How can I make my own version of IBM Watson? IBM 왓슨 제작
2016.04.30 01:30
IBM 왓슨 제작 참고
How can I make my own version of IBM Watson?
100+ million people read Quora every month
Sign in to follow people with unique insights on Artificial Intelligence
By continuing, I agree that I am at least 13 years old and have read and agree to the Terms of Service and Privacy Policy.
3 Answers
Phillip Rhodes,
Views
Building a literal duplicate of Watson would be damn tough and cost a lot of money. But you can do some similar things and get a start a few ways...
For an overview of how to build a "watson-like" you can start here:
https://www.ibm.com/devel
For more on some of the Open Source software you would use to build something like Watson, look into NLP software, semantic knowledgebases, and reasoning software. Some good starting points include:
http://opennlp.apache.org
http://uima.apache.org
GATE.ac.uk - index.html
Quepy: A Python framework to transform natural language questions to queries.
http://jena.apache.org
Watson also apparently used Prolog heavily, so you might want to look into Prolog resources. I cataloged a few here a while back:
Prolog? I'm Going To Learn Prolog??
For some interesting research on interacting with natural language queries, see:
http://start.csail.mit.ed
And you might also look at another project that bills itself as "an open source wolfram alpha", here:
SymPy Gamma
Be prepared to invest quite a lot of time and energy.
For an overview of how to build a "watson-like" you can start here:
https://www.ibm.com/devel
For more on some of the Open Source software you would use to build something like Watson, look into NLP software, semantic knowledgebases, and reasoning software. Some good starting points include:
http://opennlp.apache.org
http://uima.apache.org
GATE.ac.uk - index.html
Quepy: A Python framework to transform natural language questions to queries.
http://jena.apache.org
Watson also apparently used Prolog heavily, so you might want to look into Prolog resources. I cataloged a few here a while back:
Prolog? I'm Going To Learn Prolog??
For some interesting research on interacting with natural language queries, see:
http://start.csail.mit.ed
And you might also look at another project that bills itself as "an open source wolfram alpha", here:
SymPy Gamma
Be prepared to invest quite a lot of time and energy.
This book contains a lot of the prerquisite knowledge needed to get started.
It covers the front end search portion.
Taming Text: How to Find, Organize, and Manipulate It: Grant S. Ingersoll, Thomas S. Morton, Andrew L. Farris: 9781933988382: Amazon.com: Books
I would follow this up though with resources on interfacing PROLOG with database using a Fuzzy model.
I haven't read this one yet due to its cost but going through the table of contents this one should be high on your list.
Knowledge Management in Fuzzy Databases (Studies in Fuzziness and Soft Computing): Olga Pons, Maria A. Vila: 9783540668367: Amazon.com: Books
You won't be able to get enough performance out of the high end PC to do a large knowledge base. So you'll probably want to plan on a high end PC cluster so that you can scale your solution. Many existing programming languages do not do distributed comping without a lot of complex software infrastructure. Since NOSQL already provides you a way of scaling your knowledge out access multiple nodes as documents, I would recommend one of these as a scalable data storage mechanism.
It covers the front end search portion.
Taming Text: How to Find, Organize, and Manipulate It: Grant S. Ingersoll, Thomas S. Morton, Andrew L. Farris: 9781933988382: Amazon.com: Books
I would follow this up though with resources on interfacing PROLOG with database using a Fuzzy model.
I haven't read this one yet due to its cost but going through the table of contents this one should be high on your list.
Knowledge Management in Fuzzy Databases (Studies in Fuzziness and Soft Computing): Olga Pons, Maria A. Vila: 9783540668367: Amazon.com: Books
You won't be able to get enough performance out of the high end PC to do a large knowledge base. So you'll probably want to plan on a high end PC cluster so that you can scale your solution. Many existing programming languages do not do distributed comping without a lot of complex software infrastructure. Since NOSQL already provides you a way of scaling your knowledge out access multiple nodes as documents, I would recommend one of these as a scalable data storage mechanism.
I think that's a question which is being answered in a variety of places. Actually, not an open source version of Watson in any literal sense, but there are open source projects, which include:
OAQA which is a student project actually associated with Watson
Watsonsim another student project
YodaQA -- I built and played with this one
AKSW OpenQA
AKSW QA
AquaLog
and my own OpenSherlock (was SolrSherlock).
Most of those are at GitHub or elsewhere on the web. OpenSherlock has not really found its way to GitHub, though early components of it are at GitHub under the old name SolrSherlock. Update: OpenSherlock now has its own repo: OpenSherlock
There are many papers and slides online about those various projects. One slidedeck for SolrSherlock is here
SolrSherlock: Linkfinding among Biomolecules with Literature-based Di…
and at the end of this month, a slidedeck for a conference talk about OpenSherlock will be at my slideshare account (root of the above link).
My experience with YodaQA is interesting: it's UIMA based, uses Solr for indexing Wikipedia, and uses Fuseki for indexing DbPedia and Yago, then answers questions. It took me a while to bring it up on a Windows platform (used Cygwin for parts of it). In terms of question answering, it, and perhaps most of the open source so-called "cognitive systems" are not particularly accurate. I believe that's simply because a year or so of open source experiments is no match for the zillions of person-years invested by IBM in Watson. That's not to say open source won't rise to the occasion; I believe it will.
I did not mention:
OpenCog
DeepDive
earlier because they are not strictly question-answering platforms; but they each include powerful tools very useful in such a project; DeepDive does have an app written on it that approaches question answering.
OAQA which is a student project actually associated with Watson
Watsonsim another student project
YodaQA -- I built and played with this one
AKSW OpenQA
AKSW QA
AquaLog
and my own OpenSherlock (was SolrSherlock).
Most of those are at GitHub or elsewhere on the web. OpenSherlock has not really found its way to GitHub, though early components of it are at GitHub under the old name SolrSherlock. Update: OpenSherlock now has its own repo: OpenSherlock
There are many papers and slides online about those various projects. One slidedeck for SolrSherlock is here
SolrSherlock: Linkfinding among Biomolecules with Literature-based Di…
and at the end of this month, a slidedeck for a conference talk about OpenSherlock will be at my slideshare account (root of the above link).
My experience with YodaQA is interesting: it's UIMA based, uses Solr for indexing Wikipedia, and uses Fuseki for indexing DbPedia and Yago, then answers questions. It took me a while to bring it up on a Windows platform (used Cygwin for parts of it). In terms of question answering, it, and perhaps most of the open source so-called "cognitive systems" are not particularly accurate. I believe that's simply because a year or so of open source experiments is no match for the zillions of person-years invested by IBM in Watson. That's not to say open source won't rise to the occasion; I believe it will.
I did not mention:
OpenCog
DeepDive
earlier because they are not strictly question-answering platforms; but they each include powerful tools very useful in such a project; DeepDive does have an app written on it that approaches question answering.
[출처] https://www.quora.com/How-can-I-make-my-own-version-of-IBM-Watson
본 웹사이트는 광고를 포함하고 있습니다.
광고 클릭에서 발생하는 수익금은 모두 웹사이트 서버의 유지 및 관리, 그리고 기술 콘텐츠 향상을 위해 쓰여집니다.
광고 클릭에서 발생하는 수익금은 모두 웹사이트 서버의 유지 및 관리, 그리고 기술 콘텐츠 향상을 위해 쓰여집니다.
