Research And Application Of Recommendation Algorithm Based On Association Rules And Collaborative Filtering

The information transfer revolution brought by the intelligent Internet is reshaping the traditional service industry.Internet users produce,create and use data every day,making the problem of information overload inevitable.In the face of massive amounts of information,how to establish a direct and effective association between users and information becomes crucial.The recommendation algorithm uses data mining technology to efficiently use valuable hidden information to provide users with personalized recommendation services and improve user perception,which has become a new force to promote the development of the industry.The research content of this article mainly focuses on the following three parts:In the first part,the personal interests of users and the popularity of items are constantly changing.The traditional collaborative filtering recommendation algorithm calculates based on historical behavior data generated by users,and focuses on the past time dimension.There are timeliness issues that do not take into account the latest information of the item,and the user's interest,which may change over time.In view of the above problems,a time factor is introduced to improve the traditional similarity measurement algorithm,and a more accurate nearest neighbor set is constructed,and then the prediction score is used to fill the user-item rating matrix.Alleviate the timeliness of collaborative filtering algorithms in the recommendation process and improve the quality of recommendations.The second part is aimed at the problem that the user's historical scoring data of the item is too small compared to the item.Based on the research results of the first part,the association rules are introduced to improve the way of the collaborative filtering algorithm's prediction score considering the time factor.Through the association rule algorithm to mine the frequent co-occurrence relationship combinations between items,construct a strong association rule recommendation degree calculation method,and then predict the item score that has an association relationship with the target user's item.At the same time,a parallel hybrid association rules recommendation algorithm and collaborative filtering recommendation algorithm are tried in parallel to further optimize the user-item scoring matrix to produce hybrid recommendation results.It alleviates the problem of the decreasein recommendation accuracy caused by sparse data in collaborative filtering algorithms and improves the user experience.Finally,based on the theoretical research and experimental results of the above two parts,the architecture is designed from the user dimension,the item dimension,and the interaction dimension between the user and the article,and a music recommendation system with personalized service functions is designed and implemented.