Interlocutor Personality Perception in a Dyadic Conversation based on BFI-Profiles and Coupled Hidden Markov Models

• Main Authors: Yu-TingZheng, 鄭宇廷
• Other Authors: Chung-Hsien Wu
• Format: Others
• Language: en_US
• Published: 2014
• Online Access: http://ndltd.ncl.edu.tw/handle/50586941220512319915

碩士 === 國立成功大學 === 資訊工程學系 === 102 === In recent years, with the development of hardware and software technologies, intelligent devices offer a variety of convenient services in our daily life. Users can interact with those intelligent devices through a series of simple commands and feel that they are interacting with a real person. For intelligent devices can provide more personalized services for users, emotional intelligence computing is becoming an important issue. The responses in most of intelligent devices tend to be simple and monotonic, which makes users feel bored easily. Those intelligent devices could automatically distinguish between different users based on a brief interaction, then those intelligent devices can give a more appropriate response to the user according to the user’s personality traits. Therefore, how to identify the user's personality has become an important research topic. Recent research on personality trait detection are generally based on voice and text. Acoustic features and textual features are employed to explore the correlations between different personality traits. Although those studies have obtained significant achievements, few studies analyze mutual influence of two human personality traits in an interactive process. In this thesis, an Automatic Personality Perception method is proposed. First, we establish the single speaker turn personality perception model by using the Recurrent Neural Networks to train the relationship between linguistic features and the big-five personality traits in each speaker's turn. Second, we establish the multiple-speaker turn personality perception model by using the Coupled Hidden Markov Model to observe two speaker’s personality across many speaker’s turns in each dialogue process. In order to evaluate the proposed method, an automatic personality perception system was constructed and the overall accuracy achieved 71.9%. Compared to traditional HMM-based and SVM-based methods, the proposed approach can obtain the highest performance. The promising results confirm the usability of this system for future applications.