A sentiment-aware deep learning approach for personality detection from text

Zhancheng Ren,Qiang Shen,Xiaolei Diao,Hao Xu

DOI: https://doi.org/10.1016/j.ipm.2021.102532

2021-05-01

Abstract:<p>Personality detection based on user-generated text content analysis has a significant impact on information science, for instance, information seeking. Existing deep learning-based approaches, however, have two major limitations. Firstly, they extract only keywords for personality detection and lack the analysis of sentiment information and psycholinguistic features. Secondly, the information about the context and polysemous words are ignored. To tackle these problems, we propose a novel multi-label personality detection model based on neural networks, which combines emotional and semantic features. Specifically, we leverage Bidirectional Encoder Representation from Transformers (BERT) to generate sentence-level embedding for text semantic extraction. In addition, a sentiment dictionary is used for text sentiment analysis in order to consider sentiment information. Finally, we input the above semantic information and emotional information into the neural network to construct an automatic personality detection model. The performance of the model has been evaluated on two public personality datasets. The experiments show that we obtain average accuracy improvements of 6.91% and 6.04% on the Myers-Briggs Type Indicator (MBTI) and Big Five datasets, respectively, compared with the state-of-the-art techniques.</p>

computer science, information systems,information science & library science

What problem does this paper attempt to address?

The paper aims to address the problem of personality detection based on text analysis, particularly automated multi-label personality detection on social media text data. Existing deep learning methods have two main limitations in personality detection:

1. **Lack of emotional information and psycholinguistic features**: Most existing methods only extract keywords for personality detection and fail to consider emotional information and psycholinguistic features.

2. **Neglect of contextual information and polysemy**: Existing methods often ignore the contextual information in the text and the phenomenon of polysemy.

To overcome these limitations, the authors propose a new neural network-based multi-label personality detection model that combines emotional and semantic features. Specifically, the model uses Bidirectional Encoder Representations from Transformers (BERT) to generate sentence-level embeddings to capture the semantics of the text; at the same time, it performs sentiment analysis on the text through an emotional dictionary to consider emotional information. Finally, the above semantic information and emotional information are input into the neural network to construct an automated personality detection model.

The main contributions of the paper include:

1. Proposing a novel multi-label personality detection model that combines a pre-trained BERT model and neural networks, which can better understand sentence semantics and handle information from social media text data.

2. Proposing a method that combines semantic and emotional features, adding interpretability to personality detection and aiding in personality trait analysis.

3. Experimental results on the Myers-Briggs Type Indicator (MBTI) and Big Five personality traits datasets show that the model outperforms existing techniques in personality detection.

In summary, this study aims to improve the accuracy of personality detection from social media text by integrating emotional and semantic features and utilizing advanced models such as BERT.