EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC 4.0

Improving Neural Retrieval with Contrastive Learning

Murali Mohana Krishna Dandu, Rahul Arulkumaran, Nishit Agarwal, Anshika Aggarwal, Prof.(Dr) Punit Goel

Modern Dynamics: Mathematical Progressions · 2024

DOI: 10.36676/mdmp.v1.i2.30 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Modern Dynamics: Mathematical Progressions, 2024. Verfasst von Murali Mohana Krishna Dandu und 4 weiteren Autor:innen. Die Arbeit wurde bisher 2-mal zitiert (Crossref) und nennt 26 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

In recent years, neural retrieval models have shown remarkable progress in improving the efficiency and accuracy of information retrieval systems. However, challenges remain in effectively differentiating relevant from irrelevant documents, particularly in fine-grained distinctions. Contrastive learning, a method widely used in self-supervised learning, offers a promising approach to address this issue by enabling models to better capture the nuances between positive and negative examples. This paper explores the integration of contrastive learning into neural retrieval frameworks, with a focus on improving document ranking and relevance scoring. By applying contrastive learning, the model learns to map similar queries and documents closer in the latent space while pushing dissimilar ones apart. We highlight key advantages, such as enhanced generalization to unseen queries and better contextual understanding of user intent. The study demonstrates that contrastive learning not only improves retrieval accuracy but also reduces the computational overhead typically associated with large-scale retrieval systems. Through experiments on benchmark datasets, we show a significant improvement in retrieval performance over traditional neural retrieval methods. This work presents contrastive learning as a crucial enhancement to modern retrieval systems, offering practical insights for its implementation in various search-related applications, including web search, recommendation systems, and question-answering tasks. These findings suggest that contrastive learning can pave the way for more efficient and precise information retrieval, making it a vital tool for the future of neural search engines.

Zitieren

APA

Dandu, M. M. K., Arulkumaran, R., Agarwal, N., Aggarwal, A., & Goel, P. P. (2024). Improving Neural Retrieval with Contrastive Learning. Modern Dynamics: Mathematical Progressions. https://doi.org/10.36676/mdmp.v1.i2.30

Vancouver

Dandu MMK, Arulkumaran R, Agarwal N, Aggarwal A, Goel PP. Improving Neural Retrieval with Contrastive Learning. Modern Dynamics: Mathematical Progressions. 2024. doi:10.36676/mdmp.v1.i2.30

BibTeX
@article{Dandu2024,
  author = {Dandu, Murali Mohana Krishna and Arulkumaran, Rahul and Agarwal, Nishit and Aggarwal, Anshika and Goel, Prof.(Dr) Punit},
  title = {Improving Neural Retrieval with Contrastive Learning},
  journal = {Modern Dynamics: Mathematical Progressions},
  year = {2024},
  doi = {10.36676/mdmp.v1.i2.30}
}
Kostenlos starten →