A Survey on Deep Text Hashing: Efficient Semantic Text Retrieval with Binary Representation

TL;DR

This survey addresses scalable semantic text retrieval through deep text hashing by organizing methods into a modular taxonomy focused on semantic extraction, hash-code quality, and enabling technologies. It synthesizes reconstruction-based, pseudo-similarity, mutual-information, category-based, and relevance-based strategies, and discusses how few-bit codes, code balance, and low quantization error shape practical performance. An evaluation framework across multiple datasets is presented, highlighting unsupervised versus supervised gains and dataset-specific trends, along with open-source tools and real-world applications from general IR to code retrieval and intelligent systems. The authors also outline challenges and future directions, including benchmark realism, robustness in dynamic settings, integration with large language models, and expanding the use of binary representations beyond retrieval. Overall, the work provides a comprehensive, practice-oriented roadmap for developing and deploying deep text hashing in large-scale semantic search systems.

Abstract

With the rapid growth of textual content on the Internet, efficient large-scale semantic text retrieval has garnered increasing attention from both academia and industry. Text hashing, which projects original texts into compact binary hash codes, is a crucial method for this task. By using binary codes, the semantic similarity computation for text pairs is significantly accelerated via fast Hamming distance calculations, and storage costs are greatly reduced. With the advancement of deep learning, deep text hashing has demonstrated significant advantages over traditional, data-independent hashing techniques. By leveraging deep neural networks, these methods can learn compact and semantically rich binary representations directly from data, overcoming the performance limitations of earlier approaches. This survey investigates current deep text hashing methods by categorizing them based on their core components: semantic extraction, hash code quality preservation, and other key technologies. We then present a detailed evaluation schema with results on several popular datasets, followed by a discussion of practical applications and open-source tools for implementation. Finally, we conclude by discussing key challenges and future research directions, including the integration of deep text hashing with large language models to further advance the field. The project for this survey can be accessed at https://github.com/hly1998/DeepTextHashing.

Figures (9)

• Figure 1: The basic search framework for deep text hashing.
• Figure 2: An illustration of Hamming ball search during KNN search ($K=5$): (a) When the search distance $r=0$, it accesses the same hash bucket as the query text, returning one text. Since the required number is not met, the search range is expanded. (b) Expanding the search distance to $r=1$, it returns six texts, satisfying the required number, and the search stops.
• Figure 3: The main content flow and summary of how to construct a deep text hashing model.
• Figure 4: An illustration of two mainstream deep text hashing frameworks employing the VAE architecture. (1) VDSH kingma2013auto assumes the hash codes follow a Gaussian distribution, while (2) NASH shen2018nash assumes the hash codes follow a Bernoulli distribution.
• Figure 5: An illustration of NbrReg chaidaroon2018deep and PairRec hansen2020unsupervised: (1) NbrReg uses the original text to reconstruct the text of its neighbors, while (2) PairRec uses the text of neighbors to reconstruct the original text.