BERT-ASC

A model for implicit aspect sentiment analysis.

This is the source code for the paper: Murtadha, Ahmed, et al. "BERT-ASC: Auxiliary-Sentence Construction for Implicit Aspect Learning in Sentiment Analysis" [1].

Data

The datasets used in our experminents can be downloaded from this SemEval.

Prerequisites:

Required packages are listed in the requirements.txt file:

pip install -r requirements.txt

Pre-processing

• Generate seed words for a given dataset (e.g., semeval):
  • Go to L-LDA/ and run the following code

   python run_l_LDA.py --dataset semeval
  

The original code of L-LDA is publicly available

• Generate the semantic candidates:
  • Go to ASC_generating/
  • The processed data and embedings for restaurant is available. Note that these files were orginally proccessed by Ruidan He
  • To process your own data and embeddings, put your data file in datasets then run this code:

   python preprocessing.py
   python generate_domain_embedding.py	
  

  • Run the following code to extract the semantic candidates

   python semantic_candidate_generating.py --dataset semeval
  

• Generate the synticatic candidates:
  • Run the following code to generate the synticatic informatiom

   python ASC_generating/opinion_words_extracting.py --dataset semeval
  

Training:

• To train BERT-ASC:
  • Go to code/ and run the following code

   python code/run.py --dataset semeval 
  

  • The params could be :
    • --dataset ={semeval,sentihood}
  • Or run this scripts code/scripts

  sh training.sh 0 bert-base-uncased 
  

Evalutaion:

• To evaluate BERT-ASC:

  • Go to code/ and run the following code

   python code/evaluate.py --dataset semeval
  

  • The params could be :
    • --dataset ={semeval,sentihood}
  • Or run this scripts code/scripts

   sh evaluate.sh 0 bert-base-uncased 
  

If you use the code, please cite the paper:

@article{murtadha2022bert,
  title={BERT-ASC: Auxiliary-Sentence Construction for Implicit Aspect Learning in Sentiment Analysis},
  author={Murtadha, Ahmed and Pan, Shengfeng and Wen, Bo and Su, Jianlin and Zhang, Wenze and Liu, Yunfeng},
  journal={arXiv preprint arXiv:2203.11702},
  year={2022}
}