A Hybrid Recommendation Algorithm Based On The Mode Of User’s Thinking

With the rapid development of the Internet, a large amount of information on the internet comes together. It is difficult for users to find necessary resources from a large amount of information. The traditional search engine provides search results only according to user’s keywords, and it shows the same search results to all users. In order to meet the needs of different users and minimize the search time, a variety of personalized recommendation systems have been utilized on many Internet business platforms. However, the present recommendation algorithms have many limitations. For example, the recommendation quality-is not high, and recommendation system cannot effectively screen out the items to meet the needs of users. Now, following the increase in the number of users and information resources, the recommendation algorithms are faced with many difficulties such as data sparse, cold start, new project recommendation problem and so on. To overcome these difficulties, this paper takes into account users’thinking mode, social network and items’tag information, and modifies some recommendation algorithms, to improve the personalized character and accuracy of recommendation results.Firstly, this paper presents a recommendation algorithm based on trust relationships rebuilding and social networks transferring, named TRSP. TRSP makes full use of the relationships between users in the social network. And the main idea of TRSP algorithm is as follows:1) Because it is very simple and easy to make friends on the Internet, noise frequently appears in the relationships. A list of friends that you can trust is given directly by Social networks, but it cannot be used. In order to solve the noise problem and use the trust relationships effectively, this paper discards the false and retains the true on the circle of friends.2) Considering the data sparse problem caused by the operation in this paper, we find the use’s "potential friends", who have similar interests and hobbies with him, from his historical information, to expand his circle of friends. Then build a reliable user social network trust relationship.3) The increase in the number of users brings data sparse pressure on the social network recommendation. To solve this problem, we make use of the transmitting characteristics of the user’s trust relationship, and regard the target user’s "Friends of friends" as another data resource to complete the recommendation. We have experimented the proposed algorithm on real Epinions datasets. Results show that the reconstruction of the social network trust relationship is more accurate in recommendation:Tag information can effectively reflect the user’s own habits and interests, and it can be used to find the needs of users. Secondly, we make use of social label information and propose a popular classification method based on personalized time tag cloud, named PTTC. PTTC algorithm includes the following aspects:1) We extract the user’s preference tag cloud from the label information to represent the user’s preferences, and extract item tag cloud to describe the information of item. Then we find the best way to measure similarity between tag clouds to complete the recommendation.2) User’s interest preference varies in different time period. Considering the impact of time, this paper extracts user’s time tag cloud according to the label’s time stamp information.3) Because the user is free to label the project, there is much ambiguous or redundant information in the tag set. In this paper, we remove redundant tags based on the usage of tags. Because there is difference between itmes for one user, different items may have the same label, but the value of the label is not the same. According to this idea, the user’s label information is weighted to extract the weighted preference tag cloud and weighted time tag cloud.4) In order to solve the problem of sparse tag data caused by the increase in the number of users and items, this paper combines preference tag cloud and time tag cloud, and uses social tagging information to complete the recommendation.Both social network information and tag information can achieve better recommendation results, but single recommendation algorithm has its own defects. TRSP algorithm recommends items to users based on the advice of friends. In the process of recommendation, it does not take into account the attributes of the item itself or the user’s requirements. PTTC algorithm makes full use of the characteristics of the item attributes to recommend for users. In the process of recommendation, it does not fully consider the characteristics that users are gregarious, and there are some users who choose items directly according to the opinions of friends. In order to overcome the defects of single algorithm, and take the advantage of these two algorithms, we lastly propose a hybrid recommendation algorithm based on the mode of user’s thinking named UTMCR. Since different users have different way of thinking and method when choose items, UTMCR algorithm measures and defines the user’s way of thinking, and switches the recommendation algorithm used in the recommendation system according to the user’s way of thinking. For reference mode of thinking, we will use TRSP algorithm to recommend. Meanwhile, for the search matching way of thinking, we will use the PTTC algorithm to complete the recommendation. UTMCR algorithm uses the method of switching algorithm to complete the combination of recommendation. In this paper, we verify the advantages of the UTMCR algorithm on the dataset Last.fm from multiple aspects.