The main problem with the previous step of identifying topics, is that there are a lot of unique topics, which is normal due to the nature of generative models. We will need to combine them into categories to make it easier for analysis.

Prompt:

You are a product manager who has to research the user experience of your app.

Your task is to categorize a list of topics mentioned in user reviews.

Here is a list of topics

{{topics}}

Please output only a RAW json of following structure: [ ”$topic_category_name”, … ]

Observations:

Category numbers lowered significantly but still too much for easy analysis.

Prompt V2:

You are a product manager who has to research the user experience of your app.

Your task is to categorize a list of topics mentioned in user reviews.

Each category should be related to a single product feature related to the topic.

Categories must not intersect.

Here is a list of topics

{{topics}}

Include each topic in a single category.

Solve this problem step by step: First extract major categories. Map each topic to the best matching category.

Please output only a RAW json of following structure: [ ”$topic_category_name”, … ]

Observations:

Category numbers lowered significantly around 16 categories. From this point, we can start mapping each topic to a category.

Prompt:

You are a product manager who has to research the user experience of your app.

Your task is to map a topic into one most suitable category from given categories.

Use only the given list of categories.

Here is a json with category names

{{categories}}

Here is a topic

{{topic}}

Solve this problem step by step:

Please output only a RAW json of following structure: {{ ”$topic”: “$category_from_list_above”, … }}

Observations and metrics:

I used the platform Make to automate the process. The share of topics that were correctly categorized was 0.97 which is a good result.

Possible next steps

1. We can use the data to visualize and analyze the reviews so that we can answer business questions.
2. Develop an internal tool that periodically get new reviews then visualize and analyze it, with a tailored portals and experiences for different teams like sales, marketing, engineering, …etc.