Studies On Resource Online Sharing And Recommendation In Social Web

With the advent of Web2.0, users are allowed to produce content in the Web. The social Web is a set of relationships that link together people over the World Wide Web. Typical Social Web applications which include social networking services, social media and online communities, etc. have already become the mainstream of web application. User interaction is a pivotal characteristic of the social Web. Starting with the social media sites, such as Flickr and Youtube, user-generated contents have taken over social web. Users and contents in these sites are still rapidly increasing every day. Traditional information retrieval technologies have limits when dealing with the problem of information overload in the social Web, mainly in that large amounts of unstructured data produced by users is more difficult to classify and retrieve. In the circumstance of massive user-generated unstructured data, data sharing and recommendation approaches take a more important role than information retrieval approaches for data diffusion in the Social Web.This thesis focuses on efficient resourse sharing and recommendation. Research contents and innovations of the thesis are summarized as follows:(1) A variable granularity user classification algorithm based on multi-dimensional features of usersClassifying Web users based on multi-dimensional features is one of the foundations of realizing personalized Web applications. It could be used for user classification model, users’ multi-dimensional data analysis, potential user group discovery and personalized recommendation and so forth. This paper proposes a variable granularity user classification algorithm based on the user feature model, which classifying the users from the single feature dimension to the multiple-features dimension. The essential idea is to gather those users who have the same features in a given feature space. The proposed algorithm has the following characteristic:a) all the qualified common feature categories could be mined; b) there is no inclusion relationship between any two mined categories; c) the common feature categories with different granularity are organized with the hierarchical structure. Variable granularity classification of web users based on their features could benefit many Web applications, such as multi-dimensional users’ data analysis, potential community discovering, user classification model, and personalized recommendation services.(2) Towards an efficient data sharing and recommendation approach based on the common preference group.In the circumstance of massive user-generated unstructured data, data sharing and recommendation approaches take a more important role than information retrieval approaches for data diffusion in the Social Web. In this thesis, we propose a new approach to discover groups automatically based on user’s preference. We call the group which is automatically generating by our approach the Common Preference Group (CPG). This research is conducting under a hypothesis that a user like a data object because the user is interested in some semantic topics implied in the object. With this assumption, we switch interests of users from the objects to the semantic topics, and we group users who share common interests together as a CPG.The automatic grouping approach has the following features:i) a CPG is corresponding to a set of semantic topics, which indicate the users’preference about each semantic topic; ii) the CPG is automatically generated, and a user whose preference matches a CPG will be added to this CPG automatically; iii) The data object can be automatically added to the corresponding CPG pools as well. CPG can be used as social purposes and data recommendation. Compared with current group mechanism, the new features of CPG bring these advantages:i) the users could discover their own preferences and the other people who have the same preference with them; ii) different CPG can be recommended to users based on user’s preferences; iii) objects are distributed to different CPG without human involved.(3) An approximate approach for batch update of common preference groupUser’s preferences are not static, so the UPM is changing from time to time. With massive users, each time the user preference model changed, mining CPG from the scratch is not an option. In this paper, we analyzed the types of user behavior and their corresponding modification of UPM, and then we summarized how the two types of UPM modifications influenced CPG. We proposed an approximate batch CPG update approach to keep them up to date and also avoid CPG re-computing.(4) Implementation of CPG based advanced functions in the Web community systemThere exsit a large amount of Web communities in social Web and the number are still increasing rapidly. The user behaviors, such as upload resources or post a comment, reveal user’s preference. We developed a Web community management system prototype and provided a series of advanced functions for Web communities, services and data management based on TOTEM Object Deputy Database System. In the thesis, we explore to realize many new advanced functions based CPG in our Web community management system prototype.