Research On Collaborative Filtering Algorithm Combining User Preference And Project Association

With the advent of the information age,vast amounts of information are flooding the current social life,which is a huge challenge for the producers and users of information.The generation of recommendation system provides an effective solution to this problem.As a bridge between information producers and users,it can screen valuable information for users,and also help information producers to promote their information to It is in front of interested users.It is generally believed that the field of recommendation systems originates from the proposal of collaborative filtering algorithms,which are currently widely used recommendation algorithms.Although the collaborative filtering algorithm has been applied to many e-commerce website recommendation systems,there are still some unresolved problems,such as cold start,data sparsity,and so on.Aiming at the problem that the similarity calculation in the collaborative filtering algorithm does not consider the impact of the actual scene,the following improvements have been made to the similarity calculation method:First,an improved method for similarity calculation based on user preference is proposed.Aiming at the influence of hot and cold items on the similarity calculation,this article uses the hot item penalty factor to reduce the proportion of hot items in the similarity calculation,and introduces this factor into the cosine similarity calculation;taking into account the user's common rating item The effect of quantity on the similarity of two users,we use the combination of Jessica correlation calculation and improved cosine similarity calculation to perform user similarity calculation.Secondly,an improved method for similarity calculation based on project association is proposed.Aiming at the problem that the traditional similarity calculation only considers the item score,which leads to the irrational calculation of the item's relevance,we use a linear weighted method of the item's score similarity and the item's semantic similarity to calculate the item similarity.Then,based on the weighted Slope one algorithm as the carrier,and applying the similarity calculation method proposed above to the Slope One algorithm,a weighted Slope One algorithm UI-Slope One that combines user preferences and item associations is proposed.Finally,the most suitable values of the two parameters in the above two improved methods and the most suitable values of the neighborhood user set are determined through experiments.Substituting the two parameters into the formula,through experimental comparison,it can be proved that the improved similarity calculation method proposed in this paper has lower MAE and RMSE values than traditional similarity calculation,and has better score forecast effect.In addition,compared with the other two optimization methods proposed for the Slope One algorithm,the UI-Slope One algorithm has a better score prediction effect.