LED: A Dataset for Life Event Extraction from Dialogs

Findings of the Association for Computational Linguistics: EACL 2023, pages 384–398 May 2-6, 2023 ©2023 Association for Computational Linguistics LED: A Dataset for Life Event Extraction from Dialogs Yi-Pei Chen1, An-Zi Yen2, Hen-Hsen Huang3, Hideki Nakayama1, Hsin-Hsi Chen4 1 The University of Tokyo, Japan 2 National Yang Ming Chiao Tung University, Taiwan 3 Academia Sinica, Taiwan 4 National Taiwan University, Taiwan 1ypc@g.ecc.u-tokyo.ac.jp, 2azyen@nycu.edu.tw, 3hhhuang@iis.sinica.edu.tw 1nakayama@ci.i.u-tokyo.ac.jp, 4hhchen@ntu.edu.tw Abstract Lifelogging has gained more attention due to its wide applications, such as personalized rec- ommendations or memory assistance. The is- sues of collecting and extracting personal life events have emerged. People often share their life experiences with others through conver- sations. However, extracting life events from conversations is rarely explored. In this pa- per, we present Life Event Dialog, a dataset containing fine-grained life event annotations on conversational data. In addition, we initi- ate a novel conversational life event extraction task and differentiate the task from the pub- lic event extraction or the life event extraction from other sources like microblogs. We explore three information extraction (IE) frameworks to address the conversational life event extraction task: OpenIE, relation extraction, and event ex- traction. A comprehensive empirical analysis of the three baselines is established. The results suggest that the current event extraction model still struggles with extracting life events from human daily conversations. Our proposed life event dialog dataset and in-depth analysis of IE frameworks will facilitate future research on life event extraction from conversations. 1 Introduction Daily conversation, as a means of communication and switching information, is full of personal infor- mation, including personal background, interests and hobbies, connections to other people, and var- ious life events. Mining life events lets us better understand a person. The extracted life events can be used to construct the personal knowledge base and benefit a variety of downstream tasks, such as lifestyle understanding (Doherty et al., 2011) and memory assistance (Rahman et al., 2018). Previous research on life event extraction mainly focuses on life events from microblogs or social media platforms such as Twitter (Li et al., 2014; Yen et al., 2018, 2019). However, these events from a given fixed passage are static. In contrast, an event mentioned in a conversation might change its status dynamically throughout the chat. Besides, conversations allow participants to interact with each other and gather the information which stimu- lates participants’ interests, revealing people’s gen- eral interests in different aspects of information about a life event and expanding additional event information. For example, when a person talks about a travel event only with the destination men- tioned, the other interlocutor might ask additional information about who they are traveling with, how much the trip cost, and the period and timing of the travel. Nevertheless, life event extraction from conversations is rarely explored and existing works only detect course or ambiguous event types (Eisen- berg and Sheriff, 2020; Kao et al., 2021). The par- ticipants and status of events are not recognized, preventing more fine-grained life events analysis and limiting the applications. …

Content truncated to 500 words