hodorxxx / uabsa-symux Goto Github PK

Resources for IJCAI2022 paper: Inheriting the Wisdom of Predecessors: A Multiplex Cascade Framework for Unified Aspect-based Sentiment Analysis

See the project page for more details.

In ABSA community there are at least following seven representative subtasks:

  AE: aspect term extraction;
  OE: opinion term extraction;
  AOE: aspect-oriented opinion extraction;
  AOPE: aspect-opinion pair extraction;
  ALSC: aspect-level sentiment classification;
  AESC: aspect extraction and sentiment classification;
  TE: triplet extraction

All these subtasks are related by revolving around the predictions of three elements: <aspect, opinion, polarity>.

In this project, we consider the unfied ABSA. We try to enhance the ABSA subtasks by making full use of the interactions between all subtasks, with a multiplex cascade framework.

• Wang et al. (2017) [1] annotate the unpaired opinion terms (denoted as D_17),
• Fan et al. (2019) [2] pair the aspects with opinion terms (D_19),
• Peng et al. (2020) [3] further provide the labels for triple extraction (D_20).

To enable multi-task training, we re-ensemble the existing ABSA datasets so that most of the sentences’ annotations cover all seven subtasks.

[1] Coupled Multi-Layer Attentions for Co-Extraction of Aspect and Opinion Terms. In AAAI. 2017.

[2] Target-oriented Opinion Words Extraction with Target-fused Neural Sequence Labeling. In NAACL. 2019.

[3] Knowing What, How and Why: A Near Complete Solution for Aspect-Based Sentiment Analysis. In AAAI. 2020.

The schematic of hierarchical dependency (HD):

The multiplex cascade framework:

- python (3.8.12)
- cuda (11.4)
- numpy (1.21.4)
- torch (1.10.0)
- gensim (4.1.2)
- transformers (4.13.0)
- pandas (1.3.4)
- scikit-learn (1.0.1)
- corenlp (4.2)

First parse out the dependency trees and POS tags for each sentence, and save as json format. Recommend employ the stanfordnlp CoreNLP tool. And use the NLTK package to wrap the parsing process.

Download the RoBERTa PLM.

Configure the configs\train.conf and configs\eval.conf files.

  python main.py

If you use this work or code, please kindly cite:

@inproceedings{fei2022unifiedABSA,
  author    = {Hao Fei and Fei Li and Chenliang Li and Shengqiong Wu and Jingye Li and Donghong Ji},
  title     = {Inheriting the Wisdom of Predecessors: A Multiplex Cascade Framework for Unified Aspect-based Sentiment Analysis},
  booktitle = {Proceedings of the Thirty-First International Joint Conference on Artificial Intelligence, {IJCAI}},
  pages     = {4121--4128},
  year      = {2022},
}