Research On The Collaborative Filtering Algorithm Based On Improving Weight Calculation

Since the 21 st century came,Internet technology has developed rapidly.With the popularity of the Internet,e-commerce and Internet information dissemination has also been developed a lot.As a result various kinds of goods or Internet information overwhelmingly presented in front of us,and facing of the so much goods or information,it is difficult for users to select the most desired resource quickly.Personalized recommendation technology came up,as one of the effective ways to solve the problem,it recommend users of products of preference after collecting and analysing of the user's history browsing.A well-established recommendation system is usually composed of three parts: the recording plate,recording the user's historical behavior;analysis plate;algorithm plate.As the core of the whole recommendation system,the related research of the recommendation algorithm has become the hot research direction,because the recommendation result is closely related to the performance of the recommended algorithm.According to different needs of users,recommendations will differ from each other.At present,there here are four main kinds of recommendation algorithms: content-based recommendation algorithms,collaborative filtering recommendation algorithm,hybrid recommendation algorithm and network recommendation algorithm.Collaborative filtering recommendation technology is one of the successful technologies in the application of personalized recommendation technology.Thanks to collective wisdom,in a large number of users to it will unearthed out a small part of “neighbors” with similar hobbies among a large number of users,according to the analysis and records of these neighbors' favorite contents,and generate a sorted catalog,which is called recommended results,then push it to the “neighbors”,which will reduce the user's workload of picking process.The traditional collaborative filtering algorithm performs the similarity calculations with the user scores without considering the behavior time or the same label between the items,as a result,a lot of failures are exposed,such as the cold start problem,sparse matrix problems,recommended accuracy issues,etc.The accuracy of recommendation results is not satisfying,and will hardly meet the actual needs of users.My research is based on the item-based collaborative filtering algorithm,and add the users' behavior time and the label attributes also other information into the similarity calculation,to avoid the cold start problem,thereby improving the quality of the recommend results to meet the actual needs of different users as much as possible.The research is as follows:First,in the process of preparing the experimental data,we will introduce the behavior time of news and short video information into the data set,and integrate the time factor into the degree of popularity score by pre-processing the time decay function when calculating the similarity between the items,and then carry on the project-based collaborative filtering recommendations.Second,as the new item couldn't get the appropriate recommendation weight,we can integrate the short video tag into the similarity calculation process by using short videos' label.The label is evaluated by the Cosine Similarity after extracting the short video label.The new item will be evaluated of popularity score for predicting resources,and then carry on weight adjustment of the authority of the publisher.At last,accomplish item-based collaborative filtering.Finally,based on the user's real behavior log of news client,design the experimental scheme,and compare the results of collaborative filtering algorithm and the improved collaborative filtering.We can easily find from the experimental results that the improved collaborative filtering algorithm does improve the cold start problem,leading to the improvement of recommended accuracy to some extent.