Skip to content

Latest commit

 

History

History
36 lines (23 loc) · 2.07 KB

File metadata and controls

36 lines (23 loc) · 2.07 KB

Graduate Big Data:

Emotion Sentiment Classification of Product Reviews on Twitter

Motivation

It is considerably more difficult to discern someone's tone in writing than it is in conversation. For starters, you cannot infer an author's emotional condition from their facial expressions or body language. For instance, the tone of a message is typically not determined by a single characteristic. Numerous aspects are involved in effective communication. Because in some circumstances, what you say can be just as important as how you say it. Our approach is designed to offer feedback on a variety of keywords that represent the emotion of the message. It's also helpful in ensuring that your recipient will focus on the facts you are presenting rather than unintended consequences of your tone by making strategic judgments about how you convey your message.

Model Implementation

  • BERT
  • RoBerta
  • RNN

Proposal

To Implement two pipeline in AWS to categorise the Article/Conversation into 6 Classes (happy, angry, surprise, sad, disgust, not-relevant).

  1. Pipeline-1 : Using Spark-MLlib to implement Logistic Regression, Random Forest and SVM

SparkMl

  1. Pipeline-2 : Using Sagemaker to implement Deep Learning Model including BERT, Roberta, and RNN.

ML in Sagemaneker

Result

Achieved 97.83% accuracy and 0.83 F1-score on RNN model.

Snapshot

  1. BERT Bert Accruaracy
  2. RoBERTa RobertA Accuracy
  3. RNN RNN Accuracy