Full Record

New Search | Similar Records

Title Towards a Social Web based solution to bootstrap new domains in cross-domain recommendations:
Publication Date
Degree Level masters
University/Publisher Delft University of Technology
Abstract Most recommender systems recommend items from a single domain. However, usually users’ preferences span across multiple domains. Cross-domain recommender systems can successfully recommend items in multiple domains when there is knowledge about the user’s preferences for items in at least one of the domains and when there is knowledge about relationships between domains. But when a new domain is added to a cross-domain recommender system, this knowledge usually lacks and giving cross-domain recommendations is not a trivial problem anymore. Current approaches uses content-based relations to bootstrap new domains in cross-domain recommendations. In this thesis we propose a new model that transfers existing users’ preference based relations between domains from an auxiliary Social Web system to a cross-domain recommender system in which a new domain needs to be bootstrapped. In a case study on the Open Images dataset we researched this solution to get insight in how well the model works and whether it has potential for widespread usage.
Subjects/Keywords cross-domain recommendations; cold-start recommendations; recommender systems; Social Web; Open Images; YouTube; collaborative filtering; users' preferences
Contributors Bozzon, A.; Houben, G.J.P.M.; Hindriks, K.V.
Language en
Rights (c) 2014 Rentmeester, M.
Country of Publication nl
Record ID oai:tudelft.nl:uuid:0c5d66b1-1470-44e1-b1d6-e02de00c9b8b
Other Identifiers uuid:0c5d66b1-1470-44e1-b1d6-e02de00c9b8b
Repository delft
Date Indexed 2017-06-19

Sample Search Hits | Sample Images

…of the two most used approaches in recommender systems . A cross-domain recommendation using collaborative filtering in a situation of overlap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 13 3.1 The new proposed model…

…recommending CDs. Cold-start cross-domain recommendations can be computed in two ways, contentbased and using collaborative filtering. Current approaches use content-based methods to relate items from the new domain to items from the existing domains, for…

…instance using Semantic Web based solutions or tags. As it is known that collaborative filtering is the best method to compute recommendations in a single domain we decided to explore collaborative filtering (read: users’ preferences) to relate…

…items from the new domain to items from the existing domains. However a constraint of the situation that we research is that there is no previous knowledge of users’ preferences. That makes it impossible to do standard collaborative filtering and we…

…between domains. Most research addressed the scenario where some overlap is available. Collaborative filtering is the most used technique to solve that scenario. How this works is explained in the following example. Let us consider a cross-domain…

…recommender system containing the domains books and movies and a user that prefers the book "Harry Potter". This user wants the system to recommend movies to her. A collaborative filtering technique will look for users that also like the book "…

…Harry Potter" and look for their movie preferences. Assuming one of such user expressed a preference for "Lord of the Rings", then collaborative filtering assumes that the user that wants recommendations from the domain movies might also…

…problem that is relevant in both content-based recommender systems and collaborative filtering recommender systems; 2. New item: When a new item is added to a recommender system, no one has expressed preferences in that particular item yet. This is…