Inspiration
Social media analytics has been a fast-growing industry for the past decade, with social media becoming a viable channel for marketing and advertising campaigns. With this in mind, and the boom in interest in machine learning and AI in the past year, we chose to approach this field with a fresh lens and utilize machine learning to search for new insights in social media performance.
What it does
Our project utilizes Python machine learning libraries such as sklearn to build a machine learning model based off of a public dataset of Twitter post engagement. We attempted to make this model accessible through a web app hosted on Github pages, where users would be able to input their own Tweet characteristics and receive a prediction of how well it would perform.
How we built it
Using Anaconda and Python libraries such as Numpy, Pandas, Matplotlib, and SKLearn, we trained a Poisson regression machine-learning model based on a public Twitter dataset we found on Kaggle. After training and finetuning this model, we planned on exporting the model using the built-in Python pickle library and running it on a website utilizing a separate Python script. To build this website, we utilized React and the Vite framework, and attempted to host it using GitHub pages.
Challenges we ran into
Finding a suitable dataset for our needs took a lot of time, as many of the datasets we found were tailored for specific uses and not general cases, and we ended up having to compromise with a dataset that was originally created for different uses but we were able to trim down to fit our needs. Finetuning the model to maximize the accuracy proved to be a challenge as well, and we spent much of our time working on it. Additionally, creating the workflow to allow for Github Pages to host Vite projects proved to be a large hurdle.
Accomplishments that we're proud of
We were able to tie several important concepts into one project, such as machine learning, dynamic web development, and scripting. We were able to brainstorm and create a proof of concept within the allotted time and brought our project to a reasonable level of functionality.
What we learned
We learned how to use several new technologies such as Vite and GitHub pages, as well as an understanding of web design workflows. Additionally, we learned about many different kinds of machine-learning models in the process of training our own model.
What's next for Machine Learning Twitter Post Engagement Prediction
The next steps our project can take will be in further fine-tuning our model, refining the web UI as well as the hosting service, and finding more applicable datasets.
Log in or sign up for Devpost to join the conversation.