Research Of Recommendation Algorithm Based On The Complex Network Theory

The rapid development of the Internet allows the user to face more and more information, it is hard to timely and accurately find useful information in the huge number of information source. Personalized recommendation system gradually become one of the most effective way to overcome this problem which has been researched a lot in recent years. Personalized recommendation system can take advantage of user track record to achieve the purpose of personalized service, through the establishment of corresponding recommendation algorithm that can dig out the user features and likes.Recommendation algorithm as personalized recommendation system the most core component, its efficiency directly affects the entire recommend results of the system. Along with the development of complexity science, the research that relationship in realistic society is abstracted for the complex network has become a trend. So, the users and products of personalized recommendation system are abstracted for nodes, the relationships users choosed products are abstracted for attachment in the complex network, reserching and analysising such network structure that may bring new breakthrough for recommendation algorithm research.In order to improve the accuracy and diversity of personalized recommendation system, the paper aims to understand the insight of personalized recommendation system using complex network theory and others. In this paper, we will introduce the current situation, the meaning and value of the personalized recommendation system firstly, and then we discuss the development and evolution of recommendation algorithms that based on the network strcture and two new recommendation algorithms were proposed. The main results are as follows:The first part introduces the background of personalized recommendation system, it expounds to solve the important significance of information overload of personalized recommendation system, and gives the main research methods and research content.The second part introduces the complexity theory and personalized recommendation system. First of all, this paper outlines the concept, characteristic, the development conditions and application field of the complexity science and the complex network. Describing some kinds of the topology characteristics and introducing several kinds of classical complex network model. Secondly, the article describes the personalized recommendation system. It points out the personalized recommendation system the research significance as well as commercial value and analyses the business process, the system’s overall frame and recommend the main challenges of the recommend system.In the third part present personalized recommendation algorithm is reviewed. Mainly introducing the recommendation algorithm based on content, cooperative filter recommend algorithm, algorithm based on the complex network structure and other algorithms. Importantly, Using material diffusion of resource allocation process to deliver the similarity between the user to recommend to target users of its not collect products in the network structure recommended algorithm. The final of this part discusses the advantages and disadvantages of various algorithms.The most important part in this paper are the fourth and fifth part. The fourth part gives the new network algorithm based on random walk through fixing the taditional cooperative filter algorithm by analysising the influence ability of the degree of the product and resources spread orientation based on random walk. Pointing out that in reality of online network system, not active users occupying most of the scale, and only existing a small amount of active users, we can strengthen online social network’s and e-commerce sites’scalability and user dependency by effective using of the not equal user ratio relations. All the traditional collaborative filtering recommended considers the user’similarity from target users to the neighbor. In this paper, we add the constraint parameter to magnanimous recommended capacity of products with large degree. It would increase the appearance ratio of the unpopular. At the same time, we change the resources propagetion direction based on random walk which help us find the maximum similarty between users. The new algorithm is superior the previous algorithm through the above two steps of adjustment in the accuracy, diversity and personality.In the fifth part we put forward the new three parts network recommendation algorithm based on heat exchange which would use this process to increase the diversity. The traditional recommendation algorithm foucs on the improvement of the accuracy which is really one of the most important indexes to measure the algorithm. But the result always can’t meet the users’specific needs, for high accuracy often bring to popular product which the user would get in other various ways. In this part, we successful use the tag information and the heat process into the user-product-tag three parts network recommendation, and this new algorithm not only increase the recommended diversity, but also guarantee the accuracy.The last part is the conclusion of this paper.There are two main innovation points. The first is broking the bottleneck of the direction through the calculation of the similarity between the users. Improving the standard collaborative filtering algorithm which based on the network structure by the way used most similarity of random walk. And the result of recommend have the diversity by adding in the constrainted parameter to the products which connecting with many users. The second is introducting the heat transfer process to the original network structure of three parts which has the information of tags. We use the way of heat transfer to spreading resources between users and products would increase the differences between the two users’ lists. And it thereby improving the diversity of the recommendation algorithm. The method that adding heat teansfer to the user-product-tag network structure is of no use before.