Improving user profile with personality traits predicted from social media content

Rui Gao, Bibo Hao, Shuotian Bai, Lin Li, Ang Li, Tingshao Zhu2013 · DOI

TL;DR

This paper tackles the challenge of personalizing recommendation systems by automatically inferring users' personality traits from their social media content in Chinese. They developed a machine learning model that predicts Big Five personality traits from Sina micro blog posts, demonstrating that personality can indeed be inferred from text, paving the way for richer user profiles.

What Problem Does This Paper Address?

Recommender systems are crucial for personalized online experiences, and their effectiveness often hinges on accurate user profiles. While user preferences are a key component, incorporating personality traits could further enhance personalization. However, traditional methods of measuring personality, like psychological questionnaires, are impractical for large-scale online applications. This paper addresses the gap of automatically and efficiently obtaining personality traits from readily available social media data, specifically in the Chinese language environment, which is a rapidly growing market.

Methods

The authors propose a novel approach to automatically identify personality traits from social media content. They extracted various text features from Sina micro blog posts, including:

  • Status statistics: Number of posts, words, etc.
  • Sentence-based features: Average sentence length, declarative sentence proportion.
  • Word-based features: Counts of emotion and greeting words.
  • Character-based features: Punctuation and special character usage.
  • Industry lexicon features: Utilizing custom lexicons for specific domains (e.g., art, medical, popular network terms) derived from Sogou Cell Lexicons.
  • LIWC (Linguistic Inquiry and Word Count) features: Employing a standard tool for text analysis, adapted for Chinese.

These features were then used to train machine learning models, specifically Gaussian process, M5'Rules, and Pace Regression, to predict the Big Five personality traits (Openness to Experience, Conscientiousness, Extraversion, Agreeableness, and Neuroticism). The models were evaluated using 10-fold cross-validation and Pearson correlation.

Datasets

The primary dataset consists of Sina micro blog status texts from 1766 users who voluntarily completed the 44-item Big Five Inventory. This dataset was curated from a larger pool of Sina micro blog users, with specific filtering criteria applied to exclude inactive users and entertainment/advertisement accounts to ensure data quality and relevance.

Key Insights and Ideas

  • Personality is Predictable from Text: The core insight is that users' personality traits are significantly reflected in their social media text, making automated prediction feasible.
  • Chinese Social Media as a Data Source: The paper highlights the potential of Chinese social media platforms like Sina micro blog for personality inference, especially given the growing e-commerce market in China.
  • Feature Engineering Matters: The diverse set of text features, including custom industry lexicons and LIWC, proved effective in capturing personality-related linguistic cues.
  • Big Five Model Applicability: The study successfully applied the widely recognized Big Five personality model to social media text analysis.

Key Results

The experimental results demonstrate that personality traits can be predicted from Sina micro blog text with acceptable Pearson correlation. Specifically:

  • Conscientiousness, Extroversion, and Openness to Experience showed stronger correlations with self-reported traits compared to Agreeableness and Neuroticism.
  • The Pearson correlation between predicted and self-reported personality traits was around 0.4, indicating a fair correlation.
  • The paper provides a table (Table 2) detailing the evaluation results for each of the Big Five dimensions, with correlation coefficients ranging from 0.31 to 0.40. For example, the correlation for Conscientiousness (C) was 0.40601, and for Extroversion (E) was 0.40312.

Limitations and Future Work

The authors acknowledge several limitations and suggest future directions:

  • LIWC for Chinese: They found that the standard LIWC2007 might not process Chinese content optimally and are developing a prototype system called "TextMind" for better Simplified Chinese analysis.
  • Improving Correlation: While the correlations are acceptable, further work is needed to improve the accuracy of personality prediction and its correlation with self-reported traits.
  • Future Applications: The predicted personality traits are intended to be integrated into existing personalized recommendation systems to enhance their performance, particularly in addressing the "cold start" problem.
  • Further Research: The paper points to an open space for future research in interpreting less obvious correlations between specific text features and personality dimensions, such as the positive relationship between exclamation mark usage and conscientiousness, or the negative relationship between number usage and agreeableness.

Summary updated

Abstract

Existing studies indicate that there exists strong correlation between personality and personal preference, thus personality could potentially be used to build more personalized recommender system. Personality traits are mainly measured by psychological questionnaires, and it is hard to obtain personality traits of large amount of users in real-world scenes.In this paper, we propose a new approach to automatically identify personality traits with Social Media contents in Chinese language environments. Social Media content features were extracted from 1766 Sina micro blog users, and the predicting model is trained with machine learning algorithms.The experimental results demonstrate that users' personality traits could be predicted from Social Media contents with acceptable Pearson Correlation, which makes it possible to develop user profiles for recommender system. In future, user profiles with predicted personality traits would be used to enhance the performance of existing personalized recommendation systems.
BibTeX
@misc{gao2013improving,
  author = {Rui Gao and Bibo Hao and Shuotian Bai and Lin Li and Ang Li and Tingshao Zhu},
  title = {{Improving user profile with personality traits predicted from social media content}},
  year = {2013},
  doi = {10.1145/2507157.2507219},
}
Figure