Taxonomy enrichment using ensemble classifiers
AssigneeMicrosoft Technology Licensing, LLC
InventorsKeng-hao Chang, Srinivasa Reddy Neerudu, Sujith Vishwajith, Ruofei Zhang
Life of this patent
Estimating the term…
Abstract
A taxonomy of categories, attributes, and values can be conflated with new data triplets by identifying one or more conflation candidates among the attribute-value pairs within a category of the taxonomy that matches the category of the data triplet, and determining a suitable merge action for conflating the data triplet with each conflation candidate. The task of determining merge actions may be cast as a classification problem, and may be solved by an ensemble classifier.
Claims20
A method comprising:
- using one or more computer processors to perform operations for extending a multi-level taxonomy that comprises at least categories, attributes, and values based on data triplets each comprising an attribute-value pair and an associated category, the operations comprising, for each of the data triplets:
- identifying, among attribute-value pairs within a category of the taxonomy that matches the category of the data triplet, based on a measure of similarity of the attribute-value pairs to the attribute-value pair of the data triplet, one or more conflation candidates;
- using an ensemble classifier operating on a plurality of features associated with the conflation-candidates and the attribute-value pair of the data triplet, assigning each of the one or more conflation-candidates to one of a plurality of classes, the classes including at least duplicate matches, attribute matches, and mismatches of the attribute-value pair of the data triplet;
- and
- extending the taxonomy with the data triplets based on the classification.
A system comprising:
- one or more hardware processors;
- and
- memory storing
- a data structure representing a multi-level taxonomy comprising nodes representing categories, attributes, values, and value synonyms, and
- instructions which, when executed by the one or more hardware processors, cause the one or more hardware processors to perform operations for expanding the taxonomy with data triplets each comprising an attribute-value pair and an associated category, the operations comprising:
- identifying, among attribute-value pairs within a category of the taxonomy that matches the category of the data triplet, based on a measure of similarity of the attribute-value pairs to the attribute-value pair of the data triplet, a conflation candidate;
- using one or more classifier models to classify the conflation-candidate as one of a duplicate match, a value match, an attribute match, or a mismatch of the attribute-value pair of the data triplet;
- and
- extending the taxonomy by adding a value of the attribute-value pair of the data triplet as a new value synonym to the conflation candidate following classification of the conflation candidate as a value match, adding the value of the attribute-value pair of the data triplet as a new value to the conflation candidate following classification of the conflation candidate as an attribute match, and adding the attribute-value pair of the data triplet as a new attribute-value pair to the associated category following classification of the conflation candidate as a mismatch.
One or more machine-readable media storing instructions for execution by one or more hardware processors, execution of the instructions causing the one or more hardware processors to perform operations for merging a first attribute-value pair with a second attribute-value pair identified as a conflation candidate within a taxonomy, the operations comprising:
- using a plurality of classifier models to each classify the second attribute-value pair as one of a duplicate match, a value match, an attribute match, or a mismatch of the first attribute-value pair;
- combining classifications made by the plurality of classifier models into an overall classification by majority voting;
- and
- extending the taxonomy with the first attribute-value pair in accordance with the overall classification.
Dependencies are inferred from the claim text.