Toward Comprehensive Understanding of a Sentiment Based on Human Motives

Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 4672–4677 Florence, Italy, July 28 - August 2, 2019. c⃝2019 Association for Computational Linguistics 4672 Toward Comprehensive Understanding of a Sentiment Based on Human Motives Naoki Otani Language Technologies Institute Carnegie Mellon University Pittsburgh, PA, USA notani@cs.cmu.edu Eduard Hovy Language Technologies Institute Carnegie Mellon University Pittsburgh, PA, USA hovy@cmu.edu Abstract In sentiment detection, the natural language processing community has focused on deter- mining holders, facets, and valences, but has paid little attention to the reasons for senti- ment decisions. Our work considers human motives as the driver for human sentiments and addresses the problem of motive detec- tion as the first step. Following a study in psychology, we define six basic motives that cover a wide range of topics appearing in re- view texts, annotate 1,600 texts in restaurant and laptop domains with the motives, and re- port the performance of baseline methods on this new dataset. We also show that cross- domain transfer learning boosts detection per- formance, which indicates that these universal motives exist across different domains. 1 Introduction Understanding a person’s sentiment based on text has practical implications for improving prod- uct/service quality, along with scientific implica- tions for psychology and other fields. Despite a rich body of sentiment analysis research, a sen- timent is often simply assumed to be expressed by uni-dimensional binary or ternary labels (pos- itive, neutral, and negative), and relatively little attention has been paid to the reason for hold- ing a particular sentiment value. Aspect-based sentiment analysis (ABSA), which considers fine- grained categories (a.k.a. aspects) that may cause sentiment, partially tackles this problem. How- ever, aspects are typically limited to properties of entities such as the price of food and design of a product (e.g., (Pontiki et al., 2016)) and do not re- ally show why such aspects matter and how they cause human sentiments. For example, some peo- ple desire cheap and quick meals for saving time and money, and others desire high-grade food for enjoying the dining experience itself. Figure 1: Restaurant review texts and human motives of interest (rectangles). Following Li and Hovy (2017), we consider a sentiment as a realization of an individual’s men- tal state that relates to his/her satisfaction toward a specific event or entity. While a sentiment can be driven by a sentiment holder’s emotional, non- logical preference (like “I just don’t enjoy that kind of food”) and also conditioned by long-term plans and resources that the holder has, a senti- ment is largely triggered by whether one of the holder’s goals is satisfied or not. As Figure 1 illus- trates, one will have a negative sentiment toward a restaurant if the service is terrible because one’s basic motive for social behavior is not met. What and how many motives do we have? Decades of effort have been devoted to this ques- tion in research areas such as psychology, for ex- ample, (Maslow, 1943). A recent study by Tale- vich et al. …

Content truncated to 500 words