%0 Conference Paper %1 claypool99 %A Claypool, M. %A Gokhale, A. %A Miranda, T. %A Murnikov, P. %A Netes, D. %A Sartin, M. %B Proceedings of the ACM SIGIR '99 Workshop on Recommender Systems: Algorithms and Evaluation %C Berkeley, California %D 1999 %I ACM %K aggregation collaborative_filtering content_based hybrid_systems recommender_systems survey %T Combining content-based and collaborative filters in an online newspaper

@inproceedings{claypool99, added-at = {2009-06-22T17:28:38.000+0200}, address = {Berkeley, California}, author = {Claypool, M. and Gokhale, A. and Miranda, T. and Murnikov, P. and Netes, D. and Sartin, M.}, biburl = {https://www.bibsonomy.org/bibtex/20588ff378faea30f399acd65da69711b/lefteris8}, booktitle = {Proceedings of the ACM SIGIR '99 Workshop on Recommender Systems: Algorithms and Evaluation}, citeulike-article-id = {430353}, comment = {Collaborative filtering is inefficient for: early rater problem, sparsity problem, gray sheep. A content-based filter analyzes items rated by an individual user and uses the content of the items as well as the provided ratings to build a profile to compare to other non-rated items to recommend additional items of interest. By using a combination of content-based and collaborative filters we can realize the benefits of content-based filters which include early predictions that cover all items and users, while gaining the benefits of accurate collaborative filtering predictions as the number of users and ratings increases,. We provide a unique approach to combining content-based and collaborative filtering by basing a prediction on a weighted average of the content-based prediction and the collaborative prediction. In addition, our approach allows for the content-based and collaborative weights to be computed on a per-item basis. The weights are adjusted based on the strength of each prediction. As the number of users and ratings for the item increase, the collaborative filter is (usually) weighted more heaviy, increasing the overall accuracy of the prediction. Our approach is not a hybrid approach, meaning that the basis for the content-based and collaborative predictions are kept separate. This allows us to benefit from individual advances made to either collaborative or content-based filters since there is no inter-dependency between the two content components. For example, individual improvements to pure collaborative filtering algorithms can be fully realized. We build upon the work of the pure collaborative filtering algorithms published that compute similarities between users usaing a Pearson correlation coefficient [4, 21, 9]. Our content-based filtering algorithms match article keywords to keywords in the user profile (we use the overlap coefficient). Predicitions from the three interest indicators, explicit keywords, implicit keywords and newspaper section, are combined via matching functions with the article keywords and section. Until we determine appropriate weights for each content-match through extensive user studies, we give each an equal weight of \$1/3\$. Vogt et al showed that a simple linear combination of scores returned by different Information Retrieval agents can improve the performance of those individual systems on new documents [23]. We build upon their work by combining our collaborative filtering prediction with our content-based prediction using a weighted average. The trick is to come up with the weights that result in the most accurate prediction. The collaborative filtering predictions are more inaccurate in cases where the number, agreement or history between users is low [12]. We start by giving equal weight to both the collaborative filtering and the content-based scores for all the users. As users make ratings, we compute the absolute error from the content-based predictions and the collaborative predictions and adjust the weights so as to minimize past error. (it learns as the time passes) In order to evaluate the filtering performance of P-Tango we computed the mean absolute error between the numerical predictions provided by P-Tango and the numerical ratings entered by the user for the same articles.}, interhash = {0f83f197d906697292ef4a51f9d27242}, intrahash = {0588ff378faea30f399acd65da69711b}, keywords = {aggregation collaborative_filtering content_based hybrid_systems recommender_systems survey}, posted-at = {2009-03-11 09:30:54}, priority = {0}, publisher = {ACM}, timestamp = {2009-06-22T17:28:39.000+0200}, title = {Combining content-based and collaborative filters in an online newspaper}, year = 1999 }