LD4IE 2014 - International Workshop on Linked Data for Information Extraction

This workshop focuses on the exploitation of Linked Data for Web Scale Information Extraction (IE), which concerns extracting structured knowledge from unstructured/semi-structured documents on the Web. One of the major bottlenecks for the current state of the art in IE is the availability of learning materials (e.g., seed data, training corpora), which, typically are manually created and are expensive to build and maintain.
Linked Data (LD) defines best practices for exposing, sharing, and connecting data, information, and knowledge on the Semantic Web using uniform means such as URIs and RDF. It has so far been created a gigantic knowledge source of Linked Open Data (LOD), which constitutes a mine of learning materials for IE. However, the massive quantity requires efficient learning algorithms and the not guaranteed quality of data requires robust methods to handle redundancy and noise.
LD4IE intends to gather researchers and practitioners to address multiple challenges arising from the usage of LD as learning material for IE tasks, focusing on (i) modelling user defined extraction tasks using LD; (ii) gathering learning materials from LD assuring quality (training data selection, cleaning, feature selection etc.); (iii) robust learning algorithms for handling LD; (iv) publishing IE results to the LOD cloud.
As the second workshop of this series, a special focus this year will be on using Webpages embedded with structured data describing products, people, organizations, places, events using markup standards such as RDFa, Microdata and Microformats. We especially encourage the use of the Web Data Commons corpus of structured data [1] for experiments. The dataset contains billions of triples in terrabytes of data. Unlike the Billion Triple Challenges that typically contain triples from specific linked data sets such as DBpedia, the Web Data Commons corpus consists of specifically triples extracted from Webpages annotated with standard markup vocabularies such as RDFa, Microdata format. The provenance of triples is also recorded so the original Webpages containing those annotations can be obtained. Although it is not strictly required to use this corpora, submissions that do use this (or part of) corpora will be considered for extra credits. Authors may use this data for any IE-related tasks, although we may define some specific 'example' tasks in later calls to invite participants.
[1] http://webdatacommons.org/structureddata/
Topics of interest include, but are not limited to:
Modelling Extraction Tasks
* modelling extraction tasks (e.g. defining IE templates using LD ontologies)
* extracting and building knowledge patterns based on LD
*user friendly approaches for querying LD
Information Extraction
* selecting relevant portions of LD as training data
* selecting relevant knowledge resources from LD
* IE methods robust to noise in LD as training data
* Information Extractions tasks/applications exploiting LD (Wrapper induction, Table interpretation, IE from unstructured data, Named Entity Recognition, Relation Extraction…)
* linking extracted information to existing LD datasets
Linked Data for Learning
* assessing the quality of LD data for training
* select optimal subset of LD to seed learning
* managing heterogeneity, incompleteness, noise, and uncertainty of LD
* scalable learning methods using LD
* pattern extraction from LD
Special interest: IE using Web Data Commons corpus
* any IE tasks using (part of) the Web Data Commons corpus
FORMAT
We accept the following formats of submissions:
Full paper with a maximum of 12 pages including references
Short paper with a maximum of 6 pages including references
All submissions must be written in English and must be formatted according to the information for LNCS Authors (http://www.springer.com/computer/lncs?SGWID=0-164-....). Please submit your contributions electronically in PDF format to EasyChair at https://www.easychair.org/conferences/?conf=ld4ie2...
Accepted papers will be published online via CEUR-WS.