Context-aware Web Recommendation Algorithms

With the development of the information technologies and e-commence,people have fully entered the age of information overload from the lack of information age.In the "Large Data"age era,how to alleviate the problem of "information overload" brought about by the massive data,acquires wide concern of both the research community and academia.People put forward the notion of "personalized service",aiming at overcoming the problem of "information overload" to free people from massive redundant information.Through mining the binary relation between with users and item,it helps users to find the items of interest(such as Web information,service,online commodities,etc)from massive data,and generate personalized recommendation in order to meet the personalized needs of users.By fully utilizing contextual information in recommender systems,Context-Aware recommender systems not only help users to acquire needed information from vast resources,but also helps receive and access to information and computing resources "at any time,in any place and by any way".There's wide prospect for Context-Aware recommender technology to develop in the network applied fields,including mobile Internet,the Internet of Things,social networks,information retrieval,online travel,advertisement and so on.This thesis mainly research the Context-Aware recommender systems,and introduces the architecture of the system and explains the key techniques used in detail,including processing of user context information,contextual user preferences elicitation and Context-Aware recommender technology.And it also puts forward Recommendation Algorithm on User Temporal Preference and Location-aware Recommendation Based on Collaborative Filtering,and induces time context information and location context information in recommendation systems respectively.Time is a kind of important contextual information,and accurate capturing user preferences over time can improve the precision in recommender systems.Simple correlation over time is typically not meaningful,since users change their preferences due to different external events.Recommendation Algorithm on User Temporal Preference,by introducing the Session-based Temporal Graph(STG)in recommendation systems,and proposing the Path Fusion Algorithm(PFA)based on the STG model framework,makes accurate top-N recommendation,which allows Context-Aware recommender systems to capture user's long-term and short-term preferences,to generate User Temporal Preference.Location is also important contextual information.Users with different interests in different regions,when recommended items are spatial,users tend to those items nearer.Accurately capturing user preferences according to the users and items' location can improve the precision of recommender systems.Location-aware Recommendation Based on Collaborative Filtering introduces Pyramid Model(PM)to recommender systems to realize users partitioning and calculating travel penalty,and puts forward a Collaborative Filtering Recommendation algorithm based on Pyramid Model(PMCF)to generate Top-N recommend.Theoretical analysis and experimental results show that Recommendation Algorithm on User Temporal Preference and Location-aware Recommendation Based on Collaborative Filtering have obvious improvement on the accuracy evaluation index than traditional recommendation algorithm.