Performance Assessment of an Outlier Detection–Based Approach to Information Retrieval and Query Expansion

Abstract

Information retrieval (IR) systems face persistent challenges in addressing the complexity of user queries and large-scale document collections. This study presents a performance assessment of an outlier detection–based approach to query expansion aimed at enhancing retrieval precision and relevance. The proposed Gradient-Based Dynamic Query Expansion (GBDQE) framework integrates outlier filtering to eliminate noisy or irrelevant terms that typically reduce the effectiveness of traditional feedback mechanisms. A series of experiments on standard benchmark datasets were conducted to evaluate the model against baseline methods and established approaches such as WT2G and CWSBQE. Experimental results reveal that the outlier detection–driven GBDQE approach delivers significant improvements, achieving a 15% increase in Mean Average Precision (MAP), a 20% gain in Precision at 10 (P@10), and notable enhancements in F-measure and GM_MAP scores. These outcomes confirm the robustness of the proposed method and highlight its potential to substantially improve IR performance, thereby advancing the quality of user search experiences in modern digital environments.