Method and system for generating review summaries

Patent metadata

Field	Value
Publication number	US-11017006-B2
Application number	US-201916357901-A
Country	US
Kind code	B2
Filing date	Mar 19, 2019
Priority date	Mar 9, 2019
Publication date	May 25, 2021
Grant date	May 25, 2021

Abstract

The present disclosure relates to a method and a system for generating sentiment-based summaries for a user review. In an embodiment, a text analyzer receives a block of text indicating a user review. The text analyzer may generate one or more vectors for the plurality of words. Further, a relation is identified among the one or more vectors. A model is trained to identify a relation among the one or more vectors. Using the relation between the one or more vectors, a sentiment associated with the block of text is determined. Thereafter, one or more keywords from the block of text contributing to the determined sentiment is are identified and are classified into categories according to the sentiment contributed by the one or more words. Thereafter, the summary is generated for each category using the corresponding one or more words.

First claim

What is claimed is:

1. A method for generating sentiment-based summaries, comprising:
receiving, by a text analyzer, a block of text comprising a plurality of words indicating a user review;
generating, by the text analyser, one or more vectors respectively for the plurality of words in the block of text;
identifying, by the text analyzer, a relation among the one or more vectors using a trained model for determining at least one sentiment associated with the block of text, using a combination of Bidirectional Long Short-Term Memory (Bi-LSTM) and LSTM architecture, from a group of sentiments comprising at least a positive sentiment, a negative sentiment and a neutral sentiment, wherein one or more training vectors corresponding to a plurality of words of a training text are used for generating the trained model;
associating, by the text analyser, the one or more words to at least one of the sentiments determined;
classifying, by the text analyzer, the one or more words into one or more categories based on the determined at least one sentiment; and
generating, by the text analyzer, a summary in natural language for each of the one or more categories based on the one or more words classified in the at least one sentiment.

2. The method of claim 1 , wherein the one or more vectors indicate a context of the respective word, semantic of the respective word, syntax similarity of the respective word and a relationship of the respective word with other plurality of words in the block of text.

3. The method of claim 1 , wherein the one or more training vectors are provided as inputs for generating the trained model, wherein the trained model is at least a Long Short-Term Memory (LSTM) model and a Bidirectional-LSTM model.

4. The method of claim 3 , wherein at least the LSTM and the Bidirectional-LSTM models are trained to generate a context vector indicating a context of the user review, wherein the context vector is used to determine a sentiment associated with a plurality of block of test data comprising texts.

5. The method of claim 4 , wherein the LSTM and the Bidirectional-LSTM models use an encoder-decoder model for generating the context vector using the one or more vectors and an output sequence using the context vector, wherein the output sequence indicates the sentiment associated with the block of text.

6. A text analyzer for generating sentiment-based summaries, comprising:
a processor; and
a memory, communicatively coupled to the processor, wherein the memory stores processor-executable instructions, which on execution cause the processor to:
receive a block of text comprising a plurality of words indicating a user review;
generate one or more vectors respectively for the plurality of words in the block of text;
identify a relation among the one or more vectors using a trained model for determining at least one sentiment associated with the block of text, using a combination of Bidirectional Long Short-Term Memory (Bi-LSTM) and LSTM architecture, from a group of sentiments comprising at least a positive sentiment, a negative sentiment and a neutral sentiment, wherein one or more training vectors corresponding to a plurality of words of a training text are used for generating the trained model;
associate the one or more words to at least one sentiment determined;
classify the one or more words into one or more categories based on the determined at least one sentiment; and

generate a summary in natural language for each of the one or more categories based on the one or more words classified in the at least one sentiment.

7. The text analyzer of claim 6 , wherein the processor generates the one or more training vectors, wherein the one or more training vectors are provided as inputs for generating the trained model, wherein the trained model is at least a Long Short-Term Memory (LSTM) model and a Bidirectional-LSTM model.

8. The text analyzer of claim 7 , wherein the processor generates a context vector indicating a context of the user review using at least the LSTM and the Bidirectional-LSTM models, wherein the context vector is used to determine a sentiment associated with a plurality of block of test data comprising texts.

9. The text analyzer of claim 6 , wherein the processor generates a context vector using the one or more vectors and an output sequence using the context vector, wherein the output sequence indicates the sentiment associated with the block of text.

10. A non-transitory computer readable medium including instructions stored thereon that when processed by at least one processor cause a device to perform operations comprising:
receiving a block of text comprising a plurality of words indicating a user review;
generating one or more vectors respectively for the plurality of words in the block of text;
identifying a relation among the one or more vectors using a trained model for determining at least one sentiment associated with the block of text, using a combination of Bidirectional Long Short-Term Memory (Bi-LSTM) and LSTM architecture, from a group of sentiments comprising at least a positive sentiment, a negative sentiment and a neutral sentiment, wherein one or more training vectors corresponding to a plurality of words of a training text are used for generating the trained model;
associating the one or more words to at least one of the sentiments determined;
classifying the one or more words into one or more categories based on the determined at least one sentiment; and
generating a summary in natural language for each of the one or more categories based on the one or more words classified in the at least one sentiment.

11. The computer readable media as claimed in claim 10 , wherein the one or more vectors indicate a context of the respective word, semantic of the respective word, syntax similarity of the respective word and a relationship of the respective word with other plurality of words in the block of text.

12. The computer readable media as claimed in claim 10 , wherein the one or more training vectors are provided as inputs for generating the trained model, wherein the trained model is at least a Long Short-Term Memory (LSTM) model and a Bidirectional-LSTM model.

13. The computer readable media as claimed in claim 12 , wherein at least the LSTM and the Bidirectional-LSTM models are trained to generate a context vector indicating a context of the user review, wherein the context vector is used to determine a sentiment associated with a plurality of block of test data comprising texts.

14. The computer readable media as claimed in claim 13 , wherein the LSTM and the Bidirectional-LSTM models use an encoder-decoder model for generating the context vector using the one or more vectors and an output sequence using the context vector, wherein the output sequence indicates the sentiment associated with the block of text.

Assignees

• Wipro Ltd

Inventors

• Chatterjee Arindam
• Ramachandra Iyer Manjunath
• Bangalore Narayanamurthy Vinutha

Classifications

• G06F16/345 Primary

  Summarisation for human users · CPC title

• G06F40/30 Primary

  Semantic analysis · CPC title

• G06N3/044

  Recurrent networks, e.g. Hopfield networks · CPC title

• G06F18/214

  Generating training patterns; Bootstrap methods, e.g. bagging or boosting · CPC title

• G06N3/0442

  characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU] · CPC title