The Research Of Recommendation Algorithm Based On Hash Learning And Time Context

With the continuous progress of science and technology,people have started the era of Web2.0,and the following problems of "information overload" and "information explosion"are becoming more and more serious.Moreover,there is the phenomenon of "long tail theory" in economics,that is,80%of profits come from 20%of popular commodities,but the remaining 80%of commodities have great commercial value.How to recommend the remaining 80%of commodities to those in need is the key to solve this problem.Massive data make people need to spend a lot of time and energy to find the information they need,and the emergence of recommendation system has greatly eased this problem.The recommendation system generates a list of items that the user may be interested in by collecting the user's historical behaviors or the user's interest preferences and calculating the recommendation algorithm,and pushes the information that the user may need under the condition that people do not have a clear purpose or the purpose is vague.However,the recommendation system still has the problems of cold start,accuracy and real-time under big data for new users or new projects.In this paper,the author introduced the basic theoretical knowledge of recommendation system comprehensively through reading and collecting a lot of relevant knowledge of recommendation system,expounded the research significance of the paper and the research status at home and abroad,and explored some problems and deficiencies in the existing recommendation algorithm.The main research results of this paper are as follows:Firstly,in view of the fact that the existing recommendation algorithm based on time context only considers the influence of memory forgetting curve on the change of user's interest,and does not take the time attribute of the item and the user itself into consideration of the influence factor on the user's interest,this paper proposes an improved recommendation algorithm based on time context,which mainly integrates the attributes of the user's age and the background time of the item into the traditional time decay function.The influence of changing the user similarity weight on the prediction score of recommended items is finally verified by experiments on real data sets.Compared with the traditional recommendation algorithm,the algorithm in this paper greatly improves the accuracy and other indicators.Secondly,aiming at the equivalence between binary code similarity and user preference of traditional recommendation algorithm based on hash learning,an improved hash learning recommendation algorithm is proposed,which can better solve the similarity by removing the score bias in the processing of score information,and alleviate the cold start problem of the algorithm by adding user and item time attributes.The feasibility of the.algorithm is illustrated by experiments and result analysis.Finally,based on the above two algorithms,this paper designs and implements a movie recommendation website that combines offline recommendation and real-time recommendation.In terms of similarity calculation,the improved hash learning recommendation algorithm proposed in chapter 4 alleviates the problem of unequal scoring information and user preferences,so that the calculated similarity better reflects user preferences and further improves the accuracy of recommendation.For problems such as cold start,the improved time context recommendation algorithm in chapter 3 of this paper will add time information of users and projects to alleviate the cold start problem of new users and new projects.Then it describes the overall structure design of the website,the design of the specific process,the process of implementation,the main functional pages and the display of the recommendation effect.The website has added a real-time recommendation section to track changes in users' interest preferences in time an d improve the.accuracy of recommendation results.