US 12,591,740 B2Granted

Methods and systems for generating textual features

AssigneePanasonic Intellectual Property Management Co., Ltd.

InventorsGayathri Saranathan, Nway Nway Aung, Ariel Beck, Chandra Suwandi Wijaya, Jianyu Chen, Debdeep Paul, Sahim Yamaura, Koji Miura

Life of this patent

Estimating the term…

patentgrep

Loading patent…

Abstract

According to an embodiment, a method for generating textual features corresponding to text documents from a raw dataset is disclosed. The method includes preprocessing the text documents and determining topic probability scores (TPS) and confidence scores (CS) using unsupervised and supervised machine learning models, respectively. The combination of TPS and CS is used to generate a compound distribution score (CDS), which forms a comprehensive representation of the output of the machine learning models. The determined TPS, CS, and CDS are then used to generate a set of textual features, which serve as independent variables for a forecasting model.

Claims16

Dependent claims

A method for generating textual features corresponding to at least one text document, the method being performed by at least one processor communicably coupled to a memory and comprising:

preprocessing the at least one text document among a plurality of text documents in a raw dataset;
determining a topic probability score (TPS) corresponding to the at least one preprocessed text document, via an unsupervised machine learning (ML) model, wherein the TPS indicates a probability of each of the at least one preprocessed text document among the plurality of text documents, being associated with at least one cluster generated by the unsupervised ML model;
determining a confidence score (CS) corresponding to the at least one preprocessed text document, via a supervised ML model, wherein the CS indicates a reliability level associated with a classification of each of the at least one preprocessed text document among the plurality of text documents, predicted by the supervised ML model;
determining a compound distribution score (CDS) corresponding to each of the at least one preprocessed text document among the plurality of text documents in the raw dataset based on the TPS and the CS, wherein the CDS indicates a weighted average of the TPS and the CS to generate a comprehensive representation for output of the unsupervised ML model and the supervised ML model respectively;
generating a set of textual features corresponding to the at least one preprocessed text document among the plurality of text documents based on inputting the determined TPS, the CS, and the CDS respectively, into a multiclass classification machine learning model, wherein the set of textual features indicates independent variables for a forecasting model, and comprises time series data;
receiving a set of numerical features corresponding to the raw dataset;
inputting the set of textual features and the set of numerical features into the forecasting model;
and
outputting, to a display and via the forecasting model, a final forecast based on the set of textual features and the set of numerical features, wherein the final forecast indicates predicting a target variable using the forecasting model.

6 more elements

A system for generating textual features corresponding to at least one text document, the system comprising:

a memory;
a display;
and
at least one processor communicably coupled to the memory, the at least one processor is configured to:
preprocess the at least one text document among a plurality of text documents in a raw dataset;
determine a topic probability score (TPS) corresponding to the at least one preprocessed text document, via an unsupervised machine learning (ML) model, wherein the TPS indicates probability of each of the at least one preprocessed text document among the plurality of text documents, being associated with at least one cluster generated by the unsupervised ML model;
determine a confidence score (CS) corresponding to the at least one preprocessed text document, via a supervised ML model, wherein the CS indicates a reliability level associated with classification of each of the at least one preprocessed text document among the plurality of text documents, predicted by the supervised ML model;
determine a compound distribution score (CDS) corresponding to each of the at least one preprocessed text document among the plurality of text documents in the dataset based on the determined TPS and the determined CS, wherein the CDS indicates a weighted average of the TPS and the CS to generate a comprehensive representation for output of the unsupervised ML model and the supervised ML model respectively;
generate a set of textual features corresponding to the at least one preprocessed text document among the plurality of text documents based on inputting the determined TPS, the CS, and the CDS into a multiclass classification machine learning model, wherein the set of textual features indicates independent variables for a forecasting model, and comprises time series data;
receive a set of numerical features corresponding to the raw dataset;
input the set of textual features and the set of numerical features into the forecasting model;
and
output, to the display and via the forecasting model, a final forecast based on the set of textual features and the set of numerical features, wherein the final forecast indicates predicting a target variable using the forecasting model.

10 more elements

Dependencies are inferred from the claim text.