Recommendation method and recommendation system based on double-view correction

The invention discloses a recommendation method based on double-view correction, which comprises the following steps of: 1, quantitatively analyzing whether article popularity deviation and user activeness deviation exist in data or not, and constructing a causal graph conforming to a data generation mechanism; step 2, based on the causal graph constructed in the step 1, using a do operator in a training stage, adopting P (Ydo (U, I)) to replace traditional P (YU, I), eliminating the influence of popularity deviation of article characterization and user characterization to represent the real preference of a user to an article, and then using a backdoor adjustment technology to estimate a P (Ydo (U, I)) causal estimator through data; 3, parameterizing the causal estimator in the step 2 to enable the causal estimator to be recovered by existing data and to be trained, and improving a Bayesian personalized ranking (BPR) loss function to enable a recommendation model to be debiased by using two aspects of informa