800c ) , ( w d n \u8868\u793a\u5b57\u8a5e\u5728 \u6587\u4ef6\u4e2d\u7684\u983b\uf961\uff0c\u4e14 ) ( ) , ( d n w d n w = \u2211 \u3002\u5047\u5b9a\u6587\u96c6 D \u70ba N \u7bc7\u6587\u4ef6\u7684\u96c6\u5408\uff0c\u7d66\u5b9a\u6587\u4ef6\u6a21\u578b\uff0c \u5176\u6587\u96c6 D \u7684\u5c0d\uf969\u76f8\u4f3c\ufa01\u53ef\u4ee5\u8868\u793a\u70ba \u2211 \u2211 \u220f \u2211 \u23a5 \u23a5 \u23a5 \u23a6 \u23a4 \u23a2 \u23a2 \u23a2 \u23a3 \u23a1 \u239f \u239f \u23a0 \u239e \u239c \u239c \u239d \u239b = = = d j w w d n z j h z P z w P j h P L ) , ( TTMM ) | ( ) | ( ) ( log (17) \u5982\u540c PLSA \u4e00\u6a23\uff0c\uf96b\uf969\u4f30\u8a08\u4ea6\u53ef\u7d93\u7531 EM \u6f14\u7b97\u6cd5\u4f7f\u5f97\u5c0d\uf969\u76f8\u4f3c\ufa01\u6700\u5927\u5316\u3002\u5728 E-step \u4e2d\uff0c \u6f5b\u5728\u8b8a\uf969\u7684\u4e8b\u5f8c\u6a5f\uf961\u88ab\u4f30\u8a08\uff0c\u5982\u4e0b\u6240\u793a \u2211 \u220f \u2211 \u220f \u2211 \u23a5 \u23a6 \u23a4 \u23a2 \u23a3 \u23a1 = = \u23a5 \u23a6 \u23a4 \u23a2 \u23a3 \u23a1 = = = = j w w d n z w w d n z j h z P z w P j h P j h z P z w P j h P d j h P ) , ( ) , ( ) | ( ) | ( ) ( ) | ( ) | ( ) ( ) | (", |
|
"eq_num": "(18)" |
|
} |
|
], |
|
"section": "A", |
|
"sec_num": null |
|
}, |
|
{ |
|
"text": "EQUATION", |
|
"cite_spans": [], |
|
"ref_spans": [], |
|
"eq_spans": [ |
|
{ |
|
"start": 0, |
|
"end": 8, |
|
"text": "EQUATION", |
|
"ref_id": "EQREF", |
|
"raw_str": "\u2211 \u2032 \u2032 = \u2032 = = = z z w P j h z P z w P j h z P j h w z P ) | ( ) | ( ) | ( ) | ( ) , | (", |
|
"eq_num": "(19)" |
|
} |
|
], |
|
"section": "A", |
|
"sec_num": null |
|
}, |
|
{ |
|
"text": "\u5728 M-step \u4e0b\uff0c\u5176\u5c0d\uf969\u76f8\u4f3c\ufa01\u671f\u671b\u503c\u662f\u4f7f\u7528\u5728\u4e0a\u4e00\u968e\u6bb5\u4f30\u6e2c\u7684\u4e8b\u5f8c\u503c\uff0c\u4f7f\u5f97\u5728\u6a19\u6e96\u5316\u9650 \u5236(normalization constraint)\u689d\u4ef6\u4e0b\u6700\u5927\u5316\u3002\u6a21\u578b\uf96b\uf969\u7684\u91cd\u65b0\u4f30\u6e2c\uff0c\u53ef\u4ee5\u8868\u793a\u70ba ", |
|
"cite_spans": [], |
|
"ref_spans": [], |
|
"eq_spans": [], |
|
"section": "A", |
|
"sec_num": null |
|
}, |
|
{ |
|
"text": "EQUATION", |
|
"cite_spans": [], |
|
"ref_spans": [], |
|
"eq_spans": [ |
|
{ |
|
"start": 0, |
|
"end": 8, |
|
"text": "EQUATION", |
|
"ref_id": "EQREF", |
|
"raw_str": "N M d phi w z beta \u5716\uf9d1 BTMM \u793a\u610f\u5716 \u5728 BTMM \uf9e8\uff0c\u5047\u8a2d\u6587\u4ef6\u96c6 D \u5305\u542b\u6587\u4ef6\uf969 N \u7bc7\uff0c\u6587\u4ef6\u8868\u793a\u70ba } , , { 1 N d d K \u2208 d \uff0c\u800c\u5b57\u5178V \u76f8\u7576\u65bc\u662f M \u500b\u5b57\u8a5e\u6240\u5f62\u6210\u7684\u96c6\u5408\uff0c\u5b57\u8a5e\u4ee5 } , , { 1 M w w K \u2208 w \u8868\u793a\u3002\u672a\u89c0\u5bdf\u8b8a\uf969\u70ba\u4e3b\u984c\uff0c \u4ee5 } , , { 1 K z z K \u2208 z \u8868\u793a\u3002\u5047\u8a2d\u6587\u4ef6 d \u548c\u5b57\u8a5e w \u689d\u4ef6\u7368\uf9f7\u65bc\u7d66\u5b9a\u7684\u672a\u89c0\u5bdf\u4e3b\u984c\u8b8a\uf969 z\uff0c\u5c0d\u65bc \u6240\u7522\u751f\u7684\u6a21\u578b\uf96b\uf969\uff0c\u5b57\u8a5e\u662f\u7d93\u7531\u4e3b\u984c\u7684\u591a\u9805\u5206\u4f48\u03c6 \u6240\u7522\u751f\uff0c\u800c\u5c0d\u65bc\u5b57\u8a5e\u5206\u4f48\u7684\u5177\u9ad4\u4e3b\u984c \u591a\u9805\u5206\u4f48\u03c6 \uff0c\u53ef\u4ee5\u5f9e Dirichlet priori \uf96b\uf969 \u03b2 \u5c0d\u61c9\u7684\u4e3b\u984c z \u5f97\u5230\u3002\u53e6\u5916\uff0c\u6587\u4ef6\u662f\u5728 K \u500b\u6f5b \u5728\u4e3b\u984c\u4e0a\u4f7f\u7528 N \u500b\u6df7\u5408\uf969\u7684\u591a\u9805\u5206\u4f48\uf92d\u8868\u793a\uff0c\u4e14 1 ) | ( = \u2211 z d z P \u3002\u5728\u6a21\u578b\uf9e8\uff0c\uf96b\uf969\u96c6\u4ee5 \u96c6\u5408 )} | ( , , { d z P \u03b2 \u03c6 \uf92d\u8868\u793a\uff0c\u5728\u63a8\u6f14\u904e\u7a0b\u4e2d\uff0c\u4f7f\u7528 Dirichlet \u5206\u4f48\u65bc\u4e3b\u984c\u591a\u9805\u5206\u4f48\u4e4b\u4e0a\uff0c \u56e0\u6b64\u96b1\u85cf\uf96b\uf969\u03c6 \u53ef\u4ee5\u88ab\u5728\u5916\u7d50\u5408\u800c\uf967\u9700\u8981\u660e\u78ba\u5730\u88ab\u4f30\u8a08\uff0c\u6b64\u7c21\u5316\u904e\u7a0b\uff0c\uf967\u9700\u8981\u5728\u5c0d\u03c6 \u53d6 \u6a23\u3002\u5982\u6b64\u4e00\uf92d\uff0c\u6240\u9700\u8981\u7684\uf96b\uf969\uf97e\u5171\u6709 KN + K \u500b\u3002\u4f9d\u64da\u751f\u6210\u904e\u7a0b\uff0c\u5b57\u8a5e\u548c\u4e3b\u984c\u7684\uf997\u5408\u5206 \u4f48\u53ef\u4ee5\u8868\u793a\u70ba \u222b = \u03c6 \u03c6 \u03b2 \u03c6 \u03c6 \u03b2 d z P w P z w P ) , | ( ) | ( ) , | ( (23) \u800c\u6587\u4ef6-\u5b57\u8a5e\u5c0d ) , ( w d \u7684\uf997\u5408\u6a5f\uf961\u53ef\u4ee5\u5beb\u6210 \u2211 \u222b \u2211 = = z z d z P w P d z P d P z w P d z P d P w d P \u03c6 \u03c6 \u03b2 \u03c6 \u03c6 \u03b2 \u03b2 ) , | ( ) | ( ) | ( ) ( ) , | ( ) | ( ) ( ) | , (", |
|
"eq_num": "(24" |
|
} |
|
], |
|
"section": "A", |
|
"sec_num": null |
|
}, |
|
{ |
|
"text": "\u222b = \u00ac z d d d i dz P P z P ) , ( ) , ( ) , | ( w z w z w z (26) \u5176\u4e2d\uff0c i \u00ac z \u5b9a\u7fa9\u70ba } { i z \u2212 z \uff0c\u8868\u793a\u9664\uf9ba\u76ee\u524d\u7684\u5b57\u8a5e i w \u4e4b\u5916\uff0c\u5c0d\u6240\u6709\u5b57\u8a5e\u7684\u4e3b\u984c\u5206\u914d\u3002\u5728 BTMM \u4e2d\uff0c\uf997\u5408\u5206\u4f48\u53ef\u4ee5\u88ab\u5206\u89e3\u70ba ) | ( ) , | ( ) , | , ( d z P z w P d w z P \u03b2 \u03b2 = (27) \u7b49\u5f0f\u53f3\u908a\u7684\uf978\u500b\u5143\u7d20\u80fd\u5920\u88ab\u5206\u5225\u8655\uf9e4\uff0c\u7b2c\u4e00\u9805 ) , | ( \u03b2 z w P \u53ef\u4ee5\u7531\u7d66\u5b9a\u76f8\u95dc\u4e3b\u984c\u7684\u88ab\u89c0\u5bdf \u5b57\u8a5e\u7e3d\uf969\u4e4b\u591a\u9805\u5f0f\u5c0e\u51fa\uff0c\u5982\u5f0f(28)\u6240\u793a \u220f \u2211 \u2211 \u222b \u220f \u220f \u2211 \u220f \u222b \u0393 + \u0393 + \u0393 \u0393 \u2245 \u0393 \u0393 = = \u2212 + w w w w z w w z w w w w n z w w w w w w n n d n n d z P w P z w P w w z ) ( ) ( ) ( ) ( ) ( ) ( ! ! ) , | ( ) | ( ) , | ( ) ( ) ( 1 ) ( \u03b2 \u03b2 \u03b2 \u03b2 \u03c6 \u03c6 \u03b2 \u03b2 \u03c6 \u03b2 \u03c6 \u03c6 \u03b2 \u03c6 \u03b2 \u03c6 (28) \u5176\u4e2d\uff0c z n \u5b9a\u7fa9\u70ba\u5b57\u8a5e w \u88ab\u5206\u914d\u5230\u6f5b\u5728\u4e3b\u984c\u8b8a\uf969 z \u767c\u751f\u7684\u6b21\uf969\u3002\u5728\u5f0f(28)\u4e2d\uff0c \u220f w n w w \u03c6 \u548c \u220f \u2212 w w w 1 \u03b2 \u03c6 \u7d50 \u5408 \u662f Dirichlet \u5206 \u4f48 ) | ( w w n P \u03b2 \u03c6 + \u7684 \u672a \u6b63 \u898f \u5316 \u8b8a \u5316 \u5f62 \u5f0f \uff0c \u4e26 \uf9dd \u7528 1 ) | ( = + \u222b \u03c6 \u03b2 \u03c6 d n P w w \u63a8\u5c0e\u6240\u5f97\u3002\u904e\u7a0b\u4e2d\uff0c\uf967\u9700\u5c0e\u5165\uf96b\uf969\u03c6 \uff0c\u56e0\u70ba\u4ed6\u5011\u53ea\u662f\u5728\u88ab\u89c0\u5bdf\u8cc7\uf9be (d,w)\u548c\u5c0d\u61c9\u4e3b\u984c z \u4e4b\u99ac\u53ef\u592b\u93c8\u7684\uf9fa\u614b\u8b8a\uf969\u4e4b\u9593\u7684\u95dc\uf997\u7d71\u8a08\u3002\u8003\u616e\u5f0f(28)\u4e2d\u7684\u5206\u4f48\uff0c\u53ea\u5c0d \u5305\u542b\uf96a\u5f15 i \u4e4b\u6f5b\u5728\u8b8a\uf969 z \u4e58\u7a4d\u9805\u4fdd\uf9cd\uff0c\u5176\u4ed6\u5168\u90e8\u6d88\u53bb\u3002\uf901\u9032\u4e00\u6b65\u5730\uff0c\uf9dd\u7528\u7b49\u5f0f ) 1 ( ) 1 ( ) ( \u2212 \u0393 \u2212 = \u0393 x x x \u3002\u56e0\u6b64\uff0c\u5f0f(28)\u53ef\u4ee5\u91cd\u5beb\u70ba ' ) ( , ) ( , ' ' ) ' ( ) ( , ' ) ' ( ) ( ' ) ' ( ) ( BTMM 1 ] [ ) 1 ] ([ ) 1 ( ) ( ) ( ) | ( ) | ( ) , | ( w z i w w z i w w w z w w i z w w w z w w z w w w z w w z i i V", |
|
"cite_spans": [], |
|
"ref_spans": [], |
|
"eq_spans": [], |
|
"section": "A", |
|
"sec_num": null |
|
}, |
|
{ |
|
"text": "= = \u22c5 \u00ac \u00ac \u00ac \u00ac \u00ac \u2211 \u2211 \u2211 z z z (29) \u540c\uf9e4\uff0c\u6f5b\u5728\u4e3b\u984c\u5206\u4f48 ) | ( d z P \u53ef\u4ee5\u88ab\u63a8\u5f97\u4ee5\u4e0b\u7d50\u679c \u2211 \u00ac \u00ac \u00ac = ' ) ' ( , ) ( , BTMM ) , | ( z z i d z i d i n n d z P z (30) \u5176\u4e2d\uff0c ) , ( w d n \u8868\u793a\u5b57\u8a5e w \u5728\u6587\u4ef6 d \u4e2d\u51fa\u73fe\u7684\u500b\uf969\u3002\u6700\u5f8c\uff0c\u5c0d\u65bc\u6f5b\u5728\u8b8a\uf969\uff0c\u7531\u5f0f(29)\u3001(30) \u6211\u5011\u53ef\u4ee5\u63a8\u5c0e\u51fa\uf901\u65b0\u7b49\u5f0f\uff0c\u5176\u7d50\u679c\u70ba \u2211 \u00ac \u00ac \u22c5 \u00ac \u00ac \u00ac \u00ac \u00ac \u22c5 + + \u221d \u221d ' ) ' ( , ) ( , ) ( , ) ( , ) , | ( ) , | ( ) , , | ( z z i d z i d z z i z w z i i i i i n n V n n d z P z w P d w z P \u03b2 \u03b2 z z z (31) \u5176\u4e2d\uff0c ) ( , w z i n \u00ac \u8868\u793a\u5b57\u8a5e w \u5206\u914d\u7d66\u4e3b\u984c z \u7684\u6b21\uf969 \uff0c ) ( , z i d n \u00ac \u5305\u542b\u4e3b\u984c z \u5728\u6587\u4ef6 d \uf9e8\u88ab\u5206\u914d\u5230\u4e00 \u4e9b\u5b57\u8a5e w \u7684\u6b21\uf969\uff0c\u800c ) ( , \u22c5 \u00ac z i n \u8868\u793a\u6240\u6709\u5b57\u8a5e\u5206\u914d\u7d66\u4e3b\u984c z \u7684\u7e3d\uf969\uff0c\u6a19\u8a18 i \u00ac \u8868\u793a\u7576\u524d\u5b57\u8a5e i w \u5728 \u9019\u4e9b\u8a08\uf969\u5df2\u88ab\u79fb\u53bb\uff0c\uf967\u88ab\uf99c\u5165\u8a08\u7b97\u8003\u616e\u3002\u03b2\u8868\u793a Dirichlet priori\uff0c\u5728\u672c\u6a21\u578b\uf9e8\uff0c\u5c0d\u5168\u90e8\u5b57 \u8a5e \u03b2\u5047\u8a2d\u662f\u76f8\u540c\u7684\uff0c\u4ea6\u5373 \u03b2 \u7684\u6240\u6709\u7d44\u6210\u90e8\u5206\u90fd\u76f8\u540c\u3002 i z \u7684\u521d\u59cb\u88ab\u8a2d\u5b9a\u4ecb\u65bc\u503c 1 \u5230", |
|
"cite_spans": [], |
|
"ref_spans": [], |
|
"eq_spans": [], |
|
"section": "A", |
|
"sec_num": null |
|
} |
|
], |
|
"back_matter": [], |
|
"bib_entries": { |
|
"BIBREF0": { |
|
"ref_id": "b0", |
|
"title": "Language model adaptation based on PLSA of topics and speakers", |
|
"authors": [ |
|
{ |
|
"first": "Y", |
|
"middle": [], |
|
"last": "Akita", |
|
"suffix": "" |
|
}, |
|
{ |
|
"first": "T", |
|
"middle": [], |
|
"last": "Kawahara", |
|
"suffix": "" |
|
} |
|
], |
|
"year": 2004, |
|
"venue": "Proceedings of International Conference on Spoken Language Processing", |
|
"volume": "", |
|
"issue": "", |
|
"pages": "1045--1048", |
|
"other_ids": {}, |
|
"num": null, |
|
"urls": [], |
|
"raw_text": "Y. Akita and T. Kawahara, \"Language model adaptation based on PLSA of topics and speakers\", Proceedings of International Conference on Spoken Language Processing, pp. 1045-1048, 2004.", |
|
"links": null |
|
}, |
|
"BIBREF1": { |
|
"ref_id": "b1", |
|
"title": "Exploiting latent semantic information in statistical language modeling", |
|
"authors": [ |
|
{ |
|
"first": "J", |
|
"middle": [ |
|
"R" |
|
], |
|
"last": "Bellegarda", |
|
"suffix": "" |
|
} |
|
], |
|
"year": 2000, |
|
"venue": "Proceeding of the IEEE", |
|
"volume": "88", |
|
"issue": "", |
|
"pages": "1279--1296", |
|
"other_ids": {}, |
|
"num": null, |
|
"urls": [], |
|
"raw_text": "J. R. Bellegarda, \"Exploiting latent semantic information in statistical language modeling,\" Proceeding of the IEEE, vol. 88, No. 8, pp. 1279-1296, 2000.", |
|
"links": null |
|
}, |
|
"BIBREF2": { |
|
"ref_id": "b2", |
|
"title": "Using linear algebra for intelligent information retrieval", |
|
"authors": [ |
|
{ |
|
"first": "M", |
|
"middle": [ |
|
"W" |
|
], |
|
"last": "Berry", |
|
"suffix": "" |
|
}, |
|
{ |
|
"first": "S", |
|
"middle": [ |
|
"T" |
|
], |
|
"last": "Dumais", |
|
"suffix": "" |
|
}, |
|
{ |
|
"first": "G", |
|
"middle": [ |
|
"W" |
|
], |
|
"last": "O'brien", |
|
"suffix": "" |
|
} |
|
], |
|
"year": 1995, |
|
"venue": "SIAM Review", |
|
"volume": "37", |
|
"issue": "4", |
|
"pages": "573--595", |
|
"other_ids": {}, |
|
"num": null, |
|
"urls": [], |
|
"raw_text": "M. W. Berry, S. T. Dumais and G. W. O'Brien, \"Using linear algebra for intelligent information retrieval\", SIAM Review, vol. 37, no. 4, pp. 573-595, 1995.", |
|
"links": null |
|
}, |
|
"BIBREF3": { |
|
"ref_id": "b3", |
|
"title": "Correlated topic model", |
|
"authors": [ |
|
{ |
|
"first": "D", |
|
"middle": [ |
|
"M" |
|
], |
|
"last": "Blei", |
|
"suffix": "" |
|
}, |
|
{ |
|
"first": "J", |
|
"middle": [ |
|
"D" |
|
], |
|
"last": "Lafferty", |
|
"suffix": "" |
|
} |
|
], |
|
"year": 2006, |
|
"venue": "Advances in Neural Information Processing Systems (NIPS)", |
|
"volume": "18", |
|
"issue": "", |
|
"pages": "147--154", |
|
"other_ids": {}, |
|
"num": null, |
|
"urls": [], |
|
"raw_text": "D. M. Blei and J. D. Lafferty, \"Correlated topic model\", Advances in Neural Information Processing Systems (NIPS), vol. 18, pp. 147-154, 2006.", |
|
"links": null |
|
}, |
|
"BIBREF4": { |
|
"ref_id": "b4", |
|
"title": "Dynamic topic model", |
|
"authors": [ |
|
{ |
|
"first": "D", |
|
"middle": [ |
|
"M" |
|
], |
|
"last": "Blei", |
|
"suffix": "" |
|
}, |
|
{ |
|
"first": "J", |
|
"middle": [ |
|
"D" |
|
], |
|
"last": "Lafferty", |
|
"suffix": "" |
|
} |
|
], |
|
"year": 2006, |
|
"venue": "Proceedings of the 23rd International Conference on Machine Learning", |
|
"volume": "", |
|
"issue": "", |
|
"pages": "113--120", |
|
"other_ids": {}, |
|
"num": null, |
|
"urls": [], |
|
"raw_text": "D. M. Blei and J. D. Lafferty, \"Dynamic topic model\", Proceedings of the 23rd International Conference on Machine Learning, pp.113-120, 2006.", |
|
"links": null |
|
}, |
|
"BIBREF5": { |
|
"ref_id": "b5", |
|
"title": "Latent Dirichlet allocation", |
|
"authors": [ |
|
{ |
|
"first": "D", |
|
"middle": [ |
|
"M" |
|
], |
|
"last": "Blei", |
|
"suffix": "" |
|
}, |
|
{ |
|
"first": "A", |
|
"middle": [ |
|
"Y" |
|
], |
|
"last": "Ng", |
|
"suffix": "" |
|
}, |
|
{ |
|
"first": "M", |
|
"middle": [ |
|
"I" |
|
], |
|
"last": "Jordan", |
|
"suffix": "" |
|
} |
|
], |
|
"year": 2003, |
|
"venue": "Journal of Machine Learning Research", |
|
"volume": "3", |
|
"issue": "5", |
|
"pages": "993--1022", |
|
"other_ids": {}, |
|
"num": null, |
|
"urls": [], |
|
"raw_text": "D. M. Blei, A. Y. Ng and M. I. Jordan, \"Latent Dirichlet allocation\", Journal of Machine Learning Research, vol. 3, no. 5, pp. 993-1022, 2003.", |
|
"links": null |
|
}, |
|
"BIBREF6": { |
|
"ref_id": "b6", |
|
"title": "Topic-based document segmentation with probabilistic latent semantic analysis", |
|
"authors": [ |
|
{ |
|
"first": "T", |
|
"middle": [], |
|
"last": "Brants", |
|
"suffix": "" |
|
}, |
|
{ |
|
"first": "F", |
|
"middle": [], |
|
"last": "Chen", |
|
"suffix": "" |
|
}, |
|
{ |
|
"first": "I", |
|
"middle": [], |
|
"last": "Tsochantaridis", |
|
"suffix": "" |
|
} |
|
], |
|
"year": 2002, |
|
"venue": "Proceedings of the Eleventh International Conference on Information and Knowledge Management", |
|
"volume": "", |
|
"issue": "", |
|
"pages": "211--218", |
|
"other_ids": {}, |
|
"num": null, |
|
"urls": [], |
|
"raw_text": "T. Brants, F. Chen and I. Tsochantaridis, \"Topic-based document segmentation with probabilistic latent semantic analysis\", Proceedings of the Eleventh International Conference on Information and Knowledge Management, pp. 211-218, 2002.", |
|
"links": null |
|
}, |
|
"BIBREF7": { |
|
"ref_id": "b7", |
|
"title": "Bayesian learning for latent semantic language", |
|
"authors": [ |
|
{ |
|
"first": "J.-T", |
|
"middle": [], |
|
"last": "Chien", |
|
"suffix": "" |
|
}, |
|
{ |
|
"first": "M.-S", |
|
"middle": [], |
|
"last": "Wu", |
|
"suffix": "" |
|
}, |
|
{ |
|
"first": "C.-S", |
|
"middle": [], |
|
"last": "Wu", |
|
"suffix": "" |
|
} |
|
], |
|
"year": 2005, |
|
"venue": "Proceedings of European Conference on Speech Communication and Technology", |
|
"volume": "", |
|
"issue": "", |
|
"pages": "25--28", |
|
"other_ids": {}, |
|
"num": null, |
|
"urls": [], |
|
"raw_text": "J.-T. Chien, M.-S. Wu and C.-S. Wu, \"Bayesian learning for latent semantic language\", Proceedings of European Conference on Speech Communication and Technology, pp. 25-28, 2005.", |
|
"links": null |
|
}, |
|
"BIBREF8": { |
|
"ref_id": "b8", |
|
"title": "On latent semantic language modeling and smoothing", |
|
"authors": [ |
|
{ |
|
"first": "J.-T", |
|
"middle": [], |
|
"last": "Chien", |
|
"suffix": "" |
|
}, |
|
{ |
|
"first": "M.-S", |
|
"middle": [], |
|
"last": "Wu", |
|
"suffix": "" |
|
}, |
|
{ |
|
"first": "H.-J", |
|
"middle": [], |
|
"last": "Peng", |
|
"suffix": "" |
|
} |
|
], |
|
"year": 2004, |
|
"venue": "Proceedings of International Conference on Spoken Language Processing", |
|
"volume": "2", |
|
"issue": "", |
|
"pages": "1373--1376", |
|
"other_ids": {}, |
|
"num": null, |
|
"urls": [], |
|
"raw_text": "J.-T. Chien, M.-S. Wu and H.-J. Peng, \"On latent semantic language modeling and smoothing\", Proceedings of International Conference on Spoken Language Processing, vol. 2, pp. 1373-1376, 2004.", |
|
"links": null |
|
}, |
|
"BIBREF9": { |
|
"ref_id": "b9", |
|
"title": "Indexing by latent semantic analysis", |
|
"authors": [ |
|
{ |
|
"first": "S", |
|
"middle": [], |
|
"last": "Deerwester", |
|
"suffix": "" |
|
}, |
|
{ |
|
"first": "S", |
|
"middle": [ |
|
"T" |
|
], |
|
"last": "Dumais", |
|
"suffix": "" |
|
}, |
|
{ |
|
"first": "G", |
|
"middle": [ |
|
"W" |
|
], |
|
"last": "Furnas", |
|
"suffix": "" |
|
}, |
|
{ |
|
"first": "T", |
|
"middle": [ |
|
"K" |
|
], |
|
"last": "Landauer", |
|
"suffix": "" |
|
}, |
|
{ |
|
"first": "R", |
|
"middle": [], |
|
"last": "Harshman", |
|
"suffix": "" |
|
} |
|
], |
|
"year": 1990, |
|
"venue": "Journal of the American Society for Information Science", |
|
"volume": "41", |
|
"issue": "6", |
|
"pages": "391--407", |
|
"other_ids": {}, |
|
"num": null, |
|
"urls": [], |
|
"raw_text": "S. Deerwester, S. T. Dumais, G. W. Furnas, T. K. Landauer and R. Harshman, \"Indexing by latent semantic analysis\", Journal of the American Society for Information Science, vol. 41, no. 6, pp. 391-407, 1990.", |
|
"links": null |
|
}, |
|
"BIBREF10": { |
|
"ref_id": "b10", |
|
"title": "Maximum likelihood from incomplete data via the EM algorithm", |
|
"authors": [ |
|
{ |
|
"first": "A", |
|
"middle": [ |
|
"P" |
|
], |
|
"last": "Dempster", |
|
"suffix": "" |
|
}, |
|
{ |
|
"first": "N", |
|
"middle": [ |
|
"M" |
|
], |
|
"last": "Laird", |
|
"suffix": "" |
|
}, |
|
{ |
|
"first": "D", |
|
"middle": [ |
|
"B" |
|
], |
|
"last": "Rubin", |
|
"suffix": "" |
|
} |
|
], |
|
"year": 1977, |
|
"venue": "Journal of the Royal Statistical Society, Series B", |
|
"volume": "39", |
|
"issue": "1", |
|
"pages": "1--38", |
|
"other_ids": {}, |
|
"num": null, |
|
"urls": [], |
|
"raw_text": "A. P. Dempster, N. M. Laird and D. B. Rubin, \"Maximum likelihood from incomplete data via the EM algorithm\", Journal of the Royal Statistical Society, Series B, vol. 39, no. 1, pp. 1-38, 1977.", |
|
"links": null |
|
}, |
|
"BIBREF11": { |
|
"ref_id": "b11", |
|
"title": "Clustering documents with an exponential-family approximation of the Dirichlet compound multinomial distribution", |
|
"authors": [ |
|
{ |
|
"first": "C", |
|
"middle": [], |
|
"last": "Elkan", |
|
"suffix": "" |
|
} |
|
], |
|
"year": 2006, |
|
"venue": "Proceedings of the 23rd International Conference on Machine Learning", |
|
"volume": "", |
|
"issue": "", |
|
"pages": "289--296", |
|
"other_ids": {}, |
|
"num": null, |
|
"urls": [], |
|
"raw_text": "C. Elkan, \"Clustering documents with an exponential-family approximation of the Dirichlet compound multinomial distribution\", Proceedings of the 23rd International Conference on Machine Learning, pp. 289-296, 2006.", |
|
"links": null |
|
}, |
|
"BIBREF12": { |
|
"ref_id": "b12", |
|
"title": "On an equivalence between PLSI and LDA", |
|
"authors": [ |
|
{ |
|
"first": "M", |
|
"middle": [], |
|
"last": "Girolami", |
|
"suffix": "" |
|
}, |
|
{ |
|
"first": "A", |
|
"middle": [], |
|
"last": "Kaban", |
|
"suffix": "" |
|
} |
|
], |
|
"year": 2003, |
|
"venue": "Proceedings of the 26th Annual International ACM SIGIR Conference on Research and Development in Information Retrieval", |
|
"volume": "", |
|
"issue": "", |
|
"pages": "433--434", |
|
"other_ids": {}, |
|
"num": null, |
|
"urls": [], |
|
"raw_text": "M. Girolami and A. Kaban, \"On an equivalence between PLSI and LDA\", Proceedings of the 26th Annual International ACM SIGIR Conference on Research and Development in Information Retrieval, pp. 433-434, 2003.", |
|
"links": null |
|
}, |
|
"BIBREF13": { |
|
"ref_id": "b13", |
|
"title": "Finding scientific topics", |
|
"authors": [ |
|
{ |
|
"first": "T", |
|
"middle": [ |
|
"L" |
|
], |
|
"last": "Griffiths", |
|
"suffix": "" |
|
}, |
|
{ |
|
"first": "M", |
|
"middle": [], |
|
"last": "Steyvers", |
|
"suffix": "" |
|
} |
|
], |
|
"year": 2004, |
|
"venue": "Proceedings of the National Academy of Science", |
|
"volume": "101", |
|
"issue": "", |
|
"pages": "5228--5235", |
|
"other_ids": {}, |
|
"num": null, |
|
"urls": [], |
|
"raw_text": "T. L. Griffiths and M. Steyvers, \"Finding scientific topics\", Proceedings of the National Academy of Science, vol. 101, pp. 5228-5235, 2004.", |
|
"links": null |
|
}, |
|
"BIBREF14": { |
|
"ref_id": "b14", |
|
"title": "Overview of the Fourth Text Retrieval Conference", |
|
"authors": [ |
|
{ |
|
"first": "D", |
|
"middle": [], |
|
"last": "Harman", |
|
"suffix": "" |
|
} |
|
], |
|
"year": 1995, |
|
"venue": "", |
|
"volume": "", |
|
"issue": "", |
|
"pages": "", |
|
"other_ids": {}, |
|
"num": null, |
|
"urls": [], |
|
"raw_text": "D. Harman, Overview of the Fourth Text Retrieval Conference. 1995. Available at http://trec.nist.gov/pubs/trec4/overvies.ps.gz", |
|
"links": null |
|
}, |
|
"BIBREF15": { |
|
"ref_id": "b15", |
|
"title": "Probabilistic latent semantic analysis", |
|
"authors": [ |
|
{ |
|
"first": "T", |
|
"middle": [], |
|
"last": "Hofmann", |
|
"suffix": "" |
|
} |
|
], |
|
"year": 1999, |
|
"venue": "Proceedings of the Fifteenth Annual Conference on Uncertainty in Artificial Intelligence", |
|
"volume": "", |
|
"issue": "", |
|
"pages": "289--296", |
|
"other_ids": {}, |
|
"num": null, |
|
"urls": [], |
|
"raw_text": "T. Hofmann, \"Probabilistic latent semantic analysis\", Proceedings of the Fifteenth Annual Conference on Uncertainty in Artificial Intelligence, pp. 289-296, 1999.", |
|
"links": null |
|
}, |
|
"BIBREF16": { |
|
"ref_id": "b16", |
|
"title": "Unsupervised learning by probabilistic latent semantic analysis", |
|
"authors": [ |
|
{ |
|
"first": "T", |
|
"middle": [], |
|
"last": "Hofmann", |
|
"suffix": "" |
|
} |
|
], |
|
"year": 2001, |
|
"venue": "Machine Learning", |
|
"volume": "42", |
|
"issue": "", |
|
"pages": "177--196", |
|
"other_ids": {}, |
|
"num": null, |
|
"urls": [], |
|
"raw_text": "T. Hofmann, \"Unsupervised learning by probabilistic latent semantic analysis\", Machine Learning, vol. 42, no. 1, pp. 177-196, 2001.", |
|
"links": null |
|
}, |
|
"BIBREF17": { |
|
"ref_id": "b17", |
|
"title": "Unsupervised learning from dyadic data", |
|
"authors": [ |
|
{ |
|
"first": "T", |
|
"middle": [], |
|
"last": "Hofmann", |
|
"suffix": "" |
|
} |
|
], |
|
"year": 1999, |
|
"venue": "Advances in Neural Information Processing Systems", |
|
"volume": "11", |
|
"issue": "", |
|
"pages": "", |
|
"other_ids": {}, |
|
"num": null, |
|
"urls": [], |
|
"raw_text": "T. Hofmann, \"Unsupervised learning from dyadic data\", Advances in Neural Information Processing Systems, vol. 11. MIT Press, 1999.", |
|
"links": null |
|
}, |
|
"BIBREF18": { |
|
"ref_id": "b18", |
|
"title": "Web usage mining based on probabilistic latent semantic analysis", |
|
"authors": [ |
|
{ |
|
"first": "X", |
|
"middle": [], |
|
"last": "Jin", |
|
"suffix": "" |
|
}, |
|
{ |
|
"first": "Y", |
|
"middle": [], |
|
"last": "Zhou", |
|
"suffix": "" |
|
}, |
|
{ |
|
"first": "B", |
|
"middle": [], |
|
"last": "Mobasher", |
|
"suffix": "" |
|
} |
|
], |
|
"year": 2004, |
|
"venue": "Proceedings of the 2004 ACM SIGKDD International Conference on Knowledge Discovery and Data Mining", |
|
"volume": "", |
|
"issue": "", |
|
"pages": "197--205", |
|
"other_ids": {}, |
|
"num": null, |
|
"urls": [], |
|
"raw_text": "X. Jin, Y. Zhou and B. Mobasher, \"Web usage mining based on probabilistic latent semantic analysis\", Proceedings of the 2004 ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pp. 197-205, 2004.", |
|
"links": null |
|
}, |
|
"BIBREF19": { |
|
"ref_id": "b19", |
|
"title": "Learning in Graphical Models", |
|
"authors": [ |
|
{ |
|
"first": "M", |
|
"middle": [], |
|
"last": "Jordan", |
|
"suffix": "" |
|
} |
|
], |
|
"year": 1999, |
|
"venue": "", |
|
"volume": "", |
|
"issue": "", |
|
"pages": "", |
|
"other_ids": {}, |
|
"num": null, |
|
"urls": [], |
|
"raw_text": "M. Jordan, editor. Learning in Graphical Models. MIT Press, Cambrige, MA, 1999.", |
|
"links": null |
|
}, |
|
"BIBREF20": { |
|
"ref_id": "b20", |
|
"title": "Introduction to variational methods for graphical models", |
|
"authors": [ |
|
{ |
|
"first": "M", |
|
"middle": [], |
|
"last": "Jordan", |
|
"suffix": "" |
|
}, |
|
{ |
|
"first": "Z", |
|
"middle": [], |
|
"last": "Ghahramani", |
|
"suffix": "" |
|
}, |
|
{ |
|
"first": "T", |
|
"middle": [], |
|
"last": "Jaakkola", |
|
"suffix": "" |
|
}, |
|
{ |
|
"first": "L", |
|
"middle": [], |
|
"last": "Sail", |
|
"suffix": "" |
|
} |
|
], |
|
"year": 1999, |
|
"venue": "Machine Learning", |
|
"volume": "37", |
|
"issue": "", |
|
"pages": "183--233", |
|
"other_ids": {}, |
|
"num": null, |
|
"urls": [], |
|
"raw_text": "M. Jordan, Z. Ghahramani, T. Jaakkola, and L. Sail, \"Introduction to variational methods for graphical models\", Machine Learning, vol. 37, pp. 183-233, 1999.", |
|
"links": null |
|
}, |
|
"BIBREF21": { |
|
"ref_id": "b21", |
|
"title": "Distribution of content words and phrases in text and language modeling", |
|
"authors": [ |
|
{ |
|
"first": "S", |
|
"middle": [ |
|
"M" |
|
], |
|
"last": "Katz", |
|
"suffix": "" |
|
} |
|
], |
|
"year": 1996, |
|
"venue": "Natural Language Engineering", |
|
"volume": "2", |
|
"issue": "", |
|
"pages": "15--59", |
|
"other_ids": {}, |
|
"num": null, |
|
"urls": [], |
|
"raw_text": "S. M. Katz, \"Distribution of content words and phrases in text and language modeling\", Natural Language Engineering, vol. 2, pp. 15-59, 1996.", |
|
"links": null |
|
}, |
|
"BIBREF22": { |
|
"ref_id": "b22", |
|
"title": "Theme topic mixture model: A graphical model for document representation", |
|
"authors": [ |
|
{ |
|
"first": "M", |
|
"middle": [], |
|
"last": "Keller", |
|
"suffix": "" |
|
}, |
|
{ |
|
"first": "S", |
|
"middle": [], |
|
"last": "Bengio", |
|
"suffix": "" |
|
} |
|
], |
|
"year": 2004, |
|
"venue": "PASCAL Workshop on Learning Methods for Text Understanding and Mining", |
|
"volume": "", |
|
"issue": "", |
|
"pages": "", |
|
"other_ids": {}, |
|
"num": null, |
|
"urls": [], |
|
"raw_text": "M. Keller and S. Bengio, \"Theme topic mixture model: A graphical model for document representation\", in PASCAL Workshop on Learning Methods for Text Understanding and Mining, 2004.", |
|
"links": null |
|
}, |
|
"BIBREF23": { |
|
"ref_id": "b23", |
|
"title": "A semi-discrete matrix decomposition for latent semantic indexing in information retrieval", |
|
"authors": [ |
|
{ |
|
"first": "T", |
|
"middle": [ |
|
"G" |
|
], |
|
"last": "Kolda", |
|
"suffix": "" |
|
}, |
|
{ |
|
"first": "D", |
|
"middle": [ |
|
"P" |
|
], |
|
"last": "O'leary", |
|
"suffix": "" |
|
} |
|
], |
|
"year": 1998, |
|
"venue": "ACM Transactions on Information Systems", |
|
"volume": "16", |
|
"issue": "4", |
|
"pages": "322--346", |
|
"other_ids": {}, |
|
"num": null, |
|
"urls": [], |
|
"raw_text": "T. G. Kolda and D. P. O'Leary, \"A semi-discrete matrix decomposition for latent semantic indexing in information retrieval\", ACM Transactions on Information Systems, vol. 16, no. 4, pp. 322-346, 1998.", |
|
"links": null |
|
}, |
|
"BIBREF24": { |
|
"ref_id": "b24", |
|
"title": "Modeling word burstiness using the Dirichlet distribution", |
|
"authors": [ |
|
{ |
|
"first": "R", |
|
"middle": [], |
|
"last": "Madsen", |
|
"suffix": "" |
|
}, |
|
{ |
|
"first": "D", |
|
"middle": [], |
|
"last": "Kauchak", |
|
"suffix": "" |
|
}, |
|
{ |
|
"first": "C", |
|
"middle": [], |
|
"last": "Elkan", |
|
"suffix": "" |
|
} |
|
], |
|
"year": 2005, |
|
"venue": "Proceedings of the 22nd International Conference on Machine Learning", |
|
"volume": "", |
|
"issue": "", |
|
"pages": "545--552", |
|
"other_ids": {}, |
|
"num": null, |
|
"urls": [], |
|
"raw_text": "R. Madsen, D. Kauchak, and C. Elkan, \"Modeling word burstiness using the Dirichlet distribution\", Proceedings of the 22nd International Conference on Machine Learning, pp. 545-552, 2005.", |
|
"links": null |
|
}, |
|
"BIBREF25": { |
|
"ref_id": "b25", |
|
"title": "The Dirichlet-tree distribution", |
|
"authors": [ |
|
{ |
|
"first": "T", |
|
"middle": [], |
|
"last": "Minka", |
|
"suffix": "" |
|
} |
|
], |
|
"year": null, |
|
"venue": "", |
|
"volume": "", |
|
"issue": "", |
|
"pages": "", |
|
"other_ids": {}, |
|
"num": null, |
|
"urls": [], |
|
"raw_text": "T. Minka, \"The Dirichlet-tree distribution\", in http://research.microsoft.com/~minka/papers/dirichlet/minka-dirtree.pdf", |
|
"links": null |
|
}, |
|
"BIBREF26": { |
|
"ref_id": "b26", |
|
"title": "Estimating a Dirichlet distribution", |
|
"authors": [ |
|
{ |
|
"first": "T", |
|
"middle": [], |
|
"last": "Minka", |
|
"suffix": "" |
|
} |
|
], |
|
"year": 2000, |
|
"venue": "", |
|
"volume": "", |
|
"issue": "", |
|
"pages": "", |
|
"other_ids": {}, |
|
"num": null, |
|
"urls": [], |
|
"raw_text": "T. Minka, \"Estimating a Dirichlet distribution\", Technical Report, MIT, 2000.", |
|
"links": null |
|
}, |
|
"BIBREF27": { |
|
"ref_id": "b27", |
|
"title": "Expectation-propagation for the generative aspect model", |
|
"authors": [ |
|
{ |
|
"first": "T", |
|
"middle": [], |
|
"last": "Minka", |
|
"suffix": "" |
|
}, |
|
{ |
|
"first": "J", |
|
"middle": [], |
|
"last": "Lafferty", |
|
"suffix": "" |
|
} |
|
], |
|
"year": 2002, |
|
"venue": "Proceedings of the 18th Conference on Uncertainty in Artificial Intelligence", |
|
"volume": "", |
|
"issue": "", |
|
"pages": "352--359", |
|
"other_ids": {}, |
|
"num": null, |
|
"urls": [], |
|
"raw_text": "T. Minka and J. Lafferty, \"Expectation-propagation for the generative aspect model\", Proceedings of the 18th Conference on Uncertainty in Artificial Intelligence, pp. 352-359, 2002.", |
|
"links": null |
|
}, |
|
"BIBREF28": { |
|
"ref_id": "b28", |
|
"title": "A PLSA-based Language Model for Conversational Telephone Speech", |
|
"authors": [ |
|
{ |
|
"first": "D", |
|
"middle": [], |
|
"last": "Mrva", |
|
"suffix": "" |
|
}, |
|
{ |
|
"first": "P", |
|
"middle": [ |
|
"C" |
|
], |
|
"last": "Woodland", |
|
"suffix": "" |
|
} |
|
], |
|
"year": 2004, |
|
"venue": "Proceedings of International Conference on Spoken Language Processing", |
|
"volume": "", |
|
"issue": "", |
|
"pages": "2257--2260", |
|
"other_ids": {}, |
|
"num": null, |
|
"urls": [], |
|
"raw_text": "D. Mrva and P. C. Woodland, \"A PLSA-based Language Model for Conversational Telephone Speech\", Proceedings of International Conference on Spoken Language Processing, pp. 2257-2260, 2004.", |
|
"links": null |
|
}, |
|
"BIBREF29": { |
|
"ref_id": "b29", |
|
"title": "Unsupervised language model adaptation for mandarin broadcast conversation transcription", |
|
"authors": [ |
|
{ |
|
"first": "D", |
|
"middle": [], |
|
"last": "Mrva", |
|
"suffix": "" |
|
}, |
|
{ |
|
"first": "P", |
|
"middle": [ |
|
"C" |
|
], |
|
"last": "Woodland", |
|
"suffix": "" |
|
} |
|
], |
|
"year": 2004, |
|
"venue": "Proceedings of International Conference on Spoken Language Processing", |
|
"volume": "", |
|
"issue": "", |
|
"pages": "1961--1964", |
|
"other_ids": {}, |
|
"num": null, |
|
"urls": [], |
|
"raw_text": "D. Mrva and P. C. Woodland, \"Unsupervised language model adaptation for mandarin broadcast conversation transcription\", Proceedings of International Conference on Spoken Language Processing, pp. 1961-1964, 2004.", |
|
"links": null |
|
}, |
|
"BIBREF30": { |
|
"ref_id": "b30", |
|
"title": "Text classification from labeled and unlabeled documents using EM", |
|
"authors": [ |
|
{ |
|
"first": "K", |
|
"middle": [], |
|
"last": "Nigam", |
|
"suffix": "" |
|
}, |
|
{ |
|
"first": "A", |
|
"middle": [ |
|
"K" |
|
], |
|
"last": "Mccallum", |
|
"suffix": "" |
|
}, |
|
{ |
|
"first": "S", |
|
"middle": [], |
|
"last": "Thrun", |
|
"suffix": "" |
|
}, |
|
{ |
|
"first": "T", |
|
"middle": [], |
|
"last": "Mitchell", |
|
"suffix": "" |
|
} |
|
], |
|
"year": 2000, |
|
"venue": "Machine Learning", |
|
"volume": "39", |
|
"issue": "", |
|
"pages": "103--134", |
|
"other_ids": {}, |
|
"num": null, |
|
"urls": [], |
|
"raw_text": "K. Nigam, A. K. McCallum, S. Thrun and T. Mitchell, \"Text classification from labeled and unlabeled documents using EM\", Machine Learning, vol. 39, no. 2-3, pp. 103-134, 2000.", |
|
"links": null |
|
}, |
|
"BIBREF31": { |
|
"ref_id": "b31", |
|
"title": "Introduction to Modern Information Retrieval", |
|
"authors": [ |
|
{ |
|
"first": "G", |
|
"middle": [], |
|
"last": "Salton", |
|
"suffix": "" |
|
}, |
|
{ |
|
"first": "M", |
|
"middle": [ |
|
"J" |
|
], |
|
"last": "Mcgill", |
|
"suffix": "" |
|
} |
|
], |
|
"year": 1983, |
|
"venue": "", |
|
"volume": "", |
|
"issue": "", |
|
"pages": "", |
|
"other_ids": {}, |
|
"num": null, |
|
"urls": [], |
|
"raw_text": "G. Salton and M. J. McGill, Introduction to Modern Information Retrieval, New York: McGraw-Hill, 1983.", |
|
"links": null |
|
}, |
|
"BIBREF32": { |
|
"ref_id": "b32", |
|
"title": "Dynamic language model adaptation using variational Bayes inference", |
|
"authors": [ |
|
{ |
|
"first": "Y.-C", |
|
"middle": [], |
|
"last": "Tam", |
|
"suffix": "" |
|
}, |
|
{ |
|
"first": "T", |
|
"middle": [], |
|
"last": "Schultz", |
|
"suffix": "" |
|
} |
|
], |
|
"year": 2005, |
|
"venue": "Proceedings of European Conference on Speech Communication and Technology", |
|
"volume": "", |
|
"issue": "", |
|
"pages": "5--8", |
|
"other_ids": {}, |
|
"num": null, |
|
"urls": [], |
|
"raw_text": "Y.-C. Tam and T. Schultz, \"Dynamic language model adaptation using variational Bayes inference\", Proceedings of European Conference on Speech Communication and Technology, pp. 5-8, 2005.", |
|
"links": null |
|
}, |
|
"BIBREF33": { |
|
"ref_id": "b33", |
|
"title": "Correlated latent semantic model for unsupervised LM adaptation", |
|
"authors": [ |
|
{ |
|
"first": "Y.-C", |
|
"middle": [], |
|
"last": "Tam", |
|
"suffix": "" |
|
}, |
|
{ |
|
"first": "T", |
|
"middle": [], |
|
"last": "Schultz", |
|
"suffix": "" |
|
} |
|
], |
|
"year": 2007, |
|
"venue": "Proceedings of International Conference on Acoustics, Speech, and Signal Processing", |
|
"volume": "4", |
|
"issue": "", |
|
"pages": "41--44", |
|
"other_ids": {}, |
|
"num": null, |
|
"urls": [], |
|
"raw_text": "Y.-C. Tam and T. Schultz, \"Correlated latent semantic model for unsupervised LM adaptation\", Proceedings of International Conference on Acoustics, Speech, and Signal Processing, vol. 4, pp. 41-44, 2007.", |
|
"links": null |
|
} |
|
}, |
|
"ref_entries": { |
|
"TABREF0": { |
|
"content": "<table><tr><td>\u8868\u793a\u6a21\u578b\uf96b\uf969\uff0c\u53ef\u4ee5\u5728\u5b57\u8a5e\u5206\u914d\u5230\u4ee3\u8868\u7684\u6f5b\u5728\u4e3b\u984c\u65b9\u9762\uff0c\u7c21\u55ae\u5b9a\u7fa9\u6a21\u578b\u3002\u672c\u7814\u7a76\uf9dd\u7528\u8c9d</td></tr><tr><td>\u6c0f\u4e3b\u984c\u6df7\u5408\u6a21\u578b\u9032\ufa08\u8cc7\u8a0a\u6aa2\uf96a\u76f8\u95dc\u7814\u7a76\uff0c\u6240\u7372\u5f97\u6210\u679c\u5c0d\u65bc\u6539\u5584\u641c\u5c0b\u7cfb\u7d71\u6aa2\uf96a\u8f03\uf9e0\u5177\u6709\u76f8</td></tr><tr><td>\u7576\u7684\u61c9\u7528\u50f9\u503c\u3002\u6b64\u5916\u4e5f\u53ef\u63d0\u4f9b\u76f8\u95dc\uf9b4\u57df\u5982\u8cc7\uf9be\u63a2\u52d8\u3001\u6a5f\u5668\u5b78\u7fd2\u7b49\uf9b4\u57df\u9032\ufa08\u6df1\u5165\u63a2\u8a0e\u3002\u672c</td></tr><tr><td>\u6587\u63a5\u4e0b\uf92d\u7ae0\u7bc0\u7d44\u7e54\u5982\u4e0b\u3002\u7b2c\u4e8c\u7ae0\u63a2\u8a0e\u76ee\u524d\u6587\u737b\u4e2d\u5404\u7a2e\u76f8\u95dc\u7684\u6587\u4ef6\u6a21\u578b\u7814\u7a76\u65b9\u6cd5\u3002\u7b2c\u4e09\u7ae0</td></tr><tr><td>\u5c07\uf96f\u660e\u672c\u6587\u6240\u63d0\u51fa\u7684\u65b9\u6cd5\uff0c\u4e26\u6bd4\u8f03\u5e7e\u7a2e\u4e3b\u8981\u6a21\u578b\u7684\u5dee\uf962\u3002\u7b2c\u56db\u7ae0\u70ba\u672c\u6587\u63d0\u51fa\u7684\u65b9\u6cd5\u548c\u5176</td></tr><tr><td>\u4ed6\u4f5c\u6cd5\u6bd4\u8f03\u5be6\u9a57\u6548\u80fd\u5206\u6790\u7684\u7d50\u679c\uff0c\u7528\u4ee5\u8b49\u660e\u672c\u7814\u7a76\u65b9\u6cd5\u7684\u6548\u76ca\u53ca\u7d50\u679c\u8a0e\uf941\u3002\u6700\u5f8c\uff0c\u7b2c\u4e94</td></tr><tr><td>\u7ae0\u70ba\u672c\u6587\u7684\u7d50\uf941\u4ee5\u53ca\u672a\uf92d\u7684\u7814\u7a76\u65b9\u5411\u3002</td></tr><tr><td>\u4e8c\u3001\u76f8\u95dc\u6587\u737b\u63a2\u8a0e</td></tr><tr><td>\u5728\u8a31\u591a\u7684\u61c9\u7528\u4e0a\uff0c\u8cc7\u8a0a\u6aa2\uf96a\u548c\u6a5f\u5668\u5b78\u7fd2\u53ef\u4ee5\uf96f\u5bc6\uf967\u53ef\u5206\u3002\u672c\u7ae0\uff0c\u6211\u5011\u5c07\u63a2\uf96a\u4e00\u4e9b\u8f03</td></tr><tr><td>\u5177\u9ad4\u3001\u719f\u77e5\u7684\u6a5f\uf961\u7d71\u8a08\u6a21\u578b\u3002\u9996\u5148\uff0c\u7c21\u55ae\u63cf\u8ff0\u5728\u8cc7\u8a0a\u6aa2\uf96a\u4e2d\u8f03\u5e38\ufa0a\u7684\u6587\u4ef6\u8868\u793a\u6cd5</td></tr><tr><td>[10][32]\u3002\u63a5\u8457\uff0c\u91dd\u5c0d\u5ee3\u6cdb\u7684\u751f\u6210\u6a21\u578b\u505a\uf901\u6df1\u5165\u7684\u63a2\u8a0e\uff0c\u5176\u4e2d\u5305\u542b\u4e00\u4e9b\u6a5f\uf961\u6a21\u578b\u548c\u6df7\u5408</td></tr><tr><td>\u6a21\u578b\u7b49\u5716\u5f62\u6a21\u578b\u8868\u793a\u5f0f[6][16][17][23] [31]\u3002</td></tr><tr><td>(\u4e00)\u3001\u6587\u4ef6\u8868\u793a\u6cd5</td></tr></table>", |
|
"html": null, |
|
"type_str": "table", |
|
"num": null, |
|
"text": "12][25]\u3002\u6240\u8b02\u300c\u7a81\u767c\u73fe\u8c61\u300d\u610f\u6307\uff0c\u5b57\u8a5e\u5728\u6587\u4ef6\u4e2d\u51fa\u73fe\u904e\u4e00\u6b21\u4e4b\u5f8c\uff0c\u5f88\u6709\u53ef \u80fd\u6703\u518d\u51fa\u73fe\u7684\u60c5\u5f62[22]\u3002\u4e00\u822c\u800c\u8a00\uff0c\u5b57\u8a5e\u5728\u6587\u96c6\uf9e8\u4e00\u822c\u5206\u70ba\u4e09\u7a2e\u7bc4\u7587\uff0c\u5373\u5e38\ufa0a(common)\u3001 \u4e00\u822c(average)\u548c\u7a00\u6709(rare)\u3002\u96d6\u7136\u591a\u9805\u5f0f\u8868\u793a\u80fd\u7372\u5f97\u5e38\ufa0a\u5b57\u8a5e\u7684\u7a81\u767c\u6027\uff0c\u4f46\u662f\u5c0d\u65bc\u4e00\u822c\u548c \u7a00\u6709\u5b57\u8a5e\u7684\u7a81\u767c\u6027\u4e26\u672a\u88ab\u6b63\u78ba\u7684\u6a21\u7d44\u5316\u3002\u800c\u900f\u904e Dirichlet \u5206\u4f48\uf92d\u66ff\u4ee3\u591a\u9805\u5206\u4f48\uff0c\u53ef\u4ee5\u8da8 \u7de9\u7a81\u767c\u73fe\u8c61\u7684\u554f\u984c[25]\u3002\u5728\u672c\u7814\u7a76\u4e2d\uff0c\u5c0d\u65bc\u6a5f\uf961\u548c\u4e3b\u984c\u6df7\u5408\u6a21\u578b\u554f\u984c\u611f\u8208\u8da3\uff0c\u5c07\u63a2\u8a0e\u5e7e \u500b\u8f03\u5148\u9032\u7684\u5716\u5f62\u6a21\u578b[6][16][23][25]\uff0c\u671f\u671b\u85c9\u7531\u76f8\u95dc\u80cc\u666f\uff0c\uf92d\u6539\u5584\u73fe\u6709\u7684\u6587\u4ef6\u6a21\u578b\u67b6\u69cb\u3002 \u672c\u6587\u4e2d\u4ee5 PLSA \u6a5f\uf961\u6a21\u578b\u70ba\u57fa\u790e\uff0c\u5728\u6df7\u5408\u6a21\u5f0f\u7684\u7d50\u5408\u4e0a\uff0c\u900f\u904e\u8c9d\u6c0f\u65b9\u6cd5\u4f7f\u7528 Dirichlet \u5206\u4f48\u6c7a\u5b9a\u5404\u500b\u5206\u914d\u6240\u4f54\u7684\u6bd4\uf9b5\uff0c\u7a31\u4e4b\u70ba\u8c9d\u6c0f\u4e3b\u984c\u6df7\u5408\u6a21\u578b(Bayesian Topic Mixture Model, BTMM )\u3002\u900f\u904e Gibbs \u62bd\u6a23\u6cd5\uf92d\u4f30\u8a08\u6240\u9700\u7684\uf96b\uf969\u3002Gibbs \u62bd\u6a23\u6cd5\u7684\u512a\u52e2\u662f\uf967\u9700\u8981\u660e\u78ba\u5730" |
|
}, |
|
"TABREF3": { |
|
"content": "<table><tr><td colspan=\"4\">\u5982\u524d\u6240\u8ff0\uff0cLDA\u6a21\u578b\u8fd1\u4f3c\u63a8\uf941\u6f14\u7b97\u6cd5\u4e26\u7121\u6cd5\u5f97\u5230\u6b63\u89e3(Exact Solution)\u4e14\u8a08\u7b97\u8907\u96dc\ufa01</td></tr><tr><td colspan=\"4\">\u589e\u52a0\u3002\u70ba\uf9ba\u514b\u670d\u9019\u500b\u554f\u984c\uff0cKeller\u548cBengio[23]\u63d0\u51fa\u4e00\u500b\u6b63\u63a8\uf941\u4e14\uf9e0\u8655\uf9e4\u7684\u6a21\u578b\uff0c\u7a31\u4e4b\u70ba</td></tr><tr><td colspan=\"4\">Theme Topic Mixture Model (TTMM)\u3002\u5728TTMM\uf9e8\uff0c\u6587\u4ef6\u7a7a\u9593\u7684\u8b8a\uf969\u7a31\u70baTheme\uff0c\uf967\u540c</td></tr><tr><td colspan=\"4\">\u65bcLDA\uff0cTTMM\u5c0d\u65bctopic\u7684\u6df7\u5408\u7a0b\ufa01\u6240\u4f54\u7684\u6bd4\uf9b5\uf9dd\u7528\uf9ea\u6563\u6709\u9650\u96c6(discrete finite set)\uf92d\u4ee3</td></tr><tr><td colspan=\"4\">\u66ff\uf99a\u7e8c\u7a7a\u9593\u7684\u4f7f\u7528\u3002\u5982\u5716\u4e94\u6240\u793a\uff0c\u6b64\u6a21\u578b\u7684\u89c0\u5bdf\u8b8a\uf969\u70ba\u6587\u4ef6d\uff0c\u53ef\u8996\u70ba\u5b57\u8a5ew\u7684\u96c6\u5408\uff0c</td></tr><tr><td>\u800c\u672a\u89c0\u5bdf\u8b8a\uf969\u70batheme\uff0c\u4ee5</td><td>, 1 {</td><td>,</td><td>}</td></tr></table>", |
|
"html": null, |
|
"type_str": "table", |
|
"num": null, |
|
"text": "15) \u800c\uf96b\uf969\u03b1 \u53ef\u4ee5\u900f\u904e Newton-Raphson \u6f14\u7b97\u6cd5\u6c42\u5f97[27]\u3002Girolamin \u548c Kaban [13]\uf96f\u660e\u7576 Dirichlet \u5206\u4f48\u76f8\u540c\u6642\uff0cPLSA \u6a21\u578b\u5be6\u969b\u4e0a\u662f LDA \u7684\u4e00\u500b\u7279\uf9b5\u3002 4\u3001Theme Topic Mixture Model" |
|
}, |
|
"TABREF6": { |
|
"content": "<table><tr><td>K \u4e4b\u9593\uff0c \u6c7a\u5b9a\u99ac\u53ef\u592b\u93c8(Markov chain)\u7684\u521d\u59cb\uf9fa\u614b\u3002\u7136\u5f8c\u57f7\ufa08\u5e7e\u500b\u8fed\u4ee3\u6b21\uf969\uff0c\u76f4\u5230\u93c8\u63a5\u8fd1\u76ee\u6a19\u5206 \u4f48\uff0c i z \u76ee\u524d\u503c\u5c07\u6703\u88ab\u8a18\uf93f\u4e0b\uf92d\u3002 (\u4e09) \uf967\u540c\u6a21\u578b\u4e4b\u95dc\uf997\u548c\u6bd4\u8f03 \u5728\u672c\u7ae0\u7bc0\u4e2d\uff0c\u6211\u5011\u5c07\u8a0e\uf941\u4e26\u6bd4\u8f03\u524d\u9762\u7ae0\u7bc0\u6240\u63cf\u8ff0\u7684\u5e7e\u500b\u6a21\u578b\u3002\u5f9e\u4e3b\u8981\u7684\u65b9\u7a0b\u5f0f\u770b \uf92d\uff0c\u6a21\u578b\u4e4b\u9593\u5dee\uf962\u5927\u540c\u5c0f\uf962\u3002\u70ba\uf9ba\u5bb9\uf9e0\uf9e4\u89e3\u6587\u4ef6\u6a21\u578b\u751f\u6210\u7684\u5dee\uf962\u3002\u91dd\u5c0d\u672c\u6587\u6240\u63d0\u51fa\u7684\u65b9 \u6cd5\u548c\u7b2c\u4e8c\u7ae0\u6240\u63d0\u5230\u7684\u6a21\u578b\uff0c\u5982 PLSA\u3001LDA \u4ee5\u53ca TTMM \u7b49\uff0c\u5c0d\u5176\u7d44\u6210\u5143\u7d20(\u5b57\u8a5e\u3001\u4e3b\u984c \u53ca\u6587\u4ef6)\u4e4b\u751f\u6210\u6a5f\uf961/\u5206\u4f48\u8868\u793a\uff0c\u7c21\u55ae\u6b78\u7d0d\u5982\u4e0b\u8868\u4e00\u6240\u793a\u3002 \u8868\u4e00\u3001\uf967\u540c\u65b9\u6cd5\u4e4b\u5404\u7d44\u6210\u5143\u7d20\u6a5f\uf961\u5206\u4f48\u8868\u793a Word Topic Document PLSA ) | ( z w P ) | ( d z P ) , ( w d P LDA ) ( , | \u03b2 \u03b2 Mult z w ) ( \u03b8 Mult z ) ( \u03b1 \u03b8 Dir TTMM ) ( , | \u03b2 \u03b2 Mult z w ) ( \u03c4 Mult z ) ( \u03c0 Mult h BTMM ) ( z Mult w \u03c6 , ) ( \u03b2 \u03c6 Dir z ) | ( d z P ) , ( w d P \u5047\u8a2d\u5728\u6587\u4ef6\u96c6\uf9e8\u6709 N \u7bc7\u6587\u4ef6\uff0c\u5b57\u5178\uf969\u5927\u5c0f\u70ba M\uff0c|d|\u8868\u793a\u6587\u4ef6\u9577\ufa01\uff0c\u4ea6\u5373\u5728\u6587\u4ef6\u7684 \u5b57\u8a5e\u500b\uf969\uff0cK \u70ba\u4e3b\u984c(Topic)\u500b\uf969\uff0cJ \u70ba theme \uf969\u76ee\u4ee5\u53ca\u7fa4\u7d44\u500b\uf969\u70ba C\u3002\u5c0d\u65bc\u6a21\u578b\u7684\u7a7a\u9593 \u8907\u96dc\ufa01\u6bd4\u8f03\uff0c\u4ee5\u8868\u4e09\u505a\u4e00\u7c21\u55ae\u7684\u95e1\u8ff0\u3002\u5404\u500b\u6a21\u578b\u6240\u9700\u7684\uf96b\uf969\uf97e\uff0c\u5f9e\u8868\u4e8c\u53ef\u4ee5\u5f97\u77e5\uff0cTTMM PLSA LDA TTMM BTMM Parameters O(KN+KM) O(K+KM) O(J+JK+KM) O(KN+K) \u56db\u3001\u5be6\u9a57 (\u4e00)\u3001\u5be6\u9a57\u6587\u96c6\u53ca\u8a2d\u5b9a\uf96f\u660e \u5728\u672c\u6587\u7684\u5be6\u9a57\u4e2d\uff0c\u6211\u5011\u4f7f\u7528 TREC \u6240\u6536\u96c6\u7684\u6587\u96c6\uff0c\u5206\u5225\u70ba Associated Press newswire (AP) 88 \u548c \u8868\u4e09\u3001TREC \u6587\u96c6\u7684\u7d71\u8a08\u8cc7\u8a0a Collection Description Size (MB) #Doc. Vocabulary Size WSJ89 Wall Street Journal (1989), Disk2 36.5 12,380 17,732 AP88 Associate Press (1988), Disk1 237 79,908 8,783 (\u4e8c)\u3001\u5be6\u9a57\u7d50\u679c 1\u3001\uf967\u540c\u6a21\u578b\u5728\u6aa2\uf96a\u6548\u80fd\u7684\u5f71\u97ff \u9996\u5148\uff0c\u6bd4\u8f03\uf967\u540c\u7684\u65b9\u6cd5\u5c0d TREC \u6587\u4ef6\u96c6\u5728\u6587\u4ef6\u6aa2\uf96a\u4e0a\u6548\u80fd\u7684\u6bd4\u8f03\u3002\u5f9e\u5716\u4e03\u548c\u5716\u516b\u8868 \u793a\uf967\u540c\u6a21\u578b\u4e4b Precision-Recall \u66f2\u7dda\uff0c\u5206\u5225 WST89 \u548c AP88 \u7684\u7d50\u679c\uff0c\u800c\u8868\u56db\u70ba mAP \u5728\uf967 \u540c\u6a21\u578b\u6240\u8a08\u7b97\u7684\u7d50\u679c\u3002\u5f9e\u9019\u4e9b\u5716\u8868\u7576\u4e2d\uff0c\u53ef\u4ee5\u770b\u51fa\u4ee5\u4e3b\u984c\u70ba\u57fa\u790e\u7684\u6587\u4ef6\u6a21\u578b\uff0c\u7686\u6bd4\u8a9e\u8a00 \u6a21\u578b\u6709\uf901\u597d\u7684\u6548\u80fd\u3002BTMM \u7684\u6548\u80fd\u96d6\u7136\u6bd4 PLSA \u597d\uff0c\u7136\u800c\u6548\u679c\u4e26\uf967\u660e\u986f\u3002\u5206\u6790\u5176\u539f\u56e0\uff0c \u5176\u5f71\u97ff\u7684\u56e0\u7d20\u53ef\u80fd\uf92d\u81ea\u6f5b\u5728\u4e3b\u984c\u8b8a\uf969 k \u503c\u7684\u8a2d\u5b9a\u548c\uf96b\uf969\u503c\u521d\u59cb\u7684\u8a2d\u5b9a\u3002\u53e6\u5916\uff0c\u6587\u4ef6\u524d\u8655 \u9700\u8981 J(1 \u8868\u4e8c\u3001\u5c0d\uf967\u540c\u6a21\u578b\u4e4b\u7a7a\u9593\u8907\u96dc\ufa01\u6bd4\u8f03 \uf9e4 stemming \u4ea6\u53ef\u80fd\u9020\u6210\u5f71\u97ff\u3002\u56e0\u6b64\uff0c\u5728\u672a\uf92d\u7684\u5be6\u9a57\uff0c\u5c07\u91dd\u5c0d\u9019\u4e9b\u90e8\u5206\uf901\u9032\u4e00\u6b65\u63a2\u8a0e\u3002</td></tr></table>", |
|
"html": null, |
|
"type_str": "table", |
|
"num": null, |
|
"text": "+ K) + KM \u500b\uf96b\uf969\uff0c\u800c LDA \u53ea\u9700 K + KM \u500b\uf96b\uf969\u3002\u4e3b\u8981\u662f\u7531\u65bc\uf99a\u7e8c\u5206\u4f48\u4f7f\u7528\u4e00 \u500b\uf96b\uf969\uff0c\u5728 LDA \u7522\u751f\u6df7\u5408\u6bd4\uf9b5\u03b8 \uf96b\uf969\uff0c\u53d6\u4ee3\u5728 TTMM \uf978\u500b\uf9ea\u6563\u5206\u4f48\u3002\u9664\u6b64\uff0c\u7576\u6587\u4ef6\u900f \u904e\u4e3b\u984c(theme)\u88ab\u7fa4\u805a\u5728\u4e00\u8d77\uff0c\u5982\u6b64 J < N\uff0c\u5247 TTMM \u7684\uf96b\uf969\uf97e\u53ef\u80fd\u5c11\u65bc PLSA \u7684\uf96b\uf969\uf97e KN + KM\u3002\u5728 BTMM \u6a21\u578b\u4e2d\uff0c\u5b57\u8a5e\u662f\u7d93\u7531\u4e3b\u984c z \u7684\u591a\u9805\u5206\u4f48\u03c6 \u6240\u7522\u751f\uff0c\u800c\u5c0d\u65bc\u5b57\u8a5e\u5206 \u4f48\u7684\u5177\u9ad4\u4e3b\u984c\u591a\u9805\u5206\u4f48\u03c6 \uff0c\u53ef\u4ee5\u5f9e Dirichlet priori \uf96b\uf969 \u03b2 \u5c0d\u61c9\u7684\u4e3b\u984c z \u5f97\u5230\uff0c\u5176\uf96b\uf969\uf97e \u6bd4 PLSA \u5c11\uff0c\u53ea\u9700 KN + K \u500b\u3002 Wall Street Journal (WSJ) 89\uff0c\u8cc7\uf9be\u7684\u7d71\u8a08\u8cc7\u8a0a\uff0c\u5982\u8868\u4e09\u6240\u793a\u3002\u6211\u5011\u6240\u4f7f\u7528\u6e2c \u8a66\u7684\u67e5\u8a62\uf906\u5b50\u70ba Topics 101-150\uff0c\u4e3b\u8981\u53d6\u5404\u500b\u4e3b\u984c\u4e2d\u7684\u6a19\u984c(title)\u548c\u6558\u8ff0(description)\u90e8\u5206 \u4f5c\u70ba\u67e5\u8a62\uf906\uff0c\u6bcf\u500b\u67e5\u8a62\uf906\u7684\u5e73\u5747\u9577\ufa01\u70ba 14.48 \u500b\u5b57\u3002\u6587\u4ef6\u6703\u5148\u7d93\u904e stop word \u548c stemming \u7684\u524d\u8655\uf9e4\u3002\u672c\u6587\u5206\u5225\u5c0d\u6b64\uf978\u6587\u96c6\u4ee5\u6587\u4ef6\u6aa2\uf96a\u548c\u6587\u4ef6\u6a21\u7d44\u5316\u9a57\u8b49\u672c\u6587\u65b9\u6cd5\u7684\u6b63\u78ba\u6027\u548c\u53ef\ufa08 \u6027\u3002\u5728\u5be6\u9a57\u4e2d\u4e3b\u8981\u662f\u91dd\u5c0d Language Model (LM)\u3001PLSA\u3001LDA \u53ca\u672c\u6587\u6240\u63d0\u51fa\u7684 BTMM \u505a\u6bd4\u8f03\u3002\u5c0d\u65bc\u6f5b\u5728\u8b8a\uf969 k \u7684\u500b\uf969\uff0c\u521d\u59cb\u5be6\u9a57\u8a2d\u5b9a\u70ba 16\u3002\u5be6\u9a57\u5206\u70ba\uf978\u500b\u90e8\u5206\uff0c\u7b2c\u4e00\u8a55\u4f30 \u5404\u500b\u6a21\u578b\u61c9\u7528\u5728\u6587\u4ef6\u6aa2\uf96a\u4e0a\u7684\u6548\u80fd\uff0c\u4ee5 Precision-Recall curve \u548c mAP \u4f5c\u70ba\u8a55\u4f30\u7684\u6e96\u5247 [15]\u3002\u7b2c\u4e8c\u500b\u662f\u4ee5 perplexity \u8a55\u4f30\u6587\u4ef6\u6a21\u578b\u7684\u6548\u679c\u3002" |
|
}, |
|
"TABREF7": { |
|
"content": "<table><tr><td>LM 0.2128 0.2761 2\u3001\uf967\u540c\u6a21\u578b\u5728\u6587\u4ef6\u6a21\u7d44\u5316\u7684\u8a55\u4f30 AP88 WSJ89 \u5728\u6587\u4ef6\u6a21\u7d44\u5316\u7684\u5be6\u9a57\u904e\u7a0b\uf9e8\uff0c\u4ee5 WSJ89 \u70ba\u5be6\u9a57\u8cc7\uf9be\uff0c\u5c07\u6587\u4ef6\u5206\u70ba\uf978\u500b\u90e8\u5206\uff0c\u4e09\u5206 PLSA LDA BTMM 0.2507 0.2411 0.2536 0.3448 0.3507 0.3486 \u4e4b\u4e8c\u7684\u8cc7\uf9be\uf97e\u4f5c\u70ba\u57fa\u790e\u6a21\u578b\u7684\u8a13\uf996\u8cc7\uf9be\u96c6\uff0c\u5171 7,931 \u7bc7\u6587\u4ef6\uff0c\u53e6\u5916\uff0c\u4e09\u5206\u4e4b\u4e00\u90e8\u4efd\u505a\u6e2c LM PLSA LDA BTMM Perplexity 257.59 251.8 248.63 250.42 BTMM \u4e3b\u8981\u662f\u6539\u9032 PLSA \u4e2d\uff0c\u5b57\u8a5e\u548c\u4e3b\u984c\u4e4b\u9593\u7684\u8868\u793a\u578b\u614b\uff0c\u4ee5 Dirichlet \u5206\u4f48\u66ff\u4ee3\u539f\u59cb \u7684\u591a\u9805\u5206\u4f48\uff0c\u5728\u5b57\u8a5e\u7684\u4e3b\u984c\u5206\u4f48\u4e0a\u5c0e\u5165 Dirichlet \u4e8b\u524d\u6a5f\uf961\uff0c\u4f7f\u5f97\u8cc7\u8a0a\uf901\u5b8c\u6574\u548c\u8c50\u5bcc\u3002\u7136 \u800c\uff0c\u5f9e\u5be6\u9a57\u7d50\u679c\u6211\u5011\u53ef\u4ee5\u767c\u73fe\u6bd4 LDA \uf976\u5dee\u3002\u91dd\u5c0d\u6b64\u90e8\u5206\uff0c\u6211\u5011\u5c07\u5c0d\u5b57\u5178\u500b\uf969\u7684\u5f71\u97ff\uf901 \u9032\u4e00\u6b65\u7684\u63a2\u8a0e\u5206\u6790\u3002\u5176\u5206\u6790\u7d50\u679c\u5982\u8868\uf9d1\u6240\u793a\u3002\u6211\u5011\u5206\u5225\u9078\u53d6\u5b57\u5178\u5b57\uf969\u4e00\u842c\u3001\u4e8c\u842c\u53ca\u4e09\u842c \u5b57\uf92d\u505a\u5c0d\u7167\uff0c\u6f5b\u5728\u4e3b\u984c\u8b8a\uf969\u500b\uf969\u8a2d\u5b9a\u70ba 8\u3002 \u8868\uf9d1\u3001\uf967\u540c\u5b57\u5178\u500b\uf969\u5c0d perplexity \u503c\u7684\u5f71\u97ff 10,000 20,000 30,000 LM 247 380 511 PLSA 240 372 504 LDA 205 365 505 BTMM 232 369 495 \u5f9e\u8868\uf9d1\u53ef\u4ee5\u5f97\u77e5\uff0c\u7576\u5b57\u5178\uf969\u589e\u52a0\u6642\uff0c\u6a21\u578b\u91dd\u5c0d\u6587\u5b57\u767c\u751f\u6a5f\uf961\u7684\u9810\u6e2c\u5206\u652f\ufa01\u8d8a\u9ad8\uff0c\u6240\u4ee5 perplexity \u90fd\u5448\u73fe\u4e0a\u5347\u7684\u8da8\u52e2\u3002\u7576\u5b57\u5178\u5927\u5c0f\u7d04\u70ba 3 \u4e94\u3001\u7d50\uf941 \u672c \u6587 \u4e2d \u4e3b \u8981 \u662f \u4ee5 \u6a5f \uf961 \u6a21 \u578b \u70ba \u57fa \u790e \u63d0 \u51fa \u4e00 \u500b \u8c9d \u6c0f \uf9e4 \uf941 \u7684 \u6587 \u4ef6 \u6a21 \u578b \uff0c \u81f4 \uf98a \u89e3 \u6c7a bag-of-word \u8868\u793a\u6cd5\u7684\u554f\u984c\uff0c\u4e26\u5c0d\u73fe\u6709\u6a21\u578b\u505a\u6539\u9032\uff0c\u4ee5\u671f\u9054\u5230\uf901\u597d\u7684\u6548\u80fd\u3002\u5176\u67b6\u69cb\u5ef6\u4f38 \u539f\u59cb PLSA \u6a21\u578b\u7684\u6982\uf9a3\uff0c\u5c0d\u65bc\u4e00\u500b\u4e3b\u984c\u7684\u689d\u4ef6\u5206\u4f48\u4ee5 Dirichlet \u4ee3\u66ff\u539f\u6709\u7684\u591a\u9805\u5206\u4f48\u8868 \u793a\uff0c\u5728\u6b64\u7a31\u4e4b\u70ba\u8c9d\u6c0f\u4e3b\u984c\u6df7\u5408\u6a21\u578b\u3002\u6587\u4e2d\uf9dd\u7528 Gibbs \u62bd\u8c61\u6cd5\u4f30\u8a08\u6a21\u578b\u672a\u77e5\uf96b\uf969\uff0c\u6b64\u65b9\u6cd5 \u7684\u512a\u9ede\u662f\uf967\u9700\u8981\u660e\u78ba\u5730\u8868\u9054\u6a21\u578b\uf96b\uf969\u4e14\u5be6\u505a\u4e0a\u6bd4\u8f03\u5bb9\uf9e0\uff0c\u5c0d\u8a18\u61b6\u9ad4\u9700\u6c42\uf97e\u4e5f\u6bd4\u8f03\u5c11\u3002\u5728 \u4e3b\u984c\u6df7\u5408\u6a21\u578b\u4e2d\uff0c\u96d6\u7136\u5047\u8a2d\u6587\u4ef6\u53ef\u7531\uf967\u540c\u4e3b\u984c\u6240\u7522\u751f\uff0c\u4f46\u6587\u4ef6\u8207\u5b57\u8a5e\u5f7c\u6b64\u4e4b\u9593\u662f\u7368\uf9f7 \u7684\u3002\u7136\u800c\uff0c\u5728\u771f\u5be6\u4e16\u754c\uf9e8\uff0c\u6587\u4ef6\u4e4b\u9593\u901a\u5e38\u662f\u6709\u95dc\uf997\u7684\u3002\uf9b5\u5982\uff0c\u5728\u65b0\u805e\u7684\u6587\u4ef6\u6a19\u984c\u4e2d\uff0c\u53ef \u4ee5\u5206\u70ba\u4e3b\u8981\u4e3b\u984c\u548c\u6b21\u8981\u4e3b\u984c\u3002\u5728 Tam \u548c Schultz[34]\u7684\u7814\u7a76\u4e2d\uff0c\u4ee5 Dirichlet Tree[26]\u4ee3\u66ff LDA \u4e2d Dirichlet Prior\uff0c\u4f7f\u5f97\u6f5b\u5728\u4e3b\u984c\u53ef\u4ee5\u8868\u9054\uf901\u591a\u95dc\uf997\u3002\u5728\u672a\uf92d\u7684\u7814\u7a76\u65b9\u5411\uff0c\u5c0d\u65bc\u6587 \u4ef6\u6a21\u578b\u6f14\u7b97\u6cd5\uff0c\u6211\u5011\u64ec\u5ef6\u4f38\u81f3\u5c64\u7d1a\u6982\uf9a3\uff0c\u5c07\u6587\u4ef6\u4ee5\u5c11\uf97e\u7684\u6982\uf9a3\u6216\u662f\u4e3b\u984c\uf92d\u5448\u73fe\uff0c\u4f7f\u5f97\u6a21 \u578b\uf901\u5177\u6709\u5f37\u5065\u6027\u3002\u53e6\u5916\uff0c\u76ee\u524d\u6587\u4ef6\u7684\u6a5f\uf961\u6a21\u578b\u8868\u793a\u6cd5\uff0c\u5927\u81f4\u4ee5 Unigram \u70ba\u4e3b\uff0c\u5982\u4f55\u7d50\u5408 n-gram \u8a9e\u8a00\u6a21\u578b\uff0c\u4f7f\u5f97\u6587\u4ef6\u6a21\u578b\uf901\u5177\u5f37\u5065\u6027\uff0c\u4ea6\u662f\u672a\uf92d\u7814\u7a76\u5de5\u4f5c\u3002 \u8a66\u7684\u6587\u4ef6\u8cc7\uf9be\u96c6\u5408\uff0c\u5305\u542b 4,449 \u8868\u4e94\u3001\uf967\u540c\u6a21\u578b\u4e4b\u9593 perplexity \u4e4b\u6bd4\u8f03 \uf96b\u8003\u6587\u737b</td></tr></table>", |
|
"html": null, |
|
"type_str": "table", |
|
"num": null, |
|
"text": "\u5716\u516b\u3001Precision-recall curves \u5c0d\uf967\u540c\u65b9\u6cd5\u5728 AP88 \u6587\u96c6\u4e0a\u7684\u6bd4\u8f03 \u8868\u56db\u3001LM\u3001PLSA\u3001LDA \u4ee5\u53ca BTMM \u5728\uf967\u540c\u6587\u96c6\u4e2d mAP \u4e4b\u6bd4\u8f03 \u7bc7\u6587\u4ef6\u3002\u521d\u6b65\u5be6\u9a57\u7d50\u679c\u5982\u8868\u4e94\u6240\u793a\u3002\u5f9e\u8868\u4e2d\u53ef\u4ee5\u770b\u51fa BTMM \u6bd4 LM \u548c PLSA \u6a21\u578b\u6709\u8f03\u597d\u7684\u7d50\u679c\uff0c\u5176 perplexity \u5206\u5225\u7531 257.59 \u548c 251.8 \ufa09\u81f3 250.42\u3002 \u842c\u5b57\u6642\uff0cBTMM \u7684 perplexity \u6bd4 LDA \u4f4e\u3002\u4e3b\u8981\u539f\u56e0\u662f\u56e0\u70ba\u7576\u6211\u5011\u904e\ufa01\u5c0d\u5b57\u5178\uf969\u505a\u522a\u6e1b\u6642\uff0c\u7a81\u767c\u73fe\u8c61\u5c0d\u6a21\u578b\u7684\u5f71\u97ff\u8b8a\u5f97\u8f15\u5fae\u3002 \u800c\u7531\u65bc LDA \u6a21\u578b\u5c0d\u6587\u4ef6\u968e\u5c64\u52a0\u5165\u4e8b\u524d\u6a5f\uf961\uff0c\u4f7f\u5f97\u4f30\u7b97\u6587\u4ef6\u7684\u4e3b\u984c\u5206\u4f48\u6642\uff0c\u8f03\u8cbc\u8fd1\u771f\u5be6 \u7684\u5206\u4f48\u60c5\u5f62\u3002\u7136\u800c\uff0c\u5728\u5b57\u5178\uf969\u8f03\u5927\u6642\uff0c\u5f9e\u5be6\u9a57\uf969\u64da\uff0c\u53ef\u4ee5\u767c\u73fe\u7a81\u767c\u73fe\u8c61\u8f03\u70ba\u986f\u8457\uff0c\u4f7f\u5f97 \u5728\u6587\u4ef6\u4e2d\u8f03\u7a00\u6709\u4f46\u537b\u5177\u6709\u9451\u5225\u6027\u7684\u5b57\u8a5e\u5c0d\u6a21\u578b\u7522\u751f\u5f71\u97ff\uff0c\u7531\u65bc BTMM \u6a21\u578b\u5c0d\u5b57\u8a5e\u7684\u4e3b \u984c\u5206\u4f48\u5c0e\u5165 Dirichlet \u4e8b\u524d\u5206\u4f48\uff0c\u4f7f\u5f97\u5728 perplexity \u7684\u8a55\u4f30\u4e0a\uf976\u6bd4 LDA \u4f73\u3002" |
|
} |
|
} |
|
} |
|
} |