We have two databases with products. The first is the client's product's database and the other formed by their competitors products. Both have a field that link the perfect matches, for validation.
The job is to search for the products to find the better match, comparing informations. After that, to validate the results with the matching code of the competitor base.
Using the Laravel PHP Framework as a base, I've created a simple REST service which receives the product and compares it by the title similarity, using the PHP functions similar_text and levenshtein.
First the algorithm breaks the product title and compares each word with the competitor's product's title using similar_text function. This phase accepts only 100% word matches to calculate a percentage of similarity of the titles, I call it similarity by word.
Then it compares the whole title similarity using the similar_text function. A product can passes this phase if it have until 65% of similarity.
The last step is to get only the results which passes the tests of similarity by word and similarity by whole title and sort it using the data of similarity by word, levenshtein and similar_text, resulting in the better match to the product.
See functional version at http://mbeck.com.br/matcher/