1.5 Example : Comparison of Texts Using an ESA Model
The examples shows the FEATURE_COMPARE
function comparing texts for semantic relatedness (similarity) using the Explicit Semantic Analysis (ESA) prebuilt Wikipedia-based model, which extracts topics and compares text.
The examples shows an ESA model built against a prebuilt Wiki data set rendering over 200,000 features. The documents are analyzed as text and the document titles are given as the feature IDs. In the first example, the pair of sentence scores higher because Nick Price is a golfer born in South Africa.
Similar Texts
SELECT 1-FEATURE_COMPARE(esa_wiki_mod USING 'There are several PGA tour golfers from South Africa' text AND USING 'Nick Price won the 2002 Mastercard Colonial Open' text) similarity FROM DUAL;
The output is as follows:
SIMILARITY
----------
.110
The output metric shows distance calculation. Therefore, smaller number represent more similar texts. So, 1
minus the distance in the queries result in similarity.
Dissimilar Texts
SELECT 1-FEATURE_COMPARE(esa_wiki_mod USING 'There are several PGA tour golfers from South Africa' text AND USING 'John Elway played quarterback for the Denver Broncos' text) similarity FROM DUAL;
The output is as follows:
SIMILARITY
----------
.004