Granted Patent B2

US 11,861,918 · App. 17/147,657 · Granted Jan 2, 2024

Inventors: Bijan Kumar Mohanty (Austin, TX); Hung Dinh (Austin, TX); Devipriya Selvaraju (Bangalore, IN); Karthik Sethuramalingam (Bangalore, IN)

Assignee: Dell Products L.P.

G06V20/62G06F18/214G06F18/2411G06F40/284G06N3/04G06N20/00

Abstract

Claims (61)

1. A method, comprising:

receiving an input of at least one image associated with an issue;

extracting text from the at least one image;

determining an intent from the extracted text;

recommending a response to the issue based at least in part on the determined intent;

transmitting the recommended response to a user;

wherein the extracting, determining and recommending are performed at least in part using one or more machine learning models;

wherein the one or more machine learning models are trained with training data comprising at least a plurality of intents;

wherein recommending the response to the issue comprises:

building a first matrix comprising a plurality of words corresponding to the plurality of intents and a first plurality of term frequency scores for the plurality of words corresponding to the plurality of intents;

building a second matrix comprising a plurality of words corresponding to the determined intent and a second plurality of term frequency scores for the plurality of words corresponding to the determined intent;

comparing the first matrix with the second matrix to determine a matching intent from the plurality of intents to the determined intent; and

recommending an issue resolution corresponding to the matching intent; and

wherein the steps of the method are executed by at least one processing device operatively coupled to a memory.

2. The method of claim 1 , wherein extracting the text from the at least one image comprises identifying a plurality of regions in the at least one image comprising one or more text objects.

3. The method claim 2 , wherein extracting the text from the at least one image further comprises generating respective boundaries around respective ones of the plurality of regions.

4. The method of claim 2 , wherein extracting the text from the at least one image further comprises classifying objects in respective ones of the plurality of regions as one of text objects and non-text objects.

5. The method of claim 2 , wherein extracting the text from the at least one image further comprises recognizing text in the one or more text objects of each of the plurality of regions.

6. The method of claim 1 , wherein the one or more machine learning models comprises a mask region-based convolutional neural network (Mask-R-CNN).

7. The method of claim 1 , further comprising training the one or more machine learning models with training data comprising a plurality of text entries, wherein the plurality of intents correspond to the plurality of text entries.

8. The method of claim 1 , wherein the one or more machine learning models comprises a bi-directional recurrent neural network with long short-term memory for natural language understanding.

9. The method of claim 1 , further comprising training the one or more machine learning models with training data comprising a plurality of issue resolutions corresponding to the plurality of intents.

10. The method of claim 9 , further comprising creating a term-frequency-inverse document frequency (TF-IDF) vectorizer of the plurality of intents from the training data.

11. The method of claim 10 , wherein the first matrix comprises a TF-IDF matrix and the first plurality of term frequency scores comprises a plurality of TF-IDF scores for the plurality of words corresponding to the plurality of intents.

12. The method of claim 11 , wherein the second matrix comprises an additional TF-IDF matrix and the second plurality of term frequency scores comprises a plurality of TF-IDF scores for the plurality of words corresponding to the determined intent.

13. The method of claim 1 , further comprising maintaining a data repository comprising respective ones of at least some of the plurality of intents associated with respective ones of a plurality of images and respective ones of a plurality of issue resolutions corresponding to the respective ones of the at least some of the plurality of intents.

14. The method of claim 13 , wherein recommending the response to the issue further comprises:

comparing the determined intent with the at least some of the plurality of intents;

identifying an additional matching intent of the at least some of the plurality of intents to the determined intent based on the comparing of the determined intent with the at least some of the plurality of intents; and

recommending an additional issue resolution corresponding to the additional matching intent.

15. An apparatus comprising:

a processing device operatively coupled to a memory and configured to:

receive an input of at least one image associated with an issue;

extract text from the at least one image;

determine an intent from the extracted text;

recommend a response to the issue based at least in part on the determined intent;

transmit the recommended response to a user;

wherein the processing device is configured to at least in part use one or more machine learning models to perform the extracting, determining and recommending; and

wherein the one or more machine learning models are trained with training data comprising at least a plurality of intents;

wherein, in recommending the response to the issue, the processing device is configured to:

build a first matrix comprising a plurality of words corresponding to the plurality of intents and a first plurality of term frequency scores for the plurality of words corresponding to the plurality of intents;

build a second matrix comprising a plurality of words corresponding to the determined intent and a second plurality of term frequency scores for the plurality of words corresponding to the determined intent;

compare the first matrix with the second matrix to determine a matching intent from the plurality of intents to the determined intent; and

recommend an issue resolution corresponding to the matching intent.

16. The apparatus of claim 15 , wherein, in extracting the text from the at least one image, the processing device is configured to identify a plurality of regions in the at least one image comprising one or more text objects.

17. The apparatus of claim 16 , wherein, in extracting the text from the at least one image, the processing device is further configured to generate respective boundaries around respective ones of the plurality of regions.

18. The apparatus of claim 16 , wherein, in extracting the text from the at least one image, the processing device is further configured to classify objects in respective ones of the plurality of regions as one of text objects and non-text objects.

19. An article of manufacture 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 said at least one processing device to perform the steps of:

receiving an input of at least one image associated with an issue;

extracting text from the at least one image;

determining an intent from the extracted text;

recommending a response to the issue based at least in part on the determined intent;

transmitting the recommended response to a user;

wherein the extracting, determining and recommending are performed at least in part using one or more machine learning models; and

wherein the one or more machine learning models are trained with training data comprising at least a plurality of intents;

wherein, in recommending the response to the issue, the program code causes said at least one processing device to perform the steps of:

building a first matrix comprising a plurality of words corresponding to the plurality of intents and a first plurality of term frequency scores for the plurality of words corresponding to the plurality of intents;

building a second matrix comprising a plurality of words corresponding to the determined intent and a second plurality of term frequency scores for the plurality of words corresponding to the determined intent;

comparing the first matrix with the second matrix to determine a matching intent from the plurality of intents to the determined intent; and

recommending an issue resolution corresponding to the matching intent.

20. The article of manufacture of claim 19 , wherein, in extracting the text from the at least one image, the program code causes said at least one processing device to perform the step of identifying a plurality of regions in the at least one image comprising one or more text objects.

Assignments (9)