README.md

How to start the flask server?

Use the following command line: python3 [filename.py] (example: python3 sentimentAnalysis.py). It return a host ip with port number (Note: Acc. to our convenience, we can change host ip in flask server code).

Quick start about our service Sentiment Analysis

Sentiment Analysis (SA) is the interpretation and classification of sentiment (positive, negative and neutral) within text data using text analysis techniques. A sentiment analysis system for text analysis combines natural language processing (NLP) and machine learning techniques to assign weighted sentiment scores to the entities, topics, themes and categories within a sentence or phrase.

Our sentiment analysis model detect polarity score within a given text, based on the polarity score model assigns a sentiment label. Here for our model used the coarse-grained sentiment anaysis lables those are Positive, Negative, Neutral. In this NLP-as-a-toolkit service we are building the SA model for normal English text and Hinglish (Hindi and English Code-Mixed data) text. To build the model for nomal English text used the open-source libraries TextBlob and NLTK's VADER (Valence Aware Dictionary and sEntiment Reasoner) and for Code-Mixed text used the MultiLayer Perceptron (MLP) with TFIDF vectors.

Model for Normal Text

We'll give the input text for TextBlob and VADER. Each model predict a sentiment label score. Based on that score we will assign maximum scored label to input text.

Examples:

Input	VADER Score	TextBlob Score	Output Label
VADER is VERY SMART, handsome, and FUNNY	0.63	0.578	Positive
A really bad, horrible book.	-0.432	-0.231	Negative

positive sentiment: score >= 0.5, neutral sentiment: score > -0.5 and score < 0.5, negative sentiment: score <= -0.5

Model for Code-Mixed Text

For this text we built the Multilayer Perceptron model with word level TF-IDF vectors. And this model is trained on IIITH and SentiMix Hindi English code-mixed datasets which contains around 13000 tweets. First, We converted the each tweet in the dataset into a d-dimensional word level TF-IDF vector format. These vector representations fed to MLP to predict the label.

Examples

Input	Actual Label	Predicted Label
Trailer dhannnsu hai bhai	Positive	Positive
abe kutte tere se kon baat karega	Negative	Negative

Results

Model	precision	Recall	f1-score
TextBlob	0.75	0.74	0.76
VADER	0.8	0.8	0.8
MLP	0.6505	0.6433	0.6575

References

1. Datasets
   • IIITH dataset
   • SentiMix dataset
2. Papers
   • Code-Mixed Sentiment Analysis Using Machine Learning and Neural Network Approaches
   • Sentiment Analysis for Code-Mixed Indian Social Media Text With Distributed Representation
   • SENTIMENT ANALYSIS OF MIXED CODE FOR THE TRANSLITERATED HINDI AND MARATHI TEXTS