DECODING DIGITAL BEHAVIOR: AI-POWERED ANALYSIS AND PREDICTION OF SOCIAL MEDIA USER ENGAGEMENT PATTERNS

Abstract

Social networks in this period of hyper-connectivity provide abundant behavioral data that can be utilized for computing user intentions, sentiment and engagement. The objective of this study is to explore and predict social media user behavior through utilizing state-of the art NLP algorithms and extractive LLM for mining semantic patterns in UGC. Using a quantitative research design of >1 million deidentified public posts and comments on X (formerly Twitter) and Reddit. Textual data were preprocessed, vectorized and passed as input to transformer-based embeddings in order to analyze sentiment, topic distribution and behavioral trends over time. Finally, we developed predictive models (Random Forest, LSTM) to estimate the future engagement using linguistic, temporal and contextual features. We demonstrate that emotional tone and timing of post, as well as conversational dynamics, all have a statistically significant impact on engagement (through likes, shares and reply volume). For instance, the combination with LLM demonstrated a 14% increase in predictive accuracy against traditional NLP models. Real time audience segmentation, campaign optimization and forecasting behavioral changesOpen image in new windowDigital marketers, public policy makers and sociologists can extract cryptocurrency derived practical implications from adopting this framework. While previous research has focused on either sentiment or network influence, our work presents an end-to-end pipeline for scalable, interpretable, and holistic measurements that consider both language modeling and statistical forecasting. This contributes with a new understanding of how state of the art AI can decode big digital footprints, and offers novel ways for ethical and effective social media analytics.