Automatic Query Expansion based Document Retrieval System using Hyper Tuned Graph Enhanced Bi-directional Encoder Representation from Transformers with Manifold Ranking Algorithm

Abstract

Document retrieval system automatic query expansion (AQE) adds relevant phrases to user searches to improve search results. AQE enhances recollection but introduces irrelevant phrases, increases computing complexity, and slows retrieval. Machine learning and deep learning models may be computationally expensive and sluggish in AQE-based document retrieval. These models also need large datasets and careful tweaking, which may be resource-intensive. The suggested AQE model uses HT-GEBERT Enhanced to generate contextual query expansions and Manifold Ranking to order words by relevance. This combination improves document retrieval accuracy and reduces query drift. Start by augmenting the corpus user query using Pseudo Adversarial Embedding (PAE) to increase data variety for robust model training. To maintain model analysis consistency, the augmented text and response are pre-processed using tokenization, lemmatization, acronym expansion, stop word removal, hyperlink removal, and spell correction. Next, Eccentricity-Based Keyword Extraction (EKE) extracts key phrases. After keyword extraction, the Hyper-Tuned Graph Enhanced Bidirectional Encoder Representation from Transformers (HT-GEBERT) model vectorises words and optimizes its hyper parameters using the Artificial Gorilla Troops Optimization Algorithm. Finally, a ranking-based query expansion approach re-ranks the phrase using a manifold ranking algorithm and splits the text into pieces for cosine similarity relevance assessment to obtain the document. The suggested method achieves 94% accuracy, 94.5% PPV, and 5.5% FDR in dataset 1. The suggested method uses query expansion, embedding, and optimization to retrieve documents.