Information Recommendation In Heterogeneous Pervasive Environment

Compared with the ordinary desktop computing, pervasive computing, as a new computing model, has integrated information processing into human's daily life. The computing resources disappearing into everywhere we live can be presented to us in any way that we want, so as to meet our need for information any where, any time. Pervasive computing is deeply changing the way how people interact with the information system around them, while introducing new challenges. The first challenge refers to so-called context awareness problem. As said to be for people, pervasive computing firstly needs to know the computing environment and interaction state, and to research on collecting, modeling, reasoning of context information. Next is the network heterogeneity problem. As we know, there are distinct differences between wired network and wireless network. One of the basic problems in pervasive computing is how to dig for the core characteristics of these two different networks so as to provide an integrated API to upward applications. The last problem is seamless service integration problem, mainly about how the services can support user mobility in order to realize service mobility.Meanwhile, rapid development of Internet has brought people into a new era of information society and network economy, when we are surrounded by the sea of information. The explosive increment of the amount of information in both physical and logical world has make people in trouble to find information that they like in such a large scale. In such a situation, recommendation systems are emerging, which can provide a user personal recommendation information based on the predicted user preference. Note that, recommendation service is one kind of context-based intelligent service. In both academic and industrial community, however, most effects are put on researching centralized recommendation algorithms, mainly about how to improve the preciseness of recommendation results. Few researchers focus on how to provide consistent services in a heterogeneous environment, and how to modify the centralized algorithms to apply into a pervasive and mobile environment.Aimed at the three basic problems above from the angle of recommendation service, we design a heterogeneous recommendation framework, where one of the core components is Heterogeneous Communication Module. This module uses Bluetooth Service Discovery Protocol and Gossip-based Bluetooth Interaction mechanism to provide users'devices natural interactions just like that in human's daily life. Gossip-based Bluetooth Interaction mechanism leverages a scalable gossip protocol, namely SCAMP, with Piconet-based Bluetooth topology, so as to model the social communication of people. By integrating the Bluetooth neighbor set with the partial view of SCAMP, we design a new information distribution method, whose efficiency is proved in a simulated experiment. Using the heterogeneous recommendation framework, we design and implement a music recommendation system that adopts collaborative filtering algorithms. This prototype system uses a double-criteria recommendation strategy combining songs'specificity and artists'generality together. Without bothering users for any explicit rating, we have presented our implicit rating extraction mechanism for both songs and artists, which is used in collaborative filtering later. What's more, by taking advantage of a scalable gossip-based membership management protocol, we have designed a novel P2P recommendation algorithm, which integrates with the centralized as much as possible so as to provides consistent music services in a heterogeneous pervasive environment.