HoverboardEdgard MarxAKSW openqa.aksw.org
Links
http://openqa.aksw.org http://github.com/aksw/openqa http://bitbuket.com/emarx/openqa http://bitbuket.com/emarx/openqa/faq
Agenda
What is openQA? When should I use or not use openQA? History (idea, versions, problems) Myths and truths Hoverboard
Architecture Features
Conclusion & Future Work
What is openQA ?
Frameworks Document Retrieval
Lucene Solr Elastic (formal Elastic Search)
Question Answering ?
Not Frameworks Sina Hawk TBSL Google
It is a framework for question answering systems.
When should I use or not use openQA ?
You should use if: Beginner in QA systems Do not want to write everything from scratch Have a social life and yet publish a good paper Benefit from the work of other specialists Share your code/approach with the community
You should not use if: Does not need any of the built-in functionalities Best coder ever and you belive nobady codes better than you do Have plenty of time You are a sad person
History
Proposed in 2013 by Prof. Dr. Sören Auer Framework Open-source Combine different approaches Name from a visit to Bonn
History
v0.0.4-beta (prototype) Features
Different query parser plug-ins (SINA, TBSL) Designed for be used in two different enviroments Webserver/Standalone
Problems Fixed pipeline Difficult to deploy Bugs
v0.0.6-beta (No Websever) No Web-server
v0.0.7-beta Hoverboard Flexible pipeline Webserver no Infograph (JSF)
More than 300 downloads Used by students in Iran, Brazil, India,
Germany and Turkey!
History
2014
2015
Myths and truths
It has a heavy architecture Myth, payload for the architecture is 0 milleseconds.
It‘s huge Myth, the whole framework is 1.3 MB
very few in comparision with systems like SINA (~800Mb) and TBSL(~400)
It just does not work Myth, in fact users misunderstood problems in the plugged systems with
the framework
It‘s not like Google True
It‘s useless for industry True. openQA is useful for QA test
The plugged approaches are not mature as the industry ones (a chicken egg problem) I can not travel through time
True
Myths and truths
HoverboardArchitecture
Component
Service
Query Parser
Retriever
Synthesizer
Context
Tokenizer
Service
Old New
HoverboardArchitecture
Service 1 Service N
Service 2
Component 1
Component 2
Component 3
Component 4
Component 5
What Can I do with it?
HoverboardFeatures
Front-EndTest / Evaluation (Benchmark)Flexible pipelineBult-in JSON/RESTHybrid Architecture (Web Service, Micro(Super, Mega
Hyper)-service, Plug-in)NLP approaches (Lucene/Standford NLP)
Conclusion & Future Work
Is useful for test and benchmarkImprove the platformIntegrate other benchmarksIncrease the number of plugged approachesIt might be (future) used by ordinary
users and industry
Thanks
http://openqa.aksw.org