A context-aware framework for personalised recommendation in mobile environments

摘要

Context‐awareness has become an essential part in various personalised applications such as mobile recommender systems and mobile information retrieval. Much progress has been made in context‐aware applications. However there is alack of general framework for supporting the rapid development of context‐aware applications and enabling the sharing and dissemination of context information across different applications.This dissertation presents a novel context‐aware framework for supporting context distributions and personalised services in mobile environments. This dissertation makes four major contributions: First, it proposes a JXTA‐based Hybrid Peer‐to‐peer framework, called JHPeer, for efficient organisation, representation, retrieval and management of context data, which enables rapid development of context‐aware applications for mobile users. JHPeer is customisable and supports diverse high-level applications with a set of abstractions that are open to many possible implementations. Second, it develops an analytic hierarchy process based multi-criteria ranking approach, AHP‐MCR, to rate information and help users in finding relevant items. AHP‐MCR takes user context information into account. A general and extendible criteria hierarchy model is developed. The weights of the contexts criteria can be assigned by user or automatically adjusted via individual‐based and/or group‐based assignment. Third, it develops a Bayesian Network (BN) based user profiling method to model user’s preference. The BN model construction process is defined as being capable of handling the cold‐start issue and can be applied in multiple applications. Finally, it designs and implements a Proactive Personalised News recommender, PPNews, on top of JHPeer framework. All JHPeer components are implemented in PPNews for effective news recommendation. The BN‐based user profiling method estimates users’ preference including new users.The AHP‐MCR approach effectively ranks news articles based on the user’s preference, past click history and news attributes. The experimental results show that PPNews can proactively recommend relevant news to mobile users.