Fine-Tuned BERT Algorithm-Based Automatic Query Expansion for Enhancing Document Retrieval System

Deepak Vishwakarma

doi:10.1007/s12559-024-10354-5

2024-12-05

PRE

Deepak Vishwakarma

Suresh Kumar *

DOI：10.1007/s12559-024-10354-5

原文链接

原文求助

摘要

En 中文

Online retrieval systems are mostly web-based, which makes document collecting more dynamic or fluid than in traditional information retrieval systems. With the web growing in size every day, finding meaningful information on it using a search query consisting of only a few keywords which has become increasingly difficult. One important factor in making Internet searches better is query expansion, or QE. Manual query expansion method involves the user adding terms to the query, which takes a long time but produces good results. However, the automatic query expansion (AQE) method determines the best statements with minimal time consumption. Therefore, to improve document retrieval system, a fine-tuned BERT algorithm is developed for automatic query expansion. Initially, the input text was augmented using embedding augmentation (EA) approach. The augmented text was pre-processed using tokenization, normalization, splitting, stemming, stop word removal, as well as lemmatization. Then extracting the technical keywords from the pre-processed text using co-occurrence statistical information. After extracting the keywords, a fine-tuned BERT model is utilized for expanding the query to improve document retrieval system. The hyper parameters present in the BERT was tuned using frilled lizard optimization to enhance the performance of the BERT model. Proposed model provides 92% accuracy, 95% precision, and 95.6% recall. Thus, a fine-tuned BERT model minimizing query-document mismatch and thereby improving retrieval performance.

Keyword:

A fine-tuned BERT

Automatic query expansion

Embedding augmentation (EA)

Co-occurrence statistical information

Frilled lizard optimization

Tokenization

Normalization

Splitting