Custom Tuning: Boost most popular and latest documents

This article shows you how you can personalize results on a search client by boosting documents based on any or all of these four factors: 

  • Solved (Discussion status)
  • Title (Keyword match in titles)
  • Click Boosting Base (Popularity)
  • Decay Rate (Age)

To view any custom settings, select a search client and toggle on Custom Boosting.

Solved

This feature is exclusively available on search clients installed on communities, such as Salesforce Community and Khoros communities. When a user searches for a topic and multiple discussion threads are found, the threads with the status "solved" are prioritized and displayed before unresolved threads. The level of prioritization can be adjusted using the "Solved" slider. A multiplier value of 1 has no effect on solved threads. However, any other value, such as , increases the relevance score of solved threads by times, making them more likely to appear higher in the search results.

Title

Title takes advantage of keyword matches. You can match Title with any object field. If a keyword matches one or more words in the Title of a document, then that document can be pushed to (or near) the top of search results. This is done by inflating its relevancy score.

A multiplier value of 1 has no effect. However, any other value, such as , increases the relevance score of title match by times, making them more likely to appear higher in the search results.

Title is turned off by default. Please write to your SearchUnify representative to turn it on. Don't forget to mention the object field you wish to map Title to in your email.

Click Boosting Base

When Click Boosting Base is enabled, more popular documents rank higher in the search results. SearchUnify uses a logarithmic algorithm to rank documents, which ensures that a document with 1,500 clicks ranks significantly higher than one with only 10 clicks over the same period. Simultaneously, the algorithm ensures that documents with similar click counts—such as 1,450 and 1,500—are ranked similarly.

By default, all search queries are analyzed for intent. Click boosting is applied only when SearchUnify can identify the intent behind a query with high confidence. The confidence threshold is configured by the SearchUnify team after a discussion with the customers. If document D has been clicked 10 times for intent I, D's relevance score for that intent is increased. Administrators can modify this boost rate by changing the ML Score Multiplier (the default value is 10).

In the default configuration, clicks from all search clients are counted. For example, if you have one search client for employees and another for customers, clicks from both will influence rankings. However, administrators can restrict this by unchecking Include Clicks from All Search Clients. This is useful when you want to prevent one demographic's search behavior—such as that of customers—from affecting the results for another group, such as employees.

Fig. A snapshot of the Click Boosting Rate configuration in Custom Tuning.

Attach to Case Boosting

Attach to Case Boosting is similar to Click Boosting Rate but three crucial differences:

  1. Attach to Case Boosting works only on Salesforce
  2. Attach to Case Boosting works only when its almost eponymous SearchUnify addon is set up on a Salesforce org
  3. Instead of counting the number of clicks, Attach to Case Boosting considers how many times support rep link an article to a case. The more often an article is linked, the higher will be its probability of appearing in search.

As in Click Boosting Rate, you can select how quickly an article can rise in results.

Date Boosting

In its default state, SearchUnify serves new documents first. "Document.1.1" is more likely to be put ahead of "document.1.0" and "document.0.9", and older versions. An as admin, you can increase the probability from "more likely" to "almost always." 

The idea is to increase the relevance score gap between the latest version and old versions by reducing the relevancy of the old versions. Each newly-updated document has a score S and it remains S for a time period, called Offset. You can select an Offset value in the 1-month to 3-year range.

Once Offset has passed, S begins to drop. It continues to drop for a time period, known as Scale. Scale can be as short as a month or as long as three years. Use the dropdown to select a value for Scale.

The rate of drop of relevance score during Scale is controlled by Decay Rate, which you can enter in the Decay Rate box or which you can select by moving the slider.

At the end of first Scale, the new relevancy score S (new)  equals S x Decay Rate and the entire process starts again and goes on ad infinitum.

NOTE.

Don't use Date Boosting if the documents published months or years ago are still relevant to search users. If Date Boosting is active, then the users will not be able to find old documents easily through search.

Test Your Tuning

You can review the impact of tuning before applying Content Tuning to production. Check out Test Your Tuning.