There’s been a big back-and-forth about how big an impact QA can make on information-retrieval. A few points:
Writing a proper natural language question isn’t necessarily easier than writing a Google query. You don’t always know what you need to know, and providing feedback from the QA system is very difficult. You can iterate towards the right query fairly well, as your searching teaches you better keywords.
The percentage of queries that can be answered by a single sentence is substantial, but might be smaller than you think. People also use search engines to navigate around, to find long articles of interest, and to carry out tasks (e.g. shopping).
The strictly informational queries probably aren’t that important to Google’s revenue. The best queries to be serving are the ones where the user wants to buy something, because that’s where people will pay for advertising. If a competitor takes away the informational searches but can’t serve the commerce searches too, I doubt Google will be sweating much.
True. When I said that, I was thinking of a service that does what Watson does and gives Google-style answers.
So, if the query “What is the capital of the United States” was made, at the top it would say Washington D.C. and after that it would show search results, similar to how Google shows answers to unit conversion searches.
So, if the query “What is the capital of the United States” was made, at the top it would say Washington D.C. and after that it would show search results,
You mean like… erm… Google does at the moment if you search for “What is the capital of the United States?”?
There is a site, True Knowledge, that attempts to answer such questions using methods similar to Watson’s.
It relies on NLP and “facts”; for example, the query “What is the capital of the United States” relies on this fact:
This fact asserts that the relationship ‘”is the capital of”’ exists between “Washington, D.C.” and “the United States” at some point in time. Other facts in the knowledge base assert that this fact applies for the following time periods:
Each “fact” is assessed by a variety of sources, both human and automated. For example, the fact above has, as some of its assessments:
Fact extracted from Factacular using natural language processing of the following snippit “The United States of America has the capital Washington, D.C..”
There’s been a big back-and-forth about how big an impact QA can make on information-retrieval. A few points:
Writing a proper natural language question isn’t necessarily easier than writing a Google query. You don’t always know what you need to know, and providing feedback from the QA system is very difficult. You can iterate towards the right query fairly well, as your searching teaches you better keywords.
The percentage of queries that can be answered by a single sentence is substantial, but might be smaller than you think. People also use search engines to navigate around, to find long articles of interest, and to carry out tasks (e.g. shopping).
The strictly informational queries probably aren’t that important to Google’s revenue. The best queries to be serving are the ones where the user wants to buy something, because that’s where people will pay for advertising. If a competitor takes away the informational searches but can’t serve the commerce searches too, I doubt Google will be sweating much.
True. When I said that, I was thinking of a service that does what Watson does and gives Google-style answers.
So, if the query “What is the capital of the United States” was made, at the top it would say Washington D.C. and after that it would show search results, similar to how Google shows answers to unit conversion searches.
You mean like… erm… Google does at the moment if you search for “What is the capital of the United States?”?
Damn, I was afraid it would show that. So a more difficult query.
Although, that is a sign that Google is already experimenting with that idea, only with a far simpler algorithm.
There is a site, True Knowledge, that attempts to answer such questions using methods similar to Watson’s.
It relies on NLP and “facts”; for example, the query “What is the capital of the United States” relies on this fact:
Each “fact” is assessed by a variety of sources, both human and automated. For example, the fact above has, as some of its assessments:
and