[Developers] Search Query

[Faye Li]

Hi Arthur,

I hear you loud and clear and I think there's no dispute that relevance 
logic has a long way to go.

I think part of the issue may be that the search results don't always 
show why they are selected. Because the Wikipedia article blurb 
displayed for each topic is static, it isn't immediately observable 
that, for example, KNBR is a sister station of KSAN-FM. I'm not sure if 
this will change in the near future though.

Limiting searches to only titles and aliases is something we've tried 
and didn't like. Taking away relevance takes away the semantic part of 
the search string and reduces it to lexicon matches, where the string 
"george bush" better matches a 19th century biblical scholar than the 
current U.S. president or his father, where "beatles white album" will 
give you the album of that name by a less-known musician Rodrigo 
Constanzo before the Beatles album you probably meant (which, 
incidentally, is not called the "White Album" but "The Beatles"), and 
where "new york" will give you countless songs and albums of the same 
name before New York City. Searching for "LA, CA" (Los Angeles, 
California)? Forget it, because there's no such alias and the topic name 
is "Los Angeles" in Freebase, and if all relevance does is title/alias 
matching, if you don't know the name of a topic, you won't find it.

In fact, topic names are given a lot of weight -- than than a word match 
from the Wikipedia article. That said, Wikipedia articles contain a lot 
of valuable information. The number of times the phrase "Los Angeles" 
appears in the article for that city is about 100, and the number of 
times it appears in the article for the 1980 album of the same name by X 
is 8. That information helps relevance rank the city topic of "Los 
Angeles" ahead of the album topic of "Los Angeles" when someone enters 
that search phrase.

BTW, the newest branch of relevance server is going through the test 
cycle right now. It boosts exact title matching and yields some winners 
as well as losers. It should be released soon.

--  Faye

P.S. Metaweb is looking to hire a Key Relevance Engineer to help us with 
search: http://www.metaweb.com/jobs/relevanceengineer.html


Arthur van Hoff wrote:
> Hi Faye,
>
> It would be really nice to be able to turn the relevance off. For
> example, I would like to be able to search for a name or alias in my
> application with high relevance, and not get any of the results that are
> hits from Wikipedia. I find that the Wikipedia generated matches are
> confusing and often do not improve the search results.
>
> In general I think that matches on a name/alias should always be ranked
> higher than Wikipedia matches.
>
> Thanks.
>
> -----Original Message-----
> From: developers-bounces at freebase.com
> [mailto:developers-bounces at freebase.com] On Behalf Of Faye Li
> Sent: Friday, December 14, 2007 11:37 AM
> To: For discussions about MQL,Freebase API and apps built on Freebase
> Subject: Re: [Developers] Search Query
>
> Christopher,
>
> Topic name is just one of many things that are considered by search. 
> Other examples include topic alias and Wikipedia article, if there is 
> one. This allows users to do queries like "he who must not be named" and
>
> get "Lord Voldemort" as the top result.
>
> The first topic is the image for the "107.7 The Bone" station, topic 
> "/topic/en/ksan-fm" is the topic for that station, topic 
> "/topic/en/knbr" shows up because it's a sister-station of "The Bone", 
> and the last one in your result, "/topic/en/kfjc" shows up apparently 
> because it had  competed with "The Bone" as a radio station, with shared
>
> personal and a tangled history.
>
> -- Faye
>
>
> Christopher Dummy wrote:
>   
>> I'm just wondering how the Search Query determines what to return. I 
>> know it will do some thing based on the name. Does it also do things 
>> that are related somehow? Where relatedness might just be similar 
>> types? The query I'm looking at is:
>>
>>
>>     
> http://www.freebase.com/api/service/search?query=bone&type=/user/cotton/
> default_domain/broadcast 
>   
> <http://www.freebase.com/api/service/search?query=bone&type=/user/cotton
> /default_domain/broadcast> 
>   
>> There is a radio station called, "107.7 The bone", and it shows up as 
>> the first query. But there are some other stations that also show up, 
>> and I cannot find any reference in them to "bone". So it is a little 
>> surprising that they show up in the search results.
>>
>> Anyone have a good idea of why?
>>
>> -- 
>> Christopher Cotton
>> Zing/Dell Engineer
>> http://zing.net/
>>
>>     
> ------------------------------------------------------------------------
>   
>> _______________________________________________
>> Developers mailing list
>> Developers at freebase.com
>> http://lists.freebase.com/mailman/listinfo/developers
>>   
>>     
>
> _______________________________________________
> Developers mailing list
> Developers at freebase.com
> http://lists.freebase.com/mailman/listinfo/developers
> _______________________________________________
> Developers mailing list
> Developers at freebase.com
> http://lists.freebase.com/mailman/listinfo/developers
>
>