Multi-Instance Multi-Label Learning Networks for Aspect-Category Sentiment Analysis

Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing, pages 3550–3560, November 16–20, 2020. c⃝2020 Association for Computational Linguistics 3550 Multi-Instance Multi-Label Learning Networks for Aspect-Category Sentiment Analysis Yuncong Li 1∗, Cunxiang Yin 1∗, Sheng-hua Zhong 2†, Xu Pan 1 1Baidu Inc., Beijing, China {liyuncong,yincunxiang,panxu}@baidu.com 2College of Computer Science and Software Engineering, Shenzhen University, Shenzhen, China csshzhong@szu.edu.cn Abstract Aspect-category sentiment analysis (ACSA) aims to predict sentiment polarities of sen- tences with respect to given aspect categories. To detect the sentiment toward a particular aspect category in a sentence, most previ- ous methods first generate an aspect category- specific sentence representation for the aspect category, then predict the sentiment polarity based on the representation. These methods ignore the fact that the sentiment of an aspect category mentioned in a sentence is an aggre- gation of the sentiments of the words indicat- ing the aspect category in the sentence, which leads to suboptimal performance. In this pa- per, we propose a Multi-Instance Multi-Label Learning Network for Aspect-Category sen- timent analysis (AC-MIMLLN), which treats sentences as bags, words as instances, and the words indicating an aspect category as the key instances of the aspect category. Given a sen- tence and the aspect categories mentioned in the sentence, AC-MIMLLN first predicts the sentiments of the instances, then finds the key instances for the aspect categories, finally ob- tains the sentiments of the sentence toward the aspect categories by aggregating the key instance sentiments. Experimental results on three public datasets demonstrate the effective- ness of AC-MIMLLN 1. 1 Introduction Sentiment analysis (Pang and Lee, 2008; Liu, 2012) has attracted increasing attention recently. Aspect- based sentiment analysis (ABSA) (Pontiki et al., 2014, 2015, 2016) is a fine-grained sentiment anal- ysis task and includes many subtasks, two of which are aspect category detection (ACD) that detects the aspect categories mentioned in a sentence and ∗Equal contribution †Corresponding author 1Data and code are available at https://github.com/l294265421/AC-MIMLLN While it was large and a bit noisy, the drinks were fantastic, and the food was superb. <ambience, negative> <food, positive> food, positive food, positive ambience, negative ambience, neutral Figure 1: An example of ACD and ACSA. The under- lined words are key instances, the labels of the key in- stances are in the dotted line boxes, and the labels of the sentence are in the angle brackets. aspect-category sentiment analysis (ACSA) that predicts the sentiment polarities with respect to the detected aspect categories. Figure 1 shows an ex- ample. ACD detects the two aspect categories, am- bience and food, and ACSA predicts the negative and positive sentiment toward them respectively. In this work, we focus on ACSA, while ACD as an auxiliary task is used to find the words indicating the aspect categories in sentences for ACSA. Since a sentence usually contains one or more aspect categories, previous studies have developed various methods for generating aspect category- specific sentence representations to detect the sen- timent toward a particular aspect category in a sentence. To name a few, attention-based …

Content truncated to 500 words