| 其他摘要 | The Need for Affect (NFA) refers to the motivation to approach or avoid emotion-inducing situations and encompasses three dimensions: NFA total, NFA approach, and NFA avoidance. As a relatively stable intrinsic aspect of human nature, NFA plays a crucial role in mental health, information processing, decision-making, social attitudes, and various other social psychological processes and behaviors. However, traditional data collection and research methods, such as self-reporting, are not suitable for large-scale assessment scenarios. Consequently, finding an efficient and automatic method for identifying NFA on a large scale is both theoretically and practically significant. This study explores the relationship between users' social network behavior, particularly on Weibo, and their NFA, aiming to automate NFA identification based on social network data. The study comprises three aspects:
Study 1 examines the relationship between users' NFA and Weibo behaviors, investigating the correlation between specific Weibo behaviors and users' NFA. The results indicate that users with a high level of NFA approach are more focused on themselves, use richer language expression patterns, like to express their true feelings and emotions, are more likely to seek sensory experiences and stimulation, and are more willing to interact with others. Users with a high level of NFA avoidance are more concerned about others, like to interact with familiar people, and use more death-related vocabulary in their Weibo posts. Users with high NFA total levels tend to express their emotions and interact with others, and use death-related vocabulary less frequently. The results reveal a close relationship between Weibo behavior and NFA, demonstrating the feasibility of NFA identification based on user behavior on Weibo.
Study 2 predicts users' NFA scores based on their Weibo behaviors, using regression algorithms to model and predict these scores from selected Weibo features. The results show that Extreme Gradient Boosting (XGB) performs best among the eight machine learning algorithms used. The Pearson correlation coefficients between predicted scores and NFA questionnaire scores achieved 0.25 (NFA avoidance), 0.31 (NFA approach) and 0.34 (NFA total), and the split-half reliabilities were 0.66 (NFA total), 0.68 (NFA approach) and 0.70 (NFA avoidance).
Study 3 classifies users' NFA levels based on Weibo behavior, using selected features and classification algorithms to model and identify users' NFA levels. The results indicate that Random Forest (RF) has the best classification effect on NFA total and NFA approach. The precision rates of high and low group classification of NFA total are 0.68 and 0.64 respectively, while the precision rates of high and low group classification of NFA approach are 0.65 and 0.64 respectively. Logistic Regression (LG) performs best on NFA avoidance, with precision rates of high and low group classification being 0.63 and 0.62 respectively.
Overall, this study demonstrates a correlation between users' Weibo behavior and their NFA, suggesting the feasibility of identifying users' NFA through Weibo behavior. The proposed non-invasive method for NFA identification can be applied to mental health monitoring and other large-scale NFA measurement scenarios, complementing traditional scale measurement methods. |
修改评论