Research on performance improvement of personalized recommendation algorithm based on deep neural network optimization

The rapid development of the Internet has led to a massive surge of data, and information overload has become a major problem in modern development. In this paper, a new personalized recommendation model is proposed based on a matrix decomposition algorithm, which integrates auxiliary information into the original feature matrix and uses a deep neural network to extract its features. With the help of the Doc2Vec model, syntactic statements are extracted and the similarity between text vectors is calculated to improve the accuracy of recommendations. Personalized recommendation experiments are carried out to verify the recommendation performance of the proposed matrix decomposition personalized recommendation model in this paper, and the settings of the model parameters such as the number of MLP layers, the number of pre-training and convolution of Adam’s algorithm, and the output dimensions all have an impact on the personalized recommendation performance of the model in this paper. Comparing the BPR, eALS, MLP, and NeuMF models, the recommendation accuracy of this paper’s model is the highest on the MovieLens dataset, and the NDCG value is slightly lower than that of NeuMF only when the predictor is 8, while the NDCG value is optimal in all other cases. On the Pinterest dataset, the recommendation accuracy of this paper’s model is improved compared to other models, and the NDCG value is always the highest, reaching up to 0.559.