Supporting Serendipitous Recommendations With Knowledge Graphs

Abstract

Abstract: en Recommender systems are commonly designed and evaluated with high precision and accuracy in mind. Optimising systems for these metrics alone can, however, lead to a decrease in overall collection coverage of recommended items, and carries potential to over-emphasize popular content. Notions such as serendipitous discovery and, closely related, novelty and diversity propose that rather than replicating a user's taste or that of a cluster of others they happen to be similar to, recommendation systems should present useful suggestions to users, including novel and diverse items and thus supporting serendipitous discovery. We implement a recommender system based on a knowledge graph of musical items with serendipity, novelty and diversity in mind. Using acoustic features as contextual information for vertices in the graph, we explicitly select content dissimilar from the user's previous experience. We compare our results to a set of baseline algorithms and find that we are able to recommend diverse and novel items.

Keywords: Information Retrieval; Knowledge Graphs; Recommender Systems; serendipity; novelty; diversity; attributed graphs

Title: Supporting Serendipitous Recommendations With Knowledge Graphs
Publication type: Conference paper
Language: English
Funded by: Deutsche Forschungsgemeinschaft; EXC2052/1 Africa Multiple, DFG Project number 390713894
Presented at: 2nd Joint Conference of the Information Retrieval Communities in Europe (CIRCLE 2022), Samatan, Gers, France, 04.-07. Juli 2022
Year: July 2022
Status: Peer reviewed
Relation: Table of contents