意見探勘最常用於分析使用者於社群平台上的留言與產品評論，而面向情感分析作為意見探勘的主要任務，已成為一個重要的研究主題。現有許多面向情感分析的模型是基於神經網絡，例如遞歸神經網絡、長短期記憶模型和注意力機制，這些模型通過神經網絡來獲取意見詞彙與面向字眼的語意關係以獲得更好的模型成效。此外，一些研究指出，使用圖卷積網絡來分析依存結構樹，可有效捕捉句子中不同詞彙之間的語法關係，從而縮短面向情感分析的意見詞彙與面向字眼之間的距離。
近來，一些研究人員使用異質性網絡來保存依存結構樹中的信息並新增一些新的節點和邊。然而，此研究所提出的異質性網絡模型成效不如單一的依存結構樹。我們推測原因是圖卷積網絡不適合用來分析異質性網絡。為了解決此問題，我們提出一個新的模型於面向情感分析，使用關係注意力網絡代替圖卷積網絡來分析異質性網絡。
從四個資料集上的實驗結果，我們的模型優於現在最佳的方法。另外，我們也研究了異質性網絡中的信息流動狀況來了解圖中每個節點於面向情感分析的重要性。此外，通過注意力分析，我們發現關係注意力網絡在重要的詞彙上，有更重要的注意力權重。

Opinion mining is very popular for analyzing users' comments or product reviews on social media. Being a core subtask in opinion mining, aspect-based sentiment analysis (ABSA) has recently become an important research topic. Many existing methods for the ABSA are based on neural networks such as the Recursive Neural Network, the Long Short-Term Memory (LSTM), and the attention mechanism. Some existing methods tended to have better performance by capturing the semantic relationships between the opinion words and the target words in ABSA using a neural network. In addition, some studies used the Graph Convolutional Network (GCN) to encode the dependency tree to effectively capture the syntactical relationships between different words in a sentence to shorten the distance between the aspect terms and the opinion words for the ABSA.
More recently, some researchers used a heterogeneous graph to record the information in a dependency tree and some new conceptual nodes and relationships. However, the performance of the proposed heterogeneous graph is not as good as a pure dependency tree. We suspected that the reason is that a GCN is not suitable for encoding the heterogeneous graph. To address this problem, we propose to use the Relational Graph Attention Neural Network (RGAT) to replace the GCN to process the information in an aspect-specific heterogeneous graph for the ABSA. The experimental results on four benchmark datasets showed that our model outperformed the state-of-the-art approaches. We also studied the information flow in the aspect-specific heterogeneous graph to understand the importance of each node in the graph for the ABSA. Furthermore, by attention analysis, we showed that the RGAT has put attention on the words that are more important for the ABSA.

1. Ding, X., B. Liu, and P.S. Yu, A holistic lexicon-based approach to opinion mining, in Proceedings of the 2008 International Conference on Web Search and Data Mining. 2008, Association for Computing Machinery: Palo Alto, California, USA. p. 231–240.
2. Hu, M. and B. Liu, Mining and summarizing customer reviews, in Proceedings of the tenth ACM SIGKDD international conference on Knowledge discovery and data mining. 2004, Association for Computing Machinery: Seattle, WA, USA. p. 168–177.
3. Popescu, A.-M. and O. Etzioni, Extracting product features and opinions from reviews. Conference on Human Language Technology and Empirical Methods in Natural Language Processing, 2005: p. 9-28.
4. Dong, L., et al. Adaptive Recursive Neural Network for Target-dependent Twitter Sentiment Classification. 2014. Baltimore, Maryland: Association for Computational Linguistics.
5. Nguyen, T.H. and K. Shirai. PhraseRNN: Phrase Recursive Neural Network for Aspect-based Sentiment Analysis. 2015. Lisbon, Portugal: Association for Computational Linguistics.
6. Tang, D., et al. Effective LSTMs for Target-Dependent Sentiment Classification. 2016. Osaka, Japan: The COLING 2016 Organizing Committee.
7. Wang, Y., et al. Attention-based LSTM for Aspect-level Sentiment Classification. 2016. Austin, Texas: Association for Computational Linguistics.
8. Tang, D., B. Qin, and T. Liu. Aspect Level Sentiment Classification with Deep Memory Network. 2016. Austin, Texas: Association for Computational Linguistics.
9. Ma, D., et al., Interactive attention networks for aspect-level sentiment classification, in Proceedings of the 26th International Joint Conference on Artificial Intelligence. 2017, AAAI Press: Melbourne, Australia. p. 4068–4074.
10. Chen, P., et al. Recurrent Attention Network on Memory for Aspect Sentiment Analysis. 2017. Copenhagen, Denmark: Association for Computational Linguistics.
11. Gu, S., et al. A Position-aware Bidirectional Attention Network for Aspect-level Sentiment Analysis. 2018. Santa Fe, New Mexico, USA: Association for Computational Linguistics.
12. Yao, L., C. Mao, and Y. Luo, Graph Convolutional Networks for Text Classification. 2018.
13. Marcheggiani, D. and I. Titov. Encoding Sentences with Graph Convolutional Networks for Semantic Role Labeling. 2017. Copenhagen, Denmark: Association for Computational Linguistics.
14. Zhang, Y., P. Qi, and C.D. Manning. Graph Convolution over Pruned Dependency Trees Improves Relation Extraction. 2018. Brussels, Belgium: Association for Computational Linguistics.
15. Sun, K., et al. Aspect-Level Sentiment Analysis Via Convolution over Dependency Tree. 2019. Hong Kong, China: Association for Computational Linguistics.
16. Zhang, C., Q. Li, and D. Song. Aspect-based Sentiment Classification with Aspect-specific Graph Convolutional Networks. 2019. Hong Kong, China: Association for Computational Linguistics.
17. Bai, X., P. Liu, and Y. Zhang, Investigating Typed Syntactic Dependencies for Targeted Sentiment Classification Using Graph Attention Neural Network. IEEE/ACM Transactions on Audio, Speech, and Language Processing, 2021. 29: p. 503-514.
18. Xu, K., H. Zhao, and T. Liu, Aspect-Specific Heterogeneous Graph Convolutional Network for Aspect-Based Sentiment Classification. IEEE Access, 2020. 8: p. 139346-139355.
19. Huang, L., et al., Syntax-Aware Graph Attention Network for Aspect-Level Sentiment Classification. 2020. 799-810.
20. Zhao, P., L. Hou, and O. Wu, Modeling sentiment dependencies with graph convolutional networks for aspect-level sentiment classification. Knowledge-Based Systems, 2019. 193: p. 105443.
21. Zuo, E., et al., Context-Specific Heterogeneous Graph Convolutional Network for Implicit Sentiment Analysis. IEEE Access, 2020. 8: p. 37967-37975.
22. Shi, J., et al., Heterogeneous Graph Neural Network for Recommendation. 2020.
23. Wang, D., et al. Heterogeneous Graph Neural Networks for Extractive Document Summarization. 2020. Online: Association for Computational Linguistics.
24. Jiang, L., et al. Target-dependent Twitter Sentiment Classification. 2011. Portland, Oregon, USA: Association for Computational Linguistics.
25. Mikolov, T., et al. Efficient Estimation of Word Representations in Vector Space. in ICLR. 2013.
26. Pennington, J., R. Socher, and C. Manning. GloVe: Global Vectors for Word Representation. 2014. Doha, Qatar: Association for Computational Linguistics.
27. Socher, R., et al., Parsing Natural Scenes and Natural Language with Recursive Neural Networks. 2011. 129-136.
28. Hochreiter, S. and J. Schmidhuber, Long Short-Term Memory. Neural Comput., 1997. 9(8): p. 1735–1780.
29. Vaswani, A., et al., Attention is All you Need. ArXiv, 2017. abs/1706.03762.
30. Cho, K., et al., Learning Phrase Representations using RNN Encode' Decoder for Statistical Machine Translation. ArXiv, 2014. abs/1406.1078.
31. Devlin, J., et al. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding. in NAACL-HLT. 2019.
32. Brown, T., et al., Language Models are Few-Shot Learners. ArXiv, 2020. abs/2005.14165.
33. LeCun, Y. and Y. Bengio, Convolutional networks for images, speech, and time series, in The handbook of brain theory and neural networks. 1998, MIT Press. p. 255–258.
34. Hui, K., et al. PACRR: A Position-Aware Neural IR Model for Relevance Matching. in EMNLP. 2017.
35. Zhang, X., J. Zhao, and Y. LeCun, Character-level Convolutional Networks for Text Classification. ArXiv, 2015. abs/1509.01626.
36. Li, X., et al. Transformation Networks for Target-Oriented Sentiment Classification. 2018. Melbourne, Australia: Association for Computational Linguistics.
37. Kipf, T. and M. Welling, Semi-Supervised Classification with Graph Convolutional Networks. ArXiv, 2017. abs/1609.02907.
38. Velickovic, P., et al., Graph Attention Networks. ArXiv, 2018. abs/1710.10903.
39. Pontiki, M., et al. SemEval-2014 Task 4: Aspect Based Sentiment Analysis. 2014. Dublin, Ireland: Association for Computational Linguistics.