Recommendation Algorithm Research Combining Interrelationship Mining And Collaborative Filtering

In recent years,new technologies such as artificial intelligence and big data have developed rapidly,and the scale of the Internet has expanded year by year,resulting in a large number of Internet companies,etc.,and people's lifestyles have changed dramatically.In the era of big data with soaring information,people are surrounded by countless kinds of information,and it is difficult to accurately find what they need from massive amounts of data.Information overload is a problem that needs to be solved urgently during this period.The generation of a recommendation system can solve this problem very well,but when a new item is added to the system,it does not have any scoring information,which makes it difficult to be recommended to the user,so a cold start problem occurs.In order to solve the cold start problem of the recommendation system and achieve personalized recommendation of new items,this thesis combines interrelationship mining and collaborative filtering algorithm.The main contents are:(1)This thesis proposes a recommendation algorithm combining interrelationship mining and collaborative filtering for items cold start(IMCF).Firstly,based on the interrelationship mining,the characteristics of the items are extracted and extended.The interrelationship mining method is used to extract the three kinds of mutual interrelationships between the items,which expands the number of attributes,enriches the available attribute information of the target items,and increases the discrimination between items.Difficult to extract issues.Then,a neighbor selection method based on interrelationship mining is proposed,which improves the lack of similarity calculation between items in traditional methods,and solves the problem of a single set of neighboring items.Finally,the combination of interrelationship mining and collaborative filtering algorithms,and the introduction of adjustment parameters to optimize the results,can solve the problem of complete cold start and incomplete cold start of new items at the same time.(2)This thesis proposes a recommendation algorithm of Interrelationship Mining based on Item Weight and Dispersion Degree(IMWD).Firstly,it analyzes the defects of traditional cosine method using scoring to calculate the similarity of items,and then proposes a new similarity calculation method,which makes full use of each user's information,item scoring information,and calculates the weight of items.Then adjust parameters and merge with the similarity of interrelationship mining to find the parameter value of the optimal result.Finally,the degree of discreteness of the item scoring data is introduced to calculate the volatility of the data,and the accuracy of the system's prediction scoring stage is improved.Compared with other methods,the results prove that this method has more advantages.In this thesis,the optimal values of the two parameters are determined through a large number of experiments,and the algorithm proposed in this thesis is verified on the Movielens 100 K,Movielens 1M and Movielens 10 M datasets.Experimental results show that this thesis can not only solve the problem of cold start of new items,but also improve the accuracy,diversity,novelty,and coverage of the recommendation system.