Enhancing E-Commerce Query Expansion Using Generative Adversarial Networks (GANs)

In this study, we propose an innovative approach to query expansion (QE) in e-commerce, aiming to enhance the effectiveness of information search. Our method utilizes a generative adversarial network (GAN) called modified QE conditional GAN (mQE-CGAN) to expand queries by generating synthetic queries that incorporate semantic information from textual input. The (mQE-CGAN) framework consists of a generator and a discriminator. The generator is a sequence-to-sequence transformer model trained to produce keywords, while the discriminator is a recurrent neural network model used to classify the generator’s output in an adversarial manner. By incorporating a modified CGAN framework, we introduce various forms of semantic insights from the query-document corpus into the generation process. These insights serve as conditions for the generator model and are instrumental in improving the query expansion task. Through various preliminary experiments, we demonstrate that the utilization of condition structures within the mQE-CGAN framework significantly enhances the semantic similarity between the generated sequences and reference documents. Compared to baseline models, our approach achieves an impressive increase of approximately 5–10% in semantic similarity.