以「知識增強膠囊網路」實作意見層面情感分析

Knowledge Enhance Capsule Network for Aspect-Based Sentiment Analysis

吳雨澤(Yu-Tse Wu)

指導教授 ： 曹承礎

國立臺灣大學/管理學院/資訊管理學研究所/碩士(2022年)

https://doi.org/10.6342/NTU202201143

摘要

意見層面情感分析(ABSA)因其廣泛的應用領域而成為情感分析中的一項重要任務。此任務的目標是識別句子或文章對於給定意見的情感極性。在過去的研究中，使用傳統機器學習方法或神經網絡方法都能達成不錯的表現，而近幾年，許多研究者將膠囊網路應用於ABSA問題上，其研究結果顯示膠囊網路能夠有效的提高準確率。然而，儘管前述的研究取得了不錯的成果，當一個句子或一篇文章有多個不同的意見目標且對於各個意見有不同的情感時，如何正確擷取出有關該意見目標的情感字詞仍然是一個挑戰。在本研究中，我們提出了一個新模型「知識增強膠囊網絡（KECapsNet）」來實作ABSA任務。不同於傳統的膠囊網路，KECapsNet使用如語法結構、局部上下文關係等多種先驗知識來建構初級膠囊，然後利用情感辭典來引導這些初級膠囊並將其轉換為輸出膠囊，這些輸出膠囊將最終決定情感分類的結果。我們在多個資料集上進行實驗，其結果顯示我們所提出的模型能夠達到比現存方法更高的準確率。

關鍵字

意見層面情感分析 ； 膠囊網路 ； 卷積神經網路 ； 圖卷積神經網路 ； 長短期記憶模型

並列摘要

Aspect-based sentiment analysis (ABSA) is an important task in the field of sentiment analysis due to its wide applications.The goal of ABSA is to identify the sentiment polarities of a sentence or document toward given aspects. Previous studies using traditional machine learning methods or neural network methods have achieved good performance on ABSA task, while recent research using capsule-based methods have shown that utilizing capsule network on ABSA task can improve the accuracy effectively. However, it is still a challenge to identify the sentiment words to the correct aspects when a sentence or a paragraph expresses different emotions toward multiple aspects. In this paper, we proposed a knowledge enhance capsule network (KECapsNet) for ABSA, which use multiple prior knowledge to enhance the original capsule-based method. We utilize prior knowledge such as syntactic knowledge and local context knowledge to construct the primary capsules in KECapsNet, then the model make the sentiment classification using lexicon-guided routing mechanism, which utilize the sentiment lexicon to guide the transformation of primary capsules to output capsules. We implement the experiment on several benchmark datasets, and the results show that the proposed model outperform the state-of-the-art methods.

並列關鍵字

Aspect-based sentiment analysis ； capsule network ； convolutional neural network ； graph convolutional network ； long short term memory network

參考文獻

B. Liu, “Sentiment analysis and opinion mining,” Synthesis lectures on human language technologies, vol. 5, no. 1, pp. 1–167, 2012.

L. Jiang, M. Yu, M. Zhou, X. Liu, and T. Zhao, “Target-dependent twitter sentiment classification,” in Proceedings of the 49th annual meeting of the association for computational linguistics: human language technologies, pp. 151–160, 2011.

S. M. Mohammad, S. Kiritchenko, and X. Zhu, “Nrc-canada: Building the state-ofthe-art in sentiment analysis of tweets,” arXiv preprint arXiv:1308.6242, 2013.

S. Poria, E. Cambria, D. Hazarika, and P. Vij, “A deeper look into sarcastic tweets using deep convolutional neural networks,” arXiv preprint arXiv:1610.08815, 2016.

P. Chen, Z. Sun, L. Bing, and W. Yang, “Recurrent attention network on memory for aspect sentiment analysis,” in Proceedings of the 2017 conference on empirical methods in natural language processing, pp. 452–461, 2017.