Understanding Search Results Ranking
When you select to sort search results by relevance, ranking is the process during which results are sorted by relevance to your query from the most to the least pertinent. The Coveo Cloud Platform ranks search results by calculating a relevance score based on a series of ranking factors. The score spans from minus infinity to infinity. The higher the score, the higher the result will be in the result list.
The relevance score is a combination of the index ranking algorithm, and other relevance modifiers such as QRE (Query Ranking Expressions) and QRF (Query Ranking Functions) [see Index Ranking Phases, Managing Query Pipeline Ranking Expressions, and Ranking Functions].
Because of their nature, featured results (see Managing Query Pipeline Featured Results) and Coveo Machine Learning ART-recommended results [see Automatic Relevance Tuning (ART) Feature] are not affected by ranking and always appear at the top of the result list.
Coveo Cloud organization members with the required privileges can modify the relative weight of some factors in order to fine-tune ranking (see Managing Query Pipeline Ranking Weights).
ART models produce query ranking expressions to recommend results (see Managing Coveo Machine Learning Automatic Relevance Tuning Models in a Query Pipeline).
You perform the “Washing Machine” query against an appliance vendor, and two results are returned. To learn why the results are in that specific order, you inspect their relevance score in the Debug panel.
You first take a look at the index ranking. The first result (product A) has “Washing” and “Machine” in its title and contains several occurrences of “washing machine” in its content. Therefore, the index sets the result score at 5 000. The second result (product B) has only “machine” in its title, so the index gives the result a score of 3 000.
You then analyze how the Coveo ML ART feature impacted the ranking. Since product B is clicked more often than product A and that users usually do not return to the search page to perform another query after consulting the product page, the ART model boosts the product B score. ART adds ten times the ranking modifier set in the model configuration, and considering 250 is the default value, 2 500 is added to the product score.
So far, the score for product A is 5 000 and 5 500 for product B.
Finally, you remember your marketing team had an incentive to promote product A. The team created a query ranking expression with a ranking modifier of 100, which is also multiplied by ten. Consequently, the expression adds 1 000 points, pushing the product A score to 6 000, which is higher than the product B one at 5 500. Hence why product A is the first returned result.