Group 85 Project

We have observed that student often seek respite from online forums as compared to traditional counsellors these days.Equipped with our intertest in Machine Learning, we decided to take this matter up. We decided to use Facebook posts from NTU Confessions, a forum that many NTU students resort to.

It analyses the posts and gives us a score. A sentiment score of 0 indicates neutral emotions. A negative sentiment score indicates negative emotions. A positive sentiment score indicates positive emotions. We then found correlation between the sentiment scores and the weather at the time of posting and the effect of having holidays on the post frequency.

We acquired data from NTU Confession by web scraping the Facebook page to create our own dataset. We then cleaned up the data. We ran a VADER sentiment model on this data in a Jupyter Notebook. This gave us sentiment scores relative to different attributes such as time and date of posting, etc. We tried to find correlations between the sentiment scores and 2 factors: Weather at the time of posting and if it was posted on a holiday.

We were unable to find a correlation between holidays and sentiment scores in the given time

None of us had any prior knowlege about Machine Learning. This allowed us to lean a lot during this hackathon.

The study can be expanded to include earlier posts and can be compared to factors such as daily COVID cases, exam dates, holidays and more. The model can be further developed such that the admin of the forum can be notified if the program detects very strong negative emotions. This can enable the admin to reach out to the user. Sentiment analysis can also be run on major social media platforms such as Facebook and Instagram to detect if someone is undergoing strong negative emotions, and family or friends can be notified accordingly. Although, this should be deployed in such a way that their privacy isn’t intruded to a large extent.

Pls note that we did not have a video

Built With

• jupyter
• python