Psychic Search: Improve search results using a self learning algorithm
Google has a “I’m feeling lucky” button that takes you directly to the first search result. In a perfect world, when a user enters a search term, the first result(s) should be what the user intended to find. However, this is easier said than done. Search results are ranked by relevancy, and are dependent on a large number of variables. This means that the ranking can be changed by tweaking these variables. Here is one way to do this, and it involves [automatic] feedback from the user. When a user enters a search and subsequently clicks on one of the results, both are stored, together with the sequence number in the list. For example, see image below, the user searches for ‘hooiber’, then clicks on the 4th result. What we store is the search term, the clicked result and the sequence (which is 4).
Now, after 1000′s of searches from all users, we have a large collection of data that is being fed back into the system. For example in this case, the system detects that ‘Jobs’ are more requested than ‘Primary Conductors’ and can then tweak the search algorithm to boost ‘Jobs’ in the ranking. The end result would eventually look like this:
The end result is better search for users. And if users change their behavior or new data enters the system, the system continuously adapts to provide better results.
Comments are closed.