Ralf Schenkel

The rapid advancement of Large Language Models (LLMs) has introduced a paradigm shift in Information Retrieval (IR), moving beyond conventional keyword queries and ranked result lists. LLMs now play a critical role in the evolution of IR technologies and introduce new interaction forms like Retrieval-Augmented Generation, which is a more dynamic and interactive retrieval process that integrates various aspects of Information Access, like Question Answering, into the dialog between a searcher and the search engine. We explore the multi-faceted impact of LLMs on IR, particularly in three distinct layers where they have become an integral part of the retrieval process, namely the retrieval system and processing pipeline that can make use of a richer semantic representation using advanced language models, the interaction layer, and the broader IR ecosystem. For the latter, we focus on evaluation issues as well as bias, fairness, and ethical concerns. We also highlight some recent cases of using LLMs in the medical domain to demonstrate the impact on one specific domain.

Named entity recognition is an important task when constructing knowledge bases from unstructured data sources. Whereas entity detection methods mostly rely on extensive training data, Large Language Models (LLMs) have paved the way towards approaches that rely on zero-shot learning (ZSL) or few-shot learning (FSL) by taking advantage of the capabilities LLMs acquired during pretraining. Specifically, in very specialized scenarios where large-scale training data is not available, ZSL / FSL opens new opportunities. This paper follows this recent trend and investigates the potential of leveraging Large Language Models (LLMs) in such scenarios to automatically detect datasets and software within textual content from GitHub repositories. While existing methods focused solely on named entities, this study aims to broaden the scope by incorporating resources such as repositories and online hubs where entities are also represented by URLs. The study explores different FSL prompt learning approaches to enhance the LLMs' ability to identify dataset and software mentions within repository texts. Through analyses of LLM effectiveness and learning strategies, this paper offers insights into the potential of advanced language models for automated entity detection.

Argumentation is ubiquitous, and the development of argumentation machines could greatly assist humans in managing and navigating argumentation. However, the development of such systems is hindered by the lack of common standards and suitable tools, leading to ad-hoc solutions with little reuse value. Towards a more unified approach, we present an extensible microservice-based architecture for argumentation machines. Being built on the established gRPC framework, it provides strongly typed interfaces for the following services: (i) Argument Mining, (ii) Case-Based Reasoning on Arguments, (iii) Argument Retrieval and Ranking, and (iv) Quality Assessment of Arguments. Our system is designed to be extensible, allowing for easy integration of new tasks. We demonstrate the feasibility of our architecture via a proof-of-concept implementation and provide additional supplementary resources, such as a REST API gateway. Our contributions are publicly available on GitHub under the permissive MIT license.

Philipp Heinisch, Lorik Dumani, Philipp Cimiano, Ralf Schenkel. Findings of the Association for Computational Linguistics: NAACL 2024. 2024.

To answer queries many SPARQL query processors use different sources, e.g., various knowledge bases (KBs) or end points. RESTful Web APIs are rarely the focus of those systems as they come with many limitations, like not being able to process SPARQL queries. Moreover, most existing approaches optimize their query plans only for performance, even though users often have additional preferences, e.g., coverage, reliability, or currency. Additionally, data is often provided with different levels of quality so that not all sources should be trusted equally. In this paper, we therefore present TunA, a query engine that is able to combine RESTful Web APIs and local RDF KBs in the form of triple stores while tuning its (query) plans towards user preferences. Erroneous information from Web APIs is detected using hierarchical agglomerative clustering. Our evaluation shows that TunA outperforms current state-of-the-art systems and is less vulnerable to erroneous information, even in settings where only unreliable sources are available.

In computational argumentation, so-called quality dimensions such as coherence or rhetoric are often used for ranking arguments. However, the literature often only predicts which argument is more persuasive, but not why this is the case. In this paper, we introduce AQUAPLANE, a transparent and easy-to-extend application that not only decides for a pair of arguments which one is more convincing with respect to a statement, but also provides an explanation.