Sentiment analysis tuning

Patent metadata

Field	Value
Publication number	US-10878196-B2
Application number	US-201816149770-A
Country	US
Kind code	B2
Filing date	Oct 2, 2018
Priority date	Oct 2, 2018
Publication date	Dec 29, 2020
Grant date	Dec 29, 2020

Abstract

In one example, the present disclosure describes a device, computer-readable medium, and method for sentiment analysis tuning. In one example, the method, includes acquiring a first sentiment analysis-generated score for a first string of text, wherein the first string of text includes a first plurality of words, and wherein the first sentiment analysis-generated score is calculated using a set of first values associated with the first plurality of words, calculating a second value for at least one word of the first plurality of words, based on a non-sentiment-analysis generated score associated with the first string of text, acquiring a second string of text, wherein the second string of text includes a second plurality of words, and wherein the second plurality of words includes the at least one word, and calculating a sentiment analysis-generated score for the second string of text, using the second value for the at least one word.

First claim

What is claimed is:

1. A method, comprising:
acquiring, by a processing system including at least one processor, a first sentiment analysis-generated score for a first string of text, wherein the first string of text includes a first plurality of words, and wherein the first sentiment analysis-generated score is calculated using a set of first values associated with the first plurality of words;
calculating, by the processing system, a second value for at least one word of the first plurality of words, based on a first non-sentiment-analysis generated score associated with the first string of text;
acquiring, by the processing system, a second string of text, wherein the second string of text includes a second plurality of words, and wherein the second plurality of words includes the at least one word;
calculating, by the processing system, a second sentiment analysis-generated score for the second string of text, using the second value for the at least one word; and
predicting, by the processing system, a second non-sentiment analysis-generated score associated with the second string of text, using the second value, wherein the second non-sentiment analysis-generated score is a net promoter score, wherein the second value is used to establish a correlation between a plurality of sentiment analysis-generated scores and a plurality of net promoter scores, wherein the correlation is established by:
acquiring a plurality of strings of text and a plurality of non-sentiment analysis-generated scores, wherein each score of the plurality of non-sentiment analysis-generated scores corresponds to one string of text of the plurality of strings of text;
calculating the plurality of sentiment analysis-generated scores using the second value, wherein each sentiment analysis-generated score of the plurality of sentiment analysis-generated scores corresponds to one string of text of the plurality of strings of text;
grouping the plurality of sentiment analysis-generated scores into a plurality of clusters, wherein each cluster of the plurality of clusters represents one sentiment analysis-generated score of the plurality of sentiment analysis-generated scores; and
calculating the plurality of net promoter scores using the plurality of non-sentiment analysis-generated scores, wherein each net promoter score of the plurality of net promoter scores corresponds to one cluster of the plurality of clusters.

2. The method of claim 1 , wherein the first non-sentiment analysis generated score comprises a numerical score automatically derived from non-text features of a multi-modal input associated with the string of text.

3. The method of claim 2 , wherein the multi-modal input comprises an audible input.

4. The method of claim 2 , wherein the multi-modal input comprises a visual input.

5. The method of claim 2 , wherein the numerical score is generated by:
representing the non-text features as a vector; and
assigning the numerical score to the vector using a deep learning technique.

6. The method of claim 1 , wherein the first non-sentiment-analysis generated score comprises a numerical rating provided by a same source as the string of text.

7. The method of claim 1 , wherein each value of the set of first values comprises a weight that implies a magnitude of a sentiment expressed by a corresponding word of the first plurality of words.

8. The method of claim 7 , wherein the second value comprises an adjusted weight that implies a magnitude of a sentiment expressed by the at least one word.

9. The method of claim 7 , wherein the first sentiment analysis-generated score is calculated by combining values of the set of first values that correspond to the first plurality of words.

10. The method of claim 1 , wherein the calculating the second value for the at least one word is performed using a multidimensional optimization problem that finds a set of second values including the second value, and wherein the set of second values, when applied to the first plurality of words, maximizes a correlation between the first sentiment analysis-generated score and the first non-sentiment analysis-generated score.

11. The method of claim 10 , wherein the multidimensional optimization problem is tuned for a specific domain.

12. The method of claim 1 , wherein the second string of text comprises a survey response provided by a customer of a service provider.

13. The method of claim 1 , wherein each sentiment analysis-generated score of the plurality of sentiment analysis-generated scores that is included in a given cluster of the plurality of clusters has an identical score.

14. The method of claim 1 , wherein the calculating the plurality of net promoter scores for a given cluster of the plurality of clusters is based on non-sentiment analysis-generated scores of the plurality of non-sentiment analysis-generated scores corresponding to strings of text of the plurality of strings of text that are represented by sentiment analysis-generated scores of the plurality of sentiment analysis-generated scores that are included in the given cluster.

15. A device, comprising:
a processor; and
a non-transitory computer-readable medium storing instructions which, when executed by the processor, cause the processor to perform operations comprising:
acquiring a first sentiment analysis-generated score for a first string of text, wherein the first string of text includes a first plurality of words, and wherein the first sentiment analysis-generated score is calculated using a set of first values associated with the first plurality of words;
calculating a second value for at least one word of the first plurality of words, based on a non-sentiment-analysis generated score associated with the first string of text;
acquiring a second string of text, wherein the second string of text includes a second plurality of words, and wherein the second plurality of words includes the at least one word;
calculating a sentiment analysis-generated score for the second string of text, using the second value for the at least one word; and
predicting a second non-sentiment analysis-generated score associated with the second string of text, using the second value, wherein the second non-sentiment analysis-generated score is a net promoter score, wherein the second value is used to establish a correlation between a plurality of sentiment analysis-generated scores and a plurality of net promoter scores, wherein the correlation is established by:
acquiring a plurality of strings of text and a plurality of non-sentiment analysis-generated scores, wherein each score of the plurality of non-sentiment analysis-generated scores corresponds to one string of text of the plurality of strings of text;
calculating the plurality of sentiment analysis-generated scores using the second value, wherein each sentiment analysis-generated score of the plurality of sentiment analysis-generated scores corresponds to one string of text of the plurality of strings of text;
grouping the plurality of sentiment analysis-generated scores into a plurality of clusters, wherein each cluster of the plurality of clusters represents one sentiment analysis-generated score of the plurality of sentiment analysis-generated scores; and
calculating the plurality of net promoter scores using the plurality of non-sentiment analysis-generated scores, wherein each net promoter score of the plurality of net promoter scores corresponds to one cluster of the plurality of clusters.

16. A non-transitory computer-readable medium storing instructions which, when executed by a processor, cause the process

Assignees

• At & T Ip I Lp

Inventors

• Jacobson Guy
• Hsu Wen-Ling
• Xu Tan

Classifications

• G06V30/274

  Syntactic or semantic context, e.g. balancing · CPC title

• G06F18/23

  Clustering techniques · CPC title

• G06N5/01

  Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound · CPC title

• G06N3/045

  Combinations of networks · CPC title

• G06N3/09

  Supervised learning · CPC title