Predictive neural network with sentiment data

Patent metadata

Field	Value
Publication number	US-11238339-B2
Application number	US-201715667287-A
Country	US
Kind code	B2
Filing date	Aug 2, 2017
Priority date	Aug 2, 2017
Publication date	Feb 1, 2022
Grant date	Feb 1, 2022

Abstract

A set of vectors may be obtained. The vectors may be multi-dimensional vectors that are associated with and describe tokens from a first set of tokens from a corpus of sources. The description may be based in part on the relationship of the token to at least a portion of the remainder of the corpus. A set of sentiment scores may be obtained. The sentiment scores in the set of sentiment scores may describe a sentiment associated with a corresponding token that is described by a vector from the set of vectors. The set of vectors and the set of sentiment scores may be input into a pattern-recognizer pathway in a first neural network. A probability value of a potential future event may then be generated by the first neural network. The probability value may be based on the set of vectors and the set of sentiment scores.

First claim

What is claimed is:

1. A method comprising:
obtaining a first set of entity vectors, wherein each entity vector in the first set of entity vectors is a multi-dimensional vector that is associated with and describes a token from a first set of tokens from a corpus of sources, wherein the description is based in part on a relationship of the token to at least a portion of the remainder of the corpus of sources, and wherein the corpus of sources contains at least one natural-language source;
obtaining a first set of sentiment scores, wherein each sentiment score in the first set of sentiment scores describes a sentiment associated with a corresponding token that is described by a vector in the first set of entity vectors;
inputting the first set of entity vectors and the first set of sentiment scores into a first pattern-recognizer pathway in a first neural network; and
generating, by the first neural network and based on the first set of entity vectors and the first set of sentiment scores, a first probability value of a first potential future event
wherein the generating the first probability value comprises:
reducing, by the first pattern-recognizer pathway, a first dimension of each entity vector in the first set of entity vectors, resulting in a first set of reduced entity vectors;
reducing, by a second pattern-recognizer pathway, a second dimension of each keyword vector in a first set of keyword vectors, resulting in a first set of reduced keyword vectors;
reducing, by a third pattern-recognizer pathway, a third dimension of each concept vector in a first set of concept vectors, resulting in a first set of reduced concept vectors;
merging each of the reduced vectors in the first set of reduced entity vectors with a corresponding vector in the set of reduced keyword vectors and a corresponding vector in the set of reduced concept vectors, resulting in a set of merged vectors; and
inputting the merged vectors in the set of merged vectors into a fourth pattern-recognizer pathway in the first neural network.

2. The method of claim 1 , further comprising:
obtaining a second set of entity vectors, wherein each entity vector in the second set of vectors is a multi-dimensional entity vector that describes an entity from the corpus of sources, wherein the description is based in part on a relationship of the entity to at least a portion of the remainder of the corpus;
obtaining a second set of sentiment scores, wherein each sentiment score in the second set of sentiment scores describes a sentiment associated with a corresponding entity that is described by an entity vector in the second set of entity vectors;
inputting the second set of entity vectors and the second set of second sentiment scores into a fifth pattern-recognizer pathway in the first neural network, wherein the first pattern-recognizer pathway and the fifth pattern-recognizer pathways are separate pathways prior to generating the first probability value and the second probability value; and
generating, by the first neural network and based on the second set of entity vectors and the second set of sentiment scores, a second probability value of a second potential future event.

3. The method of claim 2 , further comprising creating, by a multiple-regression analysis, an overall projected score, wherein the creating comprises:
inputting the first probability value into a multiple-regression function;
inputting the second probability value into the multiple-regression function;
inputting at least one portion of structured data into the multiple-regression function; and
calculating the overall projected score based on the output of the multiple-regression function.

4. The method of claim 1 , further comprising:
obtaining the first set of keyword vectors, wherein each keyword vector in the first set of keyword vectors is a multi-dimensional vector that describes a keyword from the corpus of sources;
inputting the first set of keyword vectors into the second pattern-recognizer pathway in the first neural network;
obtaining the first set of concept vectors, wherein each concept vector in the first set of concept vectors is a multi-dimensional vector that describes a concept from the corpus of sources; and
inputting the first set of concept vectors into the third pattern-recognizer pathway in the first neural network.

5. The method of claim 1 , wherein the obtaining a first set of sentiment scores comprises:
obtaining a second set of sentiment scores, wherein the second set of sentiment scores comprises the first set of sentiment scores, wherein each sentiment score in the second set of sentiment scores describes the sentiment associated with a token in a second set of tokens, and wherein the second set of tokens comprises the first set of tokens;
cross-referencing the first set of tokens and the second set of tokens; and
selecting, based on the cross-referencing and from the second set of sentiment scores, the sentiment scores associated with tokens that are in the first set of tokens and second set of tokens, resulting in the first set of sentiment scores.

6. A computer program product, the computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a computer to cause the computer to:
obtain a first set of entity vectors, wherein each entity vector in the first set of entity vectors is a multi-dimensional entity vector that is associated with and describes an entity from a first set of entities from a corpus of sources, wherein the description is based in part on a relationship of the entity to at least a portion of the remainder of the corpus of sources, and wherein the corpus of sources contains at least one natural-language source;
obtain a first set of sentiment scores, wherein each sentiment score in the first set of sentiment scores describes a sentiment associated with a corresponding entity that is described by an entity vector in the first set of vectors;
obtain a second set of vectors, wherein each vector in the second set of vectors is a multi-dimensional vector that describes a token from a first set of tokens from the corpus of sources, wherein the description is based in part on a relationship of the token to at least a portion of the remainder of the corpus;
obtain a second set of sentiment scores, wherein each sentiment score in the second set of sentiment scores describes a sentiment associated with a corresponding token that is described by a vector in the second set of vectors;
input the first set of entity vectors and the first set of sentiment scores into a first pattern-recognizer pathway in a first neural network;
inputting the second set of vectors and the second set of second sentiment scores into a second pattern-recognizer pathway in a second neural network, wherein the first neural network and the second neural network are separate networks prior to generating the first probability value and the second probability value;
generate, by the first neural network and based on the first set of entity vectors and the first set of sentiment scores, a first probability value of a first potential future event; and
generate, by the second neural network and based on the second set of vectors and the second set of sentiment scores, a second probability value of a second potential future event.

7. The computer program product of claim 6 , wherein the program instructions further cause the computer to obtain a third set of sentiment scores, wherein the third set of sentiment scores comprises an average sentiment score of the corpus of sources.

8. The computer program product of claim 6 , wherein the first set of entities is identified by a third neural net

Assignees

• IBM

Inventors

• Powell Jeff
• Baughman Aaron K
• Kent John J
• Newell John C
• Provan David
• Syken Noah

Classifications

• G06N3/045 Primary

  Combinations of networks · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06N3/0499

  Feedforward networks · CPC title

• G06F40/30 Primary

  Semantic analysis · CPC title

• G06N3/082

  modifying the architecture, e.g. adding, deleting or silencing nodes or connections · CPC title