Final Report Summary - CARS (Context-Aware Recommender Systems (CARS))
Introduction
Chris Anderson in his famous book “The Long Tail” asserts that “we are leaving the age of information and entering the age of Recommendation”. Recommender Systems have become essential navigational tools for users to wade through the plethora of online content as they provide a personalized tool for information discovery.
Context in Recommender Systems Current research in Recommender Systems, while taking into account the relation between user and item, often ignores the “context” of the recommendation. “Context” is any environmental, temporal or otherwise variable that influences a decision a user might make. In order to improve the quality and relevance of a recommendation, it is essential to leverage all information available.
However, the importance of contextual information has recently been recognized. For example, work- shops on the topic were held at the last two ACM Recommender Systems Conferences1. Moreover new applications (e.g. Foursquare 2) that use contextual variables are emerging.
Research Work and Results:
The main aim of the CARS project is to build a compact context-aware recommender system for mobile and desktop computing devices that can integrate content-based data when available.
Over the two year period we have developed several novel context-aware recommendation algorithms that have been awarded with two (2) Best Research Paper award at the ACM Recommender Systems 2012 Conference and the European Conference of Machine Learning (ECML/PKDD) 2013.
Moreover we have developed a novel context aware app recommendation app for the Android Mobile OS called Frappe. Frappe , which is freely available and can be downloaded at Google Play on any Android device, uses the state-of-the-art context-aware collaborative filtering methods developed in this project.
At the end of this project we also conducted a user study to evaluate in-the-wild the usefulness and impact of the use of context on the Frappe app recommendation service.
During this two-year period the key results achieved are as follows:
a) 5 novel algorithms for context-aware collaborative filtering that outperformed the current state-of-the-art. The models developed incorporate context information an also optimize Information Retrieval metrics such as Mean Reciprocal Rank and Mean Average Precision. This leads to very significant performance improvements over the state-of-the-art. Care was taken so that the developed algorithms scale linearly with the size of the data and can thus be used on large-scale data.
b) We designed, implemented and evaluated a context-aware mobile app recommendation app. Frappe is an Android app that recommends users other apps that they might find interesting it is using the contextual information available from the mobile phone sensors to determine the apps matching the current context of the user. Frappe is based on an implementation of the algorithms developed at this project. It is available for free for all Android devices and has been downloaded over 4000 times from the Google Play store.
c) We have conducted a live user study on the Frappe mobile app recommendation service to evaluate the impact and utility of context in the app recommendation domain. This study in particular revealed
d) The research results obtained through all this work have been published at top tier conference and workshops in fields related to the Machine Learning, information retrieval, HCI and Recommender Systems. Two of our publications where awarded with Best Paper awards:
• ACM Recommender Systems 2012 Best Research Paper Award. Yue Shi, Alexandros Karatzoglou, Linas Baltrunas, Martha Larson, Nuria Oliver, and Alan Hanjalic “CLiMF: Learning to Maximize Reciprocal Rank with Collaborative Less-is-More Filtering” ACM RecSys 2012, Dublin, Ireland
• European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases 2013 (ECMLPKDD) Best Student Paper Award Julien Delporte, Alexandros Karatzoglou, Tomasz Matuszczyk, Stéphane Canu Socially Enabled Preference Learning from Implicit Feedback Data
Chris Anderson in his famous book “The Long Tail” asserts that “we are leaving the age of information and entering the age of Recommendation”. Recommender Systems have become essential navigational tools for users to wade through the plethora of online content as they provide a personalized tool for information discovery.
Context in Recommender Systems Current research in Recommender Systems, while taking into account the relation between user and item, often ignores the “context” of the recommendation. “Context” is any environmental, temporal or otherwise variable that influences a decision a user might make. In order to improve the quality and relevance of a recommendation, it is essential to leverage all information available.
However, the importance of contextual information has recently been recognized. For example, work- shops on the topic were held at the last two ACM Recommender Systems Conferences1. Moreover new applications (e.g. Foursquare 2) that use contextual variables are emerging.
Research Work and Results:
The main aim of the CARS project is to build a compact context-aware recommender system for mobile and desktop computing devices that can integrate content-based data when available.
Over the two year period we have developed several novel context-aware recommendation algorithms that have been awarded with two (2) Best Research Paper award at the ACM Recommender Systems 2012 Conference and the European Conference of Machine Learning (ECML/PKDD) 2013.
Moreover we have developed a novel context aware app recommendation app for the Android Mobile OS called Frappe. Frappe , which is freely available and can be downloaded at Google Play on any Android device, uses the state-of-the-art context-aware collaborative filtering methods developed in this project.
At the end of this project we also conducted a user study to evaluate in-the-wild the usefulness and impact of the use of context on the Frappe app recommendation service.
During this two-year period the key results achieved are as follows:
a) 5 novel algorithms for context-aware collaborative filtering that outperformed the current state-of-the-art. The models developed incorporate context information an also optimize Information Retrieval metrics such as Mean Reciprocal Rank and Mean Average Precision. This leads to very significant performance improvements over the state-of-the-art. Care was taken so that the developed algorithms scale linearly with the size of the data and can thus be used on large-scale data.
b) We designed, implemented and evaluated a context-aware mobile app recommendation app. Frappe is an Android app that recommends users other apps that they might find interesting it is using the contextual information available from the mobile phone sensors to determine the apps matching the current context of the user. Frappe is based on an implementation of the algorithms developed at this project. It is available for free for all Android devices and has been downloaded over 4000 times from the Google Play store.
c) We have conducted a live user study on the Frappe mobile app recommendation service to evaluate the impact and utility of context in the app recommendation domain. This study in particular revealed
d) The research results obtained through all this work have been published at top tier conference and workshops in fields related to the Machine Learning, information retrieval, HCI and Recommender Systems. Two of our publications where awarded with Best Paper awards:
• ACM Recommender Systems 2012 Best Research Paper Award. Yue Shi, Alexandros Karatzoglou, Linas Baltrunas, Martha Larson, Nuria Oliver, and Alan Hanjalic “CLiMF: Learning to Maximize Reciprocal Rank with Collaborative Less-is-More Filtering” ACM RecSys 2012, Dublin, Ireland
• European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases 2013 (ECMLPKDD) Best Student Paper Award Julien Delporte, Alexandros Karatzoglou, Tomasz Matuszczyk, Stéphane Canu Socially Enabled Preference Learning from Implicit Feedback Data