Different Types of Tuning in SearchUnify

Search tuning refers to fine-tuning the search engine to improve the relevance and accuracy of search results according to specific business needs and user preferences.

Types of Search Tuning

There are four types of Manual tunings in SearchUnify: Intent Tuning, Keyword Tuning, Content Tuning, and Custom Tuning.

• Intent Tuning is used to boost documents for a big (more than 8) set of keywords and its synonyms. Each keyword for which a document is boosted is referred to as an ‘Utterance’ and a group of similar utterances make an ‘Intent’. With Intent Tuning, you can boost a document for hundreds of similar keywords.

• Keyword Tuning is used to boost documents for a small set (less than 8) keywords. This feature is useful when you have a list of documents that you want to turn up at specific positions in search results for a defined set of queries.

• Content Tuning is used to personalize search. However, instead of a keyword, you can:

  • Boost Content Types is used to boost results from a content type. Example: The content type “cases” can be boosted on a Salesforce Console search client.

  • Boost Content Source is used to boost results from a content source. Example: The content source “Knowbler Articles” can be boosted on a Salesforce Communities search client.

  • Activate Field Match is used to boost results whose field value matches the search query. Example: Boosting is applied to the field “author” and the field has a value “Mark Twain”. When the user query is [Mark Twain], then all the documents authored by “Mark Twain” are boosted.

  • Boost documents based on field values is used to boost results with a specific field value. Example: If boosting is applied to the value “Mark Twain” in the field “author”. Then all the documents authored by “Mark Twain” are boosted, irrespective of the search query.

• Custom Tuning is used to enhance search results on your search client by adjusting document ranking based on these factors:

  • Solved (Discussion Status) prioritizes resolved discussions in search results.

  • Title (Keyword Match in Titles) prioritizes results whose titles match the search query.

  • Click Boosting Base (Popularity) prioritizes results with higher click-through rates.

  • Date Boosting (Age) priotizes recent results by reducing the ranking of old results over time.

Auto Tuning

Auto Tuning identifies user preferences by monitoring the user search behavior through cookies and modifies the order of the search results to offer more relevant information to the user. Auto Tuning relies on SearchUnify’s machine learning capabilities to auto-boost documents. It has five elements: Auto Boosting, Spell Corrector, Facet Interpreter, Re-Ranking, and Named Entity Recognition (NER).

• Enable Auto Boosting. Turn it on to use Searchnify’s ML capabilities.

• Enable Auto Spell Corrector. Turn it on to train SearchUnify on your business-specific terminologies.

• Enable Facet Interpreter. Turn it on and connect with a content source to automate filtering on search results.

• Re-Ranking. Turn it to shuffle the selected number of search results (ranging between 10 and 50) based on machine learning.

• Named Entity Recognition (NER). Turn it on to refine SearchUnify's ability to handle nuanced and complex queries.

Order of Preference in Tuning

Manual Tuning

The order of preference in Manual Tuning is Intent Tuning > Keyword Tuning > Content Tuning & Custom Tuning.

The preference in search is given to the Intent Tuning and Keyword Tuning, which are manual or static tuning types. This implies that as Admins you can define the rankings for a set of data. So, these will show in the search results based on the set rankings.

The Content Tuning and Custom Tuning are the superchargers. Admins set the value of each element in these tuning types. Based on the value, a score is drawn for each document and the docs are ranked in decreasing order of their score in the search results.

Each document in your data carries a base score. For example, if 5 is the value for Title-based boosting and the keywords found in the title of a document, then the base score is multiplied by five and the chances of it showing in the top search results increases.

However, this is not always the case. The other tunings also matter. For example, if one doc has 50 clicks and the other has the user query ‘keyword’ in the title, then the probability of the document with more clicks to show up in the top results is higher.

Or take another example. In document X, the keyword is only in the title but in other documents, the keywords repeat in the body of the text. In this case, the other documents will take preference.

All these calculations run at the backend. Documents with high scores are shown in the search results.

NOTE.

Don’t go overboard with Content Tuning. A high value for one content source can hide results from all other content sources.

Auto Tuning

Each element in Auto Tuning performs its specific task.

Auto Boosting takes care of boosting search results based on the user profiles and user patterns data. It depends on the actual conversions. If the search users apply a filter often, then the filtered results get a boost.

Spell Corrector suggests the right words if some user has mistyped something. All these operations run independently and have no dependency; except for NER, which works based on Taxonomy.

Preference in Manual Tuning and Auto Tuning

Manual Tuning and Auto Tuning run independently of each other.

Related:

• Test Your Tuning