Research On Multidimensional Social Information Recommending Model Integrating Contextual Factors

Blog, microblogging, weixin......, we are living in the various types of social media platforms at the Web 2.0 era in applications. Pictures, video, audio, text......, we have been surrounded by the big data which has many types and updates quickly. However, in the condition of information like flood emerging today, we are difficult to seek the information we need. Nowadays, there are mainly two methods to solve the problem of information overload:one is the user active search, the other is passive recommended. Although information retrieval is one way to solve the information overload, but for users who have the different needs and preferences, the monotony for the search results do not meet the personalized needs of users. When users retrieval information, they will spent some time on browsing and retrieving unrelated information, resulting in utilization of information reduced, and not well solving the information overload problem; For the recommendation, it can provide personalized information according to various kinds of user needs preferences, which is another method to solve the problem of information overload. In addition, there are a mass of dynamic context information and user interest information on the wide popular social media platforms but have not been better used and mined. Therefore, integrating all information resources from the different types of social media, and completing interest information mining and recommendation starting from the overall concept, may be an important direction of the development for the goal of knowledge service oriented.This paper proposes a multidimensional social information recommendation model integrating contextual factors according the characteristics of social media platforms from the perspective of the global use of social media resources. Based on dimension divided of social capital, namely structural dimension, relationship dimension, and cognitive dimension, multidimensional information recommending model in this study also includes three levels:social relationships dimension, trust relationships dimension and semantic relationships dimension. The study uses the appropriate method for the each recommending dimension based on main features on that, and integrates the main context factors into the recommending process, and completes the multidimensional information recommendation based on user-resource-context. In the experiment, each dimensional method has been applied based on different data sets those are extracted from the microblogging. By analyzing, the article finds the application scope of each method dimension, and the whole information recommending framework in application based on various social media platforms.This paper is the study of theory and approach for the practical application. The full text includes nine chapters, removing the introduction and summary at the before and behind, the main content of remaining sections are as followings:Chapter 1:This chapter clarifies the concept, and consolidates the theoretical foundation of this research. To beginning with, this part clarifies the object of this paper —social media, including the definition, types and characteristics of that, which provides the basis for the multidimensional information recommending application framework based on social media at the subsequent study; Secondly, the concept of context and social capital are sorted out, including definition, classification and application domain. The former provides basis for the research in contextual factors and the latter provides the theoretical support for the dimension divided of multidimensional information recommending model. Finally, in order to deepen the understanding of social information recommendation, this part consolidates the concept of social recommendation and social information recommendation. The content in this chapter lays a solid theoretical foundation for the study, and it is the source of research ideas.Chapter 2:The overall elaborating of the multidimensional social information recommending model integrating contextual factors is proposed in this chapter. Based on the concept and dimensions of social capital, and social media features, this study considers the information needs of users for the social media are ultimately for the social capital needs, therefore the multidimensional information recommending model are proposed. Then, the author further analyzes the multidimensional characteristics of the model, including the theoretical multidimensional and methodological multidimensional. Based on above, methodology and the concreted method used for each dimension in the model are summarized. The research of the overall framework in this chapter provides a logical sort for the future research.Chapter 3:This part discusses the main contextual factors affecting users'interests preferences. First, according to context information classification in the chapter one and the basic types of information in daily life, these factors are divided into four measured variables, respectively, user context, environmental context, the task context and interest information. Each variable is set to a number of questions under, and correlation analysis and the corresponding differences in analytical methods are used to explore the relationship between each of context factors and user interest preferences. According to whether the measured variables are ordered categorical variables or not, the measured variables are divided into two parts. For the ordinal variables, Kendall's tau-b correlation analysis is adopted, and for the remaining variables, differences analysis in cross-list is used. This chapter has important theoretical guidance significance for the future study of information recommendation integrating contextual factors.Chapter 4:One of the recommended dimensions — the social information recommendation research based on social relationships is referred in this chapter. This dimension corresponds to the structure dimension of social capital. In this dimension, there are more weak ties those are composed of interpersonal networks than strong relationships from social media interactions between friends and other acquaintances. Therefore, taking the recommendation of weak ties into account is essential. CPM method can be used to community divided, getting strong ties user sets and weak ties user set. Since most social media platforms do not have the scoring mechanism, the measure of social network and complex network are used to compute the user rating for resources. From the perspective of the two user sets divided, both sets compute resources' scores from two aspects:scoring their own resources in individual, and scoring from the center of the nodes in the network. Then the method achieving for multidimensional information recommendation based on user-resource-context is discussed, including integrating contextual factors into recommendation based on the ideological distance, and using formula of Euclidian Distance Metric to get square form with the contextual similarity and the users' similarity. Experiments show that the method of integrating contextual factors into social relations preforms better than the traditional two-dimensional recommended method.Chapter 5:This chapter researches the social information recommendation based on trust relationship. This dimension corresponds to the relationship dimension of social capital. Taking users social relationships and their posting resources into consideration, the chapter divides trust relationship into two parts, explicit trust and implicit trust. The former refers to users'social relationship, and the latter is presented from the similarity of the resources posted by users. Then trust network and context network are constructed by TF-IDF and Cosine Similarity method. The direct trust (no others between two users) and indirect trust (one or more users between two users) are calculated by the shortest path based on the social network, and integrating the context factor whose trust value isgreater than the threshold value(average value) to improve the traditional trust network. Finally, according to the maximum distance, the recommended-user set has been determined and then information is recommended based on trust value in the recommended-user set. Experimental results show that the method of information recommendation integrating contextual factors in the trust dimension preforms better than the original trust networks.Chapter 6:The social information recommendation based on semantic relationship dimension is researched in this part. This dimension corresponds to the cognitive dimension of social capital. Compared to the recommendation taking use of "relationship" and "relationship+information" in previous methods, this chapter mainly discuss the aspect of "information". Firstly, the algorithm of Latent Dirichlet Allocation(LDA) is used to complete the theme mining for resources posted by every user, getting some topics of user interests. Secondly, contextual semantic model is constructed based on the features in context to support the semantic similarity calculation. According to the randomness feature of social media information, Wikipedia, one of the social media, is chosen to construct Ontology. Then Ontology constructing based on Wikipedia is discussed to support semantic calculation between resources. Finally, recommendation based on resource semantic similarity and contextual semantic similarity is discussed, including computing semantic similarity based on hierarchy tree, and integrating the context factor into the process of semantic calculation. Experiments show that the method of integrating contextual factors into semantic relationships preforms better than the recommended method only based on resource semantic similarity.Chapter 7:This chapter discusses the empirical analysis for the multidimensional social information recommending model based on the microblogging and the recommended application framework of the model. According to the characteristics on the microblogging platform, this chapter divides microblogging dataset into three parts, namely "social type", "social+information type", "and information type". Respectively, each method is applied to the three datasets to find the application scope of each dimension method. By analysis, three types of datasets correspond to the types of social media, therefore getting the application framework of the proposed multidimensional social information recommendation model. In addition, the chapter further analyzes the application development of the model based on characteristics of the datasets, which refers to that the proposed method can also be applied to the aspect of recommendation with different types of users.In summary, information recommendation based on the social media is a complex and huge issue. This paper proposes recommending model only from the dimensions of social capital and contextual factors aspects, with hoping to provide reference for the knowledge service based on social media.