Traditional student big data analysis often neglects unstructured data, such as communication content and emotional feedback, limiting its effectiveness in personalized recommendations and learning interventions. This study addresses this gap by applying Word2Vec technology to analyze semantic information and emotional tendencies in student behaviors, enabling precise learning recommendations, real-time sentiment analysis, and timely interventions. A Word2Vec model is trained on extensive student data to understand learning behaviors, while a support vector machine (SVM) performs sentiment analysis to identify emotional states. Based on these insights, a personalized recommendation system dynamically adjusts resources and task difficulty to enhance learning outcomes. Experimental results show the system outperforms others, achieving recommendation accuracy of 0.81–0.87, sentiment analysis accuracy of 0.80–0.89, and an average performance improvement rate of 15.85%. These findings validate Word2Vec’s effectiveness in intelligent education systems, offering a novel framework for personalized learning and intervention strategies.