Acquiring Social Knowledge about Personality and Driving-related Behavior

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 2306–2315 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 2306 Acquiring Social Knowledge about Personality and Driving-related Behavior Ritsuko Iwai1,2, Daisuke Kawahara1, Takatsune Kumada1,2, Sadao Kurohashi1 1Kyoto University, 2RIKEN BSI-TOYOTA Collaboration Center 1Yoshida-honmachi, Sakyo-ku, Kyoto-shi, Kyoto, JAPAN, 2Hirosawa, Wako-shi, Saitama, JAPAN iwai.ritsuko.8n@kyoto-u.ac.jp {dk, t.kumada, kuro}@i.kyoto-u.ac.jp Abstract In this paper, we introduce our psychological approach for collecting human-specific social knowledge (particularly personality and driving-related behavior) from a text corpus, using natural language processing (NLP) techniques. Although this social knowledge is not usually explicitly described, it is often shared among people. We used the language resources that were developed based on psychological research methods: a Japanese personality dictionary (317 words) and a driving experience corpus (8,080 sentences) annotated with behavior and subjectivity. We then automatically extracted collocations of personality descriptors and driving-related behavior from a driving corpus (1,803,328 sentences after filtering) to obtain 5,334 unique collocations. Furthermore, we designed four step-by-step crowdsourcing tasks to evaluate the adequacy of the collocations as social knowledge. The crowdsourcing resulted in 266 pieces of social knowledge, which included knowledge that might be difficult to recall by crowdworkers but is easy with which to agree. Finally, we discussed the acquired social knowledge and its implementation into systems. Keywords: social knowledge, driving, psychology, personality 1. Introduction The properties of the human mind, such as personality and individual differences, have long been studied in psychology. Many psychologists have focused on understanding personality construct and human behavior. However, this interest has recently expanded to other fields. Understanding individual human differences in mind and behavior is one of the important topics in robotics, artificial intelligence (AI), and natural language processing (NLP). In the field of NLP, many studies have examined human subjectivities, such as emotions (Tokuhisa et al., 2009), personality (Golbeck et al., 2011; Nasukawa and Kamijo, 2017; Nasukawa et al., 2016; Plank and Hovy, 2015; Schwartz et al., 2013), opinion mining (Sun et al., 2017), and polarity (Higashiyama et al., 2008; Kobayashi et al., 2005). In this study, we introduce our psychology-integrative approach for acquiring human- specific knowledge, referred to as social knowledge, about personality and driving-related behavior. We acquire social knowledge from a large volume of user-generated content (UGC), especially those written by non- professional authors. UGC is posted by users and is available to the public (Moens et al., 2014). UGC, such as blog articles, includes a wide variety of descriptions of personal experiences and meanings (Higashiyama et al., 2008), although it is often deemed noisy and particularly unstructured, with the frequent omissions of subjects in Japanese because of its highly casual writing tone. Social knowledge is required for machines to interact or collaboratively work with humans to improve the quality of daily human lives. In human development, people acquire social knowledge through experiences and learning in social environments (Turiel, 1983). Social knowledge enables people to infer the intentions of other people and predict their behavior through natural communication and …

Content truncated to 500 words