Hannah Bast

Hannah Bast, Matthias Hertel, Natalie Prange

We present Elevant, a tool for the fully automatic fine-grained evaluation of a set of entity linkers on a set of benchmarks. Elevant provides an automatic breakdown of the performance by various error categories and by entity type. Elevant also provides a rich and compact, yet very intuitive and self-explanatory visualization of the results of a linker on a benchmark in comparison to the ground truth. A live demo, the link to the complete code base on GitHub and a link to a demo video are provided under https://elevant.cs.uni-freiburg.de .

Simon Ging, Philipp Arnold, Sebastian Walter et al.

Recent 3D CT vision-language models align volumes with reports via contrastive pretraining, but typically rely on limited public data and provide only coarse global supervision. We train a 3D CT vision-language model on 98k report-volume pairs (50k patients) collected at a single hospital, combined with public datasets, using SigLIP-style contrastive pretraining together with prompt-based disease supervision in the shared vision-text embedding space. On CT-RATE, our model achieves state-of-the-art text-to-image retrieval (R@10 31.5 vs. 22.2) and competitive disease classification (AUC 83.8 vs. 83.8), with consistent results on Rad-ChestCT (AUC 77.0 vs. 77.3). We further observe that radiologists routinely reference specific images within their reports (e.g., ``series X, image Y''), linking textual descriptions to precise axial locations. We automatically mine 262k such snippet-slice pairs and introduce the task of intra-scan snippet localization -- predicting the axial depth referred to by a text snippet -- reducing mean absolute error to 36.3 mm at 12 mm feature resolution, compared with 67.0 mm for the best baseline. Adding this localization objective leaves retrieval and classification broadly unchanged within confidence bounds, yielding a single unified model for retrieval, classification, and intra-scan grounding.

Sebastian Walter, Hannah Bast

We propose a new approach for generating SPARQL queries on RDF knowledge graphs from natural language questions or keyword queries, using a large language model. Our approach does not require fine-tuning. Instead, it uses the language model to explore the knowledge graph by strategically executing SPARQL queries and searching for relevant IRIs and literals. We evaluate our approach on a variety of benchmarks (for knowledge graphs of different kinds and sizes) and language models (of different scales and types, commercial as well as open-source) and compare it with existing approaches. On Wikidata we reach state-of-the-art results on multiple benchmarks, despite the zero-shot setting. On Freebase we come close to the best few-shot methods. On other, less commonly evaluated knowledge graphs and benchmarks our approach also performs well overall. We conduct several additional studies, like comparing different ways of searching the graphs, incorporating a feedback mechanism, or making use of few-shot examples.

Hannah Bast, Matthias Hertel, Natalie Prange

Existing evaluations of entity linking systems often say little about how the system is going to perform for a particular application. There are two fundamental reasons for this. One is that many evaluations only use aggregate measures (like precision, recall, and F1 score), without a detailed error analysis or a closer look at the results. The other is that all of the widely used benchmarks have strong biases and artifacts, in particular: a strong focus on named entities, an unclear or missing specification of what else counts as an entity mention, poor handling of ambiguities, and an over- or underrepresentation of certain kinds of entities. We provide a more meaningful and fair in-depth evaluation of a variety of existing end-to-end entity linkers. We characterize their strengths and weaknesses and also report on reproducibility aspects. The detailed results of our evaluation can be inspected under https://elevant.cs.uni-freiburg.de/emnlp2023 . Our evaluation is based on several widely used benchmarks, which exhibit the problems mentioned above to various degrees, as well as on two new benchmarks, which address the problems mentioned above. The new benchmarks can be found under https://github.com/ad-freiburg/fair-entity-linking-benchmarks .

Martin Potthast, Stefan Heindorf, Hannah Bast

The WSDM Cup 2017 was a data mining challenge held in conjunction with the 10th International Conference on Web Search and Data Mining (WSDM). It addressed key challenges of knowledge bases today: quality assurance and entity search. For quality assurance, we tackle the task of vandalism detection, based on a dataset of more than 82 million user-contributed revisions of the Wikidata knowledge base, all of which annotated with regard to whether or not they are vandalism. For entity search, we tackle the task of triple scoring, using a dataset that comprises relevance scores for triples from type-like relations including occupation and country of citizenship, based on about 10,000 human relevance judgements. For reproducibility sake, participants were asked to submit their software on TIRA, a cloud-based evaluation platform, and they were incentivized to share their approaches open source.

Sebastian Walter, Hannah Bast

We present GRISP (Guided Recurrent IRI Selection over SPARQL Skeletons), a novel SPARQL-based question-answering method over knowledge graphs based on fine-tuning a small language model (SLM). Given a natural-language question, the method first uses the SLM to generate a natural-language SPARQL query skeleton, and then to re-rank and select knowledge graph items to iteratively replace the natural-language placeholders using knowledge graph constraints. The SLM is jointly trained on skeleton generation and list-wise re-ranking data generated from standard question-query pairs. We evaluate the method on common Wikidata and Freebase benchmarks, and achieve better results than other state-of-the-art methods in a comparable setting.

Sebastian Walter, Hannah Bast

We present the Wikidata Query Logs (WDQL) dataset, a dataset consisting of 200k question-query pairs over the Wikidata knowledge graph. It is over 6x larger than the largest existing Wikidata datasets of similar format without relying on template-generated queries. Instead, we construct it using real-world SPARQL queries sent to the Wikidata Query Service and generate questions for them. Since these log-based queries are anonymized, and therefore often do not produce results, a significant amount of effort is needed to convert them back into meaningful SPARQL queries. To achieve this, we present an agent-based method that iteratively de-anonymizes, cleans, and verifies queries against Wikidata while also generating corresponding natural-language questions. We demonstrate the dataset's benefit for training question-answering methods. All WDQL assets, as well as the agent code, are publicly available under a permissive license.

Simon Ging, Sebastian Walter, Jelena Bratulić et al.

Training high-quality CLIP models typically requires enormous datasets, which limits the development of domain-specific models -- especially in areas that even the largest CLIP models do not cover well -- and drives up training costs. This poses challenges for scientific research that needs fine-grained control over the training procedure of CLIP models. In this work, we show that by employing smart web search strategies enhanced with knowledge graphs, a robust CLIP model can be trained from scratch with considerably less data. Specifically, we demonstrate that an expert foundation model for living organisms can be built using just 10M images. Moreover, we introduce EntityNet, a dataset comprising 33M images paired with 46M text descriptions, which enables the training of a generic CLIP model in significantly reduced time.

Hannah Bast, Johannes Kalmbach, Theresa Klumpp et al.

We show how to achieve fast autocompletion for SPARQL queries on very large knowledge bases. At any position in the body of a SPARQL query, the autocompletion suggests matching subjects, predicates, or objects. The suggestions are context-sensitive in the sense that they lead to a non-empty result and are ranked by their relevance to the part of the query already typed. The suggestions can be narrowed down by prefix search on the names and aliases of the desired subject, predicate, or object. All suggestions are themselves obtained via SPARQL queries, which we call autocompletion queries. For existing SPARQL engines, these queries are impractically slow on large knowledge bases. We present various algorithmic and engineering improvements of an existing SPARQL engine such that these autocompletion queries are executed efficiently. We provide an extensive evaluation of a variety of suggestion methods on three large knowledge bases, including Wikidata (6.9B triples). We explore the trade-off between the relevance of the suggestions and the processing time of the autocompletion queries. We compare our results with two widely used SPARQL engines, Virtuoso and Blazegraph. On Wikidata, we achieve fully sensitive suggestions with sub-second response times for over 90% of a large and diverse set of thousands of autocompletion queries. Materials for full reproducibility, an interactive evaluation web app, and a demo are available on: https://ad.informatik.uni-freiburg.de/publications .

Hannah Bast, Patrick Brosi, Markus Näther

We study the following problem: given two public transit station identifiers A and B, each with a label and a geographic coordinate, decide whether A and B describe the same station. For example, for "St Pancras International" at (51.5306, -0.1253) and "London St Pancras" at (51.5319, -0.1269), the answer would be "Yes". This problem frequently arises in areas where public transit data is used, for example in geographic information systems, schedule merging, route planning, or map matching. We consider several baseline methods based on geographic distance and simple string similarity measures. We also experiment with more elaborate string similarity measures and manually created normalization rules. Our experiments show that these baseline methods produce good, but not fully satisfactory results. We therefore develop an approach based on a random forest classifier which is trained on matching trigrams between two stations, their distance, and their position on an interwoven grid. All approaches are evaluated on extensive ground truth datasets we generated from OpenStreetMap (OSM) data: (1) The union of Great Britain and Ireland and (2) the union of Germany, Switzerland, and Austria. On all datasets, our learning-based approach achieves an F1 score of over 99%, while even the most elaborate baseline approach (based on TFIDF scores and the geographic distance) achieves an F1 score of at most 94%, and a naive approach of using a geographical distance threshold achieves an F1 score of only 75%. Both our training and testing datasets are publicly available.

Hannah Bast, Matthias Hertel, Mostafa M. Mohamed

We consider the following tokenization repair problem: Given a natural language text with any combination of missing or spurious spaces, correct these. Spelling errors can be present, but it's not part of the problem to correct them. For example, given: "Tispa per isabout token izaionrep air", compute "Tis paper is about tokenizaion repair". We identify three key ingredients of high-quality tokenization repair, all missing from previous work: deep language models with a bidirectional component, training the models on text with spelling errors, and making use of the space information already present. Our methods also improve existing spell checkers by fixing not only more tokenization errors but also more spelling errors: once it is clear which characters form a word, it is much easier for them to figure out the correct word. We provide six benchmarks that cover three use cases (OCR errors, text extraction from PDF, human errors) and the cases of partially correct space information and all spaces missing. We evaluate our methods against the best existing methods and a non-trivial baseline. We provide full reproducibility under https://ad.cs.uni-freiburg.de/publications .

Hannah Bast, Björn Buchhold, Elmar Haussmann

This paper provides an overview of the triple scoring task at the WSDM Cup 2017, including a description of the task and the dataset, an overview of the participating teams and their results, and a brief account of the methods employed. In a nutshell, the task was to compute relevance scores for knowledge-base triples from relations, where such scores make sense. Due to the way the ground truth was constructed, scores were required to be integers from the range 0..7. For example, reasonable scores for the triples "Tim Burton profession Director" and "Tim Burton profession Actor" would be 7 and 2, respectively, because Tim Burton is well-known as a director, but he acted only in a few lesser known movies. The triple scoring task attracted considerable interest, with 52 initial registrations and 21 teams who submitted a valid run before the deadline. The winning team achieved an accuracy of 87%, that is, for that fraction of the triples from the test set (which was revealed only after the deadline) the difference to the score from the ground truth was at most 2. The best result for the average difference from the test set scores was 1.50.

Hannah Bast, Florian Bäurle, Björn Buchhold et al.

We present Broccoli, a fast and easy-to-use search engine for what we call semantic full-text search. Semantic full-text search combines the capabilities of standard full-text search and ontology search. The search operates on four kinds of objects: ordinary words (e.g., edible), classes (e.g., plants), instances (e.g., Broccoli), and relations (e.g., occurs-with or native-to). Queries are trees, where nodes are arbitrary bags of these objects, and arcs are relations. The user interface guides the user in incrementally constructing such trees by instant (search-as-you-type) suggestions of words, classes, instances, or relations that lead to good hits. Both standard full-text search and pure ontology search are included as special cases. In this paper, we describe the query language of Broccoli, the main idea behind a new kind of index that enables fast processing of queries from that language as well as fast query suggestion, the natural language processing required, and the user interface. We evaluated query times and result quality on the full version of the English Wikipedia (40 GB XML dump) combined with the YAGO ontology (26 million facts). We have implemented a fully functional prototype based on our ideas and provide a web application to reproduce our quality experiments. Both are accessible via http://broccoli.informatik.uni-freiburg.de/repro-corr/ .