lucene-solr-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Jan Høydahl <>
Subject Re: Faceting
Date Tue, 22 Feb 2011 10:40:10 GMT

Even if the customer types a correct product name, how do you know that merchant A and merchant
B both have registered that exact product in the same way?

Merchant A may say as product name "White Sony LCD TV XY123" and the other says "Sony XY123
LCD TV", colour=white

If you're serious about price comparison service, I think you need to invest in finding what
products are the same before indexing, and then tagging them with some unique normalized name.
Then when after a search, you show a facet with that normalized name and first when the user
has selected the correct facet, can you be 100% certain that you're comparing apples to apples.

Jan Høydahl, search solution architect
Cominvent AS -

On 22. feb. 2011, at 07.23, Praveen Parameswaran wrote:

> Hi ,
> @Tommaso @Jan Høydahl Thanks for the response :)
> I 've done it almost similar to what Tommaso suggested and yes it's about
> 70-80% accurate.
> I understand the contradiction in the search - customer find stuff without
> the exact right wording (recall) at the same time as you want the query to
> be precise (precision).
> In my scenario both cases are there as well, but mostly a customer would
> know which product name he is searching for and he will be interested in
> comparing the prices that different marchants offer. What I feel is that ,
> may be the "Search" itself has to be classified based on the contexts.
> Will it be possible in solr to have the below:
> 1 . A customer uses the correct product name to search , get the accurate
> results
> 2.  A customer uses a keyword or without the exact name , get the most
> relevant results.
> 2nd part is fine as it's working good. 1st part is where I'm struggling.
> thanks
> Praveen
> On Mon, Feb 21, 2011 at 5:23 PM, Tommaso Teofili
> <>wrote:
>> Hi Praveen,
>> as far as I understand you have to set the type of the field(s) you are
>> searching over to be conservative.
>> So for example you won't include stemmer and lowercase filters and use only
>> a whitespace tokenizer, more over you should search with the default
>> operator set to AND.
>> Then faceting over those field(s) will depend on those type settings.
>> You may find the following wiki page useful:
>> My 2 cents,
>> 2011/2/21 Praveen Parameswaran <>
>>> Hi,
>>> Is it possible to have 100% accuracy for facet counts using solr ? Since
>>> this is for a product price comparison site I would need the search to
>>> return accurate results. for example if I search "sony lcd Tv" I do not
>>> want
>>> "sony Led Tv" to be returned int he results.  Please let me know if this
>> is
>>> possible and how?
>>> Thanks
>>> Prav

View raw message