Sentiment analysis for aspect terms extracted from documents having unstructured text data

Patent metadata

Field	Value
Publication number	US-11675823-B2
Application number	US-202117500042-A
Country	US
Kind code	B2
Filing date	Oct 13, 2021
Priority date	Oct 13, 2021
Publication date	Jun 13, 2023
Grant date	Jun 13, 2023

Abstract

An apparatus comprises at least one processing device configured to receive a query to perform sentiment analysis for a document, to generate, utilizing a first machine learning model, a first set of encodings classifying words of the document as being aspect or non-aspect terms, to generate, utilizing a second machine learning model, a second set of encodings classifying sentiment of the words of the document, and to determine, for a given aspect term, attention weights for a given subset of the words of the document surrounding the given aspect term. The processing device is also configured to generate, utilizing a third machine learning model, a given sentiment classification of the given aspect term based on the attention weights and a given portion of the second set of encodings for the given subset of the words, and to provide a response to the query comprising the given sentiment classification.

First claim

What is claimed is:

1. An apparatus comprising:
at least one processing device comprising a processor coupled to a memory;
the at least one processing device being configured to perform steps of:
receiving a query to perform sentiment analysis for a document, the document comprising unstructured text data;
generating, utilizing a first machine learning model, a first set of encodings of the unstructured text data of the document, the first set of encodings classifying each word of the unstructured text data of the document as being an aspect term or a non-aspect term;
generating, utilizing a second machine learning model, a second set of encodings of the unstructured text data of the document, the second set of encodings classifying sentiment of each word of the unstructured text data of the document;
determining, for a given aspect term corresponding to a given sequence of one or more of the words of the unstructured text data of the document classified as an aspect term in the first set of encodings, attention weights for a given subset of words in the unstructured text data surrounding the given sequence of the one or more words;
generating, utilizing a third machine learning model, a given sentiment classification of the given aspect term, the third machine learning model generating the given sentiment classification based at least in part on (i) the attention weights for the given subset of the words in the unstructured text data surrounding the given sequence of the one or more words and (ii) a given portion of the second set of encodings classifying the sentiment of the given subset of the words in the unstructured text data surrounding the given sequence of the one or more words; and
providing a response to the query, the response to the query comprising the given sentiment classification of the given aspect term.

2. The apparatus of claim 1 wherein the first machine learning model comprises a bidirectional encoder representations from transformers token classification model, and wherein the second machine learning model comprises a bidirectional encoder representations from transformers sequence classification model.

3. The apparatus of claim 2 wherein the first machine learning model is pretrained using a plurality of documents associated with a plurality of different technology domains, and wherein the second machine learning model is trained for each aspect term.

4. The apparatus of claim 1 wherein the given subset of the words in the unstructured text data surrounding the given sequence of the one or more words corresponding to the given aspect term comprises (i) a first subset of words prior to the given sequence of the one or more words corresponding to the given aspect term until a previous aspect term or a beginning of the document is reached and (ii) a second subset of words following the given sequence of the one or more words corresponding to the given aspect term until a next aspect term or an end of the document is reached.

5. The apparatus of claim 4 wherein determining the attention weights for the given subset of the words in the unstructured text data surrounding the given sequence of the one or more words corresponding to the given aspect term comprises:
assigning index values to each of the words in the first subset and the second subset according to their respective distance to the given sequence of the one or more words corresponding to the given aspect term; and
computing the attention weight for each of the words in the first subset and the second subset based at least in part on a logarithmic weight adjusting factor determined using the index values assigned to each of the words in the first subset and the second subset.

6. The apparatus of claim 1 wherein, if the sequence of the one or more words corresponding to the given aspect term comprises two or more words, the third machine learning model generates the given sentiment classification based at least in part on computing an average of the given portion of the second set of encodings for each of the two or more words.

7. The apparatus of claim 1 wherein the third machine learning model comprises a multi-level feed forward neural network classifier.

8. The apparatus of claim 7 wherein the multi-level feed forward neural network classifier comprises a three-level feed forward neural network classifier which classifies the given aspect term as having one of a positive sentiment, a neutral sentiment and a negative sentiment.

9. The apparatus of claim 1 wherein the document comprises at least one of a support chat log and a support call log associated with a given information technology asset of an information technology infrastructure, and wherein the at least one processing device is further configured to perform the steps of:
identifying, utilizing the given sentiment classification, a recommended troubleshooting action for the given information technology asset; and
performing the recommended troubleshooting action on the given information technology asset.

10. The apparatus of claim 9 wherein the recommended troubleshooting action comprises at least one of a diagnostic action and a repair action.

11. The apparatus of claim 9 wherein the given information technology asset comprises a computing device, and wherein the recommended troubleshooting action comprises modifying at least one of: one or more software components of the computing device; and one or more hardware components of the computing device.

12. The apparatus of claim 1 wherein the document comprises at least one of an article, a survey and social media content associated with one or more information technology asset types.

13. The apparatus of claim 12 wherein the at least one processing device is further configured to perform the step of adjusting investment by an entity in the one or more information technology asset types based at least in part on the given sentiment classification.

14. The apparatus of claim 12 wherein the at least one processing device is further configured to perform the step of modifying configurations of information technology assets in an information technology infrastructure having the one or more types of information technology asset types based at least in part on the given sentiment classification.

15. A computer program product comprising a non-transitory processor-readable storage medium having stored therein program code of one or more software programs, wherein the program code when executed by at least one processing device causes the at least one processing device to perform steps of:
receiving a query to perform sentiment analysis for a document, the document comprising unstructured text data;
generating, utilizing a first machine learning model, a first set of encodings of the unstructured text data of the document, the first set of encodings classifying each word of the unstructured text data of the document as being an aspect term or a non-aspect term;
generating, utilizing a second machine learning model, a second set of encodings of the unstructured text data of the document, the second set of encodings classifying sentiment of each word of the unstructured text data of the document;
determining, for a given aspect term corresponding to a given sequence of one or more of the words of the unstructured text data of the document classified as an aspect term in the first set of encodings, attention weights for a given subset of words in the unstructured text data surrounding the given sequence of the one or more words;
generating, utilizing a third machine learning model, a given sentiment classification of the g

Assignees

• Dell Products Lp

Inventors

• Kanagovi Ramakanth
• Sahoo Sumant
• Swamy Arun
• Shukla Ravi
• Sridharan Prakash
• Joisa Shrikrishna K
• Ratnakar Sandeep
• Sharma Mayank

Classifications

• G06F16/3346

  using probabilistic model · CPC title

• G06F16/353 Primary

  into predefined classes · CPC title

• G06N20/20

  Ensemble learning · CPC title

• G06N3/045

  Combinations of networks · CPC title

• G06N3/08

  Learning methods · CPC title