Generating feature vectors from RDF graphs

• US 10,235,637 B2
• Filed: 08/28/2015
• Issued: 03/19/2019
• Est. Priority Date: 08/28/2015
• Status: Active Grant

• Current Assignee: Salesforce.com, Inc.
• Original Assignee: Salesforce.com, Inc.
• Inventors: Fuchs, Matthew
• Primary Examiner(s): Misir, Dave
• Application Number:
  US14/839,784
• Publication Number:
  US 20170061320A1
• Time in Patent Office: 1,299 Days
• Field of Search: 706 12
• CPC Class Codes: G06F 16/35 Clustering; Classification
• CPC Class Codes: G06N 20/00 Machine learning

Abstract

The technology disclosed describes systems and methods for generating feature vectors from resource description framework (RDF) graphs. Machine learning tasks frequently operate on vectors of features. Available systems for parsing multiple documents often generate RDF graphs. Once a set of interesting features to be considered has been established, the disclosed technology describes systems and methods for generating feature vectors from the RDF graphs for the documents. In one example setting, a machine learning system can use generated feature vectors to determine how interesting a news article might be, or to learn information-of-interest about a specific subject reported in multiple articles. In another example setting, viable interview candidates for a particular job opening can be identified using feature vectors generated from a resume database, using the disclosed systems and methods for generating feature vectors from RDF graphs.

Claims

1. A method of preparing feature vectors suitable for machine learning or machine classification from a Resource Description Framework graph of a document relevant to a topic of interest, the method comprising:

receiving a set of identified key-attributes that are node names of interest, root-attributes of interest, additional-attributes of interest, and connection information identifying connected nodes to search for at least some of the additional-attributes of interest, the additional-attributes of interest determined from information external to the document

generating a plurality of responsive node-feature vectors for the document represented as the Resource Description Framework graph, including;

collecting the root-attributes of interest from a root node of the document;

querying for and receiving responsive nodes in the Resource Description Framework graph that include the key-attributes;

for each responsive node, creating a responsive node-feature vector, wherein the responsive node-feature vector includes;

from the root node, at least some of the collected root-attributes;

from the responsive node, the additional-attributes of interest present in the responsive node; and

as directed by the connection information, from nodes connected to the responsive node by a single edge, the additional-attributes of interest present in the connected nodes; and

applying the plurality of responsive node-feature vectors as a training set for the machine learning or the machine classification to produce computer instructions configured to determine that another document is relevant to the topic of interest.

2. The method of claim 1, further including, with the set of identified key-attributes, specification of at least one node-type of interest that includes a particular key-attribute.

3. The method of claim 1, further including, with the set of identified key-attributes, derivation of a Boolean value additional-attribute of interest from data in the responsive node or a node connected by a single edge to the responsive node.

4. The method of claim 1, further including, with the set of identified key-attributes, a derivation of an additional-attribute of interest from data in the responsive node or a node connected by a single edge to the responsive node.

5. The method of claim 4, further including:

in the connection information, second additional-attributes of interest connected by a second edge to the node connected by the single edge to the responsive node; and

generating at least one additional responsive node feature vector using information from the second additional attributes of interest.

6. The method of claim 1, further including combining data in one of the responsive nodes with data from an external data source to derive an additional-attribute of interest.

7. The method of claim 1, further including, with the set of identified additional-attributes of interest, specification of at least one connected-additional-attribute that is in a node connected by a single edge to a particular responsive node-type and also including an edge-identifier that connects the connected-additional-attribute to the particular responsive node-type.

8. The method of claim 1, further including:

with the set of identified key-attributes, specification of a node-type of interest responsive to a query for a particular key-attribute; and

with the set of identified additional-attributes of interest, specification of at least one connected-additional-attribute that is in a node connected by a single edge to the node-type of interest and an edge-identifier that connects the connected-additional-attribute to the node-type of interest.

9. The method of claim 1, further including manually curating a multitude of at least 100 of the responsive node-feature vectors, wherein the training set for the machine learning includes a curated set of the at least 100 of the responsive node-feature vectors.

10. The method of claim 1, further including using confidence scores in the Resource Description Framework graph to qualify a multitude of at least 100 of the responsive node-feature vectors, wherein the training set for the machine learning includes a qualified set of the at least 100 of the responsive node-feature vectors.

11. The method of claim 1, wherein the training set for unsupervised machine learning includes a multitude of at least 100,000 of the responsive node-feature vectors.

12. The method of claim 1, further including classifying a multitude of at least 100 of the responsive node-feature vectors, wherein the training set includes a curated set of the at least 100 of the responsive node-feature vectors.

13. A system that prepares feature vectors suitable for machine learning or machine classification from a Resource Description Framework graph of a document relevant to a topic of interest, the system including:

a processor, memory coupled to the processor, and computer instructions loaded into the memory that, when executed, cause the processor to implement a process that includes;

receipt of a set of identified key-attributes that are node names of interest, root-attributes of interest, additional-attributes of interest, and connection information identifying connected nodes to search for at least some of the additional-attributes of interest, the additional-attributes of interest determined from information external to the document

generation of a plurality of responsive node-feature vectors for the document represented as the Resource Description Framework graph, including;

collection of the root-attributes of interest from a root node of the document;

a query for and receipt of responsive nodes in the Resource Description Framework graph that include the key-attributes;

for each responsive node, creating a responsive node-feature vector, wherein the responsive node-feature vector includes;

from the root node, at least some of the collected root-attributes;

from the responsive node, the additional-attributes of interest present in the responsive node; and

as directed by the connection information, from nodes connected to the responsive node by a single edge, the additional-attributes of interest present in the connected nodes; and

application of the plurality of responsive node-feature vectors as a training set for the machine learning or the machine classification to produce computer instructions configured to determine that another document is relevant to the topic of interest.

14. The system of claim 13, further including, with the set of identified key-attributes, specification of at least one node-type of interest that includes a particular key-attribute.

15. The system of claim 13, further including, with the set of identified key-attributes, derivation of a Boolean value additional-attribute of interest from data in the responsive node or a node connected by a single edge to the responsive node.

16. The system of claim 13, further including, with the set of identified key-attributes, a derivation of an additional-attribute of interest from data in the responsive node or a node connected by a single edge to the responsive node.

17. The system of claim 13, wherein the process further includes combination of data in one of the responsive nodes with data from an external data source, to derive an additional-attribute of interest.

18. The system of claim 13, further including, with the set of identified additional-attributes of interest, specification of at least one connected-additional-attribute that is in a node connected by a single edge to a particular responsive node-type and also including an edge-identifier that connects the connected-additional-attribute to the particular responsive node-type.

19. The system of claim 13, wherein the process further includes use of confidence scores in the Resource Description Framework graph to qualify a multitude of at least 100 of the responsive node-feature vectors, wherein the training set for the machine learning includes a qualified set of the at least 100 of the responsive node-feature vectors.

20. A tangible non-transitory computer readable storage medium loaded with initial computer instructions that, when executed, cause a computer system to prepare feature vectors suitable for machine learning or machine classification from a Resource Description Framework graph of a document relevant to a topic of interest, the initial computer instructions including instructions for:

receipt of a set of identified key-attributes that are node names of interest, root-attributes of interest, additional-attributes of interest, and connection information identifying connected nodes to search for at least some of the additional-attributes of interest, the additional-attributes of interest determined from information external to the document

generation of a plurality of responsive node-feature vectors for the document represented as the Resource Description Framework graph, including;

collection of the root-attributes of interest from a root node of the document;

a query for and receipt of responsive nodes in the Resource Description Framework graph that include the key-attributes;

for each responsive node, creation of a responsive node-feature vector, wherein the responsive node-feature vector includes;

from the root node, at least some of the collected root-attributes;

from the responsive node, the additional-attributes of interest present in the responsive node; and

as directed by the connection information, from nodes connected to the responsive node by a single edge, the additional-attributes of interest present in the connected nodes; and

application of the plurality of responsive node-feature vectors as a training set for the machine learning or the machine classification to produce computer instructions configured to determine that another document is relevant to the topic of interest.

21. The tangible non-transitory computer readable storage medium of claim 20, wherein the initial computer instructions further includes, with the set of identified key-attributes, specification of at least one node-type of interest that includes a particular key-attribute.

22. The tangible non-transitory computer readable storage medium of claim 20, wherein the initial computer instructions further includes:

with the set of identified key-attributes, specification of a node-type of interest responsive to a query for a particular key-attribute; and

with the set of identified additional-attributes of interest, specification of at least one connected-additional-attribute that is in a node connected by a single edge to the node-type of interest and an edge-identifier that connects the connected-additional-attribute to the node-type of interest.

23. The tangible non-transitory computer readable storage medium of claim 20, wherein the training set for the machine learning includes a curated set of at least 100 of the responsive node-feature vectors.

24. The tangible non-transitory computer readable storage medium of claim 20, wherein the initial computer instructions further includes instructions for use of confidence scores in the Resource Description Framework graph to qualify a multitude of at least 100 of the responsive node-feature vectors, wherein the training set for the machine learning includes a qualified set of the at least 100 of the responsive node-feature vectors.

25. The tangible non-transitory computer readable storage medium of claim 20, wherein the training set for unsupervised machine learning includes a multitude of at least 100,000 of the responsive node-feature vectors.