jaaprood / fuel-search Goto Github PK

When searching we expect a relevance-percentage as input, but till now we are returning a score (or actually: the number of transformations).

One would expect the output to be accompanied by relevance again.

At the moment the scores of words are calculated per occuring word. Search terms consisting out of multiple words, however, should also be matched to combinations of words out of the fields. The problem is that we can't simply go every single possibility, as the number of iterations could go up dramatically and make searching through data very slow.

My first thought is to only get scores for combinations of words if the search term contains spaces itself. The search term "love pie" should, next to single words also get scores for the combination of adjacent words. We should never skip the normal words, as the space could be just another spelling mistake. With this strategy, we do dismiss the probability of people forgetting to use a space: searching for "lovepie" won't result in there being a score calculated for "love pie" but only the words individually.

As discussed at issue #2, it would be pretty good, especially for development purposes, to be able to get information about the search result set. I'm proposing the following syntax:

$products = Model_Product::find('all');

$products_search = Search::find('shampoo')
    ->in($products)
    ->by('name');

$found_products = $products_search->execute();
$found_products_statistics = $products_search->statistics();

My guess is that this method should return an array, with the "entry_key" as key, and information as value.

Even though the foreach loop on entry_scores will break after the limit of results has been reached, all entries before $offset are still being iterated.

By first selecting the slice of $entry_scores that will be used (which can be done by a simple array_slice after the sorting) less iterations will be needed to construct the final returned array.

As discussed in issue #1, right now the search results for multiple words are very inaccurate. As a base for multiple words searching, we're starting of with loose terms searching.

This means that when people search for more than 1 word, the search phrase will be split on spaces. A similarity score will be calculated for each single word separately after which the scores will be added up to result in one overall score.

This approach should make searching for multiple words a lot more accurate than it is now. However, with this approach, the context of how the words will appear will not matter. When searching for "we love pie" an entry containing "we really love apple pie" will score just as well as "love is awesome. we rule. let's eat some pie". Support for this will be added later on.