Study On Context-Awareness Based Multidimensional Information Recommendation

With the popularization and development of network, the network has become one important way of obtaining information. However, with the great increasing of network information,"overloading" has turned into one serious matter to be solved. Information recommendation is a kind of active information service, which achieves much attention in the theory and application fields. Information recommendation technologies can automatically select the content which the user is interested in from abundant information. Therefore, information recommendation can solve the "overloading" problem and satisfy the personalized requirement of the users. In the field of information service, information recommendation is widely applied.However, the most information recommendation approaches are two-dimensional approaches, which involve only two dimensions (user dimension and item dimesion) and ignore the context dimension. In the field of information recommendation, context is the environment which the user is in, such as time, place and so on. Related researches have proved that context can have great influences on the users’action. The same person may show different interests and demands in the same item in different contexts. So, the traditional two dimensional recommendation approaches are not adapted to the new recommendation environment. How to extending the traditional two dimensional recommendation approaches to the multidimensional recommendation approaches has become the hot point of the information recommendation field. Thus, based on the context-awareness technologies, this dissertation makes deep research on the multidimensional information recommendation. The dissertation builds up a context-awareness based multidimensional information recommendation model and develops six multidimensional information recommendation algorithms.The main body of the dissertation is composed of six chapters as follows.1. The Related Research of Information RecommendationThis chapter summarizes the existing recommendation approaches. In the chapter, the recommendation theory and process of every approach is introduced. The advantages and disadvantages of the traditional recommendation approaches are analyzed. The chapter makes a summarization of the recent information recommendation research and points out the problems and development tendency, which is the research target of the dissertation.2. The Related Research of Context-Awareness This chapter makes a summarization of context-awareness theory. The definition and classification of context is introduced firstly. Secondly the achieving and processing approaches of context are analyzed. The most important approach-context-awareness is expatiated on. The system conception model and supporting platform of context-awareness is elaborated on. During analyzing the supporting platform, the chapter discusses in details the process of how the raw context data is turned into the high level context information, which is the foundation of the context-awareness based multidimensional information recommendation model.3. Context-awareness based Multidimensional Information Recommendation ModelA context-awareness based multidimensional information recommendation model is built up in this chapter. In the first, the contextual framework in multidimensional recommendation is discussed It includes the following content: making the definition of context in the multidimensional recommendation, extracting the contextual elements in the multidimensional recommendation, analyzing the context dimension, building up the contextual profile, analyzing the aggregation computation of context dimension. In the second, based on the contextual framework and context-awareness technologies, the chapter establishes a context-awareness based multidimensional information recommendation model and elaborates on operational mechanism of the model and the function of every submodule.4. Context-awareness based Multidimensional Information Recommendation algorithmsThis chapter discusses the influences of context on the traditional recommendation algorithms. On the base of the traditional recommendation algorithms, the chapter develops six new multidimensional recommendation algorithms which are classified into contextualization of recommendation input algorithm, contextualization of recommendation output algorithm and contextualization of recommendation function algorithm. In the paradigm of contextualization of recommendation input, firstly current context information is used to filter the multidimensional ratings to select the relevant set of ratings. And, ratings can be predicted using any traditional two-dimensional recommendation algorithm on the selected data. In the paradigm of the contextualization of recommendation output, current contextual information is initially ignored, and the ratings are predicted using any traditional two-dimensional recommendation algorithm on the multidimensional ratings. Then, the prediction ratings are adjusted (contextualized) for each user using the contextual information. In the paradigm of the contextualization of recommendation function, contextual information is used directly as the variable in the recommendation function. The new multidimensional recommendation algorithms developed by the author include reduction-based contextualization of recommendation input(RBCRI) recommendation algorithm, hybrid recommendation algorithm based on the RBCR recommendation algorithm and traditional recommendation algorithm, contextualization of recommendation input recommendation algorithm based on contextual similarity, model-based contextualization of recommendation output recommendation algorithm, user-based multidimensional heuristic recommendation algorithm and item-based multidimensional heuristic recommendation algorithm.5. Experiment and Evaluation of the Context-Awareness Multidimensional RecommendationBy using network survey, this chapter collects the user ratings of fifty films. The author adopts statistics software-SPSS and database management software to make data processing. The experiment realizes the six new multidimensional recommendation algorithms developed by the author and evaluates the recommendation effectiveness. The experiment proves that most of the new multidimensional recommendation algorithms are superior to the traditional algorithm except the contextualization of recommendation input recommendation algorithm based on contextual similarity, which is partly superior to the traditional algorithm.6. The Limitation and Future Research DirectionThis chapter aims to point out the limitation of this study in terms of data collection and research contents. The author will develop or make use of more intelligent data-collecting tools to actively achieve the user ratings under more contexts to solve the "cold starting" problem and "context dimension" problem of the experiment in the dissertation. In addition, the author will make modification to the new multidimensional recommendation algorithms, using matrix operation or combining the content-based recommendation approach to solve "data sparsity" problem. Finally, the author will make deeper research on the model-based recommendation technologies in the multidimensional information recommendation field in the future.