The Web-based Systems Group at the University of Mannheim conducts research on methods for integrating data from large numbers of data sources in the context of the open Web and in corporate data lakes. Our research includes areas such as entity matching, schema matching, table annotation, information extraction, and data discovery. Our current work focuses on utilizing large language models and LLM-based agents for data integration tasks. We apply the developed methods to integrate product data from large numbers of e-shops and to construct knowledge graphs such as DBpedia. The empirical research of the group includes monitoring the adoption of schema.org annotations on the public Web by regularly extracting structured data from large Web corpora.
Web-based Systems Group @ University of Mannheim
Pinned Loading
Repositories
- LAB-EU Public
LAB-EU is a benchmark for evaluating the capabilities of LLM agents to apply German, French and other European laws and to write long-form legal document in accordance with different European laws.
- SINT-Flow Public
This repository contains the code for the paper "SINT-Flow: Schema Integration using Large Language Model Workflows". SINT-Flow is a framework consisting of five LLM-based operators that can be arranged in different sequence to form schema integration workflows. The repository also introduces SINT-Bench, a schema integration benchmark.
- MaDI-Bench Public
The Mannheim Data Integration Benchmark (MaDI-Bench) is an end-to-end benchmark for tabular data integration. It provides integration tasks across five domains, covering schema matching, value normalization, entity matching, and data fusion. It supports difficulty variants and measuring step-wise as well as end-to-end performance.
- Automatic-data-labeling Public
This repository contains the code and data for the paper 'Labeling Training Data for Entity Matching Using Large Language Models' which explores knowledge distillation for entity matching, e.g. using LLMs to label training pairs that are subsequently used to train a smaller student model, such as DITTO using RoBERTa.
- billiger-de-products Public
Billiger.de Products is a bilingual entity matching benchmark. The benchmark requires matching German product offers originating from the price comparison platform billiger.de. The benchmark provides machine-translated English versions of all artifacts, enabling the assessment of the impact of the language on entity matching performance.
- WebMall Public
This repository contains the code and data of the WebMall benchmark for evaluating the capability of Web agents to find and compare product offers from multiple e-shops.
- PyDI-AgenticDI Public Forked from wbsg-uni-mannheim/PyDI
The PyDI framework provides methods for end-to-end data integration. The framework covers all steps of the integration process, including schema matching, data translation, entity matching, and data fusion. The framework offers traditional string-based methods as well as modern LLM- and embedding-based techniques for these tasks.
- wdc-page Public
This repository contains the source files of the Web Data Commons website and is used to maintain the site. The Web Data Commons project extracts structured data from the Common Crawl
- PyDI Public
The PyDI framework provides methods for end-to-end data integration. The framework covers all steps of the integration process, including schema matching, data translation, entity matching, and data fusion. The framework offers traditional string-based methods as well as modern LLM- and embedding-based techniques for these tasks.
Top languages
Loading…
Most used topics
Loading…