Research On Context-Aware Personalized Recommendation Methods Of Users

The rapid development of information industry, on the one hand, it brings convenience to people's lives, on the other hand, it also brings the drawbacks of information overload. It's normally difficult to screen and get valuable information resources from large number and wide variety of resources for people. As an effective way of filtering information, the research on personalized recommendation technology has been taken seriously gradually. In recent years, the personalized recommendation technology has been further developed, especially under the situation of those rapid breakthroughs of techniques such as artificial intelligence, pattern recognition, distributed storage, cloud computing and big data mining. By extracting and analyzing the characteristics, patterns of behavior and social preferences and other information of users, personalized recommendation technology could find out those certain information from mass content which really match the requirements of users. However, in the study of recommendation systems field, the traditional models only focus on the relationships between users and items but the impact on the environment of users, therefore, the recommended results is difficult to fit to the presentation.This thesis proposed an improved context-aware hybrid recommendation algorithm based on the studies of content aware and modeling theory as well as the comparative analysis of the traditional recommendation algorithm. This algorithm recommended information resources for the users which fit to their surroundings through the introduction of the context factors related to those users and mining the association between users, the context and the items resources. Finally, we evaluated this algorithm on a movie rating dataset. The experimental result shows that the improved algorithm has better prediction accuracy than the existing recommendation algorithm. In this paper, the main research includes the following aspects:Firstly, this dissertation studied the acquisition and modeling of context information and other related recommended technologies, and proposed a specific definition of context for specific acquisition of different contexts. Afterwards, we did related research and exposition for context-aware modeling and context aware recommendation technology. Then, this dissertation comparative analyzed three of those major traditional personalized recommendation models which were collaborative filtering, content-based filtering and knowledge-based filtering following with the respective analysis and description of their shortcomings and advantages.As the defects of traditional recommendation algorithm, this treatise made a exploratory study of improved context-aware hybrid recommendation algorithm. At the beginning, we did related discussions of the non-feature based hybrid recommendation model which can improve the accuracy of the recommendation but still has restrictions for wide application in specific environment. Therefore, this treatise devised a content-based collaborative filtering recommendation model and uses average rating of the target user's nearest neighbor(users who possess preference similarity exceeds the threshold value) to calibration the final rating. Then we bring the context factors into the recommendation model by analyzing the behavior of current user's nearest neighbor to different items in the same situation and acquired the context based recommendation list of users to recommend for the target user. Then by calculating the similarity matching of context, got the results of user's preference attributes affected by single dimension context. Finally, this paper realized the ultimate recommendation rating of the target user according to the combination of target user's predicted rating under the effecting of different context factors. Meanwhile, the data scarcity problem exist in traditional recommended mode is researched by this treatise and another cold start problem caused by addiction of new items or users is solved by the improved hybrid recommended mode based on feature supplement.