Dynamic implicit social recommendation

Publication Type:
Issue Date:
Full metadata record
Files in This Item:
Filename Description Size
01front.pdf152.15 kB
Adobe PDF
02whole.pdf2.67 MB
Adobe PDF
Due to the potential value of social relations, social recommendation has attracted a lot of attention recently in the research communities. Modelling time drifting data is a central problem in this area. Temporal changes, such as the emergence of new products or services and the changes in users’ purchase behaviour pattern, bring unique challenges. The need for modelling time changes at the level of each individual significantly reduces the amount of available data for detecting such changes. Thus we should resort to more accurate techniques that suffice the dynamic nature of social recommendations. Besides, the essence of social recommendation methods is utilizing users explicit social connections to improve recommendation results. However, explicit social connection information is not always available in real-world recommender systems. Only few Web sites have implemented the social or trust mechanisms, like Epinions and Douban. Lacking social recommendation data greatly limits the impact and utilization of social recommendation methods. Fortunately, in case that we do not have explicit social information, we can always compute a set of implicit social information to improve the recommendation performance. To this end, we propose a new Implicit Social Recommendation(ISR) model from cascade data (a variant of time series) in Chapter 3, which makes recommendations based on the inferred latent social network. It can sufficiently mine the information contained in time by mining the cascade data and identify the dynamic changes in the users in time by using the latest updated social network to make recommendations. Further, since the temporal information is crucial in social recommendation, we explore the temporal behaviour pattern from users’ time series data in Chapter 4. Themain challenge is finding the discriminative and explainable features (shapelets) that can best represent the raw time series data. We build an economical shapelet learning model that can automatically-learn shapelets without labels. Specifically, a novel Unsupervised Discriminative Subsequence Mining (UDSM) model that automatically-learn shapelets from unlabelled time series data. Another challenge in social recommendation is the continually appeared new users and new items. In real-world application, this problem is unavoidable. How to involve this kind of trapezoidal data streams and how to handle this problem smoothly must be considered. Most of the existing research either builds a general model in a sample scenario such as ignoring the dynamic nature of the system and the problem of new users and items, or only focuses on some specific fields, such as dynamic restaurant recommendation and timely news recommendation. We propose a new Online Learning with Streaming Features (OLSF) algorithm and its two variants OLSF-I and OLSF-II for mining trapezoidal data streams in Chapter 5. OLSF and its variants combine online learning and streaming feature (i.e. continuously appeared new items) together to handle the double-streaming data. Finally, we introduce a novel General Online Dynamic Social Recommendation (GODSR) model in Chapter 6, which combines network inference from cascade data, double-streaming users and items, and collaborative filtering together in an iterative process. By inferring the latent dynamic core social networks from cascade data, identifying the drift of a users preferences and involving new users and items in the online learning process, GODSR method reacts rapidly and makes accurate recommendations to users when new data arrive.
Please use this identifier to cite or link to this item: