|
iDEA: Drexel E-repository and Archives >
Drexel Academic Community >
College of Information Science and Technology >
Research Day Posters (IST) >
Automated question answering over the web: An adaptive search and retrieval strategy
Please use this identifier to cite or link to this item:
http://hdl.handle.net/1860/1578
|
| Title: | Automated question answering over the web: An adaptive search and retrieval strategy |
| Authors: | Israel, Quinsulon L. |
| Keywords: | Pattern recognition
Question-answering systems
Query expansion
Search strategy
Strategy selection
Text mining
Web mining
Natural language processing
Search engine |
| Issue Date: | 17-Apr-2007 |
| Publisher: | Drexel University. College of Information Science and Technology. |
| Series/Report no.: | IST Research Day 2007 posters |
| Abstract: | The problem of efficiently finding answers to natural language questions over the web
has gained much attention. Currently, useful experimental models for implementing
question answering work well only for smaller, specific collections of documents and/or
they only handle short, single factoid-type questions. Other more generally focused
models retrieve and re-rank only a set of documents most likely to contain an answer.
These approaches rely on only a few specific strategies to implement question answering.
A more comprehensive and dynamic model of a question answering system may provide
better performance for both retrieving candidate answer pools and extracting specific
answers.
Such a new model will be designed that efficiently combines automatic question
reformulation, search strategy selection, query expansion, and answer extraction/pooling
techniques. The system will automatically learn question reformulation for the most
popular web search engines, based on training collections of question answer pairs, such
as FAQs. Questions will be matched against automatically learned question types and
reformulated into queries based on answer phrases likely to appear in a document
containing the answer. The semantic answer type will also be determined based on the
question type and used to recognize potential answers. During system training, the top
ranked documents retrieved by the search engine will be examined for their likelihood of
containing an appropriate answer. User answer-acceptance feedback will be collected to
re-rank documents entries and/or refine new queries as necessary during live runs. |
| URI: | http://hdl.handle.net/1860/1578 |
| Appears in Collections: | Research Day Posters (IST)
|
Items in iDEA are protected by copyright, with all rights reserved, unless otherwise indicated.
|
www.drexel.edu
