Collaborative Filtering by Using a Graph Partitioning Method on Binary Data(APCIM2009 Best Papers)

Takanobu NAKAHARA, Hiroyuki MORITA
2010 Nihon Joho Keiei Gakkaishi  
Coltaborative filtering methods are f}eq"ently usedfor pft?dicting users'ptefenences in recommender systems, such as those usedfor recommending movies, m"sic, or articles, 71hese methocts have a large impact on businesses, beca"se the votume ofcontent they qfi7eris tremendous, and it is important to sumport users'ability to make injiormed ehoiees fhom among the available content, 7b increase sales and improve customer loyalty, many e-commerce companies, such as Amazon and Aleij7ix, have adopted
more » ... recommender systems, Howeven these companies generally rely on user ratings for the content they ctffen and it is usually d(filcult or eupensive to obtain such ratings data. Ifence, we need a high-q"ality necommender system that "ses only binary data, such as historicat put=hasing data, without ratings. Binary data, howeven is vetly simple, and it is theiefore dij)icult to espress a relationship in detail between "sers and items by using only such simple methods. 711iis paperpmposes a recommender system based on a gmph-partitioning methed to sotve the probtems through a two-phase approach: Vle generate a model that eupresses the retationships between vario"s items and imptements an appropriate grouping by "sing a graph-partitioning method, i-le then use our proposed algorithms to determine accurate recommendations. A comparison ofo"r results with those obtainedfhom truditional ntethods reveals that our method is mone practicalfor businesses usage.
doi:10.20627/jsim.30.4_125 fatcat:siifgxv5ovdspbnnkuwn7wsqxe