多人對話中情緒與人際關係之分析

Abstract

在對話系統中，偵測使用者的情緒並做出適當的回覆，能夠提升使用者的滿意度。除了觀察使用者的情緒外，我們觀察到人際關係也是影響對話內容的一個重要因素。由於缺乏擁有人際關係標記的資料庫，為了做相關的實驗，我們建立一個同時擁有說話人情緒以及與聽眾之間人際關係標記的中文對話語料庫，該語料庫擁有4,142 組對話，其中包含25,548句台詞。透過該語料庫，我們進行情緒及人際關係的分類實驗，並從中觀察情緒與人際關係之間的關聯性。

In the dialogue system, detecting user emotions and generating appropriate responses can improve user satisfaction. In addition to detecting the user's emotions, we observe that interpersonal relationships are also an important factor influencing the content of the conversation. Due to the lack of a corpus with interpersonal relationship labels, in order to do the related experiments, we establish a Chinese dialogue corpus with both emotions and interpersonal relationship labels between the speaker and the targeted listener. The corpus contains 25,548 utterances from 4,142 dialogues. Through this corpus, we conduct classification experiments of emotions and interpersonal relationships, and observe the correlation between emotions and interpersonal relationships.