ECO v1: Towards Event-Centric Opinion Mining

Findings of the Association for Computational Linguistics: ACL 2022, pages 2743 - 2753 May 22-27, 2022 c⃝2022 Association for Computational Linguistics ECO v1: Towards Event-Centric Opinion Mining Ruoxi Xu1,3, Hongyu Lin1∗, Meng Liao4∗, Xianpei Han1,2, Jin Xu4, Wei Tan4, Yingfei Sun3, Le Sun1,2 1Chinese Information Processing Laboratory 2State Key Laboratory of Computer Science Institute of Software, Chinese Academy of Sciences, Beijing, China 3School of Electronic, Electrical and Communication Engineering, University of Chinese Academy of Sciences, Beijing, China 4Data Quality Team, WeChat, Tencent Inc., China {ruoxi2021,hongyu,xianpei,sunle}@iscas.ac.cn {maricoliao,jinxxu,vinnietan}@tencent.com yfsun@ucas.ac.cn Abstract Events are considered as the fundamental build- ing blocks of the world. Mining event-centric opinions can benefit decision making, people communication, and social good. Unfortu- nately, there is little literature addressing event- centric opinion mining, although which signif- icantly diverges from the well-studied entity- centric opinion mining in connotation, struc- ture, and expression. In this paper, we propose and formulate the task of event-centric opin- ion mining based on event-argument structure and expression categorizing theory. We also benchmark this task by constructing a pioneer corpus and designing a two-step benchmark framework. Experiment results show that event- centric opinion mining is feasible and challeng- ing, and the proposed task, dataset, and base- lines are beneficial for future studies. 1 Introduction Events are the fundamental building blocks of the world (Russell, 1927; Ong, 1969). We express, share and propagate our opinions about events with personal understandings, emotions and attitudes in our daily life. People can better understand, communicate and interact with each other by min- ing, sharing and exchanging event-centric opinions. And being exposed to event-centric opinions from different angles can debias people’s own emotions and attitudes about social issues (Karamibekr and Ghorbani, 2013). Therefore, mining event-centric opinions have huge social and personal impacts. Unfortunately, there is little literature address- ing event-centric opinion mining, and most of cur- rent opinion mining studies focus on entity-centric opinions, which significantly diverge from the con- cerning event-centric ones. First, entity-centric opinions mostly focus on sentimental polarity of ∗Corresponding Authors Patent dispute is harmful to both side, Samsung believes. Apple is suggested by the lawyer to consider other effective tools. Today is a win for consumer choice. The appeal may result in huge impact on Samsung’s future. Figure 1: An illustration of event-centric opinions. Given an event, people can express their judgements, beliefs, attitudes and suggestions. The opinions ori- ented to an event may not directly target at itself, but can target at its related subevents or entities. the holder (Liu, 2012), meanwhile event-centric opinions care more about the content such as non- sentimental judgments, predictions or suggestions. Second, due to the rich interactions between events, entities, and people, event-centric opinions have a complicated structure. Given an event, people can express their opinions about the event itself, as well as its subevents, related events, and the involved entities. Third, the expressions of event- centric opinions are unique. The targets of event- centric opinions are frequently implicitly referred to, which often don’t appear in the opinion expres- sions. Moreover, …

Content truncated to 500 words