Hybrid Recommendation Algorithm Based On Time Weighting

The rapid development of Internet information technology is followed by explosive growth of data.How to find the information you really want in the case of information overload? For this reason,the recommendation system appears.The recommendation system does not need to provide precise requirements like search,but discovers the user's needs based on the user's historical behavior data and recommends products to the user.Traditional collaborative filtering does not consider the impact of time factors on recommendation results.At the same time,a single recommendation algorithm has its own shortcomings.For example,collaborative filtering algorithms have cold start problems,and content-based recommendations have poor recommendation quality and single content.Case.For this reason,we propose two recommendation algorithms based on time weighting,namely recommendation algorithms based on time weight and item type,and recommendation algorithms based on time weight and item topic missing.The specific research content is as follows:(1)Recommendation algorithm about the weight of time and project type.First,we calculate the similarity between items from the data of user behavior about time weight,then use the topic vector extracted and selected from the item type with the use of the LDA to calculate the similarity between the item and the item,and finally merge the two similarities,based on the fused similarity and user behavior data based on the weight of time predict user ratings and then recommend favorite items or items for users.After doing experiments,we proved that the problem of cold start are not only solved by the algorithm,accuracy,F1 value and diversity have a certain improvement when we compared the algorithm with the traditional collaborative filtering algorithm.(2)The user's historical behavior information based on time weights and the use of item profiles to supplement the algorithm about hybrid recommendation of the types of item.The way choose a few keywords from the keyword set of the project's profile information as feature attributes.When calculating the feature value of each feature attribute,use the Word2 Vec training project introduction corpus to calculate the word vector of each keyword,and at the same time classify the words with higher relevance to the same feature attribute.This method is equivalent to aggregating the feature attributes of similar items The eigenvalue of,can effectively deal with the problem of data skew after such processing.Taking the item type as the goal,train a two-class model for each item type and use this model to predict the type of item based on the information in the profile.Then,the project type that best represents the characteristics of the project is selected from the prediction results of all the project type classifiers,and the project type that is incomplete is merged with the project type.A hybrid recommendation algorithm based on item type information and user historical behavior information considering time weight is used to separately process the similarity between items for fusion,and finally recommend items to the project based on the user's historical behavior information based on time weight and the similarity of the fused items.user.Experiments show that the method of supplementing project types can make up for the shortcomings of project types,and the algorithm has improved accuracy,F1 value,and diversity.