Detecting Speaker Personas from Conversational Texts

Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 1126–1136 November 7–11, 2021. c⃝2021 Association for Computational Linguistics 1126 Detecting Speaker Personas from Conversational Texts Jia-Chen Gu1, Zhen-Hua Ling1∗, Yu Wu2, Quan Liu1,3, Zhigang Chen3, Xiaodan Zhu4 1National Engineering Laboratory for Speech and Language Information Processing, University of Science and Technology of China, Hefei, China 2Microsoft Research Asia, Beijing, China 3State Key Laboratory of Cognitive Intelligence, iFLYTEK Research, Hefei, China 4ECE & Ingenuity Labs, Queen’s University, Kingston, Canada gujc@mail.ustc.edu.cn, {zhling, quanliu}@ustc.edu.cn, Wu.Yu@microsoft.com, zgchen@iflytek.com, xiaodan.zhu@queensu.ca Abstract Personas are useful for dialogue response prediction. However, the personas used in current studies are pre-defined and hard to obtain before a conversation. To tackle this issue, we study a new task, named Speaker Persona Detection (SPD), which aims to detect speaker personas based on the plain conversational text. In this task, a best-matched persona is searched out from candidates given the conversational text. This is a many-to-many semantic matching task because both contexts and personas in SPD are composed of multiple sentences. The long-term dependency and the dynamic redundancy among these sentences increase the difficulty of this task. We build a dataset for SPD, dubbed as Persona Match on Persona-Chat (PMPC). Furthermore, we evaluate several baseline models and propose utterance-to-profile (U2P) matching networks for this task. The U2P models operate at a fine granularity which treat both contexts and personas as sets of multiple sequences. Then, each sequence pair is scored and an interpretable overall score is obtained for a context-persona pair through aggregation. Evaluation results show that the U2P models outperform their baseline counterparts significantly. 1 Introduction Recently, human-machine conversation has achieved great success (Lowe et al., 2015; Serban et al., 2016; Wu et al., 2017; Madotto et al., 2018; Tao et al., 2019; Zhang et al., 2020; Roller et al., 2021), and has been applied to various scenarios, such as customer service and conversational recommendation engine. It is well-known that a user’s persona can help the machine to generate more appropriate responses. Many studies investigate how to predict a response if a persona ∗Corresponding author. Hello, how are you tonight? Are you watching the office? U1: Hi! I am well, thanks. I actually just got back from a hike. That sounds good, I live in the city and do not have hiking trails. U2: I am in ohio where there are lots of wooded trails. Ohio is lovely, we went hiked blue hen and buttermilk falls once. U3: Yes, very nice. My 2 sons love to hike buttermilk falls. That hill though, lol. Do you like sweets? I do not. U4: I try to not eat many sweets to set a good example for my boys. I do not like them that much. Do you have any hobbies? U5: I am single mom so other than hiking, running around boys. Persona of speaker candidate #1 Profile 1: I am from the north. Profile 2: I am raising sons all on my own. Profile 3: I enjoy nature walks. …

Content truncated to 500 words