A HYBRID RECOMMENDATION ALGORITHM BASED ON USER CHARACTERISTICS

Shop recommendation system is an important part of the e-commerce recommendation system. Shop recommendation system in this paper mainly consist of three steps. First, constructed matrix decomposition module and deep network module for tackling scores data and comment data, and then connected two modules by weight factor, trained by the same loss function, at last the comprehensive score is output by scoring prediction, analyzing fusion factor for the effect of algorithm through the preprocessed text and the parameter setup fusion model. Each experiment adopts the five-fold crossover verification method, and the prediction accuracy of this algorithm compared with other five different algorithms. Experimental results verify that the UFFSR algorithm can effectively improve the accuracy of prediction scoring and alleviate the data sparsity and cold start problems to a certain extent.