Publications

2018

 back to the year overview

• Bast H, Buchhold B, Haussmann E
  A Quality Evaluation of Combined Search on a Knowledge Base and Text
  2018 Springer, pages: 19 - 26
  » show abstract « hide abstract
  Abstract

  We provide a quality evaluation of KB+Text search, a deep integration of knowledge base search and standard full-text search. A knowledge base (KB) is a set of subject–predicate–object triples with a common naming scheme. The standard query language is SPARQL, where queries are essentially lists of triples with variables. KB+Text search extends this by a special occurs-with predicate, which can be used to express the co-occurrence of words in the text with mentions of entities from the knowledge base. Both pure KB search and standard full-text search are included as special cases. We evaluate the result quality of KB+Text search on three different query sets. The corpus is the full version of the English Wikipedia (2.4 billion word occurrences) combined with the YAGO knowledge base (26 million triples). We provide a web application to reproduce our evaluation, which is accessible via http://ad.informatik. uni-freiburg.de/publications.

2017

 back to the year overview

• Bast H, Buchhold B, Haussmann E, Heindorf S, Potthast M
  WDSM Cup 2017: Vandalism Detection and Triple Scoring
  2017
  » show abstract « hide abstract
  Abstract

  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 judgments. For reproducibility sake, participants were asked to submit their soft-ware on TIRA, a cloud-based evaluation platform, and they were incentivized to share their approaches open source.
• Bast H, Buchhold B, Haussmann E
  A Quality Evaluation of Combined Search on a Knowledge Base and Text
  2017 Künstliche Intelligenz
  » show abstract « hide abstract
  Abstract

  We provide a quality evaluation of KB+Text search, a deep integration of knowledge base search and standard full-text search. A knowledge base (KB) is a set of subject-predicate-object triples with a common naming scheme. The standard query language is SPARQL, where queries are essentially lists of triples with variables. KB+Text search extends this by a special "occurs-with" predicate, which can be used to express the co-occurrence of words in the text with mentions of entities from the knowledge base. Both pure KB search and standard full-text search are included as special cases. We evaluate the result quality of KB+Text search on three different query sets. The corpus is the full version of the English Wikipedia (2.4 billion word occurrences) combined with the YAGO knowledge base (26 million triples). We provide a web application to reproduce our evaluation, which is accessible via http://ad.informatik.uni-freiburg.de/publications.
• Bast H, Buchhold B, Haussmann E
  Overview of the Triple Scoring Task at the WSDM Cup 2017
  2017
  » show abstract « hide abstract
  Abstract

  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.
• Haussmann E
  Towards Precise and Convenient Semantic Search on Text and Knowledge Bases
  2017
  » show abstract « hide abstract
  Abstract

  In this dissertation, we consider the problem of making semantic search on text and knowledge bases more precise and convenient. In a nutshell, semantic search is search with meaning . To this respect, text and knowledge bases have different advantages and disadvantages. Large amounts of text are easily available on the web, and they contain a wealth of information in natural language. However, text represents information in an unstructured form. It follows no pre-defined schema, and without further processing, a machine can understand its meaning only on a superficial level. Knowledge bases, on the other hand, contain structured information in the form of subject predicate object triples. The meaning of triples is well defined, and triples can be retrieved precisely via a query language. However, ormulating queries in this language is inconvenient and compared to text only a small fraction of information is currently available in knowledge bases. In this document, we summarize our contributions on making semantic search on text and knowledge bases more precise and convenient. For knowledge bases, we introduce an approach to answer natural language questions. A user can pose questions conveniently in natural language and ask, for example, who is the ceo of apple? , instead of having to learn and use a specific query language. Our approach applies learning-to-rank strategies and improved the state of the art on two widely used benchmarks at the time of publication. For knowledge bases, we also describe a novel approach to compute relevance scores for triples from type-like relations like profession and nationality . For example, on a large knowledge base, a query for american actors can return a list of more than 60 thousand actors in no particular order. Relevance scores allow to sort this list so that, e.g., frequent lead actors appear before those who only had single cameo roles. In a benchmark that we generated via crowdsourcing, we show that our rankings are closer to human judgments than approaches from the literature. Finally, for text, we introduce a novel natural language processing technique that identifies which words in a sentence “semantically belong together”. For example, in the sentence Bill Gates, founder of Microsoft, and Jeff Bezos, founder of Amazon, are among the wealthiest persons in the world , the words Bill Gates ,founder , and Amazon do not belong together, but the words Bill Gates ,founder, and Microsoft do. We show that when query keywords are required to belong together in order to match, search results become more precise. Given the characteristics of text and knowledge bases outlined above, it is promising to consider a search that combines both. For example, for the query CEOs of U.S. companies who advocate cryptocurrencies , a list of CEOs of U.S. companies can be retrieved from a knowledge base. The information who is advocating cryptocurrencies is rather specific and changes frequently. It is, therefore, better found in full text. As part of this thesis, we describe how a combined search could be achieved and present and evaluate a fully func- tional prototype. All of our approaches are accompanied by an extensive evaluation which show their practicability and, where available, compare them to established approaches from the literature.

2016

 back to the year overview

• Bast H, Buchhold B, Haussmann E
  Semantic Search on Text and Knowledge Bases
  2016
  » show abstract « hide abstract
  Abstract

  This article provides a comprehensive overview of the broad area of semantic search on text and knowledge bases. In a nutshell, semantic search is “search with meaning”. This “meaning” can refer to various parts of the search process: understanding the query (instead of just finding matches of its components in the data), understanding the data (instead of just searching it for such matches), or representing knowledge in a way suitable for meaningful retrieval. Semantic search is studied in a variety of different communities with a variety of different views of the problem. In this survey, we classify this work according to two dimensions: the type of data (text, knowl- edge bases, combinations of these) and the kind of search (keyword, structured, natural language). We consider all nine combinations. The focus is on fundamental techniques, concrete systems, and benchmarks. The survey also considers advanced issues: ranking, indexing, ontology matching and merging, and inference. It also provides a succinct overview of natural language processing techniques that are useful for semantic search: POS tagging, named-entity recognition and disambiguation, sentence parsing, and word vectors. The survey is as self-contained as possible, and should thus also serve as a good tutorial for newcomers to this fascinating and highly topical field.

2015

 back to the year overview

• Bast H, Haussmann E
  More Accurate Question Answering on Freebase
  Materials
  2015 CIKM
  » show abstract « hide abstract
  Abstract

  Real-world factoid or list questions often have a simple structure, yet are hard to match to facts in a given knowledge base due to high representational and linguistic variability. For example, to answer "who is the ceo of apple" on Freebase requires a match to an abstract "leadership" entity with three relations "role", "organization" and "person", and two other entities "apple inc" and "managing director". Recent years have seen a surge of research activity on learning-based solutions for this method. We further advance the state of the art by adopting learning-to-rank methodology and by fully addressing the inherent entity recognition problem, which was neglected in recent works. We evaluate our system, called Aqqu, on two standard benchmarks, Free917 andWebQuestions, improving the previous best result for each benchmark considerably. These two benchmarks exhibit quite different challenges, and many of the existing approaches were evaluated (and work well) only for one of them. We also consider efficiency aspects and take care that all questions can be answered interactively (that is, within a second). Materials for full reproducibility are available on our website: http://ad.informatik. uni-freiburg.de/publications .
• Bast H, Buchhold B, Haussmann E
  Relevance Scores for Triples from Type-Like Relations
  Materials
  2015 SIGIR .
  » show abstract « hide abstract
  Abstract

  We compute and evaluate relevance scores for knowledge-base triples from type-like relations. Such a score measures the degree to which an entity “belongs” to a type. For example, Quentin Tarantino has various professions, including Film Director, Screenwriter, and Actor. The first two would get a high score in our setting, because those are his main professions. The third would get a low score, because he mostly had cameo appearances in his own movies. Such scores are essential in the ranking for entity queries, e.g. “american actors” or “quentin tarantino professions”. These scores are different from scores for "correctness" or "accuracy" (all three professions above are correct and accurate). We propose a variety of algorithms to compute these scores. For our evaluation we designed a new benchmark, which includes a ground truth based on about 14K human judgments obtained via crowdsourcing. Inter-judge agreement is slightly over 90%. Existing approaches from the literature give results far from the optimum. Our best algorithms achieve an agreement of about 80% with the ground truth.

2014

 back to the year overview

• Haussmann E
  An Ensemble for Query Intent Detection
  2014 CKIM CUP
• Bast H, Bäurle F, Buchhold B, Haussmann E
  Easy Access to the Freebase Dataset
  2014 WWW, pages: 95 - 98
  » show abstract « hide abstract
  Abstract

  We demonstrate a system for fast and intuitive exploration of the Freebase dataset. This required solving several non- trivial problems, including: entity scores for proper anking and name disambiguation, a unique meaningful name for every entity and every type, extraction of canonical binary relations from multi-way relations (which in Freebase are modeled via so-called mediator objects), computing the transitive hull of selected relations, and identifying and merging duplicates. Our contribution is two-fold. First, we provide for download an up-to-date version of the Freebase data, enriched and simplified as just sketched. Second, we offer a user interface for exploring and searching this data set. The data set, the user interface and a demo video are available from http://freebase-easy.cs.uni-freiburg.de.
• Bast H, Haussmann E
  More Informative Open Information Extraction via Simple Inference
  Demo
  2014 ECIR
  » show abstract « hide abstract
  Abstract

  Recent Open Information Extraction (OpenIE) systems utilize grammatical structure to extract facts with very high recall and good precision. In this paper, we point out that a significant fraction of the extracted facts is, however, not informative. For example, for the sentence 'The ICRW is a non-profit organization headquartered in Washington', the extracted fact (a non-profit organization) (is headquartered in) (Washington) is not informative. This is a problem for semantic search applications utilizing these triples, which is hard to fix once the triple extraction is completed. We therefore propose to integrate a set of simple inference rules into the extraction process. Our evaluation shows that, even with these simple rules, the percentage of informative triples can be improved considerably and the already high recall can be improved even further. Both improvements directly increase the quality of search on these triples.
• Bast H, Bäurle F, Buchhold B, Haussmann E
  Semantic full-text search with broccoli
  2014 SIGIR, pages: 1265 - 1266
  » show abstract « hide abstract
  Abstract

  We combine search in triple stores with full-text search into what we call semantic full-text search. We provide a fully functional web application that allows the incremental construction of complex queries on the English Wikipedia combined with the facts from Freebase. The user is guided by context-sensitive suggestions of matching words, instances, classes, and relations after each keystroke. We also provide a powerful API, which may be used for research tasks or as a back end, e.g., for a question answering system. Our web application and public API are available under http://broccoli.cs.uni-freiburg.de.

2013

 back to the year overview

• Bast H, Haussmann E
  Open Information Extraction via Contextual Sentence Decomposition
  2013 ICSC

2012

 back to the year overview

• Bast H, Bäurle F, Buchhold B, Haussmann E
  A Case for Semantic Full-Text Search
  2012 SIGIR-JIWES
• Bast Hannah, Bäurle Florian, Buchhold Björn, Haussmann Elmar
  Broccoli: Semantic Full-Text Search at your Fingertips
  2012 CoRR, volume: abs/1207.2615
• Aslam Faisal, Baig Ghufran, Qureshi Mubashir, Uzmi Zartash, Fennell Luminous, Thiemann Peter, Schindelhauer Christian, Haussmann Elmar
  Rethinking Java call stack design for tiny embedded devices
  2012 LCTES, pages: 1 - 10

2010

 back to the year overview

• Aslam Faisal, Fennell Luminous, Schindelhauer Christian, Thiemann Peter, Ernst Gidon, Haussmann Elmar, Rührup Stefan, Uzmi Zartash
  Optimized Java Binary and Virtual Machine for Tiny Motes
  2010 DCOSS, pages: 15 - 30