language: en
license: mit
tags:
- keyphrase-extraction
datasets:
- midas/kpcrowd
metrics:
- seqeval
widget:
- 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 deep learning
methods (such as recurrent neural networks and transformers, GANS, …),
keyphrase extraction can be improved. These new methods also focus on the
semantics and context of a document, which is quite an improvement.
example_title: Example 1
- 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.
example_title: Example 2
model-index:
- name: DeDeckerThomas/keyphrase-extraction-kbir-kpcrowd
results:
- task:
type: keyphrase-extraction
name: Keyphrase Extraction
dataset:
type: midas/kpcrowd
name: kpcrowd
metrics:
- type: seqeval
value: 0.427
name: F1-score
🔑 Keyphrase Extraction model: KBIR-KPCrowd
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 deep learning methods (such as recurrent neural networks and transformers, GANS, …), keyphrase extraction can be improved. These new methods also focus on the semantics and context of a document, which is quite an improvement.
📓 Model Description
This model is a fine-tuned KBIR model on the KPCrowd dataset. KBIR or Keyphrase Boundary Infilling with Replacement is a pre-trained model which utilizes a multi-task learning setup for optimizing a combined loss of Masked Language Modeling (MLM), Keyphrase Boundary Infilling (KBI) and Keyphrase Replacement Classification (KRC). You can find more information about the architecture in this paper: https://arxiv.org/abs/2112.08547.
The model is fine-tuned as a token classification problem where the text is labeled using the BIO scheme.
| Label | Description |
|---|---|
| B-KEY | At the beginning of a keyphrase |
| I-KEY | Inside a keyphrase |
| O | Outside a keyphrase |
Kulkarni, Mayank, Debanjan Mahata, Ravneet Arora, and Rajarshi Bhowmik. "Learning Rich Representation of Keyphrases from Text." arXiv preprint arXiv:2112.08547 (2021).
Sahrawat, Dhruva, Debanjan Mahata, Haimin Zhang, Mayank Kulkarni, Agniv Sharma, Rakesh Gosangi, Amanda Stent, Yaman Kumar, Rajiv Ratn Shah, and Roger Zimmermann. "Keyphrase extraction as sequence labeling using contextualized embeddings." In European Conference on Information Retrieval, pp. 328-335. Springer, Cham, 2020.
✋ Intended uses & limitations
🛑 Limitations
- This keyphrase extraction model is very dataset-specific. It's not recommended to use this model for other domains, but you are free to test it out.
- Only works for English documents.
- Large number of annotated keyphrases.
- For a custom model, please consult the training notebook for more information (link incoming).
❓ How to use
from transformers import (
TokenClassificationPipeline,
AutoModelForTokenClassification,
AutoTokenizer,
)
from transformers.pipelines import AggregationStrategy
import numpy as np
# Define keyphrase extraction pipeline
class KeyphraseExtractionPipeline(TokenClassificationPipeline):
def __init__(self, model, *args, **kwargs):
super().__init__(
model=AutoModelForTokenClassification.from_pretrained(model),
tokenizer=AutoTokenizer.from_pretrained(model),
*args,
**kwargs
)
def postprocess(self, model_outputs):
results = super().postprocess(
model_outputs=model_outputs,
aggregation_strategy=AggregationStrategy.SIMPLE,
)
return np.unique([result.get("word").strip() for result in results])
# Load pipeline
model_name = "DeDeckerThomas/keyphrase-extraction-kbir-kpcrowd"
extractor = KeyphraseExtractionPipeline(model=model_name)
# Inference
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 deep learning methods (such as recurrent neural networks and transformers, GANS, …),
keyphrase extraction can be improved. These new methods also focus on the semantics and context of a document, which is quite an improvement.
""".replace(
"\n", ""
)
keyphrases = extractor(text)
print(keyphrases)
# Output
['Artificial Intelligence' 'GANS' 'Keyphrase extraction'
'classical machine learning' 'deep learning methods'
'keyphrase extraction' 'linguistics' 'recurrent neural networks'
'semantics' 'statistics' 'text analysis' 'transformers']
📚 Training Dataset
KPCrowd is a keyphrase a broadcast news transcription dataset consisting of 500 English broadcast news stories from 10 different categories (art and culture, business, crime, fashion, health, politics us, politics world, science, sports, technology) with 50 docs per category. This dataset is annotated by multiple annotators that were required to look at the same news story and assign a set of keyphrases from the text itself.
You can find more information here: https://huggingface.co/datasets/midas/kpcrowd and https://github.com/LIAAD/KeywordExtractor-Datasets.
👷♂️ Training procedure
For more in detail information, you can take a look at the training notebook (link incoming).
Training parameters
| Parameter | Value |
|---|---|
| Learning Rate | 1e-4 |
| Epochs | 50 |
| Early Stopping Patience | 3 |
Preprocessing
The documents in the dataset are already preprocessed into list of words with the corresponding labels. The only thing that must be done is tokenization and the realignment of the labels so that they correspond with the right subword tokens.
# Labels
label_list = ["B", "I", "O"]
lbl2idx = {"B": 0, "I": 1, "O": 2}
idx2label = {0: "B", 1: "I", 2: "O"}
def preprocess_fuction(all_samples_per_split):
tokenized_samples = tokenizer.batch_encode_plus(
all_samples_per_split[dataset_document_column],
padding="max_length",
truncation=True,
is_split_into_words=True,
max_length=max_length,
)
total_adjusted_labels = []
for k in range(0, len(tokenized_samples["input_ids"])):
prev_wid = -1
word_ids_list = tokenized_samples.word_ids(batch_index=k)
existing_label_ids = all_samples_per_split[dataset_biotags_column][k]
i = -1
adjusted_label_ids = []
for wid in word_ids_list:
if wid is None:
adjusted_label_ids.append(lbl2idx["O"])
elif wid != prev_wid:
i = i + 1
adjusted_label_ids.append(lbl2idx[existing_label_ids[i]])
prev_wid = wid
else:
adjusted_label_ids.append(
lbl2idx[
f"{'I' if existing_label_ids[i] == 'B' else existing_label_ids[i]}"
]
)
total_adjusted_labels.append(adjusted_label_ids)
tokenized_samples["labels"] = total_adjusted_labels
return tokenized_samples
Postprocessing
For the post-processing, you will need to filter out the B and I labeled tokens and concat the consecutive B and Is. As last you strip the keyphrase to ensure all spaces are removed.
# Define post_process functions
def concat_tokens_by_tag(keyphrases):
keyphrase_tokens = []
for id, label in keyphrases:
if label == "B":
keyphrase_tokens.append([id])
elif label == "I":
if len(keyphrase_tokens) > 0:
keyphrase_tokens[len(keyphrase_tokens) - 1].append(id)
return keyphrase_tokens
def extract_keyphrases(example, predictions, tokenizer, index=0):
keyphrases_list = [
(id, idx2label[label])
for id, label in zip(
np.array(example["input_ids"]).squeeze().tolist(), predictions[index]
)
if idx2label[label] in ["B", "I"]
]
processed_keyphrases = concat_tokens_by_tag(keyphrases_list)
extracted_kps = tokenizer.batch_decode(
processed_keyphrases,
skip_special_tokens=True,
clean_up_tokenization_spaces=True,
)
return np.unique([kp.strip() for kp in extracted_kps])
📝 Evaluation results
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. The model achieves the following results on the Inspec test set:
| Dataset | P@5 | R@5 | F1@5 | P@10 | R@10 | F1@10 | P@M | R@M | F1@M |
|---|---|---|---|---|---|---|---|---|---|
| Inspec Test Set | 0.47 | 0.07 | 0.12 | 0.46 | 0.13 | 0.20 | 0.37 | 0.33 | 0.33 |
For more information on the evaluation process, you can take a look at the keyphrase extraction evaluation notebook.
🚨 Issues
Please feel free to contact Thomas De Decker for any problems with this model.