%0 Journal Article %1 citeulike:4233245 %A Lekakos, George %A Caravelas, Petros %C Hingham, MA, USA %D 2008 %I Kluwer Academic Publishers %J Multimedia Tools Appl. %K hybrid-recommendation recommender %N 1-2 %P 55--70 %R 10.1007/s11042-006-0082-7 %T A hybrid approach for movie recommendation %U http://dx.doi.org/10.1007/s11042-006-0082-7 %V 36 %X Collaborative and content-based filtering are the major methods in recommender systems that predict new items that users would find interesting. Each method has advantages and shortcomings of its own and is best applied in specific situations. Hybrid approaches use elements of both methods to improve performance and overcome shortcomings. In this paper, we propose a hybrid approach based on content-based and collaborative filtering, implemented in MoRe, a movie recommendation system. We also provide empirical comparison of the hybrid approach to the base methods of collaborative and content-based filtering and draw useful conclusions upon their performance.

@article{citeulike:4233245, abstract = {{Collaborative and content-based filtering are the major methods in recommender systems that predict new items that users would find interesting. Each method has advantages and shortcomings of its own and is best applied in specific situations. Hybrid approaches use elements of both methods to improve performance and overcome shortcomings. In this paper, we propose a hybrid approach based on content-based and collaborative filtering, implemented in MoRe, a movie recommendation system. We also provide empirical comparison of the hybrid approach to the base methods of collaborative and content-based filtering and draw useful conclusions upon their performance.}}, added-at = {2018-03-19T12:24:51.000+0100}, address = {Hingham, MA, USA}, author = {Lekakos, George and Caravelas, Petros}, biburl = {https://www.bibsonomy.org/bibtex/2d33ed224ea545a5ae80e2bab2ca41973/aho}, citeulike-article-id = {4233245}, citeulike-linkout-0 = {http://portal.acm.org/citation.cfm?id=1327423.1327424}, citeulike-linkout-1 = {http://dx.doi.org/10.1007/s11042-006-0082-7}, comment = {(private-note)Jung-Shan Lee, 2955-092 The paper ” A Hybrid Approach forMovie Recommendation” by George Lekakos and Petros Caravelas summarized thebenefits and shortcomings of the content-based and collaborative filtering,then proposed a hybrid approach to implement the movie recommendation system,MoRe, based on those two methods. Thepaper also addresses empirical comparison between the hybrid approach and thecollaborative filtering and content-based. In that, it provides the usefulconclusion for their performance. Collaborative filtering andcontent-based filtering are two of the most popular recommendation methods thatpredict items concerning users' interest with previous evaluated items anditems without evaluation. To illustrate a hybrid approach, the paper first lookedinto the drawbacks of these two recommendation methods. Content-basedrecommendation can be applied when the item can be described by feature;however, content-based filtering serves a problem that it usually becomesoverspecialized. On the other hand, the most important shortcoming incollaborative filtering is low accuracy and cold-start. User tends to rate onlysmall portion of the items, which is difficult to compute the similaritybetween users. Also, no prediction can be made toward the items without ratingor new users. Combing these two methodscan make the complementary among these two, therefore comes the hybridapproach. The paper believe that the combination of the two methods can improvethe performance and overcome the shortcomings. The movie recommender system, MoRe,is a Web-based system. It collects user rated movies on a one-to-five scalethrough a GUI interface. When the most popular movie being selected, itincreased the possibility to collect the respective ratings, and it is mostpossible that these movies have been seen by the new users. The system acquired the user ratings from theMovieLens dataset. The MovieLens dataset only contains the name and theproduction year of the movie, hence a web crawler was created to gain the existedmovie description from the Internet Movie Database (IMDB). Then, it creates thesimilar-movie set at an off-line mode to speed up the system. There are four algorithms which havebeen implemented and compared in this paper: * Collaborative Filtering * Content-Based Recommendations * Hybrid (Substitute) Recommendations * Hybrid (Switching) Recommendations The hybrid approaches, substitute andswitching, use collaborative filtering as the primary method and trigger thecontent-based recommendation when the threshold has been reached. The differentbetween these two approaches is that substitute hybrid method triggered thecontent-based recommendation while the number of similar neighbors below five,but the switching hybrid method switched to content-based approach while theavailable ratings below a certain threshold. The purpose of the two approachesis that it has been expect to increase both prediction accuracy and theprediction coverage. The purpose of the experimentalevaluation is that comparing the two hybrid algorithm against each other, aswell as against with the base algorithms. While doing the evaluation, the MeanAbsolute Error (MAE) is a suitable measure of precision for the systems thatuser numerical user rating and numerical predictions. According to theexperiment, there are two conclusions have been reached. First, the number offeatures, which representing the movie, is the main factor of the accuracy. Inother words, the more features the movies have and be used, the lesspossibility of errors may occur and the more accurate the recommendations are. Second,although for the very large data set, the scalability and similarity problemmay be significant, it still be the most accurate recommendation methods. Whichrecommendation method should be implemented into the system depends on thetrade-off between accuracy, coverage and run-time. This paper related to the ” HybridRecommendation Topic” in several aspects. First, it actually implemented thehybrid algorithms in the system. Second, it compares the performance amongbasic recommendation methods and the hybrid approaches. Third, the evaluationresults point out that although the basic recommendation methods (collaborativefiltering and content-based recommendation) have some drawbacks, the hybridapproach may not always be the best policy due to the balancing issue amongaccuracy, prediction coverage and system run-time. This can give us somethought that hybrid approach indeed solves the problems in the baserecommendation; however, it might not be suitable in every condition.}, doi = {10.1007/s11042-006-0082-7}, interhash = {e49ae41a39cdb558b5b53581cbc09958}, intrahash = {d33ed224ea545a5ae80e2bab2ca41973}, issn = {1380-7501}, journal = {Multimedia Tools Appl.}, keywords = {hybrid-recommendation recommender}, number = {1-2}, pages = {55--70}, posted-at = {2009-04-29 15:26:02}, priority = {1}, publisher = {Kluwer Academic Publishers}, timestamp = {2018-03-19T12:24:51.000+0100}, title = {{A hybrid approach for movie recommendation}}, url = {http://dx.doi.org/10.1007/s11042-006-0082-7}, volume = 36, year = 2008 }