ScienceDirect Topic Pages: A Knowledge Base of Scientific Concepts Across Various Science Domains

From undergraduate students to renowned scholars, everyone occasionally encounters unknown concepts within their field of interest, especially when reading scientific articles. ScienceDirectTopic Pages (TP) are intended to facilitate learning and to provide users with a structured overview of sources to deepen their knowledge about such unfamiliar topics. Our free service provides insight into a vast set of technical topics across 20 different scientific domains. Designed to emulate the natural flow of learning, TPs are embedded within millions of articles so that users can click on unfamiliar concepts they come across whilst reading an article. This redirects the user to a TP, consisting of a definition of the concept, which provides the user with a basic understanding of the concept. The TP further presents a collection of relevant snippets extracted from books and review articles published by ScienceDirect for users interested in references and more detailed explanations and applications of the concept. Finally, a set of related topics is provided to extend the user's knowledge even further. To build TPs, we utilize various information retrieval methods across our product. We retrieve the most relevant snippets for each topic/concept using a semantic search model fine-tuned on our scientific database. We further leverage the power of Retrieval Augmented Generation to generate reliable definitions on the topics sourced from ScienceDirect's content. To retrieve a list of relevant concepts for each topic, we use the co-occurrence statistics of concepts within books and articles.