CRDB: Commodity Recommendation Model Combining DGNN and BiGRU

As potent filters of information, recommendation systems not only streamline the process for users to identify desirable products but also uncover latent interests in potential offerings. Although the current hybrid graph neural network and the recommendation algorithm of review text have achieved good results, the influence of timing information is not fully considered, and the problem of users' long and short-term interest deviation is ignored. At the same time, the contextual emotion features of long review text are not sufficiently extracted. To solve these problems, this paper proposes a product recommendation algorithm based on Dynamic Graph Neural Network(DGNN) and Bidirectional Gated Recurrent Unit(BiGRU). The algorithm uses a DGNN to extract the long and short-term correlation features between users and products. Then BERT-BiGRU is used to mine the sentiment features in the review text fully, and an attention mechanism is introduced to increase the weight of words that contribute a lot to semantic expression, to improve the accuracy of feature extraction. Finally, the user's long-term and short-term interest features and product features extracted by DGNN are combined individually with the user's sentiment and product features mined by BERT-BiGRU for matching user features and product features, to recommend products with a high feature matching degree to users. Experiments on Amazon datasets show that the CRDB model proposed in this paper reduces the mean root mean square error of LGDL prediction by 3.20% compared with the baseline model and 2.57% compared with the newer CRRG model.