The purpose of this study is to tackle the challenge of information overload and address key limitations of collaborative filtering (CF) – specifically data sparsity and the cold start problem – that have been exacerbated by the rapid growth of digital content. To address this issue, the authors propose PTransE-CF, a hybrid recommendation framework that combines CF with knowledge graph (KG)-based semantic reasoning.
The model uses an enhanced path-based PTransE algorithm to extract multi-step semantic paths from KGs, thereby enriching item representations. It then combines these semantic embeddings with item-based CF to compute rating-based similarities. This fusion improves both the accuracy and the interpretability of recommendations.
Experiments on the MovieLens-1M data set show that the PTransE-CF model significantly outperforms traditional CF approaches, especially in cold-start and sparse-data scenarios. Moreover, the incorporation of KG semantics enhances interpretability by revealing semantic relational paths that explain the rationale behind item recommendations.
This study advances the field of knowledge management by refining path-aware representation learning and introducing a scalable recommendation framework that effectively integrates semantic knowledge with CF. The model improves both the effectiveness and the explainability of recommendations, offering practical value for enterprise e-commerce platforms as well as educational platforms.
