Personalized Recommendation Algorithm with Review Sentiments and Importance
[Objective]To address the data sparsity issue and explore the impacts of emotional expression on user feature learning,this paper proposes a personalized recommendation algorithm based on sentiment and the importance of online reviews.[Methods]First,we used the BERT pre-trained language model to generate the vector representation of review texts.Then,we fed them into a Bi-GRU network to learn their semantic features.We also assigned weights to the review vector using sentiment weights and attention mechanisms.Finally,we utilized the DeepFM algorithm for deep interaction between user and product features to predict the user's rating of the products.[Results]We examined the proposed model with the Amazon product data dataset.Our model reduced the RMSE and MAE metrics by up to 24.43％and 31.44％compared to the baseline models.Compared with the attention mechanism,our method reduced the RMSE and MAE metrics by up to 2.59％and 3.89％.[Limitations]The sentiment analysis method cannot represent the users'emotional tendencies towards the different attributes of the product.[Conclusions]The proposed method considers the influence of user sentiment on user feature expression,improving the recommendation accuracy.
万方数据
维普
