Is “hot pizza” Positive or Negative? Mining Target-aware Sentiment Lexicons

Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics, pages 608–618 April 19 - 23, 2021. ©2021 Association for Computational Linguistics 608 Is “hot pizza” Positive or Negative? Mining Target-aware Sentiment Lexicons Jie Zhou1,2, Yuanbin Wu2, Changzhi Sun3, and Liang He1,2 1Shanghai Key Laboratory of Multidimensional Information Processing East China Normal University, China 2School of Computer Science and Technology East China Normal University, China 3ByteDance AI Lab jzhou@ica.stc.sh.cn, {ybwu,lhe}@cs.ecnu.edu.cn sunchangzhi@bytedance.com Abstract Modelling a word’s polarity in different con- texts is a key task in sentiment analysis. Pre- vious works mainly focus on domain depen- dencies, and assume words’ sentiments are in- variant within a specific domain. In this pa- per, we relax this assumption by binding a word’s sentiment to its collocation words in- stead of domain labels. This finer view of sen- timent contexts is particularly useful for iden- tifying commonsense sentiments expressed in neutral words such as “big” and “long”. Given a target (e.g., an aspect), we propose an effec- tive “perturb-and-see” method to extract sen- timent words modifying it from large-scale datasets. The reliability of the obtained target- aware sentiment lexicons is extensively evalu- ated both manually and automatically. We also show that a simple application of the lexicon is able to achieve highly competitive perfor- mances on the unsupervised opinion relation extraction task. 1 Introduction Sentiments of words can be subtle. We are used to using the same word to express different emo- tions in different contexts. “Hot”, for example, suggests a negative sentiment when commenting a computer hardware and a positive sentiment when commenting a pizza, even itself alone is identified without any general orientation. In these situa- tions, it is the composition of a word, contexts, and commonsense carries an opinion. Automatically detecting such context dependent sentiments would strengthen both our understanding of implicit opin- ions in languages and improve existing sentiment analyses models, which is the main topic of this work. To handle shifts of word sentiment, prior works studied how to adapt existing sentiment lexicons to new domains (Hamilton et al., 2016; Xing et al., 0 -1.0-0.8-0.6-0.4-0.2 0.2 0.4 0.6 0.8 1.0 hot cook waitresses toppings pizza run graphics card CPU battery (a) 0 -1.0-0.8-0.6-0.4-0.2 0.2 0.4 0.6 0.8 1.0 long delivery time wait time charge cooking battery life wine list menu service life (b) Figure 1: Visualization of real-world commonsense sentiment of “hot” and “long” extracted by our frame- work. Red and blue indicate the targets in restaurant and electronic domains, respectively. 2019). By modeling differences and similarities of text topics, they can detect new sentiments of words as the domain changes. The basic assump- tion of those domain-level sentiment lexicons is that a word keeps a consistent sentiment within a domain. This assumption, however, might be strong for fine-granularity analyses of text senti- ments: words (especially, neural words such as “long”, “fast”) could exhibit different orientations even in the same domain (Figure 1). To collect more detailed information of a senti- ment, another branch …

Content truncated to 500 words