Methods for extending vector-based document embedding models and systems thereof
OwnerJulia M Rosedaleinventor
Life of this patent
Estimating the term…
Abstract
The present invention provides natural language processing methods and systems for representing a document temporally in an arbitrary well-defined semantic space. The disclosed methods and systems operate by creating a Root Document Trace, which is a novel document-representing computational object in high-dimensional semantic space. Among many uses, the Root Document Trace can be used, for example, to evaluate narrative and argument structures within and between documents.
Claims15
A computer-implemented method executed by one or more processors for representing a document temporally in an arbitrary well-defined semantic space, the method comprising:
- generating, by a text parser operating upon a plaintext document, an ordered list of tokens, wherein the order of tokens in the list matches their order in the document, and each token in the list has content in the form of a string and index in the form of a non-negative integer;
- defining, by a computational process, a frame object with attributes start, end, and size, and methods for retrieving start, end, and size attributes and altering start and end attributes;
- iteratively generating, by an increment generator using the frame object and the output of the text parser, a collection of increment objects possessing the same docID attribute value, wherein each iteration generates a single increment object through a process comprising:
- assigning, by the increment generator, a predetermined document identification number to the docID attribute and a nonnegative integer value to the position attribute of the increment object, wherein the value of the position attribute is equal to the number of increment objects created prior to the current increment;
- capturing, by the frame object using the output of the text parser, a number of tokens equal to the size attribute of the frame object, wherein all tokens with index between the start and end attributes of the frame object (inclusive) are included in the collection;
- generating, by an arbitrarily chosen vector generator using the captured tokens, a paragraph or document vector representing the captured text in semantic space, wherein the input to the vector generator is a string comprising the captured tokens in index order;
- assigning, by the increment generator, start and end attributes of the increment object corresponding to the start and end attributes of the frame object, and a vector value to the vector attribute of the increment object, wherein the output of the vector generator determines the value of this attribute;
- updating, by the increment generator, the frame object start and end attributes, wherein each attribute increases by a constant positive integer value strictly less than the value of the frame size attribute;
- determining, by the increment generator from the collection of increment objects, whether there is a need to generate an additional increment object, and optionally generating one additional increment object as specified by the increment generator;
- and
- generating, from the collection of increment objects by a function generator, a Root Document Trace.
A computer-implemented method executed by one or more processors for representing a document temporally in an arbitrary well-defined semantic space, comprising:
- generating, by a text parser operating upon a plaintext document, an ordered list of tokens;
- defining, by a computational process, a frame object with assigned attributes and object methods;
- iteratively generating, by an increment generator using the frame object and the output of the text parser, a collection of increment objects possessing identical docID attribute values;
- and
- generating, from the collection of increment objects, a Root Document Trace.
A computer-implemented method executed by one or more processors for generating component parts used in generating a Root Document Trace, comprising:
- creating, by an increment generator, an empty increment object;
- assigning, by the increment generator, a predetermined document identification number to the docID attribute and a nonnegative integer value to a position attribute of the increment object;
- capturing, by a frame object using the output of a text parser, a number of tokens equal to the size attribute of the frame object;
- generating, by an arbitrary predetermined document or paragraph vector generator using the captured tokens, a paragraph or document vector representing the captured text in semantic space;
- assigning, by the increment generator, start, end, and vector attribute values to the increment object;
- updating, by the increment generator, the frame object start and end attributes;
- and
- determining, by the increment generator from the collection of increment objects, whether there is a need to generate an additional increment object, and optionally generating one additional increment object as specified by the increment generator.
Dependencies are inferred from the claim text.