Using Generative Adversarial Networks for Relevance Evaluation of Search Engine Results

Abstract

© 2020 IEEE. In the article a new approach to the problem of relevance evaluation of the search engine results, based on generative adversarial networks (GAN), is proposed. To improve the quality of search, the generative adversarial networks are used to distinguish between relevant and irrelevant search results.We used a simplistic model based on fully automated reference results selection and multi-layered generator and discriminator networks with dense layers. The queries needed to generate the reference results were themselves generated by a GPT-2 like network using the same text corpus as a source, to make them potentially relevant to the search space.The results clearly demonstrate the principal possibility and feasibility of using the described approach, despite the fact of used models being simplistic.