Studies On Web Community Recommendation Approaches And Systems

As one of the most important components of Social Web, Web community has gained a great success and enjoyed an exponential increase all over the world. A Web community is generally a group of people who use the Internet to communicate and collaborate with each other to pursue mutual interests or goals. Web community has brought immediate access to information, pervasive connectivity to others, globalized voices and more level playing field for business. In short, Web community has had a huge impact on people's daily lives.Billions of Web communities exist on the Internet, such as that Facebook maintains over 600 million communities, so the huge amount and continuous growing trend of Web communities have brought severe information overload to people. Community recommendation is of no doubt the most effective solution for the problem. Community recommendation can sift valuable communities from massive amounts of communities for common users, so as to help them communicate and collaborate with each other more conveniently; community recommendation can promote the development of communities and create greater economic value for community providers, as it attracts users to join more communities. Consequently, studies on Web community recommendation approaches and systems have great social and economic significance.Web community has distinctive topic feature, and the topic is the reason why the community can draw people to gather. Web community is a special item being recommended, because it is recommended to users and simultaneously it is composed of users, and social relations exist among users. This paper mainly focuses on the following four parts.· Accurate Community Recommendation Based on Latent Topics and Collaborative FilteringTraditional Recommendation methods based on explicit topic associations cannot solve the data sparsity problem; methods based on implicit topic associations can better handle data sparsity, but cause inaccurate results. To overcome the above shortcominings, we propose a collaborative Web community recommendation algorithm based on latent topics. Our algorithm utilizes implicit topic associations to compute relatedness between communities and the target user on latent topics, and highly ranks communities with high relatedness in the recommendation list for the target user. Then our algorithm takes advantage of behaviors and interests of similar users of the target user, aiming to promote the recommendation accuracy by reducing inaccurate results. Because communities that considerable amount of similar users are interested in should be higher ranked in the list, than those not so popular among the similar users, according to collaborative filtering.· Novel Community Recommendation based on a User-Communtiy Total RelationTypical recommendation techniques belong to accurate recommendation, which refers to the type of techniques purely pursuing the recommendation accuracy and ignoring novelty. However, pure accuracy can hurt recommender systems. In this paper, we study the problem of novel community recommendation, and propose an approach based on a user-community total relation (i.e. user-user, community-community, and user-community interactions). Our novel recommendation approach suggests communities that the target user has not seen but is potentially interested in, in order to broaden the user's horizon and promote the development of communities. Specifically, a Weighted Latent Dirichlet Allocation (WLDA) algorithm improves recommendation accuracy utilizing strength of user-community interactions. A definition of community novelty together with an algorithm for novelty computation, are further proposed based on the total relation. Finally, a multi-objective optimization strategy improves the overall recommendation quality by combining accuracy and novelty scores of communities.·Balanced Community Recommendation based on the Transitive Closure of a User-Community Total Relation with high efficiencyIn this paper, we propose an approach "NovelRec" to conduct balanced community recommendation of high quality which combines accuracy and novelty, based on the transitive closure of a user-community total relation; the closure represents the user-user, user-community and community-community multi-order interactions in Web community. Additionally, the framework of the method is designed to contain offline modeling and online recommendation, aiming to improve the efficiency of recommendation. Based on the transitive closure, NovelRec offline models user neighborhood, topical similarity between neighborhood users and topical distances between communities. According to the user-user and user-community multi-order interactions, NovelRec idenfies candidate communities for the target user, where the candidates are joined by neighborhood users of the target user; then the approach computes accuracy scores of the candidates according to interactions between candidates and the neighborhood users. NovelRec proposes a new metric of user-communtiy distance utilizing the multi-order ineractions, further, the approach computes novelty scores of the candidates, and the computation combines the new distance, participation in the candidates from the neighborhood users, as well as attributes of the candidates. Finally, NovelRec conducts balanced community recommendation, which takes both the recommendation accuracy and novelty into consideration.·The Web Community Recommender PrototypeAccording to features of Web community and recommendation evaluation metrics, i.e., accuracy and novelty, we propose three approaches for community recommendation, and embed them into a Web community recommender system. The recommender system is established on the basis of our Web community management system, and implemented as an advanced function. The community recommender system generates accurate, novel and balanced community recommendation for users, according to their multiple behaviors in communities, in conjunction with features of communities.