Discovery: an Explainable AI Powered Academic Search Engine for Knowledge Workers

Abstract

In recent years, transformer models were able to achieve astonishing results in various natural language processing (NLP) tasks. Especially with the rise of ChatGPT and the countless alternatives following its triumphal march, using large language models (LLM) for information retrieval has established itself in both the private, work and research context. For scientific search however, semantic analysis with LLMs is an underutilized tool for enhancing the work process of literature research. We therefore propose an academic search engine called Discovery, which uses BERT to semantically analyze arbitrary text queries in order to recommend fitting search results for scientific publications. Through explainable artificial intelligence (XAI), additional information about the AI output is provided to the user with the goal to decrease the time needed for evaluating the suitability of a recommended paper.