US 2022/0019741 A1 · AN UNSUPERVISED APPROACH TO ASSIGNMENT OF PRE-DEFINED LABELS TO TEXT DOCUMENTS

Abstract

There is a need for more effective and efficient text categorization. This need can be addressed by, for example, techniques for semantic text categorization. In one example, a method includes determining an input vector-based representation of an input document; processing the input vector-based representation using a trained supervised machine learning model to generate the categorization based at least in part on the input vector-based representation, wherein: (i) the trained supervised machine learning model has been trained using automatically-generated training data, and (ii) the automatically generated training data is generated by determining an inferred semantic label for each unlabeled training document of one or more unlabeled training documents; and performing one or more categorization-based actions based at least in part on the categorization, and (iii) the labels are described by one or more short documents/short texts.

Claims20

Dependent claims

A computer-implemented method for generating a categorization for an input document, the computer-implemented method comprising:

determining an input vector-based representation of the input document;
processing the input vector-based representation using a trained supervised machine learning model to generate the categorization based at least in part on the input vector-based representation, wherein:
the trained supervised machine learning model has been trained using automatically-generated training data;
and
the automatically generated training data is generated by determining an inferred semantic label for each unlabeled training document of one or more unlabeled training documents based at least in part on:
(i) a prior semantic label for each labeled training document of the one or more labeled training documents, and (ii) a cross-document similarity measure for each document pair of a plurality of document pairs that is associated with a corresponding unlabeled training document of the one or more unlabeled training documents and a corresponding labeled training document of the one or more labeled training documents;
and
performing one or more categorization-based actions based at least in part on the categorization.

5 more elements

The computer-implemented method of claim 3, wherein determining the cross-document similarity measure for a document pair of the plurality of document pairs comprises:

determining an embedded representation for word of one or more unlabeled words of the unlabeled training document associated with the document pair;
determining embedded representation for each word of one or more words of the labeled document associated with the document pair;
determining, for each word pair of a plurality of word pairs that comprises a corresponding unlabeled word of the one or more unlabeled words and a corresponding labeled word of the one or more labeled words, a pairwise similarity measure of the unlabeled embedded representation for the corresponding word in the unlabeled document and the embedded representation for the corresponding word in the labeled document;
determining, for each word pair of the plurality of word pairs, a pairwise flow indicator based at least in part on the pairwise similarity measure of the word pair relative to other pairwise similarity measures in a subset of the plurality of word pairs that is associated with the corresponding unlabeled word for the word pair;
and
determining the cross-document similarity measure based at least in part on each pairwise similarity measure for a word pair of the plurality of word pairs as well as each pairwise flow indicator for a word pair of the plurality of word pairs.

3 more elements

An apparatus for generating a categorization for an input document, the apparatus comprising at least one processor and at least one memory including program code, the at least one memory and the program code configured to, with the processor, cause the apparatus to at least:

determine an input vector-based representation of the input document;
process the input vector-based representation using a trained supervised machine learning model to generate the categorization based at least in part on the input vector-based representation, wherein:
the trained supervised machine learning model has been trained using automatically-generated training data;
and
the automatically generated training data is generated by determining an inferred semantic label for each unlabeled training document of one or more unlabeled training documents based at least in part on:
(i) a prior semantic label for each labeled training document of the one or more labeled training documents, and (ii) a cross-document similarity measure for each document pair of a plurality of document pairs that is associated with a corresponding unlabeled training document of the one or more unlabeled training documents and a corresponding labeled training document of the one or more labeled training documents;
and
perform one or more categorization-based actions based at least in part on the categorization.

5 more elements

A computer program product for generating a categorization for an input document, the computer program product comprising at least one non-transitory computer-readable storage medium having computer-readable program code portions stored therein, the computer-readable program code portions configured to:

determine an input vector-based representation of the input document;
process the input vector-based representation using a trained supervised machine learning model to generate the categorization based at least in part on the input vector-based representation, wherein:
the trained supervised machine learning model has been trained using automatically-generated training data;
and
the automatically generated training data is generated by determining an inferred semantic label for each unlabeled training document of one or more unlabeled training documents based at least in part on:
(i) a prior semantic label for each labeled training document of the one or more labeled training documents, and (ii) a cross-document similarity measure for each document pair of a plurality of document pairs that is associated with a corresponding unlabeled training document of the one or more unlabeled training documents and a corresponding labeled training document of the one or more labeled training documents;
and
perform one or more categorization-based actions based at least in part on the categorization.

5 more elements

Dependencies are inferred from the claim text.

An unsupervised approach to assignment of pre-defined labels to text documents

Life of this patent

An unsupervised approach to assignment of pre-defined labels to text documents

Life of this patent

Abstract

Claims20