TREND MATRIX

Christmas Eve 2023, I came across a topic modeling technique called BERTopic, and instantly a passion project emerged that day that I went on to obsessively work on consistently for several months. “Using our own embeddings allows you to try out BERTopic several times until you find the topics that suit you best,” explained by the model’s creator Maarten Grootendorst.

THE TOOL

My passion when working with data is optimizing categorization to quickly monitor growth as well as emerging interest. One of the greatest issues when working with text-based data, such as product names, headlines, product reviews, product descriptions, social comments, meta captions, etc. is categorizing all the topics addressed to group more and more refined themes. The Trend Matrix is a tool I created that cleans, transforms, clusters, and then categorizes text data using AI prompt engineering. It allows any text-based data to be transformed into an interactive dashboard of category specific trends, an interactive UMAP of clustered topics. Then I added additional data visualizations such as a sankey diagram, one that is filterable to specific categories as to holistically visualize the universe of all your data.

Product Explorer

The final output is an interactive Dash app of a bar chart visualization that I hosted on Heroku for anyone to use. This example is filtered on Tops.

*Heroku app is currently under construction

BERTopic Visualizations

UMAP Visualization

Of product conversion data from Spring 2023

Topic Word Score Bar Charts

Above are a BERTopic visualization that is prebuilt in the library easily visualizes bar charts of your categorized data.

Sankey Diagrams

Filtered Sankey Thread

This example is filtered on Bottoms of product conversion data from Spring 2023.

The Slankey Diagram

Data visualized is product conversion data from Spring 2023.