Result Relevancy
Hi,
I'm trying to understand how the search results are weighted due to a client being unhappy with the results they are getting for certain key terms.
For example, we have a single channel of pages and when they search for the term 'financial advisors', they expect that the page with a title of 'financial advisors' should come first, but in fact it is 5th (of 6 total results). In the collection definition we have given the title field a weighting of 3 and all other fields have a weighting of 1.
I've tried various combinations of loose_ends and search_mode parameters and observed the queries which are generated as a result. However, it seems that when using the Mysql Fulltext functions (MATCH/AGAINST) the relevance is just not correct.
Could this be a bug in MySql or have you found the accuracy of the relevance to be poor?
Thanks
James
Replies
Low 21 May 2013 11:16
MySQL's relevance calculation with MATCH/AGAINST is a bit of a black box. I've tried to find out exactly how it calculates it when I built the add-on, but couldn't find anything useful. Instead, I found out through trial and error, you can influence the score calculation by repeating text in the same field, which is what the weighting does in Low Search.
So, whilst weighting fields in LS will influence MySQL's relevance scoring of a search result, I have no control over what an actual person might find more relevant.
Adding higher weights isn't an option, as that might turn words into "common words" (according to MySQL), which will in turn result in a lower score.
In short, not much I can do. The scoring mechanism is a black box which I can only influence so much.