Research On Recommendation Algorithm Based On Information Matching

As a basic approach to solve the information overload,Recommendation systemhas been widely used in the application of the Internet. By using the technology of information found,Recommendation system can recommend movies, music and news among the vast amounts of information for users.There has been a lot of websites using recommendation system to provide users with personalized recommendation service, such as amazon, sina microblog, Aqi and storm video product recommendation system, etc.The traditional recommendation systemmainly provides personalized recommendationfor users.With the rapid development of information technology and the Internet,the phenomenon of family groups sharing information and accounts, a group of friends searching for collective activity forrecommendation has become more and more common.Thus,providing personalized service for group is becoming one of the current main researchissues in theWeb field.Group recommendation is extended on the basis of user recommendation.With rapid increase of the number of users and items, the user's personalized recommendation is faced with:(1) The expansibility of the users for item rating matrix with high computation complexityis not so strong;(2)The largesparse data causes a decline in the accuracy of the recommendation.At the same time, traditional group recommendation dose not consider the changes of the user preferences as members of the group.This article will consider recommendation algorithm which is based on information matching as the main research direction, and describe the information of users and items by the feature respectively.Both the scores of items and the description of the users and the characteristics of itemshave a certain randomness.According to the information matching correlation concept,the characteristics of users and items can realize correlation matching,which will get a personalized recommendation probability formula, and can quickly and flexibly realize personalized recommendation.This article mainly proposed two recommendation models based on information matching:personalized recommendation modeland group recommendation model.Personalized recommendation algorithm based on information matching is mainly provide recommendation service for a single user.First, introduce the scores of items is based on user preferences,and the attraction for the user.Then,according to the scores of the user for the items,we acquire the user characteristic vector and the items characteristic vector.In view of many uncertain factors of the characteristic of the users and items,we put forward a probability model for describing the user and objects features.Introduce the concept of information matching, to match the user's and items' feature vector, get a personalized recommendation based on the information matching probability model, and with the help of the probability model provide personalized recommendation for users.Group recommendation algorithm focuses on the change of the group users' preferences.Traditional group recommendation algorithm generally has two problems, one is thedata sparseness,the other one is these algorithms do not consider the change of the group users' preferences,and the personalized recommendation lists are not precise enough.To solve the above problems, this paper puts forward a kind of group recommendation based on information matching model.First of all, according to the similarity of group members we classify the group,the correlation betweengroup and items changes with the change of the group type.And then expandthe concept of information matching to group recommendation, set the correlation between groups and items relies on the correlation between group members and items,depending on the group type, we could build different probability models of the group recommendation.According to the probability model of group calculate the correlation probability between each item and groupand sort the value of probability model,Finally,according to the recommendation strategy we could getthe recommendation lists.Finally, using the MovieLens and MoviePilot data set,werun relevant recommendation algorithm on Hadoop cluster, and get the offline recommendation results.After filtering and ranking the results, and then get the final list of personalized recommendation.Using contrast test we verify the effectiveness of the personalized recommendation model and the group recommendation model based on information matching.