Whatson

This is the repository for an experimental Question-Answering-System developed in the context of the "Special Topics in Artificial Intelligence" WS14 course at the University of Osnabrück.

Installation on Ubuntu

git clone https://github.com/mome/watson
cd watson
./install_requirements.sh

System Package Requirements :

• python2.7
• python-pip
• java
• graphviz (optional, for parsetree printing)
• mplayer (optional, for speech synthesis)

Python Requirements :

• nltk
• html2text
• rdflib
• wikipedia

External Requirements :

• Stanford Parser
• Stanford NER-Tagger
• Stanford POS-Tagger (optional, future use)

ToDo:

• introduce answer choice function
• integrate document_search() into QA
  • some function from question to document_search() call (question types)
  • answer generation (answer types)
• document_search(): recognize compound terms
• document_search(): introduce answer confidence
  • keyword to NER-match distance
  • NER-match in same sentence/subsentence (not only in paragraph)
• expand NER to more than 3 or 7 classes (with some ontology: cyc, dbpedia, ...)
• more document sources than wikipedia (schoolarpedia, wikibooks, maybe own corpus storage (netstore??))
• document indexing
• introduce intaractive learning (human gives feedback)
• have a look on dependency trees, stanford relation extractor, stanford classifier, Stanford Temporal Tagger