How many document total in your corpus? And how many do you intend to have?
My point is that if you are testing this with a small corpus, the results are very likely different than when you test on a reasonable corpus. So if you expect your "real" index will contain many more docs than what you're testing, this is likely a red herring. But something isn't making a lot of sense here. You say you've traced it to having a docfreq of 2 that changes to 1. But that means that the value is unique in your entire corpus, which kind of indicates you're trying to boost on unique values which is unusual. If you're confident in your model though, the only way to guarantee what you want is to optimize/expungeDeletes. Best, Erick On Tue, Sep 1, 2015 at 7:51 PM, Derek Poh <d...@globalsources.com> wrote: > Erick > > Yes, we see documents changing their position in the list due to having > deleted docs. > In our searchresult,weapply higher boost (bq) to a group of matched > documents to have them display at the top tier of the result. > At times 1 or 2 of these documentsare not return in the top tier, they are > relegateddown to the lower tierof the result. Wediscovered that these > documents have a lower score due to docFreq=2. > After we do an optimize, these 1-2 documents are back in the top tier result > order and their docFreqis 1. > > > > On 9/1/2015 11:40 PM, Erick Erickson wrote: >> >> Derek: >> >> Why do you care? What evidence do you have that this matters >> _practically_? >> >> If you've look at scoring with a small number of documents, you'll see >> significant >> differences due to deleted documents. In most cases, as you get a larger >> number >> of documents the ranking of documents in an index with no deletions .vs. >> indexes >> that have deletions is usually not noticeable. >> >> I'm suggesting that this is a red herring. Your specific situation may >> be different >> of course, but since scoring is really only about ranking docs >> relative to each other, >> unless the relative positions change enough to be noticeable it's not a >> problem. >> >> Note that I'm saying "relative rankings", NOT "absolute score". Document >> scores >> have no meaning outside comparisons to other docs _in the same query_. So >> unless you see documents changing their position in the list due to >> having deleted >> docs, it's not worth spending time on IMO. >> >> Best, >> Erick >> >> On Tue, Sep 1, 2015 at 12:45 AM, Upayavira <u...@odoko.co.uk> wrote: >>> >>> I wonder if this resolves it [1]. It has been applied to trunk, but not >>> to the 5.x release branch. >>> >>> If you needed it in 5.x, I wonder if there's a way that particular >>> choice could be made configurable. >>> >>> Upayavira >>> >>> [1] https://issues.apache.org/jira/browse/LUCENE-6711 >>> On Tue, Sep 1, 2015, at 02:43 AM, Derek Poh wrote: >>>> >>>> Hi Upayavira >>>> >>>> In fact we are using optimize currently but was advised to use expunge >>>> deletes as it is less resource intensive. >>>> So expunge deletes will only remove deleted documents, it will not merge >>>> all index segments into one? >>>> >>>> If we don't use optimize, the deleted documents in the index will affect >>>> the scores (with docFreq=2) of the matched documents which will affect >>>> the relevancy of the search result. >>>> >>>> Derek >>>> >>>> On 9/1/2015 12:05 AM, Upayavira wrote: >>>>> >>>>> If you really must expunge deletes, use optimize. That will merge all >>>>> index segments into one, and in the process will remove any deleted >>>>> documents. >>>>> >>>>> Why do you need to expunge deleted documents anyway? It is generally >>>>> done in the background for you, so you shouldn't need to worry about >>>>> it. >>>>> >>>>> Upayavira >>>>> >>>>> On Mon, Aug 31, 2015, at 06:46 AM, davidphilip cherian wrote: >>>>>> >>>>>> Hi, >>>>>> >>>>>> The below curl command worked without error, you can try. >>>>>> >>>>>> curl http://localhost:8983/solr/techproducts/update?commit=true -H >>>>>> "Content-Type: text/xml" --data-binary '<commit waitSearcher="false" >>>>>> expungeDeletes="true"/>' >>>>>> >>>>>> However, after executing this, I could still see same deleted counts >>>>>> on >>>>>> dashboard. Deleted Docs:6 >>>>>> I am not sure whether that means, the command did not take effect or >>>>>> it >>>>>> took effect but did not reflect on dashboard view. >>>>>> >>>>>> >>>>>> >>>>>> >>>>>> >>>>>> On Mon, Aug 31, 2015 at 8:51 AM, Derek Poh <d...@globalsources.com> >>>>>> wrote: >>>>>> >>>>>>> Hi >>>>>>> >>>>>>> I tried doing a expungeDeletes=true with the following but get the >>>>>>> message >>>>>>> 'missing content stream'. What am I missing? I need to provide >>>>>>> additional >>>>>>> parameters? >>>>>>> >>>>>>> curl >>>>>>> 'http://127.0.0.1:8983/solr/supplier/update/json?expungeDeletes=true >>>>>>> '; >>>>>>> >>>>>>> Thanks, >>>>>>> Derek >>>>>>> >>>>>>> ---------------------- >>>>>>> CONFIDENTIALITY NOTICE >>>>>>> This e-mail (including any attachments) may contain confidential >>>>>>> and/or >>>>>>> privileged information. If you are not the intended recipient or have >>>>>>> received this e-mail in error, please inform the sender immediately >>>>>>> and >>>>>>> delete this e-mail (including any attachments) from your computer, >>>>>>> and you >>>>>>> must not use, disclose to anyone else or copy this e-mail (including >>>>>>> any >>>>>>> attachments), whether in whole or in part. >>>>>>> This e-mail and any reply to it may be monitored for security, legal, >>>>>>> regulatory compliance and/or other appropriate reasons. >>>>>>> >>>>>>> >>>> >>>> ---------------------- >>>> CONFIDENTIALITY NOTICE >>>> >>>> This e-mail (including any attachments) may contain confidential and/or >>>> privileged information. If you are not the intended recipient or have >>>> received this e-mail in error, please inform the sender immediately and >>>> delete this e-mail (including any attachments) from your computer, and >>>> you must not use, disclose to anyone else or copy this e-mail (including >>>> any attachments), whether in whole or in part. >>>> >>>> This e-mail and any reply to it may be monitored for security, legal, >>>> regulatory compliance and/or other appropriate reasons. >> >> > > > ---------------------- > CONFIDENTIALITY NOTICE > This e-mail (including any attachments) may contain confidential and/or > privileged information. If you are not the intended recipient or have > received this e-mail in error, please inform the sender immediately and > delete this e-mail (including any attachments) from your computer, and you > must not use, disclose to anyone else or copy this e-mail (including any > attachments), whether in whole or in part. > This e-mail and any reply to it may be monitored for security, legal, > regulatory compliance and/or other appropriate reasons.
