Cross-domain matching system

摘要

A computer implemented method for analyzing a listing object to define a match to a candidate object among many possible candidate objects is disclosed. The method includes an operation to receive a listing object as an input. The method also includes an operation to generate a set of candidate objects based on characteristics of the listing object. The candidate objecting used to generate a listing-candidate pair defined by pairing the listing object with one of the candidate objects. The method may also include operations to process the listing-candidate pair such as an operation to normalize the listing object into a canonical form. Another operation can generate a matching feature vector for the listing-candidate pair. Where the matching feature vector includes a matching score based on a common feature between the candidate object and the canonical form of the listing object. In another operation, the method analyzes the matching feature vector with a judging committee module to render a match judgment. The match judgment based on evaluating the results of the judging committee module to determine whether the listing object and the candidate object are a match. The method also includes an operation that saves the match judgment to a computer readable media.

主权项

1. A computer implemented method for analyzing a listing object to define a match to a candidate object among many possible candidate objects, comprising;
(a) receiving a listing object as an input; (b) using a text search engine to generate a set of candidate objects based on characteristics of the listing object, each of the set of candidate objects being paired with the listing object to define a listing-candidate pair, each listing candidate pair being processed by,
(i) generating a matching feature vector for a listing-candidate pair, the matching feature vector including a matching score based on a common features between the candidate object and a canonical form of the listing object based on domain specific knowledge;(ii) analyzing the matching feature vector using a judging committee module; and(iii) rendering a match judgment by evaluating the results of the judging committee module to determine whether the listing object and the candidate object are a match, wherein the judging committee module applies a plurality of machine learning classifiers and a rule system to the matching feature vector to render the matching judgment and wherein the rule system includes a set of rules with precedence; (c) saving the match judgment to a computer readable media.