277. Characteristics of human-computer interaction behavior on social media: a longitudinal study on the correlation between usage duration, content interaction, language style, and loneliness

Abstract Background Social media has become an important scene for modern people's daily interaction, and the correlation between its human-computer interaction behavior characteristics and users' psychological health, especially loneliness, has received widespread attention. Previous cross-sectional studies have shown that passive browsing and social comparison may exacerbate feelings of loneliness, while active and supportive interaction may alleviate loneliness. However, there is still a lack of long-term tracking evidence on how these behavioral characteristics change over time and are dynamically associated with feelings of loneliness. The study adopts a longitudinal design to systematically examine the temporal relationship between the duration of social media use, content interaction behavior, language style, and changes in loneliness, providing longitudinal evidence for a deeper understanding of the psychological impact of social media use. Methods The study adopted a one-year longitudinal tracking design with three waves, including T1, T2, T3 (with a 2-month interval between each group), and four groups with a 6-month interval. A total of 320 active social media users aged 18-35 were recruited through online platforms, and 298 valid samples were ultimately included, with an average age of 26.4 years and 58% being females. Obtain objective behavioral data through backend authorization, including daily average usage duration, frequency of interaction types, and language style extraction of post texts voluntarily provided by users through language analysis tools (Linguistic Inquiry and Word Count, LIWC). Loneliness was measured using the UCLA Loneliness Scale. Statistical analysis was conducted using SPSS 26.0 and Mplus 8.3, including correlation analysis, cross lagged model testing of temporal predictive relationships between variables, and control for covariates such as gender, age, and offline social support. Results Descriptive statistics show that the average daily social media usage time of users is 2.8 hours (SD = 1.4), with likes being the most frequent interactive behavior. Cross lag model analysis revealed. (1) The difference in usage duration between T1 and T2 significantly positively predicts loneliness (β = 0.18, p<.01), but T2 loneliness cannot predict changes in T3 usage duration. (2) Those with higher frequency of content interaction T1 comments and private message interactions have significantly lower T2 loneliness (β = -0.15, p<.05), and lower T2 loneliness further predicts higher T3 active interaction behavior (β = 0.12, p<.05), showing a bidirectional protective association. (3) The higher frequency of first person singular words in language style T1 post texts significantly positively predicts T2 loneliness (β = 0.22, p<.01), while the proportion of positive emotion words negatively predicts (β = -0.14, p<.05). The proportion of negative emotion words is correlated with loneliness in the cross section, but no longitudinal predictive effect was shown. All models controlled for covariates and had good fit indicators. Discussion Research has shown that there is a dynamic and differentiated correlation between social media human-computer interaction behavior characteristics and feelings of loneliness. Long term passive use and high self attention expression styles may be risk factors for loneliness, while active, interpersonal oriented interactions may buffer loneliness and form a positive cycle. This suggests that simply reducing usage time may not be the best intervention direction, but should guide users to shift from passive browsing to meaningful active interaction, and pay attention to their self attention tendencies in online expression. Future research can combine experimental interventions to train users to adjust their interaction patterns and language styles. Funding No. 2023ZWY005.