Relation Extraction

Pytorch Implementation of Deep Learning Approach for Relation Extraction Challenge(SemEval-2010 Task #8: Multi-Way Classification of Semantic Relations Between Pairs of Nominals)

关系抽取/分类的PyTorch实现。

本项目介绍了三种模型用于此任务

1. BiLSTM + Attention + PF
2. BiLSTM + RNN + PF
3. CNN + PF

Welcome to watch, star or fork.

各模型结构如下（详见文末论文链接）

BiLSTM Attention

BiLSTM CNN

CNN

Requirements

This repo was tested on Python 3.6 + and PyTorch 1.0.0. The requirements are:

• torch >= 1.0.0
• numpy
• sklearn
• tqdm

Data: SemEval-2010 Task #8

• Given: a sentence marked with a pair of nominals
• Goal: recognize the semantic relation between these nominals.
• Example:
  • "There were apples, pears and oranges in the bowl." => Content-Container(e1,e2)
  • “The cup contained tea from dried ginseng.” => Entity-Origin(e1,e2)

The Inventory of Semantic Relations

1. Cause-Effect: An event or object leads to an effect(those cancers were caused by radiation exposures)
2. Instrument-Agency: An agent uses an instrument(phone operator)
3. Product-Producer: A producer causes a product to exist (a factory manufactures suits)
4. Content-Container: An object is physically stored in a delineated area of space (a bottle full of honey was weighed) Hendrickx, Kim, Kozareva, Nakov, O S´ eaghdha, Pad ´ o,´ Pennacchiotti, Romano, Szpakowicz Task Overview Data Creation Competition Results and Discussion The Inventory of Semantic Relations (III)
5. Entity-Origin: An entity is coming or is derived from an origin, e.g., position or material (letters from foreign countries)
6. Entity-Destination: An entity is moving towards a destination (the boy went to bed)
7. Component-Whole: An object is a component of a larger whole (my apartment has a large kitchen)
8. Member-Collection: A member forms a nonfunctional part of a collection (there are many trees in the forest)
9. Message-Topic: An act of communication, written or spoken, is about a topic (the lecture was about semantics)
10. Other: If none of the above nine relations appears to be suitable.

Distribution for Dataset

Quickstart

• Train data is located in "data/SemEval2010_task8/TRAIN_FILE.TXT".
• Vector_50d.txt is used as pre-trained word2vec model.
• We use micro-average F-score over the 18 relation labels apart from Other as our evaluation criteria.

1. Build vocabularies and parameters for your dataset by running

   python build_vocab.py --data_dir data/SemEval2010_task8

   It will write vocabulary files words.txt and labels.txt containing the words and labels in the dataset. It will also save a dataset_params.json with some extra information.

2. Your experiment We created a base_model directory for you under the experiments directory. It contains a file params.json which sets the hyperparameters for the experiment. It looks like

   {
      "learning_rate": 1e-3,
      "batch_size": 50,
      "num_epochs": 100
   }

   For every new experiment, you will need to create a new directory under experiments with a params.json file.

3. Train your experiment. Simply run

   python train.py --data_dir data/SemEval2010_task8 --model_dir experiments/base_mode

   It will instantiate a model and train it on the training set following the hyperparameters specified in params.json. It will also evaluate some metrics on the development set.

4. Evaluation on the test set Once you've run many experiments and selected your best model and hyperparameters based on the performance on the development set, you can finally evaluate the performance of your model on the test set. Run

   python evaluate.py --data_dir data/SemEval2010_task8 --model_dir experiments/base_model

Results

BiLSTM + Attention

BiLSTM + MaxPooling

CNN

References

• Attention-Based Bidirectional Long Short-Term Memory Networks forRelation Classification (ACL 2016),P Zhou,W Shi. [paper]
• Relation Classification via Recurrent Neural Network ,Dongxu Zhang, Dong Wang. [paper]
• Relation Extraction: Perspective from Convolutional Neural Networks (NAACL 2015), TH Nguyen et al. [paper]
• Relation extract review literature from blog [link]