The problem Mental Health - Suicidal Intention Detector solves

Mental health can lead to mental disorders. Our team has built a web application which detects whether a text has suicidal intents or not. The application provides a motivational quote depending on which intent the support vector machine classifier predicts. Our motive is to provide motivational quotes to users using our application. Motivation to not give up on life plays an important role in one's self-development.

Challenges we ran into

The Machine Learning model we build was underfitting the dataset due to less number of data points (around 9K only), this could be due to corpus size reduction using preprocessing techniques like (removing stopwords, stemming, lemmatization, etc). We cleaned our dataset by removing unwanted information like email, retweet tag, URLs, special characters, HTML tags, accented characters and multiple spaces. To solve this issue, we trained on classifier by just removing stopwords and lowering the text separately.
We tried to use a BERT and PEGASUS Model for our classification task, but due to lack of time and computational requirements, we couldn't train such a heavy model.
Another issue we could face was if we trained BERT using our problem, Heroku doesn't provide us with so much space in the cloud platform (in the free edition which we used) so to build the model we used a simple SVM model which was giving us decent accuracy.