Research On Recommendation Algorithm Based On Item's Complementary Substitution Network And User's Similarity Network

In the era of big data,the recommendation system can be seen everywhere,and the recommendation system as an effective tool can well improve the efficiency of users in obtaining information.Therefore,continuous research to improve the performance of the recommended model is of great significance.In this paper,we propose two corresponding methods to solve the two problems in the existing recommended models.The specific problems and solutions are summarized as follows.First of all,in the recommendation system,the users and objects we study are all related.If you ignore the relation of any party,it will bring error to the recommendation result.However,most of the existing recommendation algorithms do not make full use of the users' relation and the items' relation,which leads to the recommendation effect being affected.In response to this problem,this paper presents a recommendation model based on item's complementary substitution networks and user's similarity network.In this paper,we use the rating similarities of different users to construct a user's similarity network to reflect user's relation.Then we build a item's complementary substitution networks to reflect the relation between items.Finally,the relation of users and items are all used for recommendation.In this paper,an experiment on Amazon's published dataset shows that the proposed model is more effective than the contradistinctive recommended model due to the full use of both relation between users and items.Second,in the recommendation field,the item's complementary substitution relation has been proved to be very useful for improving the performance of the recommended models.However,at present the item's complementary substitution relation are mainly got from the two item links data Also-viewed and Bought-together which provided by Amazon platform and others.This situation led to the related recommendation model relied on off-the-shelf item links data and limit its application scenarios.In response to this problem,this paper presents a method for predicting the item's complementary substitution networks based on the similarity of texts and the probability of co-purchase.This method uses the most common scoring data and item text data to predict the item's complementary substitution networks by calculating the text similarity and the co-purchase probability of items.This article experiments on Amazon's published dataset and the results show that this method has a good predictive effect.With this approach,the recommended model based on item's complementary substitution networks does not have to rely on the off-the-shelf item links data such as Also-viewed and Bought-together which provided by Amazon and other platforms,and its applicable scenarios are greatly increased.