Back to results
US 8,285,697 B1Granted
Feedback enhanced attribute extraction
AssigneeGoogle Inc.
InventorJoel Truher
Life of this patent
Estimating the term…
Abstract
A data object submitted for storage is analyzed, and a set of values is extracted from the data object that can correspond to a set of attributes. The analysis of the data object can also identify possible new ontology terms. One or more extracted values are presented to the entity which submitted the data object for approval and feedback. This feedback can be used to characterize the data object with appropriate terms, train the extraction process for future extractions, and/or expand the set of known ontology terms.
Claims39
A method implemented by data processing apparatus, the method comprising:
- analyzing a data object using one or more trained classifiers to extract a plurality of text values from the data object and determine corresponding attributes for the text values based at least in part on an ontology, wherein each of the corresponding attributes is associated with a respective confidence value determined by one of the trained classifiers;
- assigning each of one or more of the plurality of text values to a corresponding attribute for the text value that is associated with a respective confidence value that satisfies a threshold;
- providing a user interface to a client, the user interface configured to present at least one question as to whether an extracted text value of the plurality of text values paired with its assigned attribute accurately characterizes the data object, wherein at least one extracted text value is a candidate ontology term not found in the ontology and is presented in the at least one question;
- receiving a response to the at least one question from the client indicating that the attribute-value pair accurately characterizes the data object;
- in response to receiving, storing the attribute-value pair in association with the data object;
- serving the data object to a remote computer over a network in response to a request for one or more data objects characterized by the attribute-value pair;
- and
- wherein analyzing, assigning, providing, receiving and serving are performed by the data processing apparatus.
5 more elements
A system comprising:
- a non-transitory computer readable medium having instructions stored thereon;
- and data processing apparatus programmed to execute the instructions to perform operations comprising:
- analyzing a data object using one or more trained classifiers to extract a plurality of text values from the data object and determine corresponding attributes for the text values based at least in part on an ontology, wherein each of the corresponding attributes for the text values is associated with a respective confidence value determined by one of the trained classifiers;
- assigning each of one or more of the plurality of text values to a corresponding attribute for the text value that is associated with a respective confidence value that satisfies a threshold;
- providing a user interface to a client, the user interface configured to present at least one question as to whether an extracted text value of the plurality of text values paired with its assigned attribute accurately characterizes the data object, wherein at least one extracted text value is a candidate ontology term not found in the ontology and is presented in the at least one question;
- receiving a response to the at least one question from the client indicating that the attribute-value pair accurately characterizes the data object;
- in response to receiving, storing the attribute-value pair in association with the data object;
- and
- serving the data object to a remote computer over a network in response to a request for one or more data objects characterized by the attribute-value pair.
6 more elements
A computer program product encoded on a machine-readable storage device, the program product containing instructions to cause one or more computers to perform operations comprising:
- analyzing a data object using one or more trained classifiers to extract a plurality of text values from the data object and determine corresponding attributes for the text values based at least in part on an ontology, wherein each of the corresponding attributes for the text values is associated with a respective confidence value determined by one of the trained classifiers;
- assigning each of one or more of the plurality of text values to a corresponding attribute for the text value that is associated with a respective confidence value that satisfies a threshold;
- providing a user interface to a client, the user interface configured to present at least one question as to whether an extracted text value of the plurality of text values paired with its assigned attribute accurately characterizes the data object, wherein at least one extracted text value is a candidate ontology term not found in the ontology and is presented in the at least one question;
- receiving a response to the at least one question from the client indicating that the attribute-value pair accurately characterizes the data object;
- in response to receiving, storing the attribute-value pair in association with the data object;
- and
- serving the data object to a remote computer over a network in response to a request for one or more data objects characterized by the attribute-value pair.
4 more elements
Dependencies are inferred from the claim text.