How Auto Tuning Works and Its Features

Relevancy is the common thread uniting your diverse users who, irrespective for their role and goal, expect fast and relevant results which is tricky to deliver because relevancy means different things to different users.

A developer is rarely concerned with how many documents there are on a knowledge base. If a search for "analytics" does not return API documentation on the first page, then the results are irrelevant to him. On the contrary, admins tends to be more interested in reports. If a search engine cannot navigate an admin to reports documentation, then its results are irrelevant to the admin. Auto Tuning helps admin cater both the developer and the admin simultaneously.

Auto Tuning transforms SearchUnify into a context-aware search engine that does much more than connecting a search query with documents. It takes into account a user's profile, search history, permissions, and several other factors to return results that more unique and relevant at the same time. An admin can enhance user experience significantly using Auto Tuning's four features: Auto Boosting, Enable Auto Spell Correcter, Facet Interpreter, and Rich Snippets.

How Auto Tuning Works

Auto Tuning keeps a record of user activity and analyzes the data thus collected to identify patterns. The patterns are then used to change the order of search results for individual users based on their activity.

The diagram captures user journey. User 1 creates four sessions whose data is stored and analyzed in SearchUnify's machine learning engine. The engine compares User 1's activities with other searchers to find behavioral similarity. Once a threshold for behavioral similarity is reached, SearchUnify starts personalizing results. In the figure, User 1 gets more relevant results in the fifth session. Based on the analysis of the first four sessions, SearchUnify tries to deliver more relevant results.

Then in a sixth session, the activities of the first five sessions are considered. And it goes on and on.

The activities' of User 1 have no impact on the results of User 2 unless both happen to have similar search journeys. Search results of User 2 are personalized based on the data processed from his own activity.

Features

• Enable Auto Boosting. Machine learning kicks in when Auto Tuning is turned on. Toggle Enable Auto Tuning to the right to get started.
• Enable Auto Spell Corrector. Teach the terms specific to your business to SearchUnify by connecting it with a content source. If SearchUnify is aware of your organizational terminology (which can be markedly unique), it will recognize those terms in search queries and never show incorrect suggestions. For example, "Salesforce" will never be corrected to "sales force."
• Enable Facet Interpreter. Improve search experience by automatically applying filters. When turned on and connected with a content source, SearchUnify improves user experience by applying filters behind the scenes each time a search query matches a facet value. For example, a search for "Mark Twain" on an online bookshop will fetch five extra results if "Mark Twain" is also a filter value in the Authors filter.
• Enable Rich Snippets. Provide answers to how-to kind of queries right on the search page by extracting lists or steps from relevant articles. Toggle Enable Rich Snippets to turn on. No other configuration is required.
• Re-Ranking. Turning it on shuffles the top 50 results based on a machine learning. The ranking is based on the click patterns.

• Named Entity Recognition (NER). NER works when an admin has set up Taxonomy in NLP Manager. Based on the taxonomy selected by the admin, NER generates its own model for search clients. On searching, the query hits a machine learning API which returns keywords presented in the search query and documents. Then, based on these keywords, results are filtered and presented.