Research And Application Of Recommendation Algorithm Based On Network-based Inference

With information overload becomes more serious,the recommender system plays an increasingly important role in our daily lives.Simultaneously,with the rapid growth of commerce and development of Internet technology,a large number of user consumption preferences become available for online market intelligence analysis.A critical demand is to reduce the impact of information overload by using recommendation algorithms.In the field of information filtering technology,network-based recommendation algorithms based on mass-diffusion have been popular for its simplicity and efficiency.For the problem of information loss during one-mode projection,Network-based inference(NBI)uses bipartite networks for projection.In a bipartite network,users and items can be represented with different types of nodes.This method finds the causal relationship between the items by the random walk of resources in a bipartite network,and then recommends for user.This paper studies the network-based inference based on mass-diffusion in the recommended method.To solve the problem that most network-based recommendation algorithms can't distinguish how much the user likes collected items and allocate same resources to all item nodes,we propose a new approach called biased network-based inference(BNBI).The specific research contents mainly include the following aspects:First of all,for the rating systems,we draw on the item similarity commonly used in the collaborative filtering.The proposed method makes up for the loss of negative ratings in the original NBI algorithm and preserves the complexity of model at the same time.During resource initialization,the method uses the similarity of items to explore the hidden but unused information in negative ratings to eliminate the influence of item diversity on the recommendation.The function reduces the initial resources of items that are similar to items that the user does not like,thereby discovering the user's true interests and hobbies,making the initial resource allocation more reasonable.Secondly,for the non-rating systems,we compare the differences between the rating system and the non-rating system and discuss how to determine the items that users like.In order to solve the problem that the final resource in the original NBI algorithm is non-optimal,we draw on the idea of content-based filtering to increase the final resources by the incremental function and make the final resource allocation more reasonable.Items that are similar to what the user likes will get more resources.Then,in order to verify the feasibility of the proposed method,we test our method in real datasets(MovieLens and Last.FM).At the same time,we introduce a variety of evaluation indicators and compare the proposed approach with multiple benchmark algorithms.The experimental results prove that the proposed approach is more effective.Compared with the traditional methods,the allocation of resources in the bipartite networks is more reasonable,and the recommended performance is effectively improved.The BNBI method proposed in this paper effectively solves the problem of the lack of information due to the loss of negative ratings in the original NBI algorithm and the non-optimization of final resource.Finally,to solve the social problem of "difficult parking" in daily life.In the practical application,the NBI method is used to provide users with efficient recommendation of real-time parking spaces,reduce the time to find a parking space or wait for a parking space and facilitate the convenient of residents.Moreover,the application value of NBI algorithm is demonstrated in the recommended field and the application scenario of NBI algorithm is enriched.