Patrick Gratz

Ad Hoc Collaborative Filtering for Mobile Networks

This work introduces a recommender system designed to augment the information discovery and dissemination in mobile networks. We use the system to optimize the podcast providing mechanism of the HyCast application. The presented recommender system is based on collaborative filtering and provides two different algorithms to determine similar network neighbors, which are used to incrementally build up a locally stored model for the final prediction calculation.

Collaborative filtering via epidemic aggregation in distributed virtual environments

The ever-increasing amount of available information in todays digital society necessitates inline techniques for determining the most relevant content. Collaborative filtering (CF) systems have proven to be an adequate means for reducing informational overload and generating useful recommendations. Current systems are predominantly built on centralized or, more recently, structured Peer-to-Peer (P2P) approaches. However, in order to apply collaborative filtering to large-scale distributed virtual environments (DVEs) in unstructured networks with substatially higher user numbers, different approaches are necessary. Within this paper we present a collaborative filtering algorithm for DVEs utilizing epidemic data aggregation based exclusively on local information. Designed to be extremely scalable, it creates recommendations in a transparent way by distributing an accumulated view of favorite ratings to interacting users. The algorithm is intended for deployment in the HyperVerse - a self-organizing middleware service for large-scale DVEs - for generating and managing rating predictions of object favorites. Evaluation results show that, in terms of quality, locally aggregated predictions converge well on those obtained from an idealized global view.

Self-organizing collaborative filtering in global-scale massive multi-user virtual environments

Due to the huge amount of available information in todays society, it becomes more and more difficult for the consumer to locate the most useful information for a specific topic. Recommender systems using collaborative filtering (CF) are a popular technique for reducing information overload and finding useful information on the Internet. However, in massive global-scale multi-user virtual environments different approaches are required from those used within the currently dominant centralized infrastructures or lately investigated P2P approaches. Within this paper we present a novel collaborative filtering algorithm used within the HyperVerse -- a P2P-based self-organizing middleware service for massively distributed virtual worlds -- to generate and manage recommendations for HyperVerse object favorites. Due to its global extent considering users and possible ratings, using a monolithic database-backed recommendation service or huge profile- or item-rating-matrices does not scale in our scenario. The decentralized approach presented within this paper creates per user ratings in an adaptive and transparent way by comparing public favorites of passer-by users with personal peer data, weighted by self-adjusting buddy lists.

Delay-tolerant collaborative filtering

Recommender systems using collaborative filtering are a well-established technique to overcome information overload in todays digital society. Currently, predominant collaborative filtering systems mostly depend on huge centralized databases to store user preferences and furthermore are only available when connected to Internet. In this paper, we consider an incremental recommender system for highly dynamic mobile environments where no central global knowledge is available and communication links are rather unreliable in comparison to static networks. We present an algorithm that aims to reach a reasonable prediction coverage and accuracy while keeping the amount of additional network overhead as small as possible, maximizing the performance of our system. For this purpose, the presented algorithm is based on a delay-tolerant broadcasting mechanism on top of a weighted cluster topology. Evaluation results show that in terms of accuracy and coverage the results of the presented algorithm converge on those obtained from a global knowledge scenario, even in the case of message loss.

Web Service Interface Syndication and its Application to a Collaborative Weblog Search

This paper presents the idea of Web service interface syndication - a scheme for the collaborative creation of overlay networks based on common Web service interfaces. Rather than requiring complex management of dynamic peers we picture a setting of interconnected static Web applications forming syndicates and offering services in a self-organized fashion. Among other application scenarios, this paper will present vicinitySearch, a collaborative Weblog search service which has been created in order to demonstrate the concept. This service offers a remarkable added-value to the Weblog community, making use of existing collaborative structures and the interoperability provided by XML Web services. Apart from the general concept of Web service interface syndication, the paper presents an implementation of the vicinitySearch service that has been done in terms of a Wordpress plugin