FOCUS 1

Using Machine learning (and AI) in ontology matching

Problem:

• The essence of SemWeb is structured data sets (machine readable). [Intro to SemWeb, Li Yau]
• Constructing formal structured data requires ontology as a reference.
• One of the main obstacles facing SemWeb today is the lack of robust and formal ontologies. [add reference]
• That’s because there are multiple ontologies of the same domain, eventually this introduce inconsistency and confusion (inconsistency problem).
• Ontology matching is a solutions used to address this (inconsistency) problem.
• Current methods rely on manual ontology matching which is not sufficient; so tedious and error-prone. [1]

Solutions:

• Use ML and AI techniques for automated matching.
• Onto matching using ML. [1]
• PSL for onto alignment. [2]

Return value, benefiting fields, and applications:

i.e.

• More reliable ontologies for semantic web application in general.
• ie) IBM-WATSON: Watson-Paths projectl

More details see:
[[research: engineering the problem]]

[1] Ontology Matching: A Machine Learning Approach http://homes.cs.washington.edu/~pedrod/papers/hois.pdf
[2] Probabilistic Similarity Logic http://event.cwi.nl/uai2010/papers/UAI2010_0089.pdf

FOCUS 2

Ontology Learning

Automatic creation of ontology, includes extracting the corresponding domain’s terms and relationships from corpus. [1]

Problem:

Techniques:

• Use NLP methods
• For finding similar concepts, use recommender systems’ methods for finding similar users.

[1] http://en.wikipedia.org/wiki/Ontology_learning

FOCUS 3

Extracting Knowledge Base from Corpus

Problem:

Building an ontology is a manual, error-prone, and tedious process. Finding a tool

Idea: Automated ontology generation (domain specific) using NLP techniques.

Steps:

- Build a python module or framework (name it Pyonto) to transform natural language sentences into ontology concepts. [see, [2]]

Solution:
- Use facts (axioms in natural language) and the true KB from knowledge extraction engines (i.e. NELL, FreeBase .. ) and from structured semantic databases (i.e. DBpedia, LinkedData..) to automatically generate domain specific ontologies. See [9]

• Use NLTK:

  • To create KB statements from corpora
    • Then, transform the KB into Ontology

• Unsupervised learning of relations between concepts, see [8]
  

  ---

  
  References:

[1]
NLP for onto: Applying NLP FOR BUILDING DOMAIN ONTOLOGY:
FASHION COLLECTION

[2]

• Quepy Framework: http://quepy.readthedocs.org/en/latest/index.html
• iepy Framework: https://github.com/machinalis/iepy

[3]

• Machinalis at github https://github.com/machinalis
• Quepy talk: http://www.machinalis.com/blog/quepy-talk-at-pydata-silicon-valley-2014/

[4]
Also check out python-owl Seth: http://seth-scripting.sourceforge.net/

[5] IRBook.docset from Stanford’s Intro to Info Retrieval

[6] paper: Survey on clustering methods for ontological knowledge

[7] Chapter 5: Ontology Learning Using Word Net Lexical Expansion and Text Mining read it online: http://goo.gl/Gi0JRz

[8] Generating Ontologies from Linked Data GOLD

[9] Unsupervised learning of semantic relations for molecular biology ontologies, pdf