Conversation Partner Grouping Based on Speech Contents

摘要

Conversation analysis plays an important role in social psychology, interpersonal relationship management, and human-care computing. However, few of existing studies considers speech contents for effective conversation partner grouping (abbr. CPG). In this paper, we propose a new framework for conversation partner grouping based on speech contents. Under the proposed framework, we propose two novel algorithms for effective CPG, called CPG-LDA and CPG-LSI, respectively. Both of them use voice recognition tools to convert audio-based speech data into text-based speech contents, and then apply topic modeling and k-means algorithms for CPG. However, the former is based on LDA topic modeling, while the latter is LSI. The experiments show that both CPG-LDA and CPG-LSI have good performance for GPC. More impressively, the proposed CPG-LSI algorithm archives up to 95.83% recognition rate in the experiments.