I just tried this query.
It actually worsened because of "cleaner" local results included because of some decent SEOing - and thats a huge case in point!
Obviously, keyword search is NOT going to cut it on the mobile. Users are unlikely to reframe queries too often, or spend too much time. In fact, will a purely index/repository based approach ever be enough ? Unless topical matching gets really smart, mobile search is still some time away from maturing.
Web search works this way :
There's this mother-of-all-repositories, currently called Google (given that Cuil managed to get them to respond, cannot take this for granted forever), and there's a crude keyword-based-matching over it. Yeah I know there's pagerank and all that, but that just improves a crude mechanism a little teeny bit, and if the words don't really get the page with the answer out - you try again. If you're a power user, you have a few more tricks up your sleeve (thats less than 2% users) but most users try different sets of keywords. Effort, time, and only YOU can understand your query.
Current state of "semantic" engines :
They disambiguate a bit, help with with some clusters, options etc. An improvement, sure, but quite quite far from the average Joe who "gets it" in a sec!
Powersets help, sure. Related stuff - related concepts - and you're a little closer.
Can an engine ever get close enough ?
Maybe not. But combining a few of these semantic search approached, measuring the weighted output - and refer it to someone if thats weak - is one way to go.
Or create a topic map of fragments of information, and search through the map. The map will HAVE to contain metadata about whats really jey to the topic, and what incidental. And does a set of related topics begin to match the key concepts ? If not, can people be involved in that step, to do a better match on the repository ?
Alright - thats a 20,000 feet level, but I'm not giving away more, for sure :) Some of this is happening at Ziva, some is planned. Its a long journey though.
Vanilla keyword search is surely not enough - and I'm sure a decade or so from now we'll look back and laugh at ourselves for being so in awe of the current mother-of-all-indices :)
Update: tried the same query on Zook - and got a response a response from a use in about 10 minutes (low priority pure info query, i guess) - 15 mill comps, 5 mill net connections, GoI aiming at 75mill comps, 45 mill net connections by 2010. Correct ? Will check...
It actually worsened because of "cleaner" local results included because of some decent SEOing - and thats a huge case in point!
Obviously, keyword search is NOT going to cut it on the mobile. Users are unlikely to reframe queries too often, or spend too much time. In fact, will a purely index/repository based approach ever be enough ? Unless topical matching gets really smart, mobile search is still some time away from maturing.
Web search works this way :
There's this mother-of-all-repositories, currently called Google (given that Cuil managed to get them to respond, cannot take this for granted forever), and there's a crude keyword-based-matching over it. Yeah I know there's pagerank and all that, but that just improves a crude mechanism a little teeny bit, and if the words don't really get the page with the answer out - you try again. If you're a power user, you have a few more tricks up your sleeve (thats less than 2% users) but most users try different sets of keywords. Effort, time, and only YOU can understand your query.
Current state of "semantic" engines :
They disambiguate a bit, help with with some clusters, options etc. An improvement, sure, but quite quite far from the average Joe who "gets it" in a sec!
Powersets help, sure. Related stuff - related concepts - and you're a little closer.
Can an engine ever get close enough ?
Maybe not. But combining a few of these semantic search approached, measuring the weighted output - and refer it to someone if thats weak - is one way to go.
Or create a topic map of fragments of information, and search through the map. The map will HAVE to contain metadata about whats really jey to the topic, and what incidental. And does a set of related topics begin to match the key concepts ? If not, can people be involved in that step, to do a better match on the repository ?
Alright - thats a 20,000 feet level, but I'm not giving away more, for sure :) Some of this is happening at Ziva, some is planned. Its a long journey though.
Vanilla keyword search is surely not enough - and I'm sure a decade or so from now we'll look back and laugh at ourselves for being so in awe of the current mother-of-all-indices :)
Update: tried the same query on Zook - and got a response a response from a use in about 10 minutes (low priority pure info query, i guess) - 15 mill comps, 5 mill net connections, GoI aiming at 75mill comps, 45 mill net connections by 2010. Correct ? Will check...
.jpg)
No comments:
Post a Comment