Improving Human Needs Categorization of Events with Semantic Classification

Proceedings of the Eighth Joint Conference on Lexical and Computational Semantics (*SEM), pages 198–204 Minneapolis, June 6–7, 2019. c⃝2019 Association for Computational Linguistics 198 Improving Human Needs Categorization of Events with Semantic Classification Haibo Ding∗, Ellen Riloff†, Zhe Feng∗ ∗Bosch Research and Technology Center, Sunnyvale, CA 94085, USA †School of Computing, University of Utah, Salt Lake City, UT 84112, USA {haibo.ding,zhe.feng2}@us.bosch.com, riloff@cs.utah.edu Abstract Human Needs categories have been used to characterize the reason why an affective event is positive or negative. For example, “I got the flu” and “I got fired” are both negative (unde- sirable) events, but getting the flu is a Health problem while getting fired is a Financial problem. Previous work created learning mod- els to assign events to Human Needs categories based on their words and contexts. In this paper, we introduce an intermediate step that assigns words to relevant semantic concepts. We create lightly supervised models that learn to label words with respect to 10 semantic concepts associated with Human Needs cat- egories, and incorporate these labels as fea- tures for event categorization. Our results show that recognizing relevant semantic con- cepts improves both the recall and precision of Human Needs categorization for events. 1 Introduction Affective events have a positive or negative impact on the people who experience the event. For ex- ample, being hired for a job is typically a bene- ficial (positive) event, but being fired is usually a detrimental (negative) event. Recognizing affec- tive events is critical to understand people’s mo- tivations, goals, desires, and empathy in narrative stories and conversations. Previous research has proposed several methods to recognize affective events and their polarity (e.g., (Deng et al., 2013; Vu et al., 2014; Reed et al., 2017; Ding and Riloff, 2016)). To achieve a deeper level of understand- ing, Ding and Riloff (2018a) further classified af- fective events into categories associated with the- ories of Human Needs (Maslow et al., 1970; Max- Neef et al., 1991) in psychology: Physiological, Health, Leisure, Social, Finance, and Cognition, to characterize the reason for the event’s affective polarity. For example, breaking your arm is a neg- ative event because it violates a need to maintain one’s Health, but fighting with your spouse is neg- ative because it violates a need for good Social re- lations with friends and family. Human Needs categories naturally align with several broad conceptual classes, and we hypoth- esized that learning to recognize relevant seman- tic concepts would lead to more effective Human Needs categorization. For example, the Physiolog- ical need corresponds to basic functions such as breathing, sleeping, eating, and drinking. Learn- ing to recognize FOOD/DRINK concepts should help identify events that belong to this category. Broadly, semantic concepts should help in two ways. First, semantic features are more gen- eral than words, which can suffer from sparsity. Second, given semantic features, a classifier can directly learn interactions between them, which should be more robust than interactions between individual words. In this paper, we present lightly supervised clas- sifiers that …

Content truncated to 500 words