Research And Implementation Of Item-Based Collaborative Filtering Recommendation Algorithm

The rapid development of the Internet has brought convenience to the people,at the same time also plagues the people: The bewildering complexity of the data makes people dazzling.Recommendation system arises at the historic moment.As the earliest and most successful recommendation algorithm,collaborative filtering has been widely used.It aims to tap the user's historical behavior,and on this basis to make recommendations.However,with the increase of data size,collaborative filtering also exposes many problems.This thesis is based on item-based collaborative filtering recommendation algorithm,in the process of recommending,aiming at the sparsity of the scoring matrix,the real-time problem of the nearest neighbor selection and the accuracy of the scoring prediction,the research and improvement are made.Finally,the three optimization methods are integrated.The research work of this thesis is as follows:(1)Aiming at the sparsity of the scoring matrix,a collaborative filtering recommendation algorithm based on item characteristic attributes is provided and implemented.The algorithm introduces the characteristic attributes of the item and analyses the potential value of the characteristic attributes,the characteristic attribute information is used to calculate the similarity between items,and on this basis,combined with the score data of the similar items to predictive the score of the nonscoring items.Then the sparse score matrix is filled to improve the accuracy of recommendation.(2)Aiming at the real-time problem of the nearest neighbor selection,an item clustering recommendation algorithm based on improved K-means is provided and implemented.The algorithm introduces the ABC-K clustering algorithm,analyses the design idea of the ABC-K clustering algorithm and its advantages in clustering,uses the ABC-K clustering algorithm to cluster the items.It can significantly reduce the search time and improve the real-time performance of the search only in several clusters with high similarity to the target item.(3)Aiming at the accuracy of the scoring prediction,a collaborative filtering recommendation algorithm based on time weight is provided and implemented.The algorithm uses the time weight function to optimize the scoring prediction model.First of all,analyse the influence of time on the user's interest,based on this,the time weight function is proposed,and combine it with the traditional score prediction model to improve the accuracy of recommendation.(5)Integrate the three optimization methods above.First of all,analyse the idea of integration,and then integrate them with the idea.Its recommendation real-time and accuracy is tested by experiments.