Serendipity in recommender systems
Published inJyväskylä studies in computing
The number of goods and services (such as accommodation or music streaming) offered by e-commerce websites does not allow users to examine all the available options in a reasonable amount of time. Recommender systems are auxiliary systems designed to help users ﬁnd interesting goods or services (items) on a website when the number of available items is overwhelming. Traditionally, recommender systems have been optimized for accuracy, which indicates how often a user consumed the items recommended by system. To increase accuracy, recommender systems often suggest items that are popular and suitably similar to items these users have consumed in the past. As a result, users often lose interest in using these systems, as they either know about the recommended items already or can easily ﬁnd these items themselves. One way to increase user satisfaction and user retention is to suggest serendipitous items. These items are items that users would not ﬁnd themselves or even look for, but would enjoy consuming. Serendipity in recommender systems has not been thoroughly investigated. There is not even a consensus on the concept’s deﬁnition. In this dissertation, serendipitous items are deﬁned as relevant, novel and unexpected to a user. In this dissertation, we (a) review different deﬁnitions of the concept and evaluate them in a user study, (b) assess the proportion of serendipitous items in a typical recommender system, (c) review ways to measure and improve serendipity, (d) investigate serendipity in cross-domain recommender systems (systems that take advantage of multiple domains, such as movies, songs and books) and (e) discuss challenges and future directions concerning this topic. We applied a Design Science methodology as the framework for this study and developed four artifacts: (1) a collection of eight variations of serendipity deﬁnition, (2) a measure of the serendipity of suggested items, (3) an algorithm that generates serendipitous suggestions, (4) a dataset of user feedback regarding serendipitous movies in the recommender system MovieLens. These artifacts are evaluated using suitable methods and communicated through publications. ...
PublisherUniversity of Jyväskylä
serendipisyys recommender systems serendipity relevance novelty unexpectedness personalization evaluation recommendation algorithms evaluation metrics offline experiments user study serendipity metrics Käyttäjätutkimus suosittelujärjestelmät verkkopalvelut verkkokauppa täsmämarkkinointi algoritmit arviointi sattuma relevanssi
MetadataShow full item record
- Väitöskirjat 
Showing items with similar title or keywords.
Kotkov, Denis; Veijalainen, Jari; Wang, Shuaiqiang (SCITEPRESS, 2016)Most recommender systems suggest items similar to a user profile, which results in boring recommendations limited by user preferences indicated in the system. To overcome this problem, recommender systems should suggest ...
How does serendipity affect diversity in recommender systems? A serendipity-oriented greedy algorithm Kotkov, Denis; Veijalainen, Jari; Wang, Shuaiqiang (Springer Wien, 2020)Most recommender systems suggest items that are popular among all users and similar to items a user usually consumes. As a result, the user receives recommendations that she/he is already familiar with or would find anyway, ...
Kotkov, Denis; Veijalainen, Jari; Wang, Shuaiqiang (SCITEPRESS Science And Technology Publications, 2017)Most recommender systems suggest items to a user that are popular among all users and similar to items the user usually consumes. As a result, a user receives recommendations that she/he is already familiar with or would ...
Pandey, Gaurav; Kotkov, Denis; Semenov, Alexander (ACM Press, 2018)Most recommender algorithms are designed to suggest relevant items, but suggesting these items does not always result in user satisfaction. Therefore, the efforts in recommender systems recently shifted towards serendipity, ...
Kotkov, Denis; Wang, Shuaiqiang; Veijalainen, Jari (Springer International Publishing AG, 2017)Cross-domain recommender systems use information from source domains to improve recommendations in a target domain, where the term domain refers to a set of items that share attributes and/or user ratings. Most works ...