XLNet-Caps: Personality Classification from Textual Posts

• DOI: 10.3390/electronics10111360
• Publication Date: 2021-06-07T17:25:43Z
• Authors: Ying Wang; Jiazhuang Zheng; Qing Li; Chenglong Wang; Hanyun Zhang; Jibing Gong
• Keywords: machine learning; 13. Climate action; capsule; Big Five model; 0202 electrical engineering, electronic engineering, information engineering; deep learning; XLNet; 02 engineering and technology; NLP

Personality characteristics represent the behavioral characteristics of a class of people. Social networking sites have a multitude of users, and the text messages generated by them convey a person’s feelings, thoughts, and emotions at a particular time. These social texts indeed record the long-term psychological activities of users, which can be used for research on personality recognition. However, most of the existing deep learning models for multi-label text classification consider long-distance semantics or sequential semantics, but problems such as non-continuous semantics are rarely studied. This paper proposed a deep learning framework that combined XLNet and the capsule network for personality classification (XLNet-Caps) from textual posts. Our personality classification was based on the Big Five personality theory and used the text information generated by the same user at different times. First, we used the XLNet model to extract the emotional features from the text information at each time point, and then, the extracted features were passed through the capsule network to extract the personality features further. Experimental results showed that our model can effectively classify personality and achieve the lowest average prediction error.