Risk Prediction Sentiment Analysis 3


Workplace safety is a growing issue in today's society and there have been many concerns in regards to the safety of workplace environments. This project looks to use worker narratives from a workplace to analyze general sentiment to predict risk. This can be done from a web application which allows users to upload worker narratives with CSV files and run the narratives on an NLP model to classify them with sentiment scores.

The main problem with building such a system is creating an effective sentiment analysis model geared towards worker narratives. General sentiment analysis models are built to analyze the sentiment of everyday conversations rather than the specific use case of worker narratives. The prior group used the NLP model from AWS Comprehend. We surpassed the accuracy of that model leveraging the services of Google Cloud Platform, which introduced some migration complications from the AWS to Google Cloud infrastructure. The model was trained using 1544 manually labeled worker narratives to allow it to gauge sentiment specific to work environments. We also improved the user experience and security of the application, introducing instructions and login pages.



Sentiment Analysis, Machine Learning, Full Stack Application, AWS