Relevance Ranking Documenation

Anyone know where I can find information detailing how TREX assigns relevance to documents?

Hi Eric
Even though there is no detailed information on how TREX assigns relevance to documents, you could take a look at this thread: TREX Algorithms For Relevance Ranking of Results
I believe that could enlighten you a bit.
Kind regards,
Martin

Similar Messages

Relevance ranking, how?

Hi,
I am newbi in Oracle Text and I need help.
Is there any one who can tell me how to use the Relevance Ranking in Oracle Text?
best regards
Nasser

You can use the static module in the rel rank or if you wish to stratify to boost certain records to appear at the top of your search results then you can use this approach with your current rel rank search interface. This is an example using the wine demo with cross field matching enabled on the search interface All
In order for the query * portion to respect you current rel rank you will need to construct a Ntx query like so and not use Ns.
&Ntx=mode+matchallpartial+rel+stratify(collection()/record[P_WineID="89955"],collection()/record[P_WineID="73036"],*),All

Regarding relevancy ranking

This is regarding relevancy ranking. During migration from 2.1.1 to 11 the relevancy of products are different though the configuration is same.
So I am trying to find which ranking module is affecting this. While testing this, I am seeing strange behavior.
One v11.0, I see that NUMFIELDS module is giving two different relrank scores for a record in two different but similar queries.
In one query, I am allowing relrank modules to be chosen as it is configured in recsearch_config.xml and in other query I am specifying Nrr parameter, with same relrank modules.
Here are the log lines from Dgraph log
Query without Nrr
    Nty=1&Ntx=mode+matchall&Ntk=primary&Nx=whyrank&N=0&Ntt=book%20tape
Query with Nrr
Nrr=nterms,maxfield,numfields&Nrk=primary&Nrt=book%20tape&N=0&Ntk=primary&Ntt=book%20tape&Nty=1&Ntx=mode+matchall&Nx=whyrank
Result for the Query without Nrr
DGraph.WhyRank:
[ { "nterms" : { "evaluationTime" : "0.001220703125", "stratumRank" : "100", "stratumDesc" : "Matched 2 of 2 terms" }}, { "maxfield" : { "evaluationTime" : "0.001220703125", "stratumRank" : "26", "stratumDesc" : "field match", "rankedField" : "p_name_pr" }}, { "numfields" : { "evaluationTime" : "0.001220703125", "stratumRank" : "1", "stratumDesc" : "Matched 1 of 54 fields", "fieldsMatched" : ["p_highsm_stmt"] }} ]
DGraph.WhyDidItMatch:
p_highsm_stmt: Was Gaylord® Bookcraft® 6 mil Paper Write-On Book Tape
Result for the Query with Nrr
DGraph.WhyRank:
[ { "nterms" : { "evaluationTime" : "0.001708984375", "stratumRank" : "100", "stratumDesc" : "Matched 2 of 2 terms" }}, { "maxfield" : { "evaluationTime" : "0.001708984375", "stratumRank" : "26", "stratumDesc" : "field match", "rankedField" : "p_name_pr" }}, { "numfields" : { "evaluationTime" : "0.00244140625", "stratumRank" : "4", "stratumDesc" : "Matched 4 of 54 fields", "fieldsMatched" : ["Category", "p_highsm_stmt", "p_mfg_sku_pr", "p_name_pr"] }} ]
DGraph.WhyDidItMatch:
p_highsm_stmt: Was Gaylord® Bookcraft® 6 mil Paper Write-On Book Tape
You can see that it says 4 fields matched in one results and 1 field matched in another results.
Cross field is enabled for the search interface.
Does anybody know what's happening here. I would like to get the results as they come in "With Nrr" query for "Without Nrr" case too.

Hello!
   We replaced --stat-brel with the following URL parameters to enable "whyrank" properties on the results (you can see them in a JSP reference app automatically when enabled):
   Why Rank is enabled per-query by using the existing Dx parameter (Dx=whyrank), or the new Nx parameter (Nx=whyrank) in the URL.
   Results are returned in a new Dgraph.WhyRank property.
   They provide a little bit more info than the old --stat-brel flag.

Configuring relevance ranking evaluator 2.1.2 to MDEX engine 6.4.0

Hi CAN some body explain me why is this occuring when i am trying to configure relevance ranking evaluator to my current endeca work bench
Unable to query the MDEX Engine.
Specified MDEX Engine is:
Host: localhost
Port 15000
MDEX Engine query failed: make sure that an MDEX Engine is listening on the configured host and port. Verify that you're using the appropriate API version for your Endeca installation. The jar files in WEB-INF/lib should match those of the MDEX Engine version.
If you would like to connect to a different MDEX Engine, click the "Change MDEX Engine" button above.

Hi Praveen,
Follow below steps to configure:
1. Copy rel rank evaluator in
<Endeca>/PlatformServices/6.1.3/tools/server/webapps (You can keep it any where on disk but provide correct path in relrankEvaluator.xml)
2. Update path in relrankEvaluator.xml accordingly.
3. Copy relrankEvaluator.xml to
<Endeca>/PlatformServices/workspace/conf/Standalone/localhost (you can copy this to <Endeca>/ToolsAndFrameworks/3.1.2/server/workspace/conf/Standalone/localhost mostly preferred approach)
4. Copy updated endeca libs from jspref application to relrankEvaluator lib dir
5. Restart platform services (or EndecaToolsService if you have copied/configured relrankEvaluator.xml with TF)
6. Try to access the URL,
http://<host name>:8888/relrankEvaluator/index.jsp (or http://<host name>:8006/relrankEvaluator/index.jsp if configured with TF)
If issue still persist click on change MDEX engine, enter <host name>, <port> and click ok on config box. This should work
HTH
-Pravin

IFCORE-1565: The Dgraph could fail when using a relevance ranking strategy.

Hi,
I just saw the release notes for Endeca 6.2.2 and read this bug:
IFCORE-1565: The Dgraph could fail when using a relevance ranking strategy.
Is there someone who could provide some more information on this bug? What is it and when does it occur? Reading the word 'could' would suggest a kind of unpredictability?
Thanks,
Maarten

The known trigger for the bug was doing relrank on a query that contained multiple text searches.

Relevance Ranking Module

Is the Relevance Ranking Evaluator Module in Workbench available for Oracle Tools Framework 3.1?

Hi. Yes, it's available as a Workbench Extension and is still a separate package (for now). The previous version works with 3.1, so it's still:
Oracle Endeca Relrank evaluator 2.1.2 for Generic Platform

Maxfield relevance algorithm ranking dimensions higher than properties

We are using MaxField as one of the relevance algorithm in our search interfaces and observed that dimensions with low rank is matched first than compared to property of higher rank.
Records that match with dimensions are ranked in the results though their "relevance_rank" is lower than property. But the endeca documentation explains that records will be ranked based on "relevance rank" and doesnt treat dimension and property with differnce.
Here's the example
+<SEARCH_INTERFACE CROSS_FIELD_BOUNDARY="ALWAYS" CROSS_FIELD_RELEVANCE_RANK="0" DEFAULT_RELRANK_STRATEGY="All" NAME="All">+
+     <MEMBER_NAME RELEVANCE_RANK="77">D_Type</MEMBER_NAME>+
+     ....+
+     ...+
+     <MEMBER_NAME RELEVANCE_RANK="50">P_Type</MEMBER_NAME>+
+     .....+
+     .....+
+     <MEMBER_NAME RELEVANCE_RANK="64">D_Category</MEMBER_NAME>+
+     ......+
+     ......+
+     <MEMBER_NAME RELEVANCE_RANK="1">P_Name</MEMBER_NAME>+
+</SEARCH_INTERFACE >+
The records that match for D_category are given higher rank, though it matches for the property "P_Name".
Need some validattion here, if our understanding is correct.
-M

Muthu,
Still records matching D_Category where retrieved first compared to P_Name. Currenlty have modified this search interface to include only "MaxFiled". Are you setting the Relevancy Rank module via a URL query
EXAMPLE
controller.jsp??N=0&Ntk=P_Desc&Ntt=sonoma&Nrk=All&Nrt=pear&Nrr=field&Nrm=matchall
Or via a relrank_strategies.xml. If the later have a look at your ./{APP_DIR}/data/forge_output/ directory and verify if the {app_prefix}.recsearch_config.xml contains the order you expected. If ./{APP_DIR}/data/forge_output/ does -not- include your changes, you may be using the Forge Configuration Manager (FCM), the hyEnd integration with hybris, or another customization that updates ./{APP_DIR}/data/processing and is overriding your changes.
In the example of the hyEnd integration, if the changes -can- be found in ./{APP_DIR}/data/complete_index_config you can copy your changes after ConfigurationGeneratorForge has run but before Forge runs
EXAMPLE
<script id="BaselineUpdate">
<bean-shell-script>
<![CDATA[
          // Generate instance configuration
         ConfigurationGeneratorForge.archiveLogDir();
         ConfigurationGeneratorForge.run();
          cp("/app/dir/apps/MuthuTest/data/complete_index_config/recsearch_config.xml.xml","/app/dir/apps/MuthuTest/data/processing/MuthuTest.recsearch_config.xml");
         // archive logs and run ITL
         Forge.archiveLogDir();
         Forge.run();
         Dgidx.archiveLogDir();
         Dgidx.run();
         ]]>
</bean-shell-script>
</script>
Thanks,
-=Dalyn Bell
Oracle Software Support

Endeca Relvance Ranking on Workbench - not working

Hi,
I have installed the Workbench, Platform service, Endeca Server 7.4 and now installing and deploying Relevance Ranking module. I have have followed all steps as provided in the document. ONce I open the workbench, I am able to see the Relevance application but when I open it it gives me following error. :
HTTP Status 404 -
type Status report
message
description The requested resource () is not available.
Apache Tomcat/6.0.20
I am also not able to see the image which I specified in the ws-extensions.xml
Any verification step or any correction then do let me know. I am trying to do relevance ranking on dataset in MDEX on urgent basis.
Thanks and Regards,
Ashwani

The Oracle Endeca Information Discovery product ships with a different version of the Endeca Server (aka. MDEX engine). It boasts a web service API whereas the Oracle Endeca Guided Search product still uses the Java API which the relevancy rank module was written against.
If you're ambitious enough, you could re-"plumb" the relevancy rank module to work against the Endeca Server's web services (assuming there are no search "troubleshooting" features that are no longer supported in OEID). At this point, you'd have something completely custom, and helpful, if you could get it to work.
HTH,
Dan
http://branchbird.com

Failed to start ranking update due to this error - Authoritative Pages

Hello together,
i always get an error if i add an Authoritative Web Page and use the Option "Relevance Ranking Analytics".
The error is:
Failed to start ranking update due to this error: Failed to connect to system manager. SystemManagerLocations: net.tcp://SERVER/64BFD2/AdminComponent1/Management
Does anyone have a solution?
The Search Service component´s are running!
Best Regards
Darius.K

Hi! I had this issue and I was able to resolve it by retrying from an application server (I was trying to run from an Index server and that's not where the admin component was located in my topology)

Dimension Search - Rel Rank Not Working

I see for 6.4.0, 6.4.1, and 6.4.1.1 there is a known issue:
IFCORE-1637: A dimension search query with D=* produces results ordered
according to the default dimension value ranking, regardless of the value
of the Dk (Dimension Search Rank) parameter. Relevance ranking works as
expected for other types of dimension search queries
Has anyone figured out an acceptable workaround that doesn't involve post-processing to re-sort?

Ah, I think I answered this in your other thread. D=* isn't really a search (it's actually a special feature they added in a few releases ago) to list all dimension values. My guess is that since it's not a search, the MDEX doesn't pass the results through relevancy ranking to re-order them.
Note also that D=* can be very very slow in a production environment. Building out the dimension values + ancestors can be very slow.
Also, when you start working with D=* you'll notice that it can do some strange things. For instance, if your hierarchy includes:
Electronics->Cameras->Lens
.... then in the results you will see:
- Electronics
- Electronics->Cameras
- Electronics->Cameras->Lens
So you'll probably want to do some post-processing + caching anyway because of that.

Rel Rank Not Working with Boost & Bury Catridge

Hi everybody,
I am facing problem on record boost and bury,
The individual records that are selected for boost or bury. It is my understanding that, while we can boost or bury certain target records,
the unselected records would still be ranked according to the Rel Rank strategy, unless a sort property was engaged (Price, Width, etc.).
The cartridge does have drop-downs for Sorts and Relevance Ranking, both of which I switch to default,but the ranking of the unselected records still seems incorrect.
I have set Relevance Ranking using EQL,
String relStrategy = "static(P_ImageValid,descending),nterms,maxfield,glom,static(P_VendorInventory,descending),static(P_Popularity,descending)";
ERecRank relRank = new ERecRank("Site_Search", relStrategy, "matchall");
query.setNavRelRankERecRank(relRank);
I am using PBX-0.6.
Can any one help me out, how can i get result in Relevance Ranking order.
Any help is appreciated.
Thanks,
Sunil

Sunil
You need to use the stratify as part of the rel rank strategy. Here is an example using the pbx example wine app with crossfield match turned on in the search interface 'All' . This is just the parameters but it shows you how to construct the rel rank with bury and boost.
In order for the query * portion to respect you current rel rank strategy as defined in the 'All' search interface you will need to construct a Ntx query like so and not use Ns.
&Ntx=mode+matchallpartial+rel+stratify(collection()/record[P_WineID="89955"],collection()/record[P_WineID="73036"],*),All
Hope this helps
Terry

Is it possible to implement iSupplier Portal for Employee type Vendors

We have iSupplier portal for external suppliers which we customized to show only invoices and payment advices.
Is it possible to implement the same to our internal employees without creating them as new suppliers ?
We want employees use iSupplier portal to view their expense(payments).

If you are using presentation API then as Nagarajan said, you can use Endeca Dimension Search to power your TypeAhead/Predictive search. Below are a few pointers that may help.
Trigger typeahead only after user enters a few characters (ideally after 2-3 characters).
Introduce a delay of a few milliseconds between your Endeca Searches.
Limit the Dimensions to include using the Di parameter.
Control the number of dimVals returned and other Dimension behavior using Drc and Dp
Use Drc (to evaluate InStock products) in a seperate query with D=* to get a list of all valid DimVal IDs. Cache them and use this list to remove any out of stock dimVal IDs from the real query results.
Use Dk and Dx to control match mode and your relevance ranking strategy to sort your results.
Oracle Commerce Services also offers a prepackaged service called Endeca Predictive Search (EPS) module that can be quickly implemented to power TypeAhead. EPS provides other configurable features like caching, record search, frequent matches ,etc. Reach out to your Oracle representative if that intersts you.
All the best! I hope this helps.
Best,
Kaush
P.S. Google limits the search result to 4 which, works for them and might work for you but to exploit the full potential of Predictive Search based on Endeca, try HomeDepot.com, Staples.com, etc. for inspiration.

Help using XML as Dictionary key

I'm not quite understanding the key in Dictionaries. I thought if the same key was passed twice it would still only create one pair in the dictionary. Can anyone explain this to me?
With a key of a string or number it seems to work as I would expect, but the XML as key doesn't seem to work. The line in there that changes the title was just to check if the XML was being passed as a reference or value. It is a reference. So it seems like the dictionary should know the key is the same. Why doesn't it?
I'm trying to create an index of the content in some XML Nodes and want to be sure that various nodes don't get added to the index more than once. I also My thinking was to use the XML node itself as as the key. It is the only thing I have in the data that is for sure unique. Is there a way to do what I want to do?
var xmlData:XML=<data>
<session>
<title>Document 1</title>
<desc> <![CDATA[blah blah blah]]></desc>
</session>
<session>
<title>Document 2</title>
<desc> <![CDATA[blah blah blah]]></desc>
</session>
</data>;
var d:Dictionary=new Dictionary();
var xml1:XML=xmlData.session[0];
var xml2:XML=xmlData.session[0];
addToDictionary(xml1);
addToDictionary(xml2);
xmlData.session[0].title="New title";
for(var a:* in d){
trace(a.title+": "+d[a]);
function addToDictionary(xml:XML){
d[xml]=int(Math.random()*500);
traces
New title: 2
New title: 135

If I change the assigment to:
var xml1:XML=xmlData.session[0];
var xml2:XML=xml1;
addToDictionary(xml1);
addToDictionary(xml2);
Then it does work as I had hoped it might. In any event...
I'm trying to make a (forward) index of some properties of each node in an XML—total word count, frequency of each word, etc. So later when my reverse index returns several XMLs I can quickly look up those data and compute a tf-idf for relevance ranking.
I'm thinking the conversion toString() will slow the whole thing down? I'll give it a look-see.

Using "Restrict by app" in change query causes issues with Search Results

Hello!
I am hoping some experienced SP admin can help me out. I am having issues with the relevance ranking for my search results web part.
I created a Wiki (OOTB) within a site collection and on the home page of the Wiki I was looking to apply a Search Box and Search Results web part for easy searching. I wanted to ensure the search results were restricted to the scope of the pages within the
wiki. I did the following after adding the 2 web parts:
Entered edit mode for the page
Started to edit the Search Results web part
Under Properties for Search Results clicked on "Change Query"
Navigated to BASICS and clicked "Switch to Quick Mode"
For Select Query, picked "Wiki (System)" option
For Restrict by app, picked "Specify a URL" and input the wiki URL "http://webapp.com/sites/sitecollection/wiki"
Click OK
The web part has 6 automatic results when I hit the page. These are the only 6 pages in my wiki. When I search a word or title of a page, they are not ranked in order or relevance. The ordering is static to what was being shown before the search, but the
keywords I'm searching are bold. With out the taking the steps above, I can easily search these pages and have them ranked appropriately so I'm not too sure what I'm doing wrong.
Other Details
The pages have been crawled and can be searched from anywhere else in SP.
I am only a site collection admin for this specific site collection.
I other than the steps above, I have not made any changes or customization to the search web parts; they are both out of the box SP.
The SP Farm admin we unable to pin point what the root cause is either.

Hi Mikael,
Sorry for the wait, I was unable to post pictures, which i felt would better describe my issue.
Could you output the query template you are ending up with?
If you can tell me how to accomplish this, then I can provide it. Please keep in mind that I am only a site collection admin.
Also, is the ranking/sorting for your webpart set to ranking?
When looking at the web part settings, I didn't see an option to modify the ranking. Unfortunately, some sites in my intranet are down now so I can't even double check. Once it's back up, I will verify this as well.
And how do you know they are ranked incorrectly?
I wanted to display an image showing what I meant by this. As the wiki is down, I'll do my best with words. When I visit the wiki home page, I already have the search and search results web parts there. Without searching anything, the search results
web part displays 6 pages that are available in the wiki. When I search, I expect to see the most relevant result at the top. similar to the following:
*Search result from another section of the intranet site.*
What I receive instead is the same list of wiki pages, in the same order. The most relevant result
is not moved to the top like in the image above, but I can see that the word "rush" is highlighted in bold. I hope this helps.
If you need more of a description, please let me know.
Thank you

Is there any feature in Endeca to support the natural language search ?

I am looking to evaluate Endeca for implementing search for a help-center kind of a site where there are list of questions and answers available and the requirement is to provide a search for this content. The guest user can type in question or some search terms in the box to get the right question and ultimately the answer to the question.
However; what I see from the basic tests is that there will be demand of natural language search for verbal matching. I understand that some part of this feature is supported by stemming in Endeca but that dictionary is not the complete list. It will not be a scalable solution to have business update this word forms in the stemming dictionary time and again for such requirement.
Are there any thoughts on the above problem ? What can be the possible ways to move ahead ?
Thanks,
Nit

Here's a few tips:
- First, Endeca doesn't do natural language processing. It is very literal with the words that are entered and matched.
- By default, the MDEX limits the number of search terms to 10 by default. Meaning if you entered 11 or more terms, terms #11 and so-forth would be ignored.
- You can increase this by setting the --search_max flag (typically configured in the <dgraph-defaults> section of your appConfig.xml
- Now, you can increase this to be quite large (say 200 terms). Then, when a customer enters in their problem description, it won't ignore terms.
- Use the "Match Any" match mode. Now, this might give you a lot of search results, but that goes to my next point.
- Use the NTerms match mode as your first relevancy ranking module.   This will order the results based on the number of terms
- Make sure you have a good set of "Stop Words", like "the a and it was is of", etc.
So now what will happen is that the customer will enter their problem, and you'll submit the whole thing to the mdex. The MDEX will throw out the stop words, and then order the results based on the number of terms that matched.
Some other relevancy modules to think about:
- WFreq / Weighted Frequency. This counts the frequency of words that are more important.
- Stay away from Phrase and Exact. They'll almost certainly be too slow.
- Use a Static module based on the view count of an article, so that more popular articles will show up first.
You might also want to turn off these MDEX features:
--wb_noibrk    Disables word-break insertion analysis.
--wb_norbrk    Disables word-break removal analysis.
If you're going to submit big sets of search terms, those features will probably be too slow.

Relevance Ranking Documenation

Similar Messages

Maybe you are looking for