TY - GEN

T1 - Unsupervised multi-index semantic hashing

AU - Hansen, Christian

AU - Hansen, Casper

AU - Simonsen, Jakob Grue

AU - Alstrup, Stephen

AU - Lioma, Christina

N1 - Publisher Copyright: © 2021 ACM.

PY - 2021

Y1 - 2021

N2 - Semantic hashing represents documents as compact binary vectors (hash codes) and allows both efficient and effective similarity search in large-scale information retrieval. The state of the art has primarily focused on learning hash codes that improve similarity search effectiveness, while assuming a brute-force linear scan strategy for searching over all the hash codes, even though much faster alternatives exist. One such alternative is multi-index hashing, an approach that constructs a smaller candidate set to search over, which depending on the distribution of the hash codes can lead to sub-linear search time. In this work, we propose Multi-Index Semantic Hashing (MISH), an unsupervised hashing model that learns hash codes that are both effective and highly efficient by being optimized for multi-index hashing. We derive novel training objectives, which enable to learn hash codes that reduce the candidate sets produced by multi-index hashing, while being end-to-end trainable. In fact, our proposed training objectives are model agnostic, i.e., not tied to how the hash codes are generated specifically in MISH, and are straight-forward to include in existing and future semantic hashing models. We experimentally compare MISH to state-of-the-art semantic hashing baselines in the task of document similarity search. We find that even though multi-index hashing also improves the efficiency of the baselines compared to a linear scan, they are still upwards of 33% slower than MISH, while MISH is still able to obtain state-of-the-art effectiveness.

AB - Semantic hashing represents documents as compact binary vectors (hash codes) and allows both efficient and effective similarity search in large-scale information retrieval. The state of the art has primarily focused on learning hash codes that improve similarity search effectiveness, while assuming a brute-force linear scan strategy for searching over all the hash codes, even though much faster alternatives exist. One such alternative is multi-index hashing, an approach that constructs a smaller candidate set to search over, which depending on the distribution of the hash codes can lead to sub-linear search time. In this work, we propose Multi-Index Semantic Hashing (MISH), an unsupervised hashing model that learns hash codes that are both effective and highly efficient by being optimized for multi-index hashing. We derive novel training objectives, which enable to learn hash codes that reduce the candidate sets produced by multi-index hashing, while being end-to-end trainable. In fact, our proposed training objectives are model agnostic, i.e., not tied to how the hash codes are generated specifically in MISH, and are straight-forward to include in existing and future semantic hashing models. We experimentally compare MISH to state-of-the-art semantic hashing baselines in the task of document similarity search. We find that even though multi-index hashing also improves the efficiency of the baselines compared to a linear scan, they are still upwards of 33% slower than MISH, while MISH is still able to obtain state-of-the-art effectiveness.

KW - Multi-index hashing

KW - Semantic hashing

KW - Similarity search

UR - http://www.scopus.com/inward/record.url?scp=85104220631&partnerID=8YFLogxK

U2 - 10.1145/3442381.3450014

DO - 10.1145/3442381.3450014

M3 - Article in proceedings

AN - SCOPUS:85104220631

T3 - The Web Conference 2021 - Proceedings of the World Wide Web Conference, WWW 2021

SP - 2879

EP - 2889

BT - The Web Conference 2021 - Proceedings of the World Wide Web Conference, WWW 2021

PB - Association for Computing Machinery, Inc

T2 - 2021 World Wide Web Conference, WWW 2021

Y2 - 19 April 2021 through 23 April 2021

ER -