Data augmentation method for sequence recommendation fusing explicit and implicit counterfactuals

Abstract

The application discloses a data enhancement method for sequence recommendation fusion of explicit and implicit counterfactuals, which is based on the original data set to pre-train a sampler model and an anchor model; then, the explicit and implicit counterfactual data samplers are used to intervene in the two kinds of feedback information of the user respectively, and two groups of enhanced training samples are generated; finally, the anchor model is re-optimized by using the two groups of enhanced training samples generated by the sampler, and the re-optimized anchor model will provide the final recommendation list for the user. The application simultaneously considers the explicit and implicit feedback information of the user to perform data enhancement on the sequence recommendation, can cover the unexplored input space by generating synthetic data, and thus helps to improve the recommendation performance and solve the common data sparsity problem in the sequence recommendation.

Images