Sentiment-Classifier

This is a sentiment classification API, created for as a term project for RPI Natural Language Processing Spring 2018.

There are seven possible emotions: guilt, anger, sadness, joy, shame, fear, and disgust.

Test Set	Accuracy
Train	83.46%
Test	51%

• Note: Windows users should change python3 to python in start.sh

Requirements

• Python 3.6
• Flask
• Textblob

Scripts

• Create the train and test files (already done): python clean.py
• Test the classifier without running the server python classifier.py
• Start the server ./start.sh
• Interactively classify sentences python interactive.py

API Routes

Route	HTTP Method	Return Type	Description	Example Return
/	GET	String	Classifies "This is a test sentence."	"joy"
/sentiment	POST	String	Classifies the sentence in the "text" for field.	"sadness"
/sentiments	POST	JSON	Returns the probabiltiy of each emotion over a minimum threshold over 10%.	{'sentiments': [{'disgust': 0.41}, {'sadness': 0.20}, {'shame': 0.18}]}
/accuracy	GET	JSON	Tests the accuracy of the classifier on the train and test datasets.	{"train_accuracy": 0.8346, "test_accuracy": 0.51}
/labels	GET	JSON	Returns the set of all possible classifications.	{"labels": ["guilt", "anger", "sadness", "joy", "shame", "fear", "disgust"]}

Dataset courtesy of Emotion Research