Using OpenAI to help classify Instagram Posts to understand audience engagement

I was recently asked to think about how a brand could go about understanding their audience more on social media. In particular a company wanted to better understand the audience of one of their content creator partners.

The creator’s principle medium was Instagram, and whilst there posts did contain some text a lot of the meaning and context of a post is in the image posted.

I explored the idea of using OpenAI’s vision models and their regular GPT4 text model together to analyse Instagram posts and then categorise them according to a well defined marketing taxonomy (the IAB taxonomy).

I build a prototype system that can retrieve all the posts the creator has made over a given time period, feed each post’s image and text to the Vision and Language models and get a selection of the most appropriate subjects (taxonomy terms) applicable to that post.

The resulting data points when plotted out clearly show the subjects where audience engagement is highest. In this particular instance as well as highlighting subject areas the influencer was expected to be popular for they dsicvered other areas that drove significant engagement even though the didnt post as regularly on the subject.

The client agency was so happy with the results they have now used this to perform the same analysis on many more of their social content creator partners.