[Last call][Design] Search - Relevance

[D John Anderson]

Consider this situation: Suppose you are looking for items by  
searching for "/Search John Anderson". Lucene returns all the items  
with the word John followed by the work Anderson with a higher rank  
than all the items with words between John and Anderson. This can be  
a really handy feature.

While I'd be the last person to use this single example as proof that  
Lucene rank is generally useful, I think it's equally premature to  
use Jeffrey's experience to conclude that Lucene rank is never  
useful.  I think we should leaving rank in Preview, if for no other  
reason than getting more experience with it.

However, whether or not we include rank is independent from what the  
default sort (or view) is -- after all you can always change the sort  
or view to be whatever you like. So I don't think it's necessary to  
have the default sort be rank.

Finally, I think the issue of letting users change views is  
independent of the rank/sort or search issues. I find changing views  
to be useful in lots of situations other than search.

John


On Feb 6, 2007, at 4:38 PM, Mimi Yin wrote:

> Are there any more thoughts on this issue? Is there any objection  
> to sorting search results by Triage status in the 'Dashboard view'  
> for Preview, thereby simplifying usability issues having to do with  
> the View selector menu items as well?
>
> Mimi
>
> On Feb 2, 2007, at 11:48 AM, Mimi Yin wrote:
>
>> Sorting by Triage status would clear up the confusion around the  
>> Views as well. Instead of having a Search view, we could just have:
>>
>> + Day View
>> + Week View
>> + Table View
>>
>> All are viable options, whether or not the user has searched for  
>> anything.
>>
>> Mimi
>>
>> On Feb 1, 2007, at 6:21 PM, Jeffrey Harris wrote:
>>
>>> Hi John,
>>>
>>>> Almost all search in the world today sorts results by relevance  
>>>> (e.g.
>>>> Google), so I suspect people will be familiar with the concept. Of
>>>> course that doesn't mean the relevance is very meaningful.
>>>>
>>>> If we don't sort the results by PyLucene relevance, what do we  
>>>> sort them
>>>> by? The obvious options include one of our columns in the table  
>>>> or some
>>>> new column. Which would you choose?
>>>
>>> In playing around with search briefly, I haven't found the PyLucene
>>> relevance to bear much relationship to how I'd expect relevance  
>>> to be
>>> weighted.
>>>
>>> For instance, when welcome is searched for, an item who's entire
>>> contents is a title of "Welcome welcome welcome" is sorted last  
>>> (among
>>> several items containing the word welcome), which seemed pretty  
>>> odd to me.
>>>
>>> When I search for "Welcome in both", the item titled "Welcome in  
>>> both"
>>> comes up significantly lower ranked than other items that have  
>>> neither
>>> "in" nor "both" in their title or body.
>>>
>>> So while I love the idea of getting search by relevance, I think  
>>> I'd be
>>> frustrated by the hint that relevance was a factor in my searches  
>>> if I
>>> didn't believe the relevance scores were accurate.
>>>
>>> Probably making it better would take some tweaking and testing of  
>>> the
>>> parameters we're handing to PyLucene.
>>>
>>> I'd lean towards sorting and sectioning by triage status for now,  
>>> unless
>>> other people's experience with this is better than mine.
>>>
>>> Sincerely,
>>> Jeffrey
>>> _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
>>>
>>> Open Source Applications Foundation "Design" mailing list
>>> http://lists.osafoundation.org/mailman/listinfo/design
>>
>> _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
>>
>> Open Source Applications Foundation "Design" mailing list
>> http://lists.osafoundation.org/mailman/listinfo/design
>
> _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
>
> Open Source Applications Foundation "Design" mailing list
> http://lists.osafoundation.org/mailman/listinfo/design