Introducing ORKG ASK: an AI-driven Scholarly Literature Search and Exploration System Taking a Neuro-Symbolic Approach

Introducing ORKG ASK: an AI-driven Scholarly Literature Search and Exploration System Taking a Neuro-Symbolic Approach Allard Oelen1[0000−0001−9924−9153], Mohamad Yaser Jaradeh2[0000−0001−8777−2780], and Sören Auer1,2[0000−0002−0698−2864] 1 TIB – Leibniz Information Centre for Science and Technology, Hannover, Germany {allard.oelen,auer}@tib.eu 2 L3S Research Center, Leibniz University of Hannover, Hannover, Germany jaradeh@l3s.de Abstract. As the volume of published scholarly literature continues to grow, finding relevant literature becomes increasingly difficult. With the rise of generative Artificial Intelligence (AI), and particularly Large Language Models (LLMs), new possibilities emerge to find and explore literature. We introduce ASK (Assistant for Scientific Knowledge), an AI-driven scholarly literature search and exploration system that fol- lows a neuro-symbolic approach. ASK aims to provide active support to researchers in finding relevant scholarly literature by leveraging vec- tor search, LLMs, and knowledge graphs. The system allows users to input research questions in natural language and retrieve relevant arti- cles. ASK automatically extracts key information and generates answers to research questions using a Retrieval-Augmented Generation (RAG) approach. We present an evaluation of ASK, assessing the system’s us- ability and usefulness. Findings indicate that the system is user-friendly and users are generally satisfied while using the system. Keywords: AI-Supported Digital Library · Intelligent User Interface · Large Language Models · Scholarly Search System 1 Introduction Analyzing scholarly literature is a key aspect of research. However, due to the ever-increasing body of scholarly publications, finding scholarly literature be- comes increasingly difficult [12]. Consequently, finding literature consumes a substantial portion of researchers’ time [15]. Because of the recent developments in generative Artificial Intelligence (AI), and specifically Large Language Mod- els (LLMs), new possibilities arise to extract knowledge from scholarly articles, helping researchers to find relevant literature in the flood of publications. In this article, we present ORKG ASK (Assistant for Scientific Knowledge), hereafter referred to as ASK, a next-generation scholarly search and exploration system. ASK aims to provide support to researchers in finding relevant schol- arly literature. ASK takes a Neuro-Symbolic approach which consists of three arXiv:2512.16425v1 [cs.IR] 18 Dec 2025 2 Oelen et al. Q: Why are some... A: Literature shows... Enter research question Generate embeddings Retrieval of top n relevant documents Generate answer Vector database ... ... 1 2 n LLM Literature corpus Relevant documents Research question Find research you are looking for Answer question from context Prepared prompt Prompt LLM as generator brain ~76M articles Abstracts and optionally full-text Q: Why are some... Search Augment prompts with n document context Fig. 1. Explainer depicting our RAG (Retrieval-Augmented Generation) approach for scholarly search. The Retrieval step ranks articles by their relevance to the question. The Augmented step injects the previously retrieved context in the prompt. The Gen- eration step prompts the LLM and displays the answer. key components, namely Vector Search and LLMs for the neural aspect and Knowledge Graphs (KGs) for the symbolic part. We build upon our previously presented work where we demonstrated the basic ASK infrastructure [16]. In this paper, we expand on our previous work by providing an in-depth explana- tion of the approach, technical details of the implementation, and a extensive evaluation. In brief, ASK functions as follows: a user of ASK formulates their information needs as a research question. Afterward, a list of relevant articles is displayed. For each article, an automatically extracted answer for the previ- ously asked question is displayed to the user. Finally, the symbolic aspect ensures users are able to narrow down the search space by providing semantic filters. This provides both the precision of symbolic approach and the flexibility of a neural approach. ASK is running as a publicly available production service online.3 The system takes a Retrieval-Augmented Generation (RAG) approach to support the previously described workflow. RAG [14] is commonly used to inter- twine LLM extractions with information retrieval systems, as depicted in Fig- ure 1. Firstly, the Vector Search component ranks documents based on their relevance (Retrieval) for a research question. Secondly, relevant context is col- lected (i.e., the paper abstract and, if available, full-text) from the previous step (Augmented). Finally, the LLM generates answers and displays this to the user (Generation). A screenshot showing the ranked articles, search query, and gen- erated LLM responses is displayed in Figure 2. This work introduces the follow- ing contributions: i) presents an LL

…(Full text truncated)…

This content is AI-processed based on ArXiv data.