Agent based methods for discovering and documenting user expectations
Techniques are described herein for using artificial intelligence to select, curate, normalize, enrich, and synthesize the results of user experience (UX) tests. In some embodiments, a system identifies a set of unstructured textual elements associated with one or more UX tests. The system may configure agents using generative language model services, including a reviewing agent that reviews and edit outputs of a machine learning classification model applied to the unstructured textual elements and/or a curating agent that selects unstructured textual elements to represent themes within the user experience test classified using the machine learning classification model. The outputs may be used to enhance the scalability, function, and efficiency of applications directed at improving product designs.
1. A method comprising:
receiving data that includes a plurality of unstructured textual elements;
generating, by a machine learning classification model, a set of theme assignments for at least a subset of the unstructured textual elements, wherein the set of theme assignments map each unstructured textual element of the subset to one or more themes from a theme schema;
sending the set of theme assignments to a generative language model for review;
updating at least one theme assignment made by the machine learning classification model based at least in part on an output of the generative language model; and
sending a second prompt to at least one of the generative language model or a second generative language model to select representative quotes for at least a first outcome and second outcome associated with a first theme, wherein the second prompt is generated based at least in part on the output of the generative language model.
2. The method of claim 1 , further comprising: tuning or retraining the machine learning classification model based at least in part on the output of the generative language model.
3. The method of claim 1 , wherein updating the at least one theme assignment comprises: removing the at least one theme assignment based at least in part on the output of the generative language model such that at least one of the subset of the unstructured textual elements is unclassified.
4. The method of claim 1 , wherein updating the at least one theme assignment comprises: reassigning at least one unstructured textual element to a different theme in the theme schema.
5. The method of claim 1 , further comprising: presenting, within a user interface, at least one theme from the theme schema and at least one respective outcome for at least one expectation element mapped to the at least one theme by the machine learning classification model, wherein the respective outcome is selected to present in the user interface based at least in part on a second output of the generative language model or the second generative language model that selects the representative quotes.
6. The method of claim 1 , wherein sending the set of theme assignments to the generative language model for review comprises generating at least one input prompt that includes a set of expectation elements from a user experience test that were assigned a particular theme; wherein the prompt includes a request for the generative language model to identify which expectations elements to unassign from the particular theme.
7. The method of claim 1 , wherein the plurality of unstructured textual elements include expectation quotes and outcome quotes; wherein the set of theme assignments identify predicted themes from the theme schema that are distributed in the expectation quotes; the method further comprising: determining a distribution of outcomes associated with at least a subset of the predicted themes.
8. The method of claim 1 , further comprising: configuring a first agent and a second agent using at least one generative language model service; wherein the first agent is configured to review and edit outputs of the machine learning classification model; wherein the second agent is configured to select representative unstructured text based at least in part on the outputs of the machine learning classification model reviewed by the first agent.
9. The method of claim 1 , further comprising: determining, based on the output of the generative language model, at least one insight associated with optimizing a product design; and presenting, through a user interface of a product analytics applications, the at least one insight to an analyst or product designer.
10. One or more non-transitory computer-readable media storing instructions which, when executed by one or more hardware processors cause:
receiving data that includes a plurality of unstructured textual elements;
generating, by a machine learning classification model, a set of theme assignments for at least a subset of the unstructured textual elements, wherein the set of theme assignments map each unstructured textual element of the subset to one or more themes from a theme schema;
sending the set of theme assignments to a generative language model for review;
updating at least one theme assignment made by the machine learning classification model based at least in part on an output of the generative language model; and
sending a second prompt to at least one of the generative language model or a second generative language model to select representative quotes for at least a first outcome and second outcome associated with a first theme, wherein the second prompt is generated based at least in part on the output of the generative language model.
11. The media of claim 10 , wherein the instructions further cause: tuning or retraining the machine learning classification model based at least in part on the output of the generative language model.
12. The media of claim 10 , wherein updating the at least one theme assignment comprises: removing the at least one theme assignment based at least in part on the output of the generative language model such that at least one of the subset of the unstructured textual elements is unclassified.
13. The media of claim 10 , wherein updating the at least one theme assignment comprises: reassigning at least one unstructured textual element to a different theme in the theme schema.
14. The media of claim 10 , wherein the instructions further cause: presenting, within a user interface, at least one theme from the theme schema and at least one respective outcome for at least one expectation element mapped to the at least one theme by the machine learning classification model, wherein the respective outcome is selected to present in the user interface based at least in part on a second output of the generative language model or the second generative language model that selects the representative quotes.
15. The media of claim 10 , wherein sending the set of theme assignments to the generative language model for review comprises generating at least one input prompt that includes a set of expectation elements from a user experience test that were assigned a particular theme; wherein the prompt includes a request for the generative language model to identify which expectations elements to unassign from the particular theme.
16. The media of claim 10 , wherein the plurality of unstructured textual elements include expectation quotes and outcome quotes; wherein the set of theme assignments identify predicted themes from the theme schema that are distributed in the expectation quotes; wherein the instructions further cause: determining a distribution of outcomes associated with at least a subset of the predicted themes.
17. The media of claim 10 , wherein the instructions further cause: configuring a first agent and a second agent using at least one generative language model service; wherein the first agent is configured to review and edit outputs of the machine learning classification model; wherein the second agent is configured to select representative unstructured text based at least in part on the outputs of the machine learning classification model reviewed by the first agent.
18. The media of claim 10 , wherein the instructions further cause: determining, based on the output of the generative language model, at least one insight associated with optimizing a product design; and presenting, through a user interface of a product analytics applications, the at least one insight to an analyst or product designer.
19. A system comprising:
one or more hardware processors;
one or more non-transitory computer-readable media storing instructions which, when executed by the one or more hardware processors cause:
receiving data that includes a plurality of unstructured textual elements;
generating, by a machine learning classification model, a set of theme assignments for at least a subset of the unstructured textual elements, wherein the set of theme assignments map each unstructured textual element of the subset to one or more themes from a theme schema;
sending the set of theme assignments to a generative language model for review;
updating at least one theme assignment made by the machine learning classification model based at least in part on an output of the generative language model; and
sending a second prompt to at least one of the generative language model or a second generative language model to select representative quotes for at least a first outcome and second outcome associated with a first theme, wherein the second prompt is generated based at least in part on the output of the generative language model.
20. The system of claim 19 , wherein the instructions further cause: tuning or retraining the machine learning classification model based at least in part on the output of the generative language model.