OntosMiner Family: Content Extraction from MultiLingual Document Collections

1
OntosMiner FamilyContent Extraction from
MultiLingual Document Collections

• Daniel Hladky, Ontos AG, Switzerland
• Irina Efimenko, Moscow State University
• Vladimir Khoroshevsky, Computer Center RAS
• Victor Klintsov, AviComp AG, Russia

2
Presentation Map

• Introduction
• OntosMiner Project
• OntosMiner Family
• Ontos NLPs Architecture
• Domain Ontology
• NE Extraction
• Extraction of Relations
• Cognitive Maps
• Conclusion Future Trends
• Demos

3
OntosMiner Project

• Objectives
• Combining AI IT experience within the NLP
  domain
• RD in knowledge management
• Approaches
• Usage of IE technologies in NL texts processing
• Enrichment of IE techniques with NLP on the
  basis of special linguistic models
• Representation of the NL texts meaning in the
  form of cognitive maps
• Results
• New generation of MIE-systems
• Practical usage of the results within the
  commercial non-commercial organizations
• Approaches developed for Semantic Indexing,
  Clusterization and Summarization

4
Basic Principles

• Processing of those constructions, that can be
  processed correctly, and NON-processing of those
  ones, that still can not be processed correctly.
• Development of reusable components for
  multi-platform implementation.
• Providing domain ontology-driven analysis.

5
Main Requirements

• Work with multilingual document collections (at
  present OntosMiner systems deal with English,
  French, German and Russian texts).
• Work with monothematic document collection (at
  present it is, first of all, the so-called
  Business Duties domain. Such collections
  include informational materials about
  IT-companies, analytical materials about
  founding, investing, selling, buying, merging of
  companies, top-management CVs, etc. and Crime
  for Russian texts).
• An adequate processing of relevant objects and
  relations, according to the concrete ontology.
• Representation of processing results in a form of
  a cognitive map, that is a kind of semantic
  network.
• Multi-platform implementation of all systems of
  the family.

6
OntosMiner Family Architecture
7
Domain Ontology Business Duties
8
Named Entities

• People
• Organizations
• Titles and JobTitles
• Scientific degrees
• Various kinds of Addresses
• Money
• Percent
• URL, e-mail, phone (international style)
• Locations
• Cars
• Dates and Periods of Time.

9
Semantic Relations

• Affiliate
• Buy-Sell
• Employ
• Found
• Graduate
• Invest
• JointVenture
• Own
• Rival
• LocatedIn
• EarnDegree

• Employ
• Own
• LocatedIn
• Reside
• Belong
• Hijack
• Investigate
• Petition
• Crime Place Observation

10
OntosMiner family key point components
11
Evaluation

• Materials of the companies from KM Top100
  outlined by Knowledge Management Journal have
  been used as evaluation corpus for OntosMiner
  family systems.
• There were at about 150 texts in the corpus and
  all of them were manually marked by linguists
  from Moscow State University and the received
  markup was controlled by the specialists from
  OntosMiner Project.
• In addition to such documents, some Russian
  texts from the Business Duties domain, as well as
  documents from Police reports collection, were
  added into corpus.

12
OntosMiner/English
13
OntosMiner/French
14
OntosMiner/Russian
15
Conclusion Future Trends

• Multilingual CE-System has been developed
• Achieved results are going to be used in other
  tasks resolving, such as Summarization, Semantic
  Indexing, QA
• We are interested in collaboration and mutual
  projects with other teams

16

• Thank You
• for Your Attention