基於類神經多任務學習做屬性詞提取和情感分類

現今網路上的文字資料量越來越多，也越來越容易獲取，因此需要一個技術能妥善的總結這些文字資料，對個人來說能更輕鬆的獲取資訊，對公司來說則能從中找到改進產品或服務的方向。因此，屬性為基礎的情感分析則成為一個近期被高度關注的技術。屬性為基礎的情感分析是一個分析粒度較精細的情感分析任務，希望能辨別出在一篇文章或句子中，作者對不同屬性的情感表達。這個任務可以被分成兩個子任務：屬性詞抽取以及對特定屬性的情感分析。大部分現有的研究多朝這兩個子任務分別發展相關技術，完整的任務便能藉由結合兩個子任務的答案而獲得解答。然而，我們認為兩個子任務之間是高度相關的，用來預測而抽取出的特徵應要能相互使用。因此，我們將此任務視為序列標記問題，並設計一個類神經多任務學習的整合模型，搭配自我注意力機制以及結合傳統特徵工程的方式，以及使用了與過去多數研究不同的標記規範以提升研究成果。此外，我們也模擬了實際情境做實驗，分別是當訓練資料的領域是不同的、或者是混合的情境下，特別是當資料量比較不足的時候。最後，實驗結果顯示我們提出的方法在三個資料集上的表現都優於現今最先進的方法。

關鍵字

深度學習；屬性為基礎的情感分析；屬性詞抽取；情感分析；序列標記；多任務學習

並列摘要

Aspect-based sentiment analysis (ABSA) is a fine-grained task aiming to identify the sentiment expression toward a specific aspect mentioned in a sentence. A complete ABSA involves two subtasks: aspects extraction and aspect sentiment classification. Most of the existing studies solve ABSA by pipelining the solution of each subtask. In contrast, in this study, we treat ABSA as a sequence labeling problem, and develop an integrated model so that the information can be shared across these two subtasks when learning the integrated model. We improve the performance by using deep multi-task learning framework with self-attention mechanism, and integrating traditional feature engineering in our proposed method. In addition, a more effective tagging scheme is employed. We also conducted experiments on mixed domain and cross domain scenario to simulate the practical situation, especially when there are insufficient training data. Experimental results over three benchmark datasets demonstrate that our method can outperform the state-of-the-art approach.

並列關鍵字

Deep learning ； Aspect-based sentiment analysis ； Aspect extraction ； Aspect sentiment classification ； Sequence labeling ； Multi-task learning

參考文獻

C, C. T. and Joseph, S. (2015). A syntactic approach for aspect based opinion mining. In Proceedings of the 2015 IEEE 9th International Conference on Semantic Computing (IEEE ICSC 2015), pages 24–31.

Google Scholar

Cheng, J., Dong, L., and Lapata, M. (2016). Long short-term memory-networks for machine reading. In Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing, pages 551–561, Austin, Texas. Association for Computational Linguistics.

Google Scholar

Chinsha, T. and Joseph, S. (2015). A syntactic approach for aspect based opinion mining. In Proceedings of the 2015 IEEE 9th International Conference on Semantic Computing (IEEE ICSC 2015), pages 24–31. IEEE.

Google Scholar

Devlin, J., Chang, M.-W., Lee, K., and Toutanova, K. (2018). Bert: Pre-training of deep bidirectional transformers for language understanding. arXiv preprint arXiv:1810.04805.

Google Scholar

Ding, X., Liu, B., and Yu, P. S. (2008). A holistic lexicon-based approach to opinion mining. In Proceedings of the 2008 International Conference on Web Search and Data Mining, WSDM ’08, pages 231–240, New York, NY, USA. ACM.

Google Scholar

國際替代計量

基於類神經多任務學習做屬性詞提取和情感分類

全文下載

主題瀏覽