Towards Personalized Similarity Search for Vector Databases

TY - GEN

T1 - Towards Personalized Similarity Search for Vector Databases

AU - Mahrík, Marek

AU - Šikyňa, Matúš

AU - Mic, Vladimir

AU - Zezula, Pavel

N1 - Conference code: 17

PY - 2025

Y1 - 2025

N2 - The importance of similarity search has become prominent in the fast-evolving vector databases, which apply content embedding techniques on complex data to produce and manage large collections of high-dimensional vectors. Processing of such data is only possible by using a similarity function for storage, structure, and retrieval. However, if multiple users access the collection, their views on similarity can differ as similarity, in general, is subjective and context-dependent. In this article, we elaborate on the problem of a similarity search engine implementation, where users use a common index but search with personalised views of similarity, implemented by a possibly different similarity model. Specifically, we define a foundational theoretical framework and conduct experiments on real-life data to confirm the viability of such an approach. The experiments also indicate future research directions needed to propose and implement an effective and efficient personalised similarity search engine.

AB - The importance of similarity search has become prominent in the fast-evolving vector databases, which apply content embedding techniques on complex data to produce and manage large collections of high-dimensional vectors. Processing of such data is only possible by using a similarity function for storage, structure, and retrieval. However, if multiple users access the collection, their views on similarity can differ as similarity, in general, is subjective and context-dependent. In this article, we elaborate on the problem of a similarity search engine implementation, where users use a common index but search with personalised views of similarity, implemented by a possibly different similarity model. Specifically, we define a foundational theoretical framework and conduct experiments on real-life data to confirm the viability of such an approach. The experiments also indicate future research directions needed to propose and implement an effective and efficient personalised similarity search engine.

KW - Personalized similarity

KW - Similarity search

KW - Vector databases

UR - http://www.scopus.com/inward/record.url?scp=105002715222&partnerID=8YFLogxK

U2 - 10.1007/978-3-031-75823-2_11

DO - 10.1007/978-3-031-75823-2_11

M3 - Article in proceedings

SN - 978-3-031-75822-5

T3 - Lecture Notes in Computer Science

SP - 126

EP - 139

BT - Similarity Search and Applications - 17th International Conference, SISAP 2024, Proceedings

A2 - Chávez, Edgar

A2 - Kimia, Benjamin

A2 - Lokoč, Jakub

A2 - Patella, Marco

A2 - Sedmidubsky, Jan

PB - Springer

CY - Cham

T2 - 17th International Conference, SISAP 2024

Y2 - 4 June 2024 through 6 June 2024

ER -