The Design And Implementation Of Personalized Recommender System Based On The Analysis Of User’s Behaviors

With the rapid spread of Internet, human society is going to enter "information overload" era.How can the users quickly find relevant information from the vast multitude of Internet data? This question has become a popular topic.of Internet technology. Commonly, there are tow ways that can address this problem. They are searching engine and information filtering.Personalized recommendation is a new kind of information filtering technology.lt can find users’interest preference from the users’historical data, and filter the information of user’s interest from the irrelevant Information, then, give the information to user with the way of "Push".While, the e-commerce platform can use the technology of personalized recommendation to improve "long tail" advantage of the platform, increase the relevant interests party benefits.This paper brings the personalized recommendation technology into vertical search engine upgrade design of soufun.com". Analyzing the historical behaviors of users, extracting the interesting model of historical users, using the way based on the user collaborative filtering to discovery the current user’s interest, find out the item of the current would be like from the database of item.So that, the "excessive screening" problem on Vertical Search Engine may be alleviated. And this paper main work is as follows:Firstly, the paper summarizes the personalized recommendation classical algorithms, theories, hot spot of this field and the related technologies. Compares the recommendation algorithms based on rules, content and collaborative filtering,and show the advantages and disadvantages of these algorithms. It also briefly introduces the theory of related HMM (hidden markov model).Secondly, based on the behavior characteristics that users operate the search engine of soufang.com",which analyzed the logs of search engine to define what the user’s behaviors and the sequence of user’s behaviors.Besides,it designed an algorithm of sequence fusion to extract the sequence of user’s behaviors.And puts forward a similarity calculation method to calculate the similarity between two user behavior sequences.Thirdly, according to user behavior sequence, this paper does modeling for user. And based on fundamental theory of HMM, it designed a model to predict users’ behaviors. As well as, it gave the methods to estimate the parameters of prediction model.Then.it desigined a set of recommendation algorithms that considered many factors,such as the timeliness of item,the collaboration of users, the users’ behavior sequences and so on. Except for these, it also made out the "cold start" strategies both for new users and new items.Finally, combined with the actual demand of "soufun.com", this paper designed and implemented a house information personalized recommender system. And it designed the related experiments to evaluated performance of prediction of users’behaviors.The combined with the characteristics of HMM, it analyzes the limitations of the part of system about behaviors prediction.As well as, it also discussed the evaluation indicators about the sort of recommend item and the correlation between recommend item and user’s interest. And it designed experiments to check these indicators and effect of system.At the end of this paper, it based on the analysis of the system’s shortages, put forward some tasks on the next step.