WP4: Conceptual Mining from Text for Knowledge Engineering

1
WP4 Conceptual Mining from Text for Knowledge
Engineering

• State of the Art
• WP CoordinatorsAlfonso ValenciaCarlos Rodriguez

2
Why Concept/Semantic Mining?

• Knowledge Acquisition Bottleneck
• Top-Down, manually-designed Ontologies are
• sparse (non-exhaustive)
• shallow (not fine-grained)
• not mappable (to terms or other ontologies)
• not easily updated or customized
• Text-based ontologies reflect better diversity in
  knowledge as reflected by the literature and
  domain terminology

3
Information for Ontology Learning
4
State of the Art Methods

• implicit relations
• Corpus Distribuition
• Machine Learning Algorithms
• explicit relations
• Symbolic (rule and syntax-based)
• Hybrid, combining some or all
• Bootstrap the ontology-learning process using
  existing resources

5
An example
Meiosis Cyclin Checkpoint Interphase Nucleoplasma
Division Histone Replication Chromatid
Blaschke, et al., Funct. Integ. Genomics 2001
Cell cycle
Words
17 genes PCNA CDC2 MSH2 LBR TOP2A ...
GO codes
DNA replication DNA metabolism Cell Cycle control
PCNA-MSH2The binding of PCNA to MSH2 may reflect
linkage between mismatch repair and
replication. LBR-CDC2 LBR undergoes mitotic
phosphorylation mediated by p34(cdc2) protein
kinase.
Sentences
24 genes ABCA5 CAT ELF2 PIM1 WNT2 ...
Dipeptidyl Prolyl nmr Collagen-binding
Words
Unknown
6
Induce rules at different linguistic levels
7
Lexical- and syntax-derived relationships from
text

• Complex relationships in CCO
• degradates
• participate_in
• catalyses
• adjacent_to
• agent_in
• What new ones can be learnt?
• LBR undergoes mitotic phosphorylation mediated
  by p34(cdc2) protein kinase.
• mitotic phosphorylation mediated_by protein
  kinase
• Can it be subsumed by others?
• Are there other subcategories?

8
Beyond the State of the Art

• Optimal hybrid methodology for
• Extracting entities
• Discovering relations
• Providing ontology-relevant information(But what
  and how ?)
• Comparing top-down with bottom-up ontologies
• Providing definitional information
• Application to CC-cancer domains (and possibly
  to gene regulation)

9
In the context of project and other WPs

• Reasoning with text-generated ontologies
  competing or complementing?
• Reduction of lexical and semantic relationships
  to ontological relation inventory
• How to present and use Text-Mined information for
  ontology design (especially for database
  annotation)?
• How to curate, evaluate and compare ontologies?

10
Information for Ontology Engineers

• New Classes (ontology) and Instances (KB)
• Definitions and glosses
• Concept usage and entity examples
• Terms and synonyms
• Hierarchical and non-hierarchical relations
• Possible reasoning rules

11
To and Fro other WPs