Recommendation system recommendation method and device, recommendation system and storage medium

Abstract

The invention discloses a recommendation system recommendation method and device, a recommendation system and a storage medium. The method comprises the steps of constructing a DeepFM recommendation system network; calculating a surprise characteristic factor of each characteristic domain of the DeepFM recommendation system network according to the characteristic domain data of the DeepFM recommendation system network updated in the last training and the current characteristic domain data, the surprise characteristic factor reflecting user behavior changes; when the surprise characteristic factor is smaller than a preset replacement threshold, carrying out random replacement on the embed vector of the characteristic domain corresponding to the surprise characteristic factor; and the DeepFMrecommendation system network calculates a prediction probability by adopting the replaced embed vector, and obtains a user recommendation result by reasoning. The recommendation system comprises a recommendation system network construction module, a surprise characteristic factor generation module, an embed vector replacement module and a recommendation result generation module. According to themethod, the probability of discovering surprises is improved on the basis that the memory ability and the generalization ability of a recommendation system are reserved.

Description

technical field [0001] The invention relates to the field of recommendation system design, in particular to a recommendation system recommendation method, equipment, recommendation system and storage medium. Background technique [0002] The recommendation system is an information retrieval tool in the Internet era. Since the 1990s, people have realized the value of the recommendation system. After more than 20 years of accumulation and precipitation, the recommendation system has gradually become an independent subject. A lot has been achieved in both research and industry applications. [0003] The recommendation system can be regarded as a multi-classification problem of machine learning. How to find out the kind of products that customers like in a wide variety of products has become the main goal of the algorithm design of the recommendation system. The early recommendation system mainly used the collaborative filtering algorithm (Collaborative Filtering Recommendation)...

AI Technical Summary

Problems solved by technology

If this attribute is not handled properly, some serious mistakes may occur. For example, a person who often watches "Wang Jing"'s "funny movies" may really intend to like watching "Wang Jing" movies, not because of watching "Wang Jing" movies. "Jiupin Zhimaguan" and "Deer and Ding Ji 2" belong to "funny movies" and do not watch "Chasing the Dragon", but if this new user only watches "Nine Pins of Sesame Official" and "Deer and Ding Ji 2", the recommendation system does not recommend "Chasing the Dragon", the user will think that the recommendation system does not "understand" him, and he does not realize that "Wang Jing" can also make movies with other themes, and the "surprise" is lost. This is because the recommendation system does not recognize the main attributes. Misreading user intent

Images