Loading...

A Novel Metric for Evaluation of Recommender Systems

Izadi, Maliheh | 2014

1489 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 46849 (19)
  4. University: Sharif University of Technology
  5. Department: Computer Engineering
  6. Advisor(s): Jalili, Mahdi
  7. Abstract:
  8. The World Wide Web has been experiencing a massive growth regarding its content and users in recent years; therefore the need for effective means of accessing and processing available items has attracted a wide range of researchers and industries. Recommender systems has emerged to help both users to find what they may be interested in and the producers to sell their products more efficiently. As the number of these techniques grow, the need to evaluate them properly increases as well. However the proposed evaluation metrics are very diverse and often inconsistent with each other. Although there had been immense research in this field, there is no united and proper approach for evaluation of these systems to assess their performance from different points of view including precision, diversity, novelty and coverage of their recommendations. In other words, despite the divers range of available evaluation metrics, their relation or distinction has not been studied explicitly yet. Therefore it is only logical that there is a desperate need for a novel measure or framework which would balances the trade-off between the evaluation metrics and utilizes their suitable features to increase users’ satisfaction of the system. In this thesis, we provide a novel and extensible framework for evaluation of recommender techniques by carefully studying the concept of available measures. Our method consist of decreasing the multi-dimensional problem to one dimension by fixing the value of one metric and calculating maximum and/or minimum bounds of other metrics in the best case scenario, in order to measure how well they perform considering the performance of an ideal recommender for each specific dataset. Then we combine the normalized measures to provide a novel unified metric for evaluation of recommender systems considering various aspects of these systems
  9. Keywords:
  10. Recommender System ; Accuracy ; Freshness ; Coverage ; Diversity Recommender System ; Collaborative Filtering ; Evaluation Metrics

 Digital Object List

 Bookmark

...see more