Research on Recommendation Algorithm Based on TrustSVD++ and Deep Matrix Decomposition

In the recommendation algorithm, due to the incompleteness of the interaction between users and items, there are often a large number of missing data in the interaction matrix, which affects the quality of recommendation. To address this challenge, researchers have introduced matrix factorization techniques combined with machine learning algorithms to fill in missing data. However, the bilinear mapping technology adopted by the traditional matrix factorization method cannot fully capture the nonlinear relationship between users and items. To this end, this study proposes a recommendation algorithm combining TrustSVD++ and deep matrix factorization. Firstly, the explicit and implicit trust information between users is integrated into the SVD++ model to fill the interaction matrix more accurately. Then, in order to preserve the linear relationship and prevent network degradation, the residual network was used as the core module, and the parameter adjustment was optimized by adding channels directly connecting the input and output. Experimental results on public datasets show that compared with DMF1, DMF2, MC+DMF and BDMF models, the proposed TDMF model has the hit rate (HR@10) and normalized discounted cumulative gain (NDCG@10) increased by 2.1%and 3.7%respectively, which proves the effectiveness and practicability of the proposed model.