Status Presentation - WebKnox: Approaches for Automatically Building a Knowledge Base from the Web (Statusvortrag)

Date

Sep 6, 2011

Time

2:00 PM - 3:00 PM

Speaker

Dipl.-Medieninf. David Urbansky

Affiliation

Institut für Systemarchitektur, Lehrstuhl Rechnernetze

Language

en

Main Topic

Informatik

Other Topics

Informatik

Description

Recent studies have shown that more than half of the queries on search engines are about entities such as people, products, or places. Today's search engines, however, do not excel in answering those queries with entity-centric results, but rather with documents that are about those entities. For this reason, the user has the burden of clicking through different search results to gather information about the entity in which he is interested. Having structured information about entities is also of great use in the context on the Web of Data where information is stored in a machine-readable manner and thus users' search intents can be answered more precisely. In order to provide users with aggregated information about these entities of interest a large collection of entities has to be built. This collection must be updated continuously because new entities (for example, products such as mobile phones) are released almost on a daily basis. Building and maintaining such a knowledge base manually requires substantial effort and does not scale well when entities from many different domains are targeted. Today only a few aggregators exist that extract entity names from web pages, enrich them with facts, and publish them as Linked Data. Two such aggregators are DBpedia and Freebase. These systems rely, however, on very few sources (mostly Wikipedia), on manually curated data, or on direct user input. The goal of WebKnox is to extract entity names from different domains from the Web with as little manual effort as possible. Each entity is then enriched with more information, such as facts, questions/answers, and multimedia objects to provide a good overview of what each entity resembles. The contributions of WebKnox are extraction and assessment techniques to automatically create a large database of entities from the World Wide Web. The results can be used in multiple practical applications, including question answering, resolving entity-centric search questions, and improving named entity recognition Betreuer: Prof. Dr. rer. nat. habil. Dr. h. c. Alexander Schill Fachreferent: Prof. Dr.-Ing. Michael Schroeder

Last modified: Sep 6, 2011, 9:35:34 AM