DeDeckerThomas commited on
Commit
d23943e
β€’
1 Parent(s): 31531b7

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +57 -27
README.md CHANGED
@@ -6,7 +6,17 @@ tags:
6
  datasets:
7
  - midas/openkp
8
  widget:
9
- - text: "Keyphrase extraction is a technique in text analysis where you extract the important keyphrases from a text. Since this is a time-consuming process, Artificial Intelligence is used to automate it. Currently, classical machine learning methods, that use statistics and linguistics, are widely used for the extraction process. The fact that these methods have been widely used in the community has the advantage that there are many easy-to-use libraries. Now with the recent innovations in NLP, transformers can be used to improve keyphrase extraction. Transformers also focus on the semantics and context of a document, which is quite an improvement."
 
 
 
 
 
 
 
 
 
 
10
  example_title: "Example 1"
11
  - text: "In this work, we explore how to learn task specific language models aimed towards learning rich representation of keyphrases from text documents. We experiment with different masking strategies for pre-training transformer language models (LMs) in discriminative as well as generative settings. In the discriminative setting, we introduce a new pre-training objective - Keyphrase Boundary Infilling with Replacement (KBIR), showing large gains in performance (up to 9.26 points in F1) over SOTA, when LM pre-trained using KBIR is fine-tuned for the task of keyphrase extraction. In the generative setting, we introduce a new pre-training setup for BART - KeyBART, that reproduces the keyphrases related to the input text in the CatSeq format, instead of the denoised original input. This also led to gains in performance (up to 4.33 points inF1@M) over SOTA for keyphrase generation. Additionally, we also fine-tune the pre-trained language models on named entity recognition(NER), question answering (QA), relation extraction (RE), abstractive summarization and achieve comparable performance with that of the SOTA, showing that learning rich representation of keyphrases is indeed beneficial for many other fundamental NLP tasks."
12
  example_title: "Example 2"
@@ -34,19 +44,20 @@ model-index:
34
  name: F1@O (Absent)
35
  ---
36
  # πŸ”‘ Keyphrase Generation model: T5-small-OpenKP
37
- Keyphrase extraction is a technique in text analysis where you extract the important keyphrases from a text. Since this is a time-consuming process, Artificial Intelligence is used to automate it. Currently, classical machine learning methods, that use statistics and linguistics, are widely used for the extraction process. The fact that these methods have been widely used in the community has the advantage that there are many easy-to-use libraries. Now with the recent innovations in NLP, transformers can be used to improve keyphrase extraction. Transformers also focus on the semantics and context of a document, which is quite an improvement.
 
38
 
39
 
40
  ## πŸ““ Model Description
41
- This model is a fine-tuned [T5-small model](https://huggingface.co/t5-small) on the OpenKP dataset.
42
 
43
- ## βœ‹ Intended uses & limitations
44
  ### πŸ›‘ Limitations
45
  * Only works for English documents.
46
  * For a custom model, please consult the training notebook for more information (link incoming).
47
  * Sometimes the output doesn't make any sense.
48
 
49
- ### ❓ How to use
50
  ```python
51
  # Model parameters
52
  from transformers import (
@@ -78,19 +89,25 @@ class KeyphraseGenerationPipeline(Text2TextGenerationPipeline):
78
  # Load pipeline
79
  model_name = "ml6team/keyphrase-generation-t5-small-openkp"
80
  generator = KeyphraseGenerationPipeline(model=model_name)
 
81
 
82
  ```python
83
  text = """
84
- Keyphrase extraction is a technique in text analysis where you extract the important keyphrases from a text.
85
- Since this is a time-consuming process, Artificial Intelligence is used to automate it.
86
- Currently, classical machine learning methods, that use statistics and linguistics,
87
- are widely used for the extraction process. The fact that these methods have been widely used in the community
88
- has the advantage that there are many easy-to-use libraries. Now with the recent innovations in NLP,
89
- transformers can be used to improve keyphrase extraction. Transformers also focus on the semantics
90
- and context of a document, which is quite an improvement.
91
- """.replace(
92
- "\n", ""
93
- )
 
 
 
 
 
94
 
95
  keyphrases = generator(text)
96
 
@@ -104,14 +121,13 @@ print(keyphrases)
104
  ```
105
 
106
  ## πŸ“š Training Dataset
107
- OpenKP is a large-scale, open-domain keyphrase extraction dataset with 148,124 real-world web documents along with 1-3 most relevant human-annotated keyphrases.
108
-
109
- You can find more information here: https://github.com/microsoft/OpenKP.
110
 
111
- ## πŸ‘·β€β™‚οΈ Training procedure
112
- For more in detail information, you can take a look at the training notebook (link incoming).
113
 
114
- ### Training parameters
115
 
116
  | Parameter | Value |
117
  | --------- | ------|
@@ -120,9 +136,18 @@ For more in detail information, you can take a look at the training notebook (li
120
  | Early Stopping Patience | 1 |
121
 
122
  ### Preprocessing
123
- The documents in the dataset are already preprocessed into list of words with the corresponding keyphrases. The only thing that must be done is tokenization and joining all keyphrases into one string with a certain seperator of choice(;).
124
  ```python
125
- def pre_process_keyphrases(text_ids, kp_list):
 
 
 
 
 
 
 
 
 
126
  kp_order_list = []
127
  kp_set = set(kp_list)
128
  text = tokenizer.decode(
@@ -141,7 +166,6 @@ def pre_process_keyphrases(text_ids, kp_list):
141
  else:
142
  present_kp.append(kp)
143
  return present_kp, absent_kp
144
-
145
  def preprocess_fuction(samples):
146
  processed_samples = {"input_ids": [], "attention_mask": [], "labels": []}
147
  for i, sample in enumerate(samples[dataset_document_column]):
@@ -151,7 +175,7 @@ def preprocess_fuction(samples):
151
  padding="max_length",
152
  truncation=True,
153
  )
154
- present_kp, absent_kp = pre_process_keyphrases(
155
  text_ids=inputs["input_ids"],
156
  kp_list=samples["extractive_keyphrases"][i]
157
  + samples["abstractive_keyphrases"][i],
@@ -171,7 +195,13 @@ def preprocess_fuction(samples):
171
  processed_samples[key].append(inputs[key])
172
  processed_samples["labels"].append(targets["input_ids"])
173
  return processed_samples
 
 
 
 
 
174
  ```
 
175
  ### Postprocessing
176
  For the post-processing, you will need to split the string based on the keyphrase separator.
177
  ```python
@@ -179,9 +209,9 @@ def extract_keyphrases(examples):
179
  return [example.split(keyphrase_sep_token) for example in examples]
180
  ```
181
 
182
- ## πŸ“ Evaluation results
183
 
184
- One of the traditional evaluation methods is the precision, recall and F1-score @k,m where k is the number that stands for the first k predicted keyphrases and m for the average amount of predicted keyphrases.
185
  The model achieves the following results on the OpenKP test set:
186
 
187
 
 
6
  datasets:
7
  - midas/openkp
8
  widget:
9
+ - text: "Keyphrase extraction is a technique in text analysis where you extract the important keyphrases from a document.
10
+ Thanks to these keyphrases humans can understand the content of a text very quickly and easily without reading
11
+ it completely. Keyphrase extraction was first done primarily by human annotators, who read the text in detail
12
+ and then wrote down the most important keyphrases. The disadvantage is that if you work with a lot of documents,
13
+ this process can take a lot of time.
14
+
15
+ Here is where Artificial Intelligence comes in. Currently, classical machine learning methods, that use statistical
16
+ and linguistic features, are widely used for the extraction process. Now with deep learning, it is possible to capture
17
+ the semantic meaning of a text even better than these classical methods. Classical methods look at the frequency,
18
+ occurrence and order of words in the text, whereas these neural approaches can capture long-term semantic dependencies
19
+ and context of words in a text."
20
  example_title: "Example 1"
21
  - text: "In this work, we explore how to learn task specific language models aimed towards learning rich representation of keyphrases from text documents. We experiment with different masking strategies for pre-training transformer language models (LMs) in discriminative as well as generative settings. In the discriminative setting, we introduce a new pre-training objective - Keyphrase Boundary Infilling with Replacement (KBIR), showing large gains in performance (up to 9.26 points in F1) over SOTA, when LM pre-trained using KBIR is fine-tuned for the task of keyphrase extraction. In the generative setting, we introduce a new pre-training setup for BART - KeyBART, that reproduces the keyphrases related to the input text in the CatSeq format, instead of the denoised original input. This also led to gains in performance (up to 4.33 points inF1@M) over SOTA for keyphrase generation. Additionally, we also fine-tune the pre-trained language models on named entity recognition(NER), question answering (QA), relation extraction (RE), abstractive summarization and achieve comparable performance with that of the SOTA, showing that learning rich representation of keyphrases is indeed beneficial for many other fundamental NLP tasks."
22
  example_title: "Example 2"
 
44
  name: F1@O (Absent)
45
  ---
46
  # πŸ”‘ Keyphrase Generation model: T5-small-OpenKP
47
+ Keyphrase extraction is a technique in text analysis where you extract the important keyphrases from a document. Thanks to these keyphrases humans can understand the content of a text very quickly and easily without reading it completely. Keyphrase extraction was first done primarily by human annotators, who read the text in detail and then wrote down the most important keyphrases. The disadvantage is that if you work with a lot of documents, this process can take a lot of time ⏳.
48
+ Here is where Artificial Intelligence πŸ€– comes in. Currently, classical machine learning methods, that use statistical and linguistic features, are widely used for the extraction process. Now with deep learning, it is possible to capture the semantic meaning of a text even better than these classical methods. Classical methods look at the frequency, occurrence and order of words in the text, whereas these neural approaches can capture long-term semantic dependencies and context of words in a text.
49
 
50
 
51
  ## πŸ““ Model Description
52
+ This model uses [T5-small model](https://huggingface.co/t5-small) as its base model and fine-tunes it on the [OpenKP dataset](https://huggingface.co/datasets/midas/openkp). Keyphrase generation transformers are fine-tuned as a text-to-text generation problem where the keyphrases are generated. The result is a concatenated string with all keyphrases separated by a given delimiter (i.e. β€œ;”). These models are capable of generating present and absent keyphrases.
53
 
54
+ ## βœ‹ Intended Uses & Limitations
55
  ### πŸ›‘ Limitations
56
  * Only works for English documents.
57
  * For a custom model, please consult the training notebook for more information (link incoming).
58
  * Sometimes the output doesn't make any sense.
59
 
60
+ ### ❓ How To Use
61
  ```python
62
  # Model parameters
63
  from transformers import (
 
89
  # Load pipeline
90
  model_name = "ml6team/keyphrase-generation-t5-small-openkp"
91
  generator = KeyphraseGenerationPipeline(model=model_name)
92
+ ```
93
 
94
  ```python
95
  text = """
96
+ Keyphrase extraction is a technique in text analysis where you extract the
97
+ important keyphrases from a document. Thanks to these keyphrases humans can
98
+ understand the content of a text very quickly and easily without reading it
99
+ completely. Keyphrase extraction was first done primarily by human annotators,
100
+ who read the text in detail and then wrote down the most important keyphrases.
101
+ The disadvantage is that if you work with a lot of documents, this process
102
+ can take a lot of time.
103
+ Here is where Artificial Intelligence comes in. Currently, classical machine
104
+ learning methods, that use statistical and linguistic features, are widely used
105
+ for the extraction process. Now with deep learning, it is possible to capture
106
+ the semantic meaning of a text even better than these classical methods.
107
+ Classical methods look at the frequency, occurrence and order of words
108
+ in the text, whereas these neural approaches can capture long-term
109
+ semantic dependencies and context of words in a text.
110
+ """.replace("\n", " ")
111
 
112
  keyphrases = generator(text)
113
 
 
121
  ```
122
 
123
  ## πŸ“š Training Dataset
124
+ [OpenKP](https://github.com/microsoft/OpenKP) is a large-scale, open-domain keyphrase extraction dataset with 148,124 real-world web documents along with 1-3 most relevant human-annotated keyphrases.
125
+ You can find more information in the [paper](https://arxiv.org/abs/1911.02671).
 
126
 
127
+ ## πŸ‘·β€β™‚οΈ Training Procedure
128
+ For more in detail information, you can take a look at the [training notebook]().
129
 
130
+ ### Training Parameters
131
 
132
  | Parameter | Value |
133
  | --------- | ------|
 
136
  | Early Stopping Patience | 1 |
137
 
138
  ### Preprocessing
139
+ The documents in the dataset are already preprocessed into list of words with the corresponding keyphrases. The only thing that must be done is tokenization and joining all keyphrases into one string with a certain seperator of choice( ```;``` ).
140
  ```python
141
+ from datasets import load_dataset
142
+ from transformers import AutoTokenizer
143
+ # Tokenizer
144
+ tokenizer = AutoTokenizer.from_pretrained("t5-small", add_prefix_space=True)
145
+ # Dataset parameters
146
+ dataset_full_name = "midas/inspec"
147
+ dataset_subset = "raw"
148
+ dataset_document_column = "document"
149
+ keyphrase_sep_token = ";"
150
+ def preprocess_keyphrases(text_ids, kp_list):
151
  kp_order_list = []
152
  kp_set = set(kp_list)
153
  text = tokenizer.decode(
 
166
  else:
167
  present_kp.append(kp)
168
  return present_kp, absent_kp
 
169
  def preprocess_fuction(samples):
170
  processed_samples = {"input_ids": [], "attention_mask": [], "labels": []}
171
  for i, sample in enumerate(samples[dataset_document_column]):
 
175
  padding="max_length",
176
  truncation=True,
177
  )
178
+ present_kp, absent_kp = preprocess_keyphrases(
179
  text_ids=inputs["input_ids"],
180
  kp_list=samples["extractive_keyphrases"][i]
181
  + samples["abstractive_keyphrases"][i],
 
195
  processed_samples[key].append(inputs[key])
196
  processed_samples["labels"].append(targets["input_ids"])
197
  return processed_samples
198
+ # Load dataset
199
+ dataset = load_dataset(dataset_full_name, dataset_subset)
200
+ # Preprocess dataset
201
+ tokenized_dataset = dataset.map(preprocess_fuction, batched=True)
202
+
203
  ```
204
+
205
  ### Postprocessing
206
  For the post-processing, you will need to split the string based on the keyphrase separator.
207
  ```python
 
209
  return [example.split(keyphrase_sep_token) for example in examples]
210
  ```
211
 
212
+ ## πŸ“ Evaluation Results
213
 
214
+ Traditional evaluation methods are the precision, recall and F1-score @k,m where k is the number that stands for the first k predicted keyphrases and m for the average amount of predicted keyphrases. In keyphrase generation you also look at F1@O where O stands for the number of ground truth keyphrases.
215
  The model achieves the following results on the OpenKP test set:
216
 
217