Boost Documents for Specific Keywords

Keyword allows you to fix the rank of a document for one or more search queries. This feature is useful when the goal is to return a list of documents at specific positions in search results for a defined set of queries.

PREREQUISITES

The keyword (or one of its synonyms) for which a document is boosted should occur somewhere in the doc.

To use Keyword Tuning, navigate to Search Tuning > Manual Tuning > Keyword.

Select a Search Client for Manual Tuning

Manual Tuning is not universal. You can apply it to a search client without impacting other search clients. To get started, pick the search client where tuning is going to be applied.

Go to Search Tuning > Manual.
Select a search client.

NOTE.

When a search client is bookmarked, it's always selected as the default search client in Select Search Client. To bookmark a search client, click .
Pick a tuning type from Intent, Keyword, Content, and Custom.

Enable Keyword Tuning

In Keyword Tuning, if a document has been assigned, let's assume, rank five for one or more queries, then the document will always turn up on that position provided that the keyword(s) or its synonym(s) are available in the boosted document.

NOTE.

Keyword tuning is incompatible with Re-Ranking in Auto Tuning. You can use either Keyword Tuning or Re-Ranking but not both.

Only for Salesforce. When a boosted article is updated, its status changes to inactive and the boosting no longer works. After updating an article in Salesforce, delete the keyword boosting on the old version and apply it to the new version.

Turn on Boost Documents for Keywords to view settings and run a search.

NOTE.

Turn on Apply Bypass Formula if the query returns no results or few results. Turning it on allows admins to see all the results for a search query irrespective of the current tuning settings.
Find a document for boosting and click Add Keywords.
A dialog opens. Enter the same query or a query pattern in Regex and click Add.
Move the slider to select a rank between one and 10. In the image, the rank four has been selected.
When Bypass All Filters is checked, keyword and intent tuning are force applied, irrespective of the selected filters.
- Here's an illustration of what happens when Bypass All Filters is checked.
  The highlighted result has a Comment Count of zero but the filter is applied to return results only with the comment count of one. Bypass All Filters ignores the applied filter.
  The highlighted result has no answers but the filter is applied to return only Answered results. Bypass All Filters ignores the applied filter.
Click Save.

Click Test Your Tuning to verify and apply the changes in search results.

Regex and Keyword Tuning

Documents can be boosted for queries and regular expressions.

When document_x is boosted for the query key at position five, then it may not show up on position five if a user searches keys. In Keyword Boosting, key and keys are different keywords.

On the other hand, a regular expression, such as /key?/i covers both variants and the document_x is ranked correctly for both of them.

NOTE.

When a document has been boosted for a query and a regular expression, then query ranking takes precedence.

An admin can boost document_x for /key?/i at position eight. If another admin boosts document_x now for the query keys at position five, then document_x will show up at position five for the query keys.

The precedence is in the order:

A way to get the most out of regular expressions is to narrow the scope of patterns.

.* matches all queries. Such a pattern is not of much use.

A pattern, such as /search/, is little better than a query.

The ideal regex is somewhere between these extremes, where scope of pattern expansion is limited to a couple of dozen keywords and not hundreds or thousands of potential queries. Here are a few examples.

Pattern	Boosts
^(.?)reporting 2.0.$	Documents that contain “reporting 2.0”. Example: <any_word> reporting 2.0 <any_word>.
^.(reporting 2.0\|working).$	Documents that contain either “reporting 2.0” or “working” or both. Example: <any word> working <any word>
^.(?=.reporting 2.0)(?=.working).$	Documents that contain both “reporting 2.0” and “working” in any order. Either or none doesn't work. Example: <any_word> working reporting 2.0 <any_word> OR <any_word> reporting 2.0 working <any_word>
(.?)reporting 2.0(.)(?=working).*$	Documents in which "reporting 2.0" comes first, followed by "working". Example: <any_word> reporting 2.0 <any_word> working
^.(?i)(Broadway\|Organizational Units).$	An OR condition. Example: <any_word> broadway <any_word> and <any_word> organizational <any_word>
^.(?=.language)(?=.disable\|.remove\|.deactivate).$	Documents that contain "language" and one or all of the 3 keywords: "disable", "remove", or "deactivate." Example: <any_word> language <any_word> disable
^.(?=.language)(.)(?i)(enable\|add\|activate\|install).$	"Language" must be present. Other keywords in search query must contain at least one of three keywords: "enable", "add", "activate", or "install." Two keywords must be separated by at least one character. Example: <any_word> Enable language <any_word>
^.(?=.upgrade\|.1297)(?=.instance\|.cdm).$	Documents that contain either "upgrade" or "1297" along with either "instance" or "cdm". Example: <any_word> 1297 cdm <any_word>
(.?)CENG([-])([1-9][0-9]).$ (.?)([])Version([#])([1-9][0-9]).*$	"CENG" is followed by hyphen followed by numeric digits. Example: <any_word> CENG-123 <any_word>

Import Tuning

Keyword boosting is a time-consuming affair. An admin analyzes search behavior and organizational goals before curating a list of results to boost. During the course of a few months or years, the admin can end up boosting hundreds of documents on a search client. Great as it is for end users, without Import Tuning the entire thing can turn into a maintenance nightmare if the Moderator, Admin, or Super Admin has to boost exactly the same documents on another search client.

When a content source is linked with multiple search clients, then an instance user can copy the keyword tuning settings from one search client to another.

Example 1. When keyword boost settings can be imported.

An article with the title "Unable to Login" has been boosted for the keyword login for rank one on search client . The same article is boosted for the keyword unable on the search client to rank two. On import, the article will be boosted for two different keywords for two separate ranks.

Example 2. When keyword boost settings cannot be imported.

An article with the title "Unable to Login" has been boosted for the keyword login for rank one on search client and for the rank two on the same search client. You cannot import boost settings in this case.

Import Tuning saves a lot of repetitive work. It works excellently for content sources shared between two or more search clients. With the theory clear, you can now proceed to import keyword-tuning settings from one search client into another.

Click Import Boosting.
From Select Search Client, pick a search client from which tuning settings will be imported. A search client named "Keyword Tuning" has been selected in the image.
In the Select column, check the content sources and content types. In the image, the tuning settings applied on the documents stored in the content source StackOverflow with Tags and content type Question will be imported.
Click Import.

Boosted Documents Tab

Synonyms Boost

Boosted Documents offers a featured called Apply the same Keyword Tuning on the added Synonyms/Abbreviations.

Checking it extends the scope of keyword boosting to its synonyms. A document boosted for a keyword, let's say, SU, and it's boosted for all its added synonyms, such as searchunify and sfy.

NOTE.

Imagine a situation where 3 documents have been boosted for the query SU, 7 documents for the query searchunify and 4 documents for the query sfy.

In this case, searching for SU will bring more results from searchunify because it has the most documents.

When there is a parity between two synonyms, the results for a synonym are picked at random.

Also, when the query and the boosted keyword match is partial, then the keyword synonyms are ignored.

For the synonym boost to work, a precondition is to use a string instead of a regex pattern. Keyword boosting a document for the pattern Search* will not be automatically boosted for synonyms such as SearchUnify and SU.

Boosted Documents

The list of boosted documents can quickly spiral into a maintenance nightmare. In such scenarios, instance users can resort to Boosted Documents where they can look up boosted documents, change their rank, and add more keyword to the boosting.

The tab features a list of docs (column Document) along with the keyword(s) (column Search Keyword) for which they are boosted, and the boosted rank (column Rank). You can remove a boosting with the delete button in column Action. When Bypass All Filters is checked, keyword and intent tuning are force applied irrespective of the selected filters.

After making the changes, click Test Your Tuning and then Save the settings.

Boosted Queries

This tab features a list of keywords for which one or more documents have been boosted.

The tab features a list of keywords (column Search Keyword) and the documents boosted for that keyword (column Document). The Document column names the Content Source and Content Type where the document is stored. The rank for which the docs are boosted is in the column Rank.

You can remove a boosting with the delete button in column Action.

When Bypass All Filters is checked, keyword and intent tuning are force applied irrespective of the selected filters.

After making the changes, click Test Your Tuning and then Save the settings.

Deindexed Documents

If you no longer see the boosted document in the search results, it is possible that the document been removed from the SearchUnify index. You can use the Deindexed Documents to see the list of documents that have been deindexed. You can also see the Boosted Keywords for which the document was boosted.

To remove the document, use the delete icon from Action column.

Top Search Queries with Unranked Result

This report is right next to Export Tuning and contains suggestions documents that can be boosted. Click to open it.

The report presents a list of documents (Clicked Results) which users find useful (measured by the number of clicks in #Clicked) but which are buried deep inside search results (captured in Avg. Click Position).

Example: Assume that users clicked "Linux server file-based projection job reports" 18 times. Each time the users navigated beyond page one. A way to improve user experience is to boost the document from Action for the keyword listed in Search Query.

In the dialog that opens, assign the document a rank. You can boost up to 10 documents for a keyword. When you try to boost the 11th document, an error appears: "Top 10 results for this query are already boosted. Kindly, unrank one of those to be able to boost this result."

Once a document has been boosted, its status changes from "Boost" to "Boosted" in Action.

You cannot assign two documents the same rank but you can assign a document the same rank for more than one keywords.