{
    "paper_id": "O07-2008",
    "header": {
        "generated_with": "S2ORC 1.0.0",
        "date_generated": "2023-01-19T08:08:15.708108Z"
    },
    "title": "",
    "authors": [],
    "year": "",
    "venue": null,
    "identifiers": {},
    "abstract": "",
    "pdf_parse": {
        "paper_id": "O07-2008",
        "_pdf_hash": "",
        "abstract": [],
        "body_text": [
            {
                "text": "\u3002\u7136\u800c\u7576\u4e2d\u6709\u8a31\u591a \u7814\u7a76\u504f\u91cd\u5728\u55ae\u7d14\u7684\u6524\u5e73\u5f0f\u76ee\u9304\u6574\u5408\u6a5f\u5236\u4e4b\u4e0a\uff0c\u8a0e\u8ad6\u5c07\u76ee\u9304\u985e\u5225\u5168\u90e8\u6241\u5e73\u5316\u4ee5\u5f8c\uff0c\u6574\u5408\u5230 \u6524\u5e73\u5f0f\u7684\u76ee\u7684\u76ee\u9304\u4e2d\uff0c\u4e26\u4e0d\u8003\u616e\u76f4\u63a5\u6574\u5408\u5230\u968e\u5c64\u5f0f\u76ee\u9304\u4e2d [3, 10, 14, 16, 17] \u3002\u56e0\u6b64\u9019\u6a23\u7684 \u904e\u7a0b\u4e2d\u7121\u6cd5\u8003\u616e\u5230\u76ee\u7684\u76ee\u9304\u4e2d\uff0c\u985e\u5225\u8207\u5b50\u985e\u5225\u4e4b\u9593\u7684\u5f9e\u5c6c\u95dc\u4fc2\u3002\u7531\u65bc\u8a31\u591a\u5be6\u969b\u7684\u76ee\u9304\u7686 \u662f\u968e\u5c64\u5f0f\u7684\u67b6\u69cb\uff0c\u56e0\u6b64\u968e\u5c64\u5f0f\u76ee\u9304\u6574\u5408\u6a5f\u5236\u5f88\u9700\u8981\u9032\u4e00\u6b65\u63a2\u8a0e\u3002 \u904e\u5f80\u968e\u5c64\u5f0f\u6587\u4ef6\u5206\u985e\u7814\u7a76\u986f\u793a [8, 11] \uff0c\u5229\u7528\u968e\u5c64\u5f0f\u67b6\u69cb\u78ba\u5be6\u80fd\u6709\u6548\u7684\u52a0\u5f37\u539f\u672c\u6524\u5e73 \u5f0f\u5206\u985e\u7684\u6e96\u78ba\u6027\u3002\u5176\u4e2d\uff0cMcCallum \u7b49\u4eba\u767c\u73fe\u5728\u968e\u5c64\u5f0f\u67b6\u69cb\u4e0b\uff0c\u5229\u7528\u6a5f\u7387\u6a21\u578b\u8207\u805a\u6582 (shrinkage)\u7684\u65b9\u6cd5\uff0c\u53ef\u4ee5\u6709\u6548\u5730\u63d0\u5347\u6587\u4ef6\u5206\u985e\u7684\u6e96\u78ba\u6027\u3002\u4ed6\u5011\u7684\u5be6\u9a57\u7d50\u679c\u4e26\u986f\u793a\u7576\u8cc7\u6599 \u96c6\u4e2d\u7684\u7279\u5fb5\u8cc7\u8a0a\u6578\u91cf\u8db3\u5920\u591a\u6642\uff0c\u968e\u5c64\u5f0f\u67b6\u69cb\u7684\u5206\u985e\u6548\u679c\u660e\u986f\u9ad8\u65bc\u63a1\u7528\u6524\u5e73\u5f0f\u67b6\u69cb [11] \u3002 \u7136\u800c\uff0c\u4ed6\u5011\u7684\u7814\u7a76\u4e5f\u986f\u793a\uff0cshrinkage \u7684\u6548\u679c\u4e26\u975e\u7d55\u5c0d\u80fd\u63d0\u5347\u968e\u5c64\u5f0f\u5206\u985e\u7684\u6e96\u78ba\u6027\u3002\u7576 \u8a13\u7df4\u8cc7\u6599\u91cf\u8f03\u5927\u6642\uff0c\u53cd\u800c\u53ef\u80fd\u9020\u6210\u67d0\u4e9b\u76ee\u9304\u7684\u6e96\u78ba\u7387\u4e0b\u964d [11] \u3002 \u5728\u968e\u5c64\u5f0f\u76ee\u9304\u6574\u5408\u7684\u76f8\u95dc\u7814\u7a76\u4e0a\uff0c\u6709\u4e0d\u5c11\u76f8\u95dc\u7814\u7a76\u7686\u8b49\u5be6\u968e\u5c64\u5f0f\u67b6\u69cb\u5728\u76ee\u9304\u6574\u5408\u4e0a \u7684\u597d\u8655 [6, 7, 9, 13] ",
                "cite_spans": [
                    {
                        "start": 74,
                        "end": 77,
                        "text": "[3,",
                        "ref_id": "BIBREF1"
                    },
                    {
                        "start": 78,
                        "end": 81,
                        "text": "10,",
                        "ref_id": "BIBREF8"
                    },
                    {
                        "start": 82,
                        "end": 85,
                        "text": "14,",
                        "ref_id": "BIBREF12"
                    },
                    {
                        "start": 86,
                        "end": 89,
                        "text": "16,",
                        "ref_id": "BIBREF14"
                    },
                    {
                        "start": 90,
                        "end": 93,
                        "text": "17]",
                        "ref_id": "BIBREF15"
                    },
                    {
                        "start": 183,
                        "end": 186,
                        "text": "[8,",
                        "ref_id": "BIBREF6"
                    },
                    {
                        "start": 187,
                        "end": 190,
                        "text": "11]",
                        "ref_id": "BIBREF9"
                    },
                    {
                        "start": 335,
                        "end": 339,
                        "text": "[11]",
                        "ref_id": "BIBREF9"
                    },
                    {
                        "start": 412,
                        "end": 416,
                        "text": "[11]",
                        "ref_id": "BIBREF9"
                    },
                    {
                        "start": 460,
                        "end": 463,
                        "text": "[6,",
                        "ref_id": "BIBREF4"
                    },
                    {
                        "start": 464,
                        "end": 466,
                        "text": "7,",
                        "ref_id": "BIBREF5"
                    },
                    {
                        "start": 467,
                        "end": 469,
                        "text": "9,",
                        "ref_id": "BIBREF7"
                    },
                    {
                        "start": 470,
                        "end": 473,
                        "text": "13]",
                        "ref_id": "BIBREF11"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            },
            {
                "text": "\u5728\u76ee\u9304\u6574\u5408\u7684\u904e\u7a0b\u4e2d\uff0c\u6211\u5011\u9810\u8a2d\u4f7f\u7528\u8005\u4e43\u662f\u91dd\u5c0d\u5169\u500b\u540c\u8cea\u6027\u6bd4\u8f03\u9ad8\u7684\u76ee\u9304\u4f86\u9032\u884c\u6574 \u5408\u3002\u6240\u8b02\u300c\u540c\u8cea\u6027\u8f03\u9ad8\u300d\u662f\u6307\u5169\u500b\u76ee\u9304\u5177\u6709\u985e\u4f3c\u7684\u5206\u985e\u610f\u7fa9\uff0c\u4e14\u5169\u8005\u4e4b\u9593\u6709\u90e8\u5206\u6587\u4ef6\u662f \u76f8\u540c\u7684\u3002\u5169\u500b\u76ee\u9304\u5206\u5225\u662f\u4f86\u6e90\u7aef\u76ee\u9304 S \u8207\u76ee\u7684\u7aef\u76ee\u9304 D\u3002S \u7576\u4e2d\u6709 n \u500b\u76f8\u4f3c\u800c\u4e0d\u540c\u7684 \u985e\u5225\u76ee\u9304 S 1 ,S 2 ,...,S n \uff0c\u800c\u6bcf\u4e00\u500b\u985e\u5225\u5e95\u4e0b\u6216\u8005\u6709\u6700\u591a m \u500b\u5b50\u985e\u5225\u76ee\u9304 S 11 ,S 12 ,...,S 1m \uff0c\u5176 \u4e2d m \u662f\u8a72\u5c64\u5b50\u985e\u5225\u7684\u6700\u5927\u503c\u3002\u5176\u4ed6\u5c64\u6b21\u4ee5\u6b64\u985e\u63a8\u3002\u53e6\u4e00\u65b9\u9762\uff0cD \u7576\u4e2d\u6709 p \u500b\u76f8\u4f3c\u800c\u4e0d \u540c \u7684 \u985e \u5225 \u76ee \u9304 D 1 ,D 2 ,...,D p \uff0c \u800c \u6bcf \u4e00 \u500b \u985e \u5225 \u5e95 \u4e0b \u6216 \u8005 \u6709 \u6700 \u591a q \u500b \u5b50 \u985e \u5225 \u76ee \u9304 D 11 ,D 12 ,...,D",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            },
            {
                "text": "= {p(x 1 ), p(x 2 ), \u2026, p(x n )}, x i \u2208X\uff0c\u5247\u5c0d\u6a5f\u7387 p \u800c\u8a00\uff0c\u5b83\u7684 Entropy \u88ab\u5b9a\u7fa9\u70ba\uff1a ) ( p H ( ) 3.1 ) ( log ) ( ) ( \u2211 \u2208 \u2200 = X x i i i x p x p p H - \u5982\u679c\u8003\u616e\u4e00\u7d44\u689d\u4ef6\u6a5f\u7387\u5206\u4f48\u6a21\u578b p(y|x)\u7684 Conditional Entropy\uff0c\u5247\u53ef\u5b9a\u7fa9\u70ba\uff1a ( ) 3.2 ) | ( log ) | ( ) ( ) ( \u2211 \u2208 \u2200 = X x x y p x y p x p p H - \u5176\u4e2d ) ( x p \u662f\u7531\u5be6\u969b\u73fe\u8c61\u6240\u89c0\u5bdf\u51fa\u4e4b\u7d93\u9a57\u6a5f\u7387(empirical probability) \u3002 2\u3001Maximum Entropy \u7684\u9650\u5236 \u5728\u627e\u51fa\u4e00\u7d44\u4f7f\u5f97\u71b5\u503c\u70ba\u6700\u5927\u7684\u6a5f\u7387\u524d\uff0c\u5fc5\u9808\u5b9a\u7fa9\u51fa\u7279\u5fb5\u51fd\u5f0f(feature function)f(x,y) \u4f86\u8868\u793a\u6240\u8981\u89c0\u5bdf\u7684\u73fe\u8c61\u3002\u4f8b\u5982\u5f0f(3.3)\u5c31\u662f\u4e00\u500b\u5e38\u898b\u7684\u4e8c\u5143\u8868\u793a\u6cd5\u3002\u82e5\u5176\u4e2d x \u70ba\u7279\u5fb5\u8a5e\uff0c y \u4ee3\u8868\u6587\u4ef6\u7684\u96c6\u5408\uff0c\u5247 f(x,y)=1 \u6642\u6240\u6eff\u8db3\u7684\u689d\u4ef6\u662f\u8868\u793a\u7279\u5fb5\u8a5e x \u51fa\u73fe\u5728\u6587\u4ef6\u96c6\u5408 y \u4e2d\u3002 ( ) ( ) { (3.3) , y x, 1 0 \u6eff\u8db3\u689d\u4ef6 \u5982\u679c \u5176\u5b83 = y x f \u91dd\u5c0d\u6240\u7d66\u4e88\u7684\u8cc7\u6599\u96c6\uff0c\u6211\u5011\u53ef\u4f9d\u9700\u6c42\u6c7a\u5b9a\u51fa f (x,y)\u4e4b\u6eff\u8db3\u689d\u4ef6\u3002\u5728\u6b64\u8cc7\u6599\u96c6\u4e2d\uff0c\u6211 \u5011\u5e0c\u671b\u91dd\u5c0d\u7d93\u9a57\u6a5f\u7387 ) , ( y x p \u7b97\u51fa\u7279\u5fb5\u671f\u671b(feature expectation) \uff0c\u5982\u5f0f(3.4)\u3002\u4f46\u5be6\u969b\u89c0 \u5bdf\u6240\u5f97\u5230\u7684\u689d\u4ef6\u5206\u4f48\u5247\u5982\u5f0f(3.5)\u7684\u8fd1\u4f3c\u51fd\u5f0f\uff0c\u5176\u4e2d ) ( x p } { f E p \u662f\u7531\u8a13\u7df4\u6587\u4ef6\u6240\u89c0\u5bdf\u51fa\u4e4b\u7d93\u9a57 \u6a5f\u7387\u3002\u7279\u5fb5\u671f\u671b\u61c9\u8207\u7d93\u9a57\u671f\u671b\u4e00\u81f4\uff0c\u56e0\u6b64\u5fc5\u9808\u6eff\u8db3 \u2261 } { f E p \u4e4b\u9650\u5236\u3002\u6b64\u5916\uff0c\u6bcf\u4e00 \u7d44\u8a08\u7b97\u51fa\u4f86\u7684\u6a5f\u7387\u503c\u7e3d\u5408\u5fc5\u9808\u70ba 1\uff0c\u5982\u5f0f(3.6)\u3002 ( ) 3.4 ) , ( ) , ( } { , \u2211 \u2261 y x p y x f y x p f E ( ) 3.5 ) , ( ) | ( ) ( } { , \u2211 \u2261 y x p y x f x y p x p f E ( ) ( ) 3.6 1 | \u2211 = y x y p 3\u3001\u6700\u5927\u71b5\u6a21\u578b\u4e4b\u89e3 Maximum Entropy \u7684\u539f\u7406\u662f\u5f9e\u4e00\u500b\u53d7\u9650\u5236\u4e4b\u689d\u4ef6\u6a5f\u7387\u5206\u4f48\u96c6\u5408 C \u4e2d\uff0c\u627e\u51fa\u4e00\u500b\u6a5f\u7387 \u6a21\u578b \uff0c\u4f7f Entropy \u5f97\u5230\u6700\u5927\u503c\u3002\u5f0f(3.7) \u7684 * p * p \u5373\u70ba Maximum Entropy \u7684\u89e3\u6cd5\u3002 ( ) 3.7 ) ( max arg C p p H p \u2208 * = \u56e0\u6b64\u53ea\u8981\u78ba\u5b9a\u51fa \u5c31\u53ef\u5f97\u5230 Maximum Entropy\u3002\u5f9e * p * p \u7684\u5f0f(3.7)\u8207 Entropy \u672c\u8eab\u7684 \u5169\u500b\u9650\u5236\uff0c\u5e36\u5165 Lagrange Multipliers \u4f86\u8655\u7406(\u63a8\u6f14\u904e\u7a0b\u53ef\u53c3\u8003[5]) \uff0c\u53ef\u4ee5\u5f97\u5230\u5f0f(3.8) \u4f86\u8a08\u7b97\u6587\u4ef6\u5206\u985e\u7684\u689d\u4ef6\u6a5f\u7387\u3002\u900f\u904e Maximum Entropy \u8a08\u7b97 p(y|x)\u7684\u6a5f\u7387\u503c\uff0c\u5176\u4e2d y \u662f\u6587 \u4ef6\u96c6\u5408\uff0cx \u70ba\u7279\u5fb5\u8a5e\u7684\u96c6\u5408\uff0cf i (x,y)\u8868\u793a\u662f\u7b2c i \u500b\u7279\u5fb5\u51fd\u5f0f\uff0cz(x)\u7684\u8a08\u7b97\u5982\u5f0f(3.9)\u3002 ( ) ( ) ( ) (3.8 , exp 1 | 1 \uf8f7 \uf8f8 \uf8f6 \uf8ec \uf8ed \uf8eb = \u2211 = k i i i y x f \u03bb x z x y p ) ( ) ( ) ( ) 3.9 , exp 1 \uf8f7 \uf8f8 \uf8f6 \uf8ec \uf8ed \uf8eb = \u2211 \u2211 = k i i i y y x f \u03bb x z \u56e0\u6b64\u53ea\u8981\u8a08\u7b97\u51fa\u6700\u5408\u7406\u7684\u03bb\u503c\uff0c\u5c31\u53ef\u4ee5\u5f97\u5230\u6700\u5927\u7684 Entropy \u7684\u6a5f\u7387\u503c\u3002\u6a5f\u7387\u6a21\u578b\u4e2d \u7684\u6bcf\u500b\u7279\u5fb5\u8a5e\u90fd\u6703\u6709\u4e00\u500b\u03bb\u503c\uff0c\u5176\u6b0a\u91cd\u7531\u4e00\u500b Improved Iterative scaling(IIS)\u7684\u65b9\u7a0b \u5f0f\u8a08\u7b97\u6240\u5f97\u3002\u4e3b\u8981\u662f\u70ba\u4e86\u8981\u4f7f\u6bcf\u500b\u03bb\u503c\u6eff\u8db3\u4ee5\u4e0b\u65b9\u7a0b\u5f0f\uff0c\u76f8\u95dc\u5167\u5bb9\u53ef\u53c3\u8003[4]\u3002\u5716\u4e09\u70ba IIS \u7684\u6f14\u7b97\u6cd5\uff0c\u5728\u4e00\u958b\u59cb\u7684\u6642\u5019\uff0c\u7d66\u03bb i \u4e00\u4e9b\u96a8\u6a5f\u7522\u751f\u8da8\u8fd1\u65bc 0 \u7684\u6578\u503c\u3002\u63a5\u8457\u5728\u8ff4\u5708\u4e2d \u91cd\u8907\u4f5c\u5fae\u5206\u7684\u52d5\u4f5c\uff0c\u76f4\u5230\u7d50\u679c\u6536\u6582\u70ba 0\u3002\u56e0\u6b64\uff0cIIS \u6f14\u7b97\u6cd5\u7684\u4e2d\u9700\u8981\u8abf\u6574\u03bb i \u7684\u503c\uff0c\u4f7f \u5176\u6eff\u8db3\u5fae\u5206\u5f0f\u80fd\u7b49\u65bc 0\u3002\u7576\u5fae\u5206\u7d50\u679c\u6536\u6582\u70ba 0 \u6642\uff0c\u5c31\u5c07\u03bb i \u4ee3\u5165\u03bb i + \u03b4 i \uff0c\u4e26\u7522\u751f\u9810\u6e2c\u7684 \u03bb i \u503c\u8207\u6700\u5927\u71b5\u503c\u3002 \u5716\u4e09\u3001Improved",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            },
            {
                "text": "\u5728\u904e\u5f80\u7814\u7a76\u4e2d\u767c\u73fe\uff0c\u968e\u5c64\u76ee\u9304\u4e2d\u6bcf\u4e00\u985e\u5225\u7684\u6a19\u7c64\u8cc7\u8a0a\u67b6\u69cb\uff0c\u4f8b\u5982 Google \u76ee\u9304\u4e2d\u7684 \"recreation/autos/\" \u8a72\u985e\u5225\u7684\u8aaa\u660e\uff0c\u53ef\u8996\u70ba\u8a72\u76ee\u9304\u7684\u4e00\u500b\u7d22\u5f15\u5178(thesaurus) \uff0c\u5728\u76ee\u9304\u6574 \u5408\u4e0a\u76f8\u7576\u6709\u5e6b\u52a9 [6, 9] \u3002\u6b64\u8cc7\u8a0a\u53ef\u4ee5\u662f\u8a72\u985e\u5225\u7684\u540d\u7a31\uff0c\u4ea6\u6216\u8005\u662f\u8a72\u985e\u5225\u8aaa\u660e\u3002\u5728\u5be6\u9a57\u4e2d\uff0c \u6211\u5011\u4ee5\u8a72\u985e\u5225\u7684\u8aaa\u660e\u4f86\u7d44\u6210\u5176\u7d22\u5f15\u5178\u3002 ",
                "cite_spans": [
                    {
                        "start": 106,
                        "end": 109,
                        "text": "[6,",
                        "ref_id": "BIBREF4"
                    },
                    {
                        "start": 110,
                        "end": 112,
                        "text": "9]",
                        "ref_id": "BIBREF7"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "6\u3001\u968e\u5c64\u67b6\u69cb\u7684\u7d22\u5f15\u5178\u8cc7\u8a0a\u8a08\u7b97",
                "sec_num": null
            },
            {
                "text": "\u56e0\u6b64\uff0c\u91dd\u5c0d\u4e00\u500b\u6587\u4ef6 d\uff0c\u6211\u5011\u5c07\u5b83\u5728\u76ee\u9304\u7684\u7d22\u5f15\u5178\u7279\u5fb5\u5411\u91cf L d \uff0c\u8207\u7d93\u7531\u95dc\u9375\u8a5e\u8a9e\u7fa9 \u64f4\u5c55\u5f8c\u6240\u5f97\u7684\u4e0a\u4f4d\u8a5e\u7279\u5fb5\u5411\u91cf H d \uff0c\u4ee5\u53ca\u539f\u5148\u6587\u4ef6\u4e2d\u7684\u7279\u5fb5\u5411\u91cf F d \u4e00\u8d77\u6574\u5408\uff0c\u5982\u5f0f (3.13) \u6240\u793a\u3002\u6b64\u5916\uff0c\u900f\u904e \u03bb \u3001\u03b1 \u4f86\u53d6\u5f97\u6b0a\u91cd\u7684\u5e73\u8861\uff0c\u6b64\u65b9\u6cd5\u6240\u8a08\u7b97\u51fa\u4f86\u7684\u7279\u5fb5\u5411\u91cf FE d \uff0c\u5373\u53ef \u52a0\u4ee5\u63d0\u5347\u6574\u5408\u6b63\u78ba\u6027\u3002\u7576\u4e2d\u7684 \u03bb \u4e3b\u8981\u662f\u8abf\u6574 [6,9] \u6240\u63d0\u51fa\u52a0\u5165\u968e\u5c64\u5f0f\u76ee\u9304\u8cc7\u8a0a\u3002\u5982\u8868\u4e00 \u4e2d\uff0c\u6587\u4ef6\u6240\u5b58\u5728\u7684\u76ee\u9304\u8cc7\u8a0a\u6b0a\u91cd\u70ba 1/2 1 \uff0c\u518d\u4e0a\u4e00\u5c64\u7684\u76ee\u9304\u8cc7\u8a0a\u6b0a\u91cd\u70ba 1/2 2 \uff0c\u4ee5\u6307\u6578\u905e\u6e1b \u7684\u65b9\u6cd5\u4f86\u8a08\u7b97\u3002\u53e6\u4e00\u65b9\u9762\uff0c\u03b1 \u662f\u8abf\u6574\u6587\u4ef6\u7576\u4e2d\u672c\u8eab\u7279\u5fb5\u8a5e\u8207\u5f9e\u5916\u90e8\u8a9e\u7fa9\u5eab\u6240\u53d6\u51fa\u7684\u4e0a\u4f4d \u8a5e\u8cc7\u8a0a\u3002\u5c07\u6bcf\u500b\u7279\u5fb5\u8a5e\u7684\u6b0a\u91cd\u8a08\u7b97\u51fa\u4f86\uff0c\u5728\u76ee\u9304\u6574\u5408\u6642\uff0c\u53ef\u4ee5\u4f9d\u7167\u8f03\u9ad8\u6b0a\u91cd\u7684\u7279\u5fb5\u8a5e\u52a0 \u5f37\u4f86\u6e90\u7aef\u7684\u8cc7\u8a0a\uff0c\u9032\u800c\u63d0\u5347\u6574\u5408\u6548\u80fd\u3002 [ ] ( ) 3.13 ) (1 ) (1 0 d d d n i i i d F H L L L FE \u00d7 \u00d7 \u2211 = \u03b1 \u03b1 \u03bb \u03bb - \uff0b - \uff0b \uff1d \u8868\u4e00\u3001\u968e\u5c64\u5f0f\u76ee\u9304\u6a19\u7c64\u6b0a\u91cd Hierarchical Label Weight Document Level L 0 1/2 0 One Level Upper L 1 1/2 1 Two Level Upper L 2 1/2 2 \uff0e \uff0e \uff0e \uff0e \uff0e \uff0e nLevel Upper L n 1/2 n 7\u3001\u76ee\u9304\u6574\u5408 \u6700\u5f8c\uff0c\u5c07\u4e0a\u8ff0\u7684\u7279\u5fb5\u503c\u8f49\u63db\u6210 Maximum",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "6\u3001\u968e\u5c64\u67b6\u69cb\u7684\u7d22\u5f15\u5178\u8cc7\u8a0a\u8a08\u7b97",
                "sec_num": null
            }
        ],
        "back_matter": [],
        "bib_entries": {
            "BIBREF0": {
                "ref_id": "b0",
                "title": "Information mapping project",
                "authors": [],
                "year": null,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "\"Information mapping project,\" Computational Semantics Laboratory, Standford University. [Online]. Available: http://infomap.stanford.edu/.",
                "links": null
            },
            "BIBREF1": {
                "ref_id": "b1",
                "title": "On Integrating Catalogs",
                "authors": [
                    {
                        "first": "R",
                        "middle": [],
                        "last": "Agrawal",
                        "suffix": ""
                    },
                    {
                        "first": "R",
                        "middle": [],
                        "last": "Srikant",
                        "suffix": ""
                    }
                ],
                "year": 2001,
                "venue": "Proceedings of the 10 th WWW Conference. (WWW10)",
                "volume": "",
                "issue": "",
                "pages": "603--612",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "R. Agrawal and R. Srikant. \"On Integrating Catalogs.\" In Proceedings of the 10 th WWW Conference. (WWW10), pp. 603-612, Hong Kong, May 2001.",
                "links": null
            },
            "BIBREF2": {
                "ref_id": "b2",
                "title": "The improved Iterative Scaling Algorithm: A Gentle Introduction",
                "authors": [
                    {
                        "first": "A",
                        "middle": [],
                        "last": "Berger",
                        "suffix": ""
                    }
                ],
                "year": 1997,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "A. Berger. \"The improved Iterative Scaling Algorithm: A Gentle Introduction.\" Technical report, 1997.",
                "links": null
            },
            "BIBREF3": {
                "ref_id": "b3",
                "title": "A Maximum Entropy Approach to Natural Language Processing",
                "authors": [
                    {
                        "first": "A",
                        "middle": [
                            "L"
                        ],
                        "last": "Berger",
                        "suffix": ""
                    },
                    {
                        "first": "V",
                        "middle": [
                            "J D"
                        ],
                        "last": "Pietra",
                        "suffix": ""
                    },
                    {
                        "first": "S",
                        "middle": [
                            "A D"
                        ],
                        "last": "Pietra",
                        "suffix": ""
                    }
                ],
                "year": 1996,
                "venue": "Computational Linguistics",
                "volume": "",
                "issue": "",
                "pages": "39--71",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "A. L. Berger, V. J. D. Pietra, and S. A. D. Pietra. \"A Maximum Entropy Approach to Natural Language Processing.\" Computational Linguistics, pp. 39-71, 1996.",
                "links": null
            },
            "BIBREF4": {
                "ref_id": "b4",
                "title": "On Hierarchical Web Catalog Integration with Conceptual Relationships in Thesaurus",
                "authors": [
                    {
                        "first": "I.-X",
                        "middle": [],
                        "last": "Chen",
                        "suffix": ""
                    },
                    {
                        "first": "J.-C",
                        "middle": [],
                        "last": "Ho",
                        "suffix": ""
                    },
                    {
                        "first": "C.-Z",
                        "middle": [],
                        "last": "Yang",
                        "suffix": ""
                    }
                ],
                "year": 2006,
                "venue": "Proceedings of the 29 th International ACM SIGIR (SIGIR 2006)",
                "volume": "",
                "issue": "",
                "pages": "635--636",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "I.-X. Chen, J.-C. Ho, and C.-Z. Yang. \"On Hierarchical Web Catalog Integration with Conceptual Relationships in Thesaurus.\" In Proceedings of the 29 th International ACM SIGIR (SIGIR 2006), pp. 635-636, Seattle, Washington, USA, 2006.",
                "links": null
            },
            "BIBREF5": {
                "ref_id": "b5",
                "title": "Learning to Map between Ontologies on the Semantic Web",
                "authors": [
                    {
                        "first": "A",
                        "middle": [],
                        "last": "Doan",
                        "suffix": ""
                    },
                    {
                        "first": "J",
                        "middle": [],
                        "last": "Madhavan",
                        "suffix": ""
                    },
                    {
                        "first": "P",
                        "middle": [],
                        "last": "Domingos",
                        "suffix": ""
                    },
                    {
                        "first": "A",
                        "middle": [],
                        "last": "Halevy",
                        "suffix": ""
                    }
                ],
                "year": 2002,
                "venue": "Proceedings of the 11th WWW Conf. (WWW2002)",
                "volume": "",
                "issue": "",
                "pages": "662--673",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "A. Doan, J. Madhavan, P. Domingos, and A. Halevy. \"Learning to Map between Ontologies on the Semantic Web.\" In Proceedings of the 11th WWW Conf. (WWW2002), pp.662-673, Honolulu, Hawaii, 2002.",
                "links": null
            },
            "BIBREF6": {
                "ref_id": "b6",
                "title": "Hierarchical Classification of Web Content",
                "authors": [
                    {
                        "first": "S",
                        "middle": [],
                        "last": "Dumais",
                        "suffix": ""
                    },
                    {
                        "first": "H",
                        "middle": [],
                        "last": "Chen",
                        "suffix": ""
                    }
                ],
                "year": 2000,
                "venue": "Proceedings of the 23 rd Annual ACM Conf. on Research and Development in Information Retrieval (SIGIR'00)",
                "volume": "",
                "issue": "",
                "pages": "256--263",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "S. Dumais, and H. Chen. \"Hierarchical Classification of Web Content.\" In Proceedings of the 23 rd Annual ACM Conf. on Research and Development in Information Retrieval (SIGIR'00), pp.256-263, Athens, Greece, 2000.",
                "links": null
            },
            "BIBREF7": {
                "ref_id": "b7",
                "title": "Learning to Integrate Web Catalogs with Conceptual Relationships in Hierarchical Thesaurus",
                "authors": [
                    {
                        "first": "J.-C",
                        "middle": [],
                        "last": "Ho",
                        "suffix": ""
                    },
                    {
                        "first": "I.-X",
                        "middle": [],
                        "last": "Chen",
                        "suffix": ""
                    },
                    {
                        "first": "C.-Z",
                        "middle": [],
                        "last": "Yang",
                        "suffix": ""
                    }
                ],
                "year": 2006,
                "venue": "Proceedings of the 3 rd Asia Information Retrieval (AIRS 2006)",
                "volume": "",
                "issue": "",
                "pages": "217--229",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "J.-C. Ho, I.-X. Chen, and C.-Z. Yang. \"Learning to Integrate Web Catalogs with Conceptual Relationships in Hierarchical Thesaurus.\" In Proceedings of the 3 rd Asia Information Retrieval (AIRS 2006), pp. 217-229, Singapore, 2006.",
                "links": null
            },
            "BIBREF8": {
                "ref_id": "b8",
                "title": "Web Taxonomy Integration through Co-Bootstrapping",
                "authors": [
                    {
                        "first": "W",
                        "middle": [
                            "S"
                        ],
                        "last": "Lee",
                        "suffix": ""
                    },
                    {
                        "first": "D",
                        "middle": [],
                        "last": "Zhang",
                        "suffix": ""
                    }
                ],
                "year": 2004,
                "venue": "Proceedings of the 27 th Annual International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR 2004)",
                "volume": "",
                "issue": "",
                "pages": "410--417",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "W. S. Lee and D. Zhang. \"Web Taxonomy Integration through Co-Bootstrapping.\" In Proceedings of the 27 th Annual International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR 2004), pp. 410-417, 2004.",
                "links": null
            },
            "BIBREF9": {
                "ref_id": "b9",
                "title": "Improving Text Classification by Shrinkage in a Hierarchy of Classes",
                "authors": [
                    {
                        "first": "A",
                        "middle": [],
                        "last": "Macallum",
                        "suffix": ""
                    },
                    {
                        "first": "R",
                        "middle": [],
                        "last": "Rosenfeld",
                        "suffix": ""
                    },
                    {
                        "first": "T",
                        "middle": [],
                        "last": "Mitchell",
                        "suffix": ""
                    },
                    {
                        "first": "A",
                        "middle": [],
                        "last": "Ng",
                        "suffix": ""
                    }
                ],
                "year": 1998,
                "venue": "Proceedings of the 15th International Conference on Machine Learning (ICML-98)",
                "volume": "",
                "issue": "",
                "pages": "359--367",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "A. MaCallum, R. Rosenfeld, T. Mitchell, and A. Ng. \"Improving Text Classification by Shrinkage in a Hierarchy of Classes.\" In Proceedings of the 15th International Conference on Machine Learning (ICML-98), pp. 359-367, Madison, Wisconsin, 1998.",
                "links": null
            },
            "BIBREF10": {
                "ref_id": "b10",
                "title": "Using Maximum Entropy for Text Classification",
                "authors": [
                    {
                        "first": "K",
                        "middle": [],
                        "last": "Nigam",
                        "suffix": ""
                    },
                    {
                        "first": "J",
                        "middle": [],
                        "last": "Lafferty",
                        "suffix": ""
                    },
                    {
                        "first": "A",
                        "middle": [],
                        "last": "Mccallum",
                        "suffix": ""
                    }
                ],
                "year": 1999,
                "venue": "IJCAI-99 Workshop on Machine Learning for Information Filtering",
                "volume": "",
                "issue": "",
                "pages": "61--67",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "K. Nigam, J. Lafferty, and A. McCallum. \"Using Maximum Entropy for Text Classification.\" In IJCAI-99 Workshop on Machine Learning for Information Filtering, pp. 61-67, Oct. 1999.",
                "links": null
            },
            "BIBREF11": {
                "ref_id": "b11",
                "title": "A Maximum Likelihood Framework for Integrating Taxonomies",
                "authors": [
                    {
                        "first": "S",
                        "middle": [],
                        "last": "Rajan",
                        "suffix": ""
                    },
                    {
                        "first": "K",
                        "middle": [],
                        "last": "Punera",
                        "suffix": ""
                    },
                    {
                        "first": "J",
                        "middle": [],
                        "last": "Ghosh",
                        "suffix": ""
                    }
                ],
                "year": null,
                "venue": "Proceedings of the 20 th National Conference on Artificial Intelligence (AAAI-05)",
                "volume": "",
                "issue": "",
                "pages": "856--861",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "S. Rajan, K. Punera, and J. Ghosh. \"A Maximum Likelihood Framework for Integrating Taxonomies.\" In Proceedings of the 20 th National Conference on Artificial Intelligence (AAAI-05), pp.856-861, Pittsburgh, Pennsylvania.",
                "links": null
            },
            "BIBREF12": {
                "ref_id": "b12",
                "title": "Cross-Training: Learning Probabilistic Mappings between Topics",
                "authors": [
                    {
                        "first": "S",
                        "middle": [],
                        "last": "Sarawagi",
                        "suffix": ""
                    },
                    {
                        "first": "S",
                        "middle": [],
                        "last": "Chakrabarti",
                        "suffix": ""
                    },
                    {
                        "first": "S",
                        "middle": [],
                        "last": "Godbole",
                        "suffix": ""
                    }
                ],
                "year": 2003,
                "venue": "Proc. of the 9 th ACM SIGKDD Int'l Conf. on Knowledge Discovery and Data Mining",
                "volume": "",
                "issue": "",
                "pages": "177--186",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "S. Sarawagi, S. Chakrabarti, and S. Godbole. \"Cross-Training: Learning Probabilistic Mappings between Topics.\" In Proc. of the 9 th ACM SIGKDD Int'l Conf. on Knowledge Discovery and Data Mining, pp. 177-186, 2003.",
                "links": null
            },
            "BIBREF13": {
                "ref_id": "b13",
                "title": "Toward Generic Title Generation for Clustered Documents",
                "authors": [
                    {
                        "first": "Y.-H",
                        "middle": [],
                        "last": "Tseng",
                        "suffix": ""
                    },
                    {
                        "first": "C.-J",
                        "middle": [],
                        "last": "Lin",
                        "suffix": ""
                    },
                    {
                        "first": "H.-H",
                        "middle": [],
                        "last": "Chen",
                        "suffix": ""
                    },
                    {
                        "first": "Y.-I",
                        "middle": [],
                        "last": "Lin",
                        "suffix": ""
                    }
                ],
                "year": 2006,
                "venue": "Proceedings of the 3 rd Asia Information Retrieval (AIRS 2006)",
                "volume": "",
                "issue": "",
                "pages": "145--157",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Y.-H. Tseng, C.-J Lin, H.-H Chen, and Y.-I. Lin. \"Toward Generic Title Generation for Clustered Documents.\" In Proceedings of the 3 rd Asia Information Retrieval (AIRS 2006), pp. 145-157, 2006.",
                "links": null
            },
            "BIBREF14": {
                "ref_id": "b14",
                "title": "Learning to Integrate Web Taxonomies with Fine-Grained Relations: A Case Study Using Maximum Entropy Model",
                "authors": [
                    {
                        "first": "C.-W",
                        "middle": [],
                        "last": "Wu",
                        "suffix": ""
                    },
                    {
                        "first": "T.-H",
                        "middle": [],
                        "last": "Tsai",
                        "suffix": ""
                    },
                    {
                        "first": "W.-L",
                        "middle": [],
                        "last": "Hsu",
                        "suffix": ""
                    }
                ],
                "year": 2005,
                "venue": "Proceedings of the 2 nd Asia Information Retrieval Symposium 2005 (AIRS 2005)",
                "volume": "",
                "issue": "",
                "pages": "190--205",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "C.-W. Wu, T.-H. Tsai, and W.-L. Hsu, \"Learning to Integrate Web Taxonomies with Fine-Grained Relations: A Case Study Using Maximum Entropy Model.\" In Proceedings of the 2 nd Asia Information Retrieval Symposium 2005 (AIRS 2005), pp. 190-205, Jeju Island, Korea, Oct. 2005.",
                "links": null
            },
            "BIBREF15": {
                "ref_id": "b15",
                "title": "Web Taxonomy Integration using Support Vector Machines",
                "authors": [
                    {
                        "first": "D",
                        "middle": [],
                        "last": "Zhang",
                        "suffix": ""
                    },
                    {
                        "first": "W",
                        "middle": [
                            "S"
                        ],
                        "last": "Lee",
                        "suffix": ""
                    }
                ],
                "year": 2004,
                "venue": "Proceedings of the 13 th WWW Conference (WWW2004)",
                "volume": "",
                "issue": "",
                "pages": "472--481",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "D. Zhang and W.S. Lee. \"Web Taxonomy Integration using Support Vector Machines.\" In Proceedings of the 13 th WWW Conference (WWW2004), pp.472-481, New York, NY, May 2004.",
                "links": null
            }
        },
        "ref_entries": {
            "TABREF3": {
                "num": null,
                "html": null,
                "content": "<table><tr><td>\u8a9e\u7fa9\u5eab InfoMap \u7684\u4e0a\u4f4d\u8a5e\u8cc7\u8a0a\uff0c\u9032\u800c\u5c07\u9019\u4e9b\u4e0a\u4f4d\u8a5e\u7b97\u51fa\u6b0a\u91cd\u5f8c\u52a0\u5165\u7279\u5fb5\u503c\u7a7a\u9593\u4e2d\u3002\u4f8b\u5982</td></tr><tr><td>\u67d0\u4e00\u76ee\u9304\u985e\u5225\u6700\u9ad8\u7684 5 ( ) 3.12 0 i = i HF \u2211 = n i i HF HW</td></tr><tr><td>\u5176\u4e2d HF i</td></tr><tr><td>Iterative scaling\u6f14\u7b97\u6cd5</td></tr><tr><td>\u671f\u671b\u503c\u65b9\u7a0b\u5f0f(3.9)\u7684\u529f\u80fd\u662f\u8981\u6eff\u8db3\u6bcf\u500b\u689d\u4ef6\u6a5f\u7387\uff0c\u4f7f\u5f0f(3.8)\u8a08\u7b97\u51fa\u4f86\u7684\u503c\uff0c\u5728\u5f0f(3.6</td></tr><tr><td>\u52a0\u7e3d\u70ba 1\u3002\u56e0\u6b64 Maximum Entropy \u5206\u985e\u5668\u80fd\u5920\u5728\u76ee\u9304\u6574\u5408\u6642\uff0c\u4fdd\u8b49\u6bcf\u500b\u7279\u5fb5\u8a5e\u80fd\u5920\u6eff\u8db3</td></tr><tr><td>\u6a5f\u7387\u503c\u7e3d\u548c\u70ba 1\u3002</td></tr><tr><td>(\u4e09) \u3001\u95dc\u9375\u8a5e\u8a9e\u7fa9\u64f4\u5c55 \u5716\u56db\u3001\u5f15\u7528\u5916\u90e8\u8a9e\u7fa9\u5eab\u9032\u884c\u968e\u5c64\u5f0f\u76ee\u9304\u6574\u5408\u6d41\u7a0b\u5716</td></tr><tr><td>1\u3001\u76ee\u9304\u6574\u5408\u6d41\u7a0b</td></tr><tr><td>\u76ee\u9304\u6574\u5408\u7684\u6d41\u7a0b\u5982\u5716\u56db\u3002\u76ee\u9304\u6574\u5408\u7684\u6b65\u9a5f\u5305\u542b\u7db2\u9801\u6587\u4ef6\u8655\u7406\uff0c\u7279\u5fb5\u8a5e\u9078\u53d6\u3002\u7136\u5f8c\u5f9e \u7279\u5fb5\u503c\u4e2d\u9032\u884c\u95dc\u9375\u8a5e\u7684\u8a9e\u7fa9\u64f4\u5c55\uff0c\u81ea\u5916\u90e8\u8a9e\u7fa9\u5eab\u4e2d\u53d6\u51fa\u9069\u7576\u7684\u4e0a\u4f4d\u8a5e(hypernyms) \uff0c\u4e26 4\u3001\u985e\u5225\u95dc\u9375\u8a5e\u9078\u53d6</td></tr><tr><td>\u52a0\u5165\u968e\u5c64\u67b6\u69cb\u7684\u7d22\u5f15\u5178\u8cc7\u8a0a\uff0c\u8207\u6587\u4ef6\u539f\u6709\u7279\u5fb5\u8a5e\u5171\u540c\u7d44\u5408\u6210\u64f4\u5c55\u6587\u4ef6\u7279\u5fb5\u3002\u6700\u5f8c\u5c07\u6b64\u64f4 \u5982\u679c\u5c07\u6587\u4ef6\u4e2d\u6240\u6709\u7684\u7279\u5fb5\u8a5e\u90fd\u5c07\u5176\u4e0a\u4f4d\u8a5e\u52a0\u5165\uff0c\u5c07\u6703\u4f7f\u6587\u4ef6\u5145\u6eff\u904e\u591a\u7121\u95dc\u91cd\u8981\u7684\u8cc7</td></tr><tr><td>\u5c55\u6587\u4ef6\u7279\u5fb5\u8f49\u63db\u70ba ME \u5206\u985e\u5668\u683c\u5f0f\uff0c\u9032\u884c\u76ee\u9304\u6574\u5408\u3002\u4ee5\u4e0b\u5c07\u9032\u4e00\u6b65\u8aaa\u660e\u5404\u500b\u6b65\u9a5f\u3002 \u8a0a\u800c\u5f71\u97ff\u6574\u5408\u6548\u679c\u3002\u56e0\u6b64\u5728\u9032\u884c\u8a9e\u7fa9\u64f4\u5c55\u7684\u6642\u5019\uff0c\u5fc5\u9808\u5148\u5c0d\u6587\u4ef6\u4e2d\u7684\u7279\u5fb5\u8a5e\u7be9\u9078\u51fa\u91cd\u8981</td></tr><tr><td>\u7684\u95dc\u9375\u8a5e\uff0c\u518d\u91dd\u5c0d\u9019\u4e9b\u5177\u6709\u4ee3\u8868\u6027\u7684\u95dc\u9375\u8a5e\u4f86\u9032\u884c\u8a9e\u7fa9\u64f4\u5145\uff0c\u4e5f\u5c31\u662f\u5c07\u4ed6\u5011\u7684\u4e0a\u4f4d\u8a5e\u52a0</td></tr><tr><td>2\u3001\u6587\u4ef6\u8655\u7406 \u5165\u7279\u5fb5\u8a5e\u96c6\u5408\u7576\u4e2d\uff0c\u4f86\u8f14\u52a9\u6574\u5408\u6548\u679c\u3002\u540c\u6642\uff0c\u70ba\u907f\u514d\u5728\u540c\u4e00\u985e\u5225\u4e2d\uff0c\u4e0d\u540c\u6587\u4ef6\u4e4b\u9593\u7684\u95dc</td></tr><tr><td>\u5728\u6b64\u6b65\u9a5f\u4e2d\uff0c\u6211\u5011\u5c07\u5c0d Web \u7db2\u9801\u6587\u4ef6\u505a\u4e00\u4e9b\u524d\u7f6e\u8655\u7406\u3002\u9019\u4e9b\u524d\u7f6e\u8655\u7406\u5305\u542b\uff1a\u79fb\u9664 \u9375\u8a5e\u4ecd\u53ef\u80fd\u5b58\u6709\u5206\u6b67\uff0c\u56e0\u6b64\u6211\u5011\u91dd\u5c0d\u4e00\u500b\u76ee\u9304\u985e\u5225\u4f86\u9078\u53d6\u53ef\u4ee5\u4ee3\u8868\u8a72\u985e\u5225\u7684\u95dc\u9375\u8a5e\u3002</td></tr><tr><td>Web \u7db2\u9801\u6587\u4ef6\u4e2d\u7684 HTML \u6a19\u7c64\uff0cscript \u7db2\u9801\u57f7\u884c\u78bc\u548c\u6587\u5b57\u3002\u79fb\u9664\u6587\u7ae0\u4e2d\u7684 stop word\uff0c\u4e26 \u5728\u904e\u5f80\u7814\u7a76\u4e2d\u53ef\u4ee5\u767c\u73fe\uff0c\u4ee5 Correlation Coefficient \u7684\u65b9\u5f0f\uff0c\u53ef\u4ee5\u64f7\u53d6\u5230\u5177\u6709\u4ee3\u8868\u6027</td></tr><tr><td>\u7528 Porter \u7684\u6f14\u7b97\u6cd5\u5c0d\u6bcf\u500b\u8a5e\u505a stemming \u8655\u7406\u3002\u8655\u7406\u5f8c\u7684\u55ae\u8a5e\u505a\u70ba\u6587\u4ef6\u7684\u7279\u5fb5\u8a5e\u3002 \u7684\u95dc\u9375\u8a5e [15]\u3002\u56e0\u6b64\u6211\u5011\u5229\u7528 Correlation Coefficient \u7684\u65b9\u5f0f\u4f86\u62bd\u53d6\u95dc\u9375\u8a5e\uff0c\u518d\u9032\u884c\u8a9e\u7fa9</td></tr><tr><td>\u64f4\u5c55\u3002\u4e5f\u5c31\u662f\u662f\u91dd\u5c0d\u540c\u4e00\u500b\u76ee\u9304\u5e95\u4e0b\u6240\u6709\u6587\u4ef6\u4e2d\u7684\u7279\u5fb5\u8a5e\uff0c\u900f\u904e\u5e95\u4e0b\u65b9\u7a0b\u5f0f(3.11)\u8a08\u7b97 3\u3001\u6587\u4ef6\u4e2d\u7279\u5fb5\u8a5e\u6b0a\u91cd\u8a08\u7b97 \u7279\u5fb5\u8a5e\u6b0a\u91cd\u7684\u8a08\u7b97\u65b9\u5f0f\uff0c\u4e00\u822c\u5728\u5be6\u4f5c\u4e0a\u6709 TF-IDF\u3001TF \u8207 TF/sum(TF)\u7b49\u65b9\u6cd5\u3002\u6211\u5011 \u8003\u616e\u5230 TF-IDF \u5728\u8a08\u7b97\u6642\u6703\u56e0\u76ee\u9304\u7684\u8b8a\u52d5\u800c\u5e38\u5e38\u9700\u8981\u66f4\u65b0\uff0c\u5728\u5be6\u969b\u4f7f\u7528\u4e0a\u5c07\u6703\u82b1\u8cbb\u8a31\u591a \u8a08\u7b97\u6642\u9593\uff0c\u56e0\u6b64\u5728\u672c\u7814\u7a76\u4e2d\uff0c\u6211\u5011\u4f7f\u7528\u5982\u5f0f (3.10)\u4f86\u8a08\u7b97 TF/sum(TF) \u7279\u5fb5\u8a5e\u6b0a\u91cd\u8a08\u7b97\u3002 ( ) ( ) ( ) 3.10 , , 0 \u2211 = = n i i i i d w TF d w TF f \u51fa\u80fd\u5920\u4ee3\u8868\u6b64\u76ee\u9304\u7576\u4e2d\u6bcf\u4e00\u500b\u7279\u5fb5\u8a5e\u7684\u6b0a\u91cd\u3002\u5728 (3.11) ( ) 3.11 TN) FP)(FN TN)(TP FN)(FP (TP FP) FN TN (TP C) Co(T, + + + + \u00d7 + \u00d7 =</td></tr><tr><td>\u5176\u4e2d w i \u8868\u793a\u662f\u7b2c i \u500b\u7279\u5fb5\u8a5e\uff0cd \u70ba\u6587\u4ef6\uff0cTF(w i ,d)\u70ba w i \u9019\u55ae\u5b57\u6216\u7247\u8a9e\u5728\u6587\u4ef6 d \u4e2d\u6240\u51fa\u73fe</td></tr><tr><td>\u7684\u983b\u7387\uff0c\u7e3d\u5171\u6709 n \u500b\u7279\u5fb5\u8a5e\u3002 5\u3001\u95dc\u9375\u8a5e\u8a9e\u7fa9\u64f4\u5c55</td></tr><tr><td>\u7d93\u7531\u4ee5\u4e0a\u5f0f (3.11) \u6240\u8a08\u7b97\u51fa\u4f86\u7684\u6b0a\u91cd\uff0c\u53cd\u6620\u51fa\u8a72\u7279\u5fb5\u8a5e\u5728\u8a72\u985e\u5225\u4e2d\u7684\u95dc\u9023\u6027\uff0c\u56e0</td></tr><tr><td>\u6b64\u6211\u5011\u64f7\u53d6\u51fa\u6b0a\u91cd\u6700\u9ad8\u7684 5 \u500b\u7279\u5fb5\u8a5e\uff0c\u8996\u70ba\u53ef\u4ee5\u4ee3\u8868\u8a72\u985e\u5225\u7684\u95dc\u9375\u8a5e\uff0c\u67e5\u8a62\u4ed6\u5011\u5728\u5916\u90e8</td></tr></table>",
                "type_str": "table",
                "text": "\u5f0f\u7576\u4e2d\u7684 TP \u8868\u793a\u985e\u5225 C \u4e4b\u6587 \u4ef6\u542b\u6709\u7279\u5fb5\u8a5e T \u7684\u6587\u4ef6\u6578\uff0cFP \u8868\u793a C \u4ee5\u5916\u5176\u4ed6\u985e\u5225\u4e4b\u6587\u4ef6\u542b\u6709\u7279\u5fb5\u8a5e T \u7684\u6587\u4ef6\u6578\uff0cFN \u8868\u793a C \u4ee5\u5916\u5176\u4ed6\u985e\u5225\u4e4b\u6587\u4ef6\u4e0d\u542b\u6709\u7279\u5fb5\u8a5e T \u7684\u6587\u4ef6\u6578\uff0cTN \u8868\u793a\u985e\u5225 C \u4e4b\u6587\u4ef6\u4e0d\u542b\u6709\u7279 \u5fb5\u8a5e T \u7684\u6587\u4ef6\u6578\u8868\u793a\u5728\u76ee\u9304\u5e95\u4e0b\u6709\u5305\u542b\u6216\u8005\u6c92\u6709\u5305\u542b\u6b64\u7279\u5fb5\u8a5e\u7684\u6b63\u8ca0\u5411\u6587\u4ef6\u500b\u6578\u3002\u6240 \u7b97\u51fa\u4f86\u7684 Co(T,C) \u8868\u793a\u5728\u985e\u5225 C \u4e2d\u7279\u5fb5\u8a5e T \u8207\u985e\u5225 C \u7684 CorrelationCoefficient\u3002\u900f\u904e Correlation Coefficient \u65b9\u6cd5\u53ef\u4ee5\u7cbe\u78ba\u7684\u9078\u64c7\u51fa\u4e00\u4e9b\u91dd\u5c0d\u6b64\u76ee\u9304\u8f03\u5177\u4ee3\u8868\u6027\u7684\u7279\u5fb5\u8a5e\u3002 \u500b\u7279\u5fb5\u8a5e\u70ba output, signal, circuit, input, frequency\uff0c\u900f\u904e InfoMap\uff0c \u6211\u5011\u53ef\u5f97\u5230 signal, signaling, sign, communication, abstraction, relation \u7b49 6 \u500b\u4e0a\u4f4d\u8a5e\uff0c\u518d \u7528\u4e0b\u5f0f (3.12) \u8a08\u7b97\u51fa\u7b2c i \u500b\u4e0a\u4f4d\u8a5e\u7684\u6b0a\u91cd HW i \u3002\u7531\u9019\u4e9b\u6b0a\u91cd\uff0c\u6211\u5011\u53ef\u4ee5\u5c0d\u4e00\u500b\u6587\u4ef6 d \u4f86\u6c7a\u5b9a\u5b83\u7684\u4e0a\u4f4d\u8a5e\u7279\u5fb5\u5411\u91cf H d \u3002 \u4ee3\u8868\u5f9e\u9019 5 \u500b\u7279\u5fb5\u8a5e\u6240\u53d6\u51fa\u7684\u6240\u6709 n \u500b\u4e0a\u4f4d\u8a5e\u4e2d\uff0c\u7b2c i \u500b\u4e0a\u4f4d\u8a5e\u51fa\u73fe\u7684\u983b\u7387\u3002"
            },
            "TABREF4": {
                "num": null,
                "html": null,
                "content": "<table><tr><td colspan=\"5\">\uff1e\u70ba\u6587\u4ef6\u7576\u4e2d\u7684\u7279\u5fb5\u8a5e\uff0c\uff1cvalue\uff1e\u70ba\u7279\u5fb5\u8a5e\u7684\u6b0a\u91cd\u3002 2\u3001\u6e2c\u91cf\u65b9\u5f0f</td><td/><td/></tr><tr><td colspan=\"8\">\u56db\u3001\u76ee\u9304\u6574\u5408\u5be6\u9a57 \u5728\u5be6\u9a57\u7576\u4e2d\u6211\u5011\u7684\u6e2c\u91cf\u6574\u5408\u6642\u7684\u7cbe\u78ba\u7387 (P\uff0cprecision)\u8207\u53ec\u56de\u7387 (R\uff0crecall)\uff0c\u4ee5\u53ca</td></tr><tr><td colspan=\"8\">F 1 \uff1d2PR/(P+R)\uff0c\u4f86\u6bd4\u8f03\u4e0d\u540c\u7cfb\u7d71\u7684\u6210\u6548\u3002\u5728\u76ee\u524d\u5be6\u9a57\u4e2d\uff0c\u6211\u5011\u5141\u8a31\u4e00\u4efd\u6587\u4ef6\u53ef\u4ee5\u88ab\u6574 \u70ba\u4e86\u77ad\u89e3\u9019\u4e9b\u6539\u9032\u65b9\u5f0f\u7684\u6548\u80fd\uff0c\u6211\u5011\u4ee5\u771f\u5be6\u7684\u76ee\u9304\u4f86\u9032\u884c\u5be6\u9a57\uff0c\u5206\u5225\u81ea Google \u548c Yahoo!\u53d6\u5f97\u90e8\u4efd\u76ee\u9304\u3002\u5728\u5be6\u9a57\u4e2d\uff0c\u6211\u5011\u6240\u63a1\u7528\u7684\u5916\u90e8\u8a9e\u7fa9\u5eab\u662f InfoMap\u3002\u5728\u5be6\u9a57\u4e0a\uff0c\u6211 \u5011\u4f9d\u64da ECI \u65b9\u6cd5 [9] \uff0c\u4e26\u52a0\u4ee5\u904b\u7528\u5230 SVM (ECI-SVM) \u8207 ME (ECI-ME) \u5206\u985e\u5668\u4e2d\u3002\u6b64 \u5916\uff0c\u5728 ECI-ME \u4e0a\uff0c\u518d\u9032\u884c\u95dc\u9375\u8a5e\u8a9e\u7fa9\u64f4\u5c55 (KSE-ME)\uff0c\u4ee5\u4e0b\u5c07\u9032\u4e00\u6b65\u8aaa\u660e\u5404\u9805\u7d30\u7bc0\u3002 (\u4e00) \u3001\u5be6\u9a57\u74b0\u5883 1\u3001\u8cc7\u6599\u96c6 \u5728\u5be6\u9a57\u7684\u90e8\u4efd\uff0c\u6211\u5011\u5f9e\u5169\u500b\u5be6\u969b\u76ee\u9304\uff1aGoogle \u548c Yahoo!\uff0c\u5206\u5225\u53d6\u5f97 5 \u500b\u5206\u9805\u968e\u5c64 \u76ee\u9304\u4f86\u7576\u4f5c\u8cc7\u6599\u96c6\uff0c\u8868\u4e8c\u548c\u8868\u4e09\u662f\u9019\u4e9b\u968e\u5c64\u76ee\u9304\u7684\u6839\u7bc0\u9ede\u76ee\u9304\u540d\u7a31\u3002\u8868\u56db\u662f\u9019 5 \u500b\u968e\u5c64 \u76ee\u9304\u7576\u4e2d\u6240\u64f7\u53d6\u4e0b\u4f86\u6587\u4ef6\u7684\u6578\u91cf\u548c\u985e\u5225\u6578\u91cf\u3002\u5176\u6578\u91cf\u4f9d 5 \u500b\u76ee\u9304 Autos\u3001Movies\u3001 Outdoors\u3001Photo \u548c Software \u4f86\u5340\u5206\u3002 \u8868\u4e8c\u3001Yahoo!\u4e2d\u76ee\u9304\u7684\u6839\u7bc0\u9ede\u4f4d\u5740 Category Matched URL Autos http://dir.yahoo.com/recreation/automotive/ Movies http://dir.yahoo.com/entertainment/movies and film/ Outdoors http://dir.yahoo.com/recreation/outdoors/ Photos http://dir.yahoo.com/arts/visual arts/photography/ Software http://dir.yahoo.com/computers and internet/software/ \u8868\u4e09\u3001Google \u4e2d\u76ee\u9304\u7684\u6839\u7bc0\u9ede\u4f4d\u5740 Category Matched URL Autos http://dir.google.com/top/recreation/autos/ Movies http://dir.google.com/top/arts/movies/ Outdoors http://dir.google.com/top/recreation/outdoors/ Photos http://dir.google.com/top/arts/photography/ Software http://dir.google.com/top/computers/software/ \u900f\u904e\u6bcf\u500b\u7db2\u9801\u76ee\u9304\u7bc0\u9ede\u4e2d\u6240\u5305\u542b\u7684\u6b21\u76ee\u9304\u548c\u5c0d\u5916\u9023\u7d50\uff0c\u4f9d\u6b64\u5efa\u7acb\u76ee\u9304\u4e4b\u9593\u7684\u4e0a\u4e0b\u5c64 \u95dc\u4fc2\u548c\u5411\u5916\u7684\u9023\u7d50\u3002\u6574\u5408\u7684\u904e\u7a0b\u6703\u4f9d\u5411\u5916\u7684\u9023\u7d50\uff0c\u53d6\u5f97\u6587\u4ef6\u3002\u4ee5 Google \u76ee\u9304\u70ba\u4f8b\uff0c\u6211 \u5408\u5230\u591a\u500b\u76ee\u9304\u985e\u5225\u4e2d\u3002\u56e0\u6b64 Precision \u7684\u7b97\u6cd5\u662f(\u6b63\u78ba\u5206\u985e\u7684\u6587\u4ef6\u6578/\u6240\u6709\u5206\u81f3\u8a72\u985e\u7684\u6587 \u4ef6\u6578) \uff0cRecall \u7684\u7b97\u6cd5\u662f(\u6b63\u78ba\u5206\u985e\u7684\u6587\u4ef6\u6578/\u61c9\u8a72\u5206\u81f3\u8a72\u985e\u7684\u6587\u4ef6\u6578) \u3002Recall \u4e5f\u53ef\u770b\u6210 \u662f\u904e\u5f80\u7814\u7a76\u4e2d\u7684\u6574\u5408\u7cbe\u78ba\u5ea6\u3002 \u9664\u6b64\u4e4b\u5916\uff0c\u6211\u5011\u4e5f\u540c\u6642\u8003\u91cf\u591a\u500b\u985e\u5225\u7684\u6574\u9ad4\u8868\u73fe\uff0c\u56e0\u6b64\u4e5f\u4f7f\u7528 micro-average \u8207 macro-average \u5169\u7a2e\u5e73\u5747\u65b9\u6cd5\u3002Micro-average \u7531\u65bc\u662f\u5168\u90e8\u6587\u4ef6\u4e00\u8d77\u7d2f\u52a0\u7d71\u8a08\uff0c\u4e0d\u5206\u985e\u5225\uff0c \u56e0\u6b64\u5bb9\u6613\u53d7\u5230\u4f54\u5927\u591a\u6578\u7684\u5927\u4ef6\u985e\u5225\u5f71\u97ff\u3002\u76f8\u5c0d\u5730\uff0cMacro-average \u8003\u616e\u6bcf\u500b\u985e\u5225\u7684\u6210\u6548 \u5f8c\u518d\u505a\u5e73\u5747\uff0c\u56e0\u6b64\u5bb9\u6613\u53d7\u5230\u5927\u91cf\u7684\u5c0f\u985e\u5225\u800c\u5f71\u97ff\u3002 3\u3001\u5916\u90e8\u8a9e\u7fa9\u5eab \u5728\u76ee\u524d\u7814\u7a76\u4e2d\uff0c\u6211\u5011\u4f7f\u7528\u7684\u5916\u90e8\u8a9e\u7fa9\u5eab\u662f InfoMap [2] \u4f86\u5c0b\u627e\u4e0a\u4f4d\u8a5e\u3002\u4e8b\u5be6\u4e0a\uff0c\u5176 \u4ed6\u7684\u5916\u90e8\u8a9e\u7fa9\u5eab\u4e5f\u53ef\u4ee5\u4f7f\u7528\uff0c\u4f8b\u5982 WordNet\uff0c\u7136\u800c\u5728\u904e\u5f80\u7814\u7a76\u4e2d\u767c\u73fe\uff0c\u4f7f\u7528 InfoMap \u6240 \u64f4\u5c55\u7684\u7d50\u679c\u8207\u4f7f\u7528 WordNet \u7684\u7d50\u679c\u6240\u5f97\u7684\u6210\u6548\u76f8\u7576\u63a5\u8fd1 [15]\u3002\u56e0\u6b64\u63db\u7528 WordNet \u53ef\u80fd \u4e5f\u6703\u6709\u8207\u76ee\u524d\u985e\u4f3c\u7684\u5206\u985e\u8868\u73fe\u3002\u672a\u4f86\u5728\u6211\u5011\u7684\u7814\u7a76\u8a08\u756b\u4e2d\uff0c\u9810\u5099\u5c07\u9032\u4e00\u6b65\u63a2\u8a0e\u5916\u90e8\u8a9e\u7fa9 \u5eab\u7684\u54c1\u8cea\u5c0d\u65bc\u6574\u5408\u54c1\u8cea\u7684\u5f71\u97ff\u3002 4\u3001\u5206\u985e\u5668 (\u4e8c) \u3001\u5be6\u9a57\u7d50\u679c\u8207\u8a0e\u8ad6 \u5728\u5be6\u9a57\u4e2d\uff0c\u6211\u5011\u8a2d\u5b9a\u4e0d\u540c\u7684\u03bb\u503c\u3002\u70ba\u4e86\u89e3\u4e0d\u540c\u03bb\u503c\u7684\u5f71\u97ff\uff0c\u5728\u4f86\u6e90\u76ee\u9304\u4e2d\uff0c\u6211\u5011\u5c07 \u03bb s \u5011\u7684\u6210\u6548\u5982\u4f55\u3002\u5982\u6b64\u53d6\u7684\u539f\u56e0\u662f\uff0c\u5728\u904e\u5f80\u7814\u7a76\u4e2d\u767c\u73fe\u03bb\u503c\u904e\u5927\u5176\u5be6\u6703\u5c07 Recall \u503c\u964d\u4f4e [6,9]\u3002\u5728\u6211\u5011\u5be6\u9a57\u4e2d\u4e5f\u78ba\u5be6\u6709\u5982\u6b64\u60c5\u6cc1\u3002\u53e6\u4e00\u500b\u53c3\u6578\u03b1\u503c\uff0c\u6211\u5011\u76ee\u524d\u53ea\u4f5c\u4e86\u4e00\u4e9b\u521d\u6b65\u6e2c \u8a66\uff0c\u7531\u65bc\u7bc7\u5e45\u7684\u95dc\u4fc2\uff0c\u6b64\u8655\u50c5\u5831\u544a\u03b1=0.1 \u7684\u60c5\u6cc1(KSE-ME)\u3002\u5728\u8868\u516b\u8207\u8868\u4e5d\u7684\u7d50\u679c\u4e2d\uff0c \u03bb d ECI-SVM ECI-ME KSE-ME macroP 0.00 0.01 0.05 0.00 0.01 0.05 0.00 0.01 0.05 0.00 1.08% 1.09% 1.24% 1.11% 1.12% 1.21% 1.32% 1.32% 1.42% 0.10 1.18% 3.12% 1.28% 2.77% 2.36% 9.05% 0.20 1.34% 8.37% 1.74% 14.80% 4.93% 26.55% 0.30 1.82% 13.18% 3.04% 30.04% 10.09% 36.34% 0.40 3.50% 15.32% 5.98% 35.96% 17.62% 42.42% 0.50 6.87% 16.25% 12.98% 38.62% 22.82% 47.55% 0.60 9.45% 16.81% 20.91% 40.42% 30.08% 49.37% 0.70 12.68% 17.20% 23.85% 41.34% 33.96% 50.12% 0.80 14.69% 17.52% 24.95% 41.81% 35.31% 51.01% 0.90 16.86% 17.90% 24.38% 42.17% 35.52% 51.96% \u03bb s 1.00 17.64% 17.93% 24.93% 42.40% 36.05% 52.72% \u8868\u4e03\u3001\u968e\u5c64\u5f0f\u76ee\u9304\u6574\u5408\u7684 Micro-Precision \u03bb d ECI-SVM ECI-ME KSE-ME microP 0.00 0.01 0.05 0.00 0.01 0.05 0.00 0.01 0.05 0.00 1.08% 1.09% 1.23% 1.11% 1.12% 1.18% 1.29% 1.30% 1.40% 0.10 1.18% 2.43% 1.28% 2.26% 1.75% 4.80% 0.20 1.34% 6.34% 1.71% 9.82% 2.84% 18.52% 0.30 1.80% 9.86% 2.84% 23.47% 4.81% 29.10% 0.40 3.29% 11.33% 5.58% 30.47% 6.94% 35.71% 0.50 6.21% 12.24% 10.93% 34.10% 11.32% 39.81% 0.60 8.90% 12.87% 18.58% 37.24% 19.35% 40.81% 0.70 11.71% 13.46% 20.91% 38.79% 25.17% 41.34% 0.80 13.27% 13.95% 21.36% 39.32% 26.97% 41.97% 0.90 14.76% 14.26% 20.29% 39.67% 27.37% 42.41% \u03bb s 1.00 15.27% 14.33% 20.33% 39.92% 28.13% 43.61% \u8868\u516b\u3001\u968e\u5c64\u5f0f\u76ee\u9304\u6574\u5408\u7684 Macro-Recall \u6211\u5011\u53ef\u4ee5\u770b\u5230\u96a8\u8457\u03bb s \u7684\u589e\u52a0\uff0cMacro-Recall \u8207 \u8868\u516d\u3001\u968e\u5c64\u5f0f\u76ee\u9304\u6574\u5408\u7684 Macro-Precision \u03bb d \u5011\u9032\u884c\u5be6\u9a57\u7684\u6587\u4ef6\u7576\u4e2d\uff0c\u5dee\u96c6\u7684\u90e8\u4efd\uff5cGoogle-Yahoo!-\u662f\u8a13\u7df4\u6587\u4ef6(G-Y) \uff0c\u4ea4\u96c6\u7684 ECI-SVM ECI-ME KSE-ME</td></tr><tr><td colspan=\"5\">\u90e8\u4efd\uff5cGoogle\u2229Yahoo!-\u662f\u6e2c\u8a66\u6587\u4ef6(G Test) \u3002 macroR 0.00 0.01 0.05 0.00 0.01</td><td>0.05</td><td>0.00</td><td>0.01</td><td>0.05</td></tr><tr><td colspan=\"8\">0.00 94.0.05 \u8868\u56db\u3001\u5be6\u9a57\u4e2d\u6240\u7528\u5230\u7684\u76ee\u9304\u985e\u5225\u6578\uff0c\u4ee5\u53ca\u8a13\u7df4\u8207\u6e2c\u8a66\u6587\u4ef6\u6578\u91cf 95.83% 94.31% 95.96% 92.09% 96.99% 94.80% Yahoo! 0.10 95.67% 92.44% 92.98% 85.16% 93.80% 89.98% Google 0.20 92.41% 86.91% 87.57% 79.65% 89.21% 83.12%</td></tr><tr><td>0.30</td><td>Y-G</td><td>Y Class 89.58% 83.66%</td><td>Y Test</td><td colspan=\"2\">G-Y 84.40% 76.92%</td><td>G Class</td><td>G Test 85.06% 81.02%</td></tr><tr><td colspan=\"8\">Entropy \u6a21\u578b\u5206\u985e\u5668\u7684\u8cc7\u6599\u683c\u5f0f\uff0c\u9032\u884c\u76ee 5829 27 641 74.67% 72.94% 73.49% 76.00% \u9304\u6574\u5408\u5de5\u4f5c\u3002\u6bcf\u4e00\u4efd\u6587\u4ef6\u90fd\u6703\u6709\u81ea\u5df1\u672c\u8eab\u7684\u76ee\u9304\u8cc7\u8a0a\u548c\uff1cfeature, weight\uff1e\u914d\u5c0d\u800c\u6210\u7acb\u3002 Software 1876 15 710 0.90 77.18% 77.82% Autos 1681 24 436 1096 12 426 Movies 7255 27 1344 5188 26 Outdoors 1579 19 210 2396 16 Photo 1304 23 218 615 9 0.80 77.69% 77.89% 76.22% 73.49% 74.08% 76.07% 235 0.70 79.37% 78.53% 76.70% 74.35% 75.42% 76.12% 208 0.60 80.72% 79.28% 78.20% 74.87% 76.79% 76.46% 1422 0.40 86.54% 81.54% 81.89% 76.40% 81.68% 78.32% 0.50 84.11% 80.12% 79.85% 75.54% 78.66% 76.99% \u03bb s</td></tr><tr><td colspan=\"8\">\u56e0\u6b64\u6bcf\u4e00\u4efd\u6587\u4ef6\u88ab\u5b9a\u7fa9\u70ba\uff1cline\uff1e=\uff1ctarget\uff1e\uff1cfeature\uff1e\uff1a \uff1cvalue\uff1e...\uff1cfeature\uff1e\uff1a\uff1c Total 13965 108 2918 15124 90 2932 1.00 76.56% 77.79% 73.22% 72.83% 73.31% 75.93%</td></tr><tr><td colspan=\"8\">value\uff1e\u3002\u7576\u4e2d\u7684 target \u5b9a\u7fa9\u70ba 1 (\u8868\u793a\u662f\u6b63\u5411\u6587\u4ef6) \u6216\u662f-1 (\u8868\u793a\u662f\u8ca0\u5411\u6587\u4ef6) \uff0c\uff1cfeature</td></tr></table>",
                "type_str": "table",
                "text": "\u5728\u5206\u985e\u5668\u4e0a\uff0c\u6211\u5011\u4f7f\u7528 SVM light \u4f86\u5be6\u4f5c ECI-SVM \u7684\u6574\u5408\u6a5f\u5236\uff0c\u4f7f\u7528 linear kernel \u4ee5 \u53ca\u9810\u8a2d\u7684\u53c3\u6578\uff0c\u7248\u672c\u70ba 5.00 \u7248\u3002ME \u5206\u985e\u5668\u5247\u4f7f\u7528 Edinburgh \u5927\u5b78\u7684 ME \u5de5\u5177\u3002\u6240\u4f7f\u7528 \u7684 ME \u5de5\u5177\u7684\u7248\u672c\u70ba 20041229 \u7248\u3002 \u503c\u8a2d\u5b9a\u5f9e 0.00 \u5230 1.00\uff0c\u800c\u5728\u76ee\u7684\u76ee\u9304\u4e2d\uff0c\u03bb d \u503c\u8a2d\u5b9a\u70ba 0.00\uff0c0.01\uff0c0.05 \u5206\u5225\u4f86\u770b\u5b83 Micro-Recall \u90fd\u540c\u6642\u4e0b\u964d\u3002 \u8868\u516d\u5230\u8868\u5341\u4e00\u986f\u793a\u7531 Google \u5230 Yahoo!\u968e\u5c64\u5f0f\u76ee\u9304\u7684\u6548\u80fd\u3002\u5728\u8868\u516d\u8207\u8868\u4e03\u7576\u4e2d\uff0c\u53ef \u767c\u73fe Macro-Precision \u8207 Micro-Precision \u7684\u6548\u679c\u90fd\u4e0d\u5920\u826f\u597d\u3002\u9019\u662f\u56e0\u70ba\u5728\u5be6\u9a57\u4e2d\uff0c\u6211\u5011 \u5141\u8a31\u4e00\u4efd\u6587\u4ef6\u53ef\u4ee5\u5206\u81f3\u591a\u500b\u76ee\u9304\u985e\u5225\u6240\u81f4\u3002\u6240\u4ee5\u5728\u03bb s \u503c\u8f03\u4f4e\u6642\uff0c\u7531\u65bc\u7f3a\u4e4f\u968e\u5c64\u67b6\u69cb\u7d22 \u5f15\u5178\u7684\u8f14\u52a9\uff0cFalse-positive \u7684\u6bd4\u4f8b\u6703\u666e\u904d\u5347\u9ad8\uff0c\u4f7f Precision \u8868\u73fe\u4e0d\u4f73\u3002\u4f46\u6211\u5011\u53ef\u4ee5\u770b\u51fa\uff0c \u96a8\u8457\u03bb s \u503c\u63d0\u9ad8\uff0cFalse-positive \u7684\u6bd4\u4f8b\u9010\u6b65\u4e0b\u964d\uff0c\u4f7f Precision \u9010\u6b65\u5347\u9ad8\u3002\u540c\u6642\uff0c\u6211\u5011\u4e5f \u53ef\u4ee5\u767c\u73fe\uff0c\u63a1\u7528\u95dc\u9375\u8a5e\u8a9e\u7fa9\u64f4\u5c55\u7684 KSE-ME\uff0c\u5728 Precision \u4e0a\u6709\u6700\u597d\u7684\u8868\u73fe\u3002 \u5728\u8868\u516b\u8207\u8868\u4e5d\u4e2d\uff0c\u6211\u5011\u53ef\u4ee5\u767c\u73fe\uff0cKSE-ME \u5728 Recall \u7684\u8868\u73fe\u4e0a\uff0c\u666e\u904d\u90fd\u8981\u6bd4 ECI-ME \u70ba\u4f73\u3002\u81f3\u65bc ECI-SVM\uff0c\u96d6\u7136\u5176 Precision \u7684\u8868\u73fe\u4e0d\u5982 ECI-ME \u8207 KSE-ME\uff0c\u4f46\u5728 Recall \u7684\u8868\u73fe\u4e0a\uff0c\u53cd\u800c\u666e\u904d\u6709\u6700\u597d\u7684\u8868\u73fe\u3002\u4f46\u53ef\u5f9e\u8868\u4e2d\u770b\u51fa\uff0c\u7576\u03bb s =0.05 \u9019\u500b\u5e38\u5728\u5be6\u9a57\u4e2d\u4f7f\u7528 \u7684\u503c\u6642\uff0cKSE-ME \u4f9d\u7136\u6709\u9818\u5148\u7684\u8868\u73fe\u3002 \u82e5\u5f9e F 1 \u7684\u8868\u73fe\u4e0a\u4f86\u770b\uff0c\u8868\u5341\u548c\u8868\u5341\u4e00\u986f\u793a KSE-ME \u90fd\u6709\u4e0d\u932f\u7684\u6210\u6548\uff0c\u6bd4 ECI-ME \u8207 ECI-SVM \u7684 F 1 \u8868\u73fe\u90fd\u4f86\u5f97\u597d\u3002\u56e0\u6b64\uff0c\u5f9e\u76ee\u524d\u521d\u6b65\u7684\u5be6\u9a57\u53ef\u4ee5\u5f97\u77e5\uff0c\u5229\u7528\u968e\u5c64\u67b6\u69cb\u7d22 \u5f15\u5178\u8cc7\u8a0a\u4ee5\u53ca\u95dc\u9375\u8a5e\u8a9e\u7fa9\u64f4\u5c55\uff0c\u968e\u5c64\u5f0f\u76ee\u9304\u6574\u5408\u7684\u6548\u80fd\u53ef\u4ee5\u6709\u6548\u7684\u63d0\u5347\u3002 66% 94.66% 88.48% 95.80% 96.05% 94.42% 97.43% 97.69% 97.69%"
            }
        }
    }
}