Long-term Recommendation Based On Bipartite Network And The Predictability Of Diffusion-Based Method

Today,with the rapid development of human society,the information on the Internet has shown explosive growth,which has become the main obstacle for people to quickly and effectively obtain the information they are interested in.At the same time,more and more online platforms hope to improve revenue by improving user loyalty and providing better services for users.The emergence of recommendation system can not only help solve the increasingly serious problem of information overload,but also provide personalized recommendation service for users according to their historical records without clear goals.Therefore,it has gradually become an area of widespread concern.The core of recommendation system is recommendation algorithm.In many algorithms,a kind of recommendation algorithm based on diffusion over binary network has received a lot of attention.At present,a large number of studies have improved and supplemented this kind of algorithm to improve the recommendation accuracy.Relevant research shows that this kind of algorithm tends to recommend popular products in one-step recommendation.Although the accuracy of one-step recommendation is very high,its long-term performance is poor.In addition,the improved diffusion based recommendation algorithm can improve the accuracy,but this kind of algorithm has the maximum recommendation accuracy.How to improve the maximum recommendation accuracy to obtain more space for accuracy improvement is a kind of problem worthy of consideration.Therefore,this paper studies the diffusion based recommendation algorithm to improve the maximum accuracy.At present,in the research of recommendation algorithm based on diffusion,more attention is paid to the short-term performance of recommendation.In real life,recommendation is a long-term process.For example,online network will develop with the passage of time,and users often have the consumption psychology of novelty when shopping,so the long-term performance of recommendation algorithm also needs more attention.In order to solve the above problems,this paper first attempts to apply the classic algorithms that perform well in short-term recommendation to long-term recommendation based on a time-based evolutionary model.The long-term performance of these classic algorithms is observed,and the results show that the diversity and accuracy of long-term recommendation gradually deteriorate.In order to improve the performance of long-term recommendation,this paper proposes a time factor to enhance the similarity of two historical records generated by users in a short time and weaken the similarity of two historical records with a long time interval.This time factor is integrated into the diffusion based recommendation algorithm and applied to long-term recommendation.The experimental results show that the diversity of each step of the improved algorithm is better than that of the classical algorithm without losing the recommendation accuracy,especially in the late stage of evolution,the value of the diversity index is increased by more than half;observing the diversity of long-term recommendation,it is found that in the early stage of evolution,the diversity curve will decline rapidly,and the decline rate is much higher than that of the traditional algorithm The diversity curve of classical algorithms tends to be stable in the later stage of evolution.According to this algorithm,the recommended list of goods will be enriched,the user’s choice will be more diversified,and the recommendation system will be more healthy.On the other hand,when researchers improve and supplement the diffusion based recommendation algorithm,they are more committed to improving the accuracy of the recommendation algorithm,and a crucial problem will appear,that is,the accuracy of this kind of recommendation algorithm will have a maximum,so whether the maximum recommendation accuracy can be improved through some methods to provide more information for improving the accuracy of the algorithm It’s space.To solve this problem,this paper first introduces the method of quantifying the maximum recommendation accuracy,and finds that the accuracy and diversity of the recommendation algorithm depend on the resource diffusion width;then proposes a method of adding virtual edge,through which more products can get resources to increase the resource diffusion width of the recommendation algorithm,which has not appeared in the user history At the same time,the experimental results also show that the proposed method also improves the recommendation accuracy,which improves the ranking of the products in the test set in the recommendation list based on the target users,while for the target products,it improves the ranking of the products in the user’s recommendation list The average position will advance.This method can not only improve the accuracy of the algorithm to get more room to improve,but also can recommend products that have not appeared in the user’s history to improve the accuracy and diversity of recommendation.