Research on Personalized Recommendation Algorithm Integrating Cross-Grained Sentiment and Rating Interaction Features

To investigate the impact of cross-grained sentiments on user feature representation and address the issue of data sparsity, this paper proposes a Personalized Recommendation Algorithm Integrating Cross-Grained Sentiment and Rating Interaction Features (ICSR). The algorithm begins by employing a pre-trained BERT (Bidirectional Encoder Representations from Transformers) model and a Bi-GRU (Bidirectional Gated Recurrent Units) network to derive feature vectors from user and item reviews. Sentiment dictionaries and attention mechanisms are then applied to assign appropriate weights to the review features of users and items, respectively. To capture a richer set of sentiment features, a cross-grained sentiment feature fusion module is introduced. This module leverages an LDA (Latent Dirichlet Allocation) model and dependency syntactic analysis techniques to extract fine-grained sentiment features, while a word2vec pre-trained model and sentiment dictionaries are used to capture coarse-grained sentiment features. These features are then fused to form comprehensive cross-grained sentiment representations. Finally, rating interaction features are extracted using matrix factorization techniques, and all features are integrated and fed into a DeepFM model for rating prediction. Experimental results on Amazon datasets demonstrate that the proposed ICSR algorithm significantly outperforms baseline algorithms in terms of recommendation performance.