Social Media Post Approval Analytics For Marky

Inspiration

The business case for deriving insights regarding user approval was an interesting problem to solve. Hence our team decided to tackle it head on and gained a lot of insights regarding features that impact the overall approval.

What it does

Built classifiers to successfully predict the likelihood a user will approve a post. Additionally, we determined what features are most important for Marky to improve the approval percentage of their future posts.

How we built it

We used google colab, writing code using python. We first loaded in the data. Then cleans the dataset by getting rid of all null values, getting rid of stop words, emojis, and more in order to be able to evaluate it cleanly and easily. We also created various other statistics like the similarity using cosine similarity and tested it using different models and libraries in order to predict the social media approval rating to the best we can, these were built using python library sk learn.

Challenges we ran into

Our initial plan was to integrate both text and image features into our models through stacking both the features sets with each other for model consideration. However, we ran into an incompatibility between the vectorized text and images as they were not the same dimensionality and caused TensorFlow errors. We had to cut short the scope of our feature base to account for the time limit.

Accomplishments that we're proud of

We are proud of the various features we implemented in the time we had and even though some couldn't work it worked.

What we learned

We learned how to built a machine learning model. Starting with cleaning the data by removing stopwords and making the data binary, and for the strings we turned them into TF_IDF vectoriser and the models using svm, randomforest, linear regression. We gained valuable insights into constructing a machine learning model, beginning with the crucial step of data preprocessing. This involved the removal of stopwords and the binarization of our dataset. Subsequently, we experimented with various models, including Support Vector Machines, Random Forest, and Linear Regression, to find the optimal approach for our project.

What's next for Social Media Post Approval Analytics For Marky

By using a personalized approval system, we take in what the user approved before to tailor inform of the model in order to better predict what the individual user would prefer.