Project: Context-aware Recommender Systems in Mobile Scenarios

People: Dr. Wolfgang Wörndl, Florian Schulze, Prof. Dr. Johann Schlichter

The goal of the project is to apply different kind of context-aware recommender systems in various mobile scenarios. We are working on contextualizing individual, content-based and collaborative recommenders including item-based collaborative filtering. Application scenarios include mobile tourist guides and recommendation for car drivers where contextual information is very important.

Related projects: Mobile User Modeling & Mobile Interaction Design for Adaptive Services

List of currently available thesis topics by our chair.

Bachelor or Master Thesis, or Guided Research in this area.

Subproject: Proactivity in Mobile Recommender Systems

People: Dr. Wolfgang Wörndl, Florian Schulze, Daniel Gallego Vico (Universidad Politécnica de Madrid)

Traditional recommender systems usually follow a request-response pattern, i.e. these systems only return item suggestions when a user makes an explicit request. In mobile environments, user experience could possibly be improved by delivering recommendations without any user request or query. Proactivity means that the system pushes recommendations to the user when the current situation seems appropriate. We have developed a model for proactivity in mobile recommender systems. The model relies on domain-dependent context modeling in several categories. The recommendation process is divided into two phases to first analyze the current situation and then examine the suitability of particular items.

Important research questions are whether users would accept proactive recommendations, how to present recommended items and possibly notify users. To investigate these questions, we have designed two options for the user interaction with a proactive recommender for Android smartphones: a widget- and a notification-based solution. The approach was evaluated in a survey with good results regarding usefulness and effectiveness. The results also showed that test users preferred the widget-based solution. Current work includes determining user context for proactive recommendations from activity logging and sensor data on the mobile device.

Subproject: Context-aware Collaborative Filtering in a Mobile City Guide

People: Dr. Wolfgang Wörndl, Korbinian Mögele

Voxcity In this subproject, we are working on applying context-aware recommender systems for mobile guides. The fi rst one is a mobile tourist guide developed by voxcity s.r.o. and jomedia s.r.o. The idea is to rent out a mobile device with GPS positioning capabilities to support tourists. The guide is currently available for the Czech city of Prague (see The mobile application plays audio, video, pictures and (HTML) text of tourist attractions based on the current position, traveling direction and speed.

This application has been extended with options to rate point-of-interest (POI) information and recommend new items based on decentralized, item-based collaborative filtering. We also use implicit ratings that are derived from the usage of audio files. We have evaluated the approach in two separate field studies. In the first one, 30 participants – real tourists visiting Prague – used the recommender function and were asked to fill out a questionnaire with promising results. In a second field study analyzing usage log files, we discovered an improvement of recommendations based on the collaborative filter in comparison to the pure location-based filter used before. In addition, we noticed that recommendations based on implicit ratings derived from audio playback duration outperformed the model based on explicit ratings.

Current work includes the integration of other media types. So far, the information about POI is mostly an image with optional text and a corresponding audio file. It is planned to integrate more information about the sights, e.g. videos, and also other POI types such as restaurants. We assume that with a more diverse item set, the improvement of the collaborative filter with regard to recommendation quality could turn out more significant. We also intend to refine the user interface of the mobile guide and improve the overall user experience when utilizing the system.


(Concluded) Subproject: Context-Aware Recommender Systems in Automotive Environments

People: Dr. Wolfgang Wörndl, Roland Bader (BMW Research and Technology GmbH)

This subproject focuses on proactive behaviour of recommendations in automotive environments. In this kind of environment, proactivity can provide an important contribution for more convenience and safety, because information overload and restricted user interfaces are major problems in finding information while driving a car. So far, proactivity has not gained much attention in recommender system research or has been put into practice. To enable proactive recommendation systems, context plays a fundamental role, not only in adapting items to context but also in making decisions based on context. Our goal in this subproject is to recommend points-of-interest (POIs) in a proactive manner.

The core aspect of proactive systems is to percept, comprehend and predict situations to assess the relevance of an item and avoid annoying the user by irrelevant and badly timed recommendations. Our approach for situation-awareness is to exploit the route of a user as context. It gives ths system the opportunity to not only take into consideration where the user has driven, but also look ahead in the future and understand where the user is going next. Not only the current route, but also possible alternative routes and formerly driven routes can lead to an enhancement of relevance assessment concerning POIs. In this subproject, we have also worked on explaining recommendations and design a corresponding user interface for an in-car navigation system, that has been evaluated in a user study.


(Concluded) Subproject: Recommening Mobile Applications for the Framework

People: Dr. Wolfgang Wörndl, Christian Schüller (UnternehmerTUM), Tobias Ullmann (UnternehmerTUM)

This recommender system is integrated in the framework by UnternehmerTUM supporting the development of mobile applications. Part of the framework is a deployment server where developers of mobile applications can register their services and end users can browse and search for relevant and interesting gadgets, e.g. a mobile tourist guide for the city the user is currently traveling in. We have designed and implemented a recommender system to suggest interesting and – with regard to their current context – relevant applications to users on their mobile device in this framework.

Users can choose between several content-based and collaborative filtering components:

  • LocationAppRecommender: recommends applications that were used in a similar location by other users
  • CFAppRecommender: applies existing collaborative filtering algorithms to generate results
  • PoiAppRecommender: recommends mobile applications based in points-of-interests (POIs) in the vicinity of the user using triggers


(Concluded) Subproject: A Decentral Recommender System for PDAs

People: Dr. Wolfgang Wörndl, Henrik Mühe

Mobile User InterfaceDecentral recommender systems are a promising approach in mobile scenarios such as exhibition visitors with Personal Digital Assistants (PDAs), yet have not been investigated thorougly so far. Therefore we have designed and implemented an approach to recommend items such pictures to PDA users. The approach is based on item-based collaborative filtering. The system calculates item similarities based on users' ratings and exchanges the resulting item-item matrices of similarity among PDAs. Enhancements compared to existing approaches include the extensibility of the model by introducing versioned ratings vectors. In addition, our system optimizes the storage requirements of the model which is important on PDAs with limited capacities.

We have also integrated shared displays for the recommendation to user groups. We have implemented and successfully tested the approach in a small scale user test. The screenshot on the right shows the user interface on the mobile device. The scalability was tested using a standard data set for recommender systems. Our system reduced the storage requirements significantly. Ongoing work includes refining the algorithm for the group recommendations, and also improving users' anonymity by investigating the use of asymmetric encryption when exchanging ratings.



Selected Publications

 Inference of User Context from GPS Logs for Proactive Recommender Systems

 Autoren: Benjamin Lerchenmueller, Wolfgang Woerndl
 Beschreibung: Workshop Activity Context Representation: Techniques and Languages, Twenty-Sixth Conference on Artificial Intelligence (AAAI-12), Toronto, CA, Jul. 2012

 Experiences from Integrating Collaborative Filtering in a Mobile City Guide

 Autoren: Wolfgang Woerndl, Korbinian Moegele, Vivian Prinz
 Beschreibung: In: A.V. Senthil Kumar, Hakikur Rahman (Eds.): Mobile Computing Techniques in Emerging Markets: Systems, Applications and Services. Idea Group Reference, Jan. 2012

 A Study on Proactive Delivery of Restaurant Recommendations for Android Smartphones

 Autoren: Daniel Gallego Vico, Wolfgang Woerndl, Roland Bader
 Beschreibung: Workshop Personalization in Mobile Applications, ACM Recommender Systems Conference, Chicago, IL, USA, Oct. 2011

 A Model for Proactivity in Mobile, Context-aware Recommender Systems

 Autoren: Wolfgang Woerndl, Johannes Huebner, Roland Bader, Daniel Gallego Vico
 Beschreibung: 5th ACM International Conference on Recommender Systems, Chicago, IL, USA, Oct. 2011

 Context-Aware POI Recommendations in an Automotive Scenario using Multi-Criteria Decision Making Methods

 Autoren: Roland Bader, Eugen Neufeld, Wolfgang Woerndl, Vivian Prinz
 Beschreibung: Proc. Workshop on Context-aware Retrieval and Recommendation (CaRR 2011), Conference on Intelligent User Interfaces (IUI 2011), Palo Alto, CA, USA, Feb. 2011

 Situation Awareness for Proactive In-Car Recommendations of Points-Of-Interest (POI)

 Autoren: Roland Bader, Wolfgang Woerndl, Vivian Prinz
 Beschreibung: Proc. Workshop Context Aware Intelligent Assistance (CAIA 2010), 33rd Annual German Conference on Artificial Intelligence (KI 2010), Karlsruhe, Germany, Sep. 2010

 Context-Aware Recommendations in Decentralized, Item-Based Collaborative Filtering on Mobile Devices

 Autoren: Wolfgang Woerndl, Henrik Muehe, Stefan Rothlehner, Korbinian Moegele
 Beschreibung: Proc. Workshop on Innovative Mobile User Interactivity, MobiCASE Conf., San Diego, USA, Oct. 2009

 Decentral Item-based Collaborative Filtering for Recommending Images on Mobile Devices.

 Autoren: Wolfgang Woerndl, Henrik Muehe, Vivian Prinz
 Beschreibung: Proc. Workshop on Mobile Media Retrieval (MMR 09), MDM 2009 Conference, Taipeh, Taiwan, May 2009

 A Hybrid Recommender System for Context-aware Recommendations of Mobile Applications

 Autoren: Wolfgang Woerndl, Christian Schueller, Rolf Wojtech
 Beschreibung: In Proc. IEEE 3rd International Workshop on Web Personalisation, Recommender Systems and Intelligent User Interfaces (WPRSIUI 07), Istanbul, Apr. 2007

All publications of our chair