A Personalized Recommendation Framework for Online Legal Education Integrating Context Awareness and Mobile Interaction

Abstract

Significant limitations in existing personalized recommendation systems for online legal education include the neglect of contextual heterogeneity, lack of affective association, the fragmentation of interaction feedback loops, and insufficient domain-specific sentiment-labeled data. To address these issues, a collaborative recommendation framework integrating transfer learning, a sentiment-enhanced knowledge graph (SE-KG), and a knowledge graph convolutional network (KGCN) was proposed. Methodologically, a richly annotated generaldomain film review corpus was employed as the source domain, through which transfer learning was applied to optimize a long short-term memory (LSTM) network, quantifying and accurately annotating sentiment tendencies in textual evaluations of legal education resources. A core knowledge graph (KG) encompassing “knowledge points – legal provisions – cases – courseware” was then constructed, into which the quantified sentiment outputs were embedded to generate the SE-KG, with additional relational triplets such as “sentiment similarity” and “user–preference sentiment” incorporated. Finally, context-aware data and mobile interaction sequences were fused, and the KGCN’s neighbor aggregation mechanism was refined with an attention strategy to enable dynamic user preference prediction. Experiments show that the framework mitigates the shortage of sentiment-labeled legal data and significantly improves recommendation accuracy, recall, and users’ mastery of knowledge points compared with traditional collaborative filtering and baseline KGCN models. This work offers a new approach for personalized resource delivery in online legal education through cross-domain sentiment transfer and KG enhancement, significantly improving contextual adaptability and semantic relevance in recommendation systems.