Multi-channel feedback analytics for presentation generation

1. A computer-implemented method in which one or more processing devices perform operations comprising:

receiving, from a client computing device, a request to generate a presentation;

accessing one or more feedback datasets of feedback data, the feedback data comprising unstructured data available from multiple data stores;

generating, for each feedback dataset, a respective feedback text and a respective sentiment score indicating a degree of negativity associated with the respective feedback text;

for a combination of a plurality of generated feedback texts, selecting a set of themes based at least on a plurality of generated sentiment scores, wherein each sentiment score of the plurality of generated sentiment scores is associated with one of the plurality of generated feedback texts, wherein the theme selection comprises:

extracting a set of phrases from the combination of the plurality of generated feedback texts;

applying a phrase normalization process to the set of phrases to generate a distribution of normalized phrases, wherein the phrase normalization process comprises:

applying a language model to each of the phrases to generate a respective phrase embedding; and

determining a group of phrases having similar phrase embeddings and removing phrases from the group except for a representative phrase for the group, the representative phrase determined by:

ranking the group of phrases according to a character length of the respective phrases; and

selecting a longest of the ranked group of phrases as the representative phrase for the group; and

selecting the set of themes from the normalized phrases;

generating a presentation file that indicates the set of themes; and

causing the presentation file to be transmitted to the client computing device.

2. The computer-implemented method of claim 1 , the operations further comprising:

determining a theme score for each of the themes based on one or more of a part of speech pattern, a location of the theme within its respective feedback text, and a number of similar themes associated with the respective theme; and

ranking the themes according to theme score.

3. The computer-implemented method of claim 1 , wherein selecting the set of themes from the normalized phrases comprises:

ranking each normalized phrase of the distribution of normalized phrases according to the sentiment score of the feedback text from which the normalized phrase is determined; and

selecting a set of normalized phrases having sentiment scores indicating a greatest degree of negativity.

4. The computer-implemented method of claim 1 , wherein the theme selection further comprises applying a noise removal process to generate, from the set of phrases, a reduced set of phrases, wherein the phrase normalization process is applied to the reduced set of phrases.

5. The computer-implemented method of claim 4 , wherein the noise removal process includes determining an inverse document frequency (IDF) for each phrase of the set of phrases and removing phrases having an IDF of less than a threshold IDF.

6. The computer-implemented method of claim 1 , wherein the theme selection further comprises:

generating a ranked list of themes by ranking the normalized phrases based at least in part on the respective sentiment score associated with the respective feedback text from which the normalized phrase is determined, wherein selecting the set of themes from the normalized phrases comprises selecting a predefined number of top-ranked themes from the ranked list of themes.

7. The computer-implemented method of claim 1 , wherein the phrase normalization process further comprises:

removing, except for the representative phrase, phrases associated with a group of phrase embeddings having a common root form.

8. The computer-implemented method of claim 1 , further comprising, for the combination of the plurality of generated feedback texts, generating a set of actionable insights including one or more issues an entity can improve or otherwise address by taking action and a set of questions.

9. The computer-implemented method of claim 8 ,

wherein the request indicates a presentation template to present one or more combinations of theme information, actionable insights information, and question information; and

wherein the presentation file includes one or more combinations of the themes, actionable insights, and questions according to the presentation template.

10. A system comprising:

a processor; and

a non-transitory computer readable medium storing computer-readable program instructions that, when executed by the processor, cause the system to:

receive, from a client computing device, a request to generate a presentation;

access one or more feedback datasets of feedback data, the feedback data comprising unstructured data available from multiple data stores;

generate, for each feedback dataset, a respective feedback text and a respective sentiment score indicating a degree of negativity associated with the respective feedback text;

for a combination of a plurality of generated feedback texts, select a set of themes based at least on a plurality of generated sentiment scores, wherein each sentiment score of the plurality of generated sentiment scores is associated with one of the plurality of generated feedback texts, wherein the theme selection comprises:

extract a set of phrases from the combination of the plurality of generated feedback texts;

apply a phrase normalization process to the set of phrases to generate a distribution of normalized phrases, wherein the phrase normalization process comprises:

apply a language model to each of the phrases to generate a respective phrase embedding; and

determine a group of phrases having similar phrase embeddings and removing phrases from the group except for a representative phrase for the group, the representative phrase determined by:

rank the group of phrases according to a character length of the respective phrases; and

select a longest of the ranked group of phrases as the representative phrase for the group; and

select the set of themes from the normalized phrases;

generate a presentation file that indicates the set of themes; and

cause the presentation file to be transmitted to the client computing device.

11. The system of claim 10 , the non-transitory computer readable medium further storing computer-readable program instructions that, when executed by the processor, cause the system to:

determine a theme score for each of the themes based on one or more of a part of speech pattern, a location of the theme within its respective feedback text, and a number of similar themes associated with the respective theme; and

rank the themes according to theme score.

12. The system of claim 10 , wherein selecting the set of themes from the normalized phrases comprises:

ranking each normalized phrase of the distribution of normalized phrases according to the sentiment score of the feedback text from which the normalized phrase is determined; and

selecting a set of normalized phrases having sentiment scores indicating a greatest degree of negativity.

13. A non-transitory computer-readable medium having program code that is stored thereon, the program code executable by one or more processing devices for performing operations comprising:

receiving, from a client computing device, a request to generate a presentation;

accessing one or more feedback datasets of feedback data, the feedback data comprising unstructured data available from multiple data stores;

generating, for each feedback dataset, a respective feedback text and a respective sentiment score indicating a degree of negativity associated with the respective feedback text;

for a combination of a plurality of generated feedback texts, selecting a set of themes based at least on a plurality of generated sentiment scores, wherein each sentiment score of the plurality of generated sentiment scores is associated with one of the plurality of generated feedback texts, wherein the theme selection comprises:

extracting a set of phrases from the combination of the plurality of generated feedback texts;

applying a phrase normalization process to the set of phrases to generate a distribution of normalized phrases, wherein the phrase normalization process comprises:

applying a language model to each of the phrases to generate a respective phrase embedding; and

determining a group of phrases having similar phrase embeddings and removing phrases from the group except for a representative phrase for the group, the representative phrase determined by:

ranking the group of phrases according to a character length of the respective phrases; and

selecting a longest of the ranked group of phrases as the representative phrase for the group; and

selecting the set of themes from the normalized phrases;

generating a presentation file that indicates the set of themes; and

causing the presentation file to be transmitted to the client computing device.

14. The non-transitory computer-readable medium of claim 13 , the operations further comprising:

determining a theme score for each of the themes based on one or more of a part of speech pattern, a location of the theme within its respective feedback text, and a number of similar themes associated with the respective theme; and

ranking the themes according to theme score.

15. The non-transitory computer-readable medium of claim 13 , wherein the theme selection further comprises generating a ranked list of themes by ranking the normalized phrases based at least in part on the respective sentiment score associated with the respective feedback text from which the normalized phrase is determined, wherein selecting the set of themes from the normalized phrases comprises selecting a predefined number of top-ranked themes from the ranked list of themes.