Salvato in:
Dettagli Bibliografici
Autore principale: Merrick, Luke
Natura: Preprint
Pubblicazione: 2024
Soggetti:
Accesso online:https://arxiv.org/abs/2407.18887
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866914888895430656
author Merrick, Luke
author_facet Merrick, Luke
contents Recent studies of large-scale contrastive pretraining in the text embedding domain show that using single-source minibatches, rather than mixed-source minibatches, can substantially improve overall model accuracy. In this work, we explore extending training data stratification beyond source granularity by leveraging a pretrained text embedding model and the classic k-means clustering algorithm to further split training data apart by the semantic clusters within each source. Experimentally, we observe a notable increase in NDCG@10 when pretraining a BERT-based text embedding model on query-passage pairs from the MSMARCO passage retrieval dataset. Additionally, we conceptually connect our clustering approach to both the Topic Aware Sampling (TAS) aspect of the TAS-B methodology and the nearest-neighbor-based hard-negative mining aspect of the ANCE methodology and discuss how this unified view motivates future lines of research on the organization of contrastive pretraining data.
format Preprint
id arxiv_https___arxiv_org_abs_2407_18887
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Embedding And Clustering Your Data Can Improve Contrastive Pretraining
Merrick, Luke
Machine Learning
Computation and Language
Recent studies of large-scale contrastive pretraining in the text embedding domain show that using single-source minibatches, rather than mixed-source minibatches, can substantially improve overall model accuracy. In this work, we explore extending training data stratification beyond source granularity by leveraging a pretrained text embedding model and the classic k-means clustering algorithm to further split training data apart by the semantic clusters within each source. Experimentally, we observe a notable increase in NDCG@10 when pretraining a BERT-based text embedding model on query-passage pairs from the MSMARCO passage retrieval dataset. Additionally, we conceptually connect our clustering approach to both the Topic Aware Sampling (TAS) aspect of the TAS-B methodology and the nearest-neighbor-based hard-negative mining aspect of the ANCE methodology and discuss how this unified view motivates future lines of research on the organization of contrastive pretraining data.
title Embedding And Clustering Your Data Can Improve Contrastive Pretraining
topic Machine Learning
Computation and Language
url https://arxiv.org/abs/2407.18887