DEQA: Deep Web Extraction for Question Answering

Despite decades of effort, intelligent object search remains elusive. Neither search engine nor semantic web technologies alone have managed to provide usable systems for simple questions such as “Find me a flat with a garden and more than two bedrooms near a supermarket.” We introduce DEQA, a conceptual framework that achieves this elusive goal through combining state-of-the-art semantic technologies with effective data extraction. To that end, we apply DEQA to the UK real estate domain and show that it can answer the majority of such questions correctly. DEQA achieves this by mapping natural language questions to SPARQL patterns. These patterns are then evaluated on an RDF database of current real estate offers. The offers are obtained using OXPATH, a state-of-the-art data extraction system, on the major agencies in the Oxford area and linked through LIMES to background knowledge such as the location of supermarkets.

Demo

AutoSPARQL prototype user interface: http://autosparql-tbsl.dl-learner.org

General Approach

DEQA provides a conceptual framework for enhancing classic information retrieval and search techniques using recent advances in web extraction, data integration and question answering. The overall approach is illustrated in the figure above: Given a particular domain, such as real estate, the first step consists of identifying relevant websites and extracting data from those. This previously tedious task can now be reduced to the rapid creation of OXPath wrappers. In DEQA, data integration is performed through a triple store using a common base ontology. Hence, the first phase may be a combination of the extraction of unstructured and structured data. For instance, websites may already expose data as RDFa, which can then be transformed to the target schema, e.g.using R2R, if necessary. This basic RDF data is enriched, e.g. via linking, schema enrichment, geo-coding or post-processing steps on the extracted data. This is particularly interesting, since the LOD cloud contains a wealth of information across different domains which allows users to formulate queries in a more natural way (e.g., using landmarks rather than postcodes or coordinates). For instance, in our analysis of the real estate domain, over 100k triples for 2,400 properties were extracted and enriched by over 100k links to the LOD cloud. Finally, question answering or semantic search systems can be deployed on top of the created knowledge. One of the most promising research areas in question answering in the past years is the conversion of natural language to SPARQL queries, which allows a direct deployment of such systems on top of a triple store. Finally, DEQA first attempts to convert a natural language query to SPARQL, yet can fall back to standard information retrieval, where this fails.

Use Case: Application to the Real Estate Domain

he domain-specific implementation of the conceptual framework, which we used for the real estate domain, is depicted in the figure above. It covers the above described steps by employing state-of-the-art tools in the respective areas, OXPath for data extraction to RDF, LIMES for linking to the linked data cloud, and TBSL for translating natural language questions to SPARQL queries. Below are the configuration files necessary to set up the system and a pointer to a user interface for testing it:

Members

DIADEM

  • Dr. Tim Furche, http://furche.net
  • Dr. Giovanni Grasso, http://www.giovannigrasso.it/
  • Dr. Christian Schallhart, http://www.cs.ox.ac.uk/people/christian.schallhart/
  • Dr. Andrew Sellers, http://www.cs.ox.ac.uk/people/andrew.sellers/
  • David Liu

CITEC

  • Dr. Christina Unger, http://www.sc.cit-ec.uni-bielefeld.de/people/cunger/

News

AKSW at ISWC2017 ( 2017-07-30T05:57:57+02:00 Muhammad Saleem)

2017-07-30T05:57:57+02:00 Muhammad Saleem

We are very pleased to announce that AKSW will be presenting 2 papers at ISWC 2017, which will be held on 21-24 October in Vienna, Austria. The demo and workshops papers have to be announced. Read more about "AKSW at ISWC2017"

AKSW Colloquium, 07.07.2017, Two paper presentations concerning Link Discovery and Knowledge Base Reasoning ( 2017-07-06T21:24:36+02:00 by Daniel Obraczka)

2017-07-06T21:24:36+02:00 by Daniel Obraczka

At the AKSW Colloquium on Friday 7th of July, at 10:40 AM there will be two paper presentations concerning genetic algorithms to learn linkage rules, and differentiable learning of logical rules for knowledge base reasoning. Read more about "AKSW Colloquium, 07.07.2017, Two paper presentations concerning Link Discovery and Knowledge Base Reasoning"

SANSA 0.2 (Semantic Analytics Stack) Released ( 2017-06-13T18:18:28+02:00 by Prof. Dr. Jens Lehmann)

2017-06-13T18:18:28+02:00 by Prof. Dr. Jens Lehmann

The AKSW and Smart Data Analytics groups are happy to announce SANSA 0.2 – the second release of the Scalable Semantic Analytics Stack. Read more about "SANSA 0.2 (Semantic Analytics Stack) Released"

AKSW at ESWC 2017 ( 2017-06-12T10:53:35+02:00 Christopher Schulz)

2017-06-12T10:53:35+02:00 Christopher Schulz

Hello Community! The ESWC 2017 just ended and we give a short report of the course at the conference, especially regarding the AKSW-Group. Our members Dr. Muhammad Saleem, Dr. Mohamed Ahmed Sherif, Claus Stadler, Michael Röder, Prof. Dr. Read more about "AKSW at ESWC 2017"

Four papers accepted at WI 2017 ( 2017-06-10T15:01:31+02:00 Christopher Schulz)

2017-06-10T15:01:31+02:00 Christopher Schulz

Hello Community! We proudly announce that The International Conference on Web Intelligence (WI) accepted four papers by our group. The WI takes place in Leipzig between the 23th – 26th of August. Read more about "Four papers accepted at WI 2017"