TY - GEN
T1 - The impact of recommender systems on item-, user-, and rating-diversity
AU - Kowalczyk, W.
AU - Szlávik, Z.
AU - Schut, M. C.
PY - 2012
Y1 - 2012
N2 - Recommender systems are increasingly used for personalised navigation through large amounts of information, especially in the e-commerce domain for product purchase advice. Whilst much research effort is spent on developing recommenders further, there is little to no effort spent on analysing the impact of them - neither on the supply (company) nor demand (consumer) side. In this article, we investigate the diversity impact of a movie recommender. We define diversity for different parts of the domain and measure it in different ways. The novelty of our work is the usage of real rating data (from Netflix) and a recommender system for investigating the (hypothetical) effects of various configurations of the system and users' choice models. We consider a number of different scenarios (which differ in agents' choice models), run extensive simulations, analyse various measurements regarding experimental validation and diversity, and report on selected findings. The scenarios are an essential part of our work, since these can be influenced by the owner of the recommender once deployed. This article contains an overview of related work on data-mining, multi-agent systems, movie recommendation and measurement of diversity; we explain different agents' choice models (which are used to simulate how users react to recommenders); and we report on selected findings from an extensive series of simulation experiments that we ran with real usage data (from Netflix).
AB - Recommender systems are increasingly used for personalised navigation through large amounts of information, especially in the e-commerce domain for product purchase advice. Whilst much research effort is spent on developing recommenders further, there is little to no effort spent on analysing the impact of them - neither on the supply (company) nor demand (consumer) side. In this article, we investigate the diversity impact of a movie recommender. We define diversity for different parts of the domain and measure it in different ways. The novelty of our work is the usage of real rating data (from Netflix) and a recommender system for investigating the (hypothetical) effects of various configurations of the system and users' choice models. We consider a number of different scenarios (which differ in agents' choice models), run extensive simulations, analyse various measurements regarding experimental validation and diversity, and report on selected findings. The scenarios are an essential part of our work, since these can be influenced by the owner of the recommender once deployed. This article contains an overview of related work on data-mining, multi-agent systems, movie recommendation and measurement of diversity; we explain different agents' choice models (which are used to simulate how users react to recommenders); and we report on selected findings from an extensive series of simulation experiments that we ran with real usage data (from Netflix).
UR - https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=84255192296&origin=inward
UR - https://www.ncbi.nlm.nih.gov/pubmed/37313295
U2 - https://doi.org/10.1007/978-3-642-27609-5_17
DO - https://doi.org/10.1007/978-3-642-27609-5_17
M3 - Conference contribution
C2 - 37313295
VL - 7103 LNAI
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 261
EP - 287
BT - Agents and Data Mining Interaction - 7th International Workshop, ADMI 2011, Revised Selected Papers
T2 - 7th International Workshop on Agents and Data Mining Interaction, ADMI 2011
Y2 - 2 May 2011 through 6 May 2011
ER -