Skip to content

Sentiment Analysis on Covid 19 Tweets using Deep Neural Network i.e. LSTM and BiLSTM

Notifications You must be signed in to change notification settings

Sampad-Hegde/Sentiment-Analysis-on-Covid19-Tweets

Repository files navigation

Sentment-Analysis-on-Covid-19-Tweets

DataSet is Attached with python Notebook file. Dataset is already labelled (annoted) so you can directly use it for your project.

My Approch On Sentiment Analysis Using Deep Nueral Network

First Model is based on

  • Uses LSTM based Network with Input, Embedding, GlobalMaxPooling1D, Dense Layers
  • And its padding Sequence = 96 (max words in a tweet in this dataset)

Second Model is based on

  • Uses Bi-LSTM based Network with Input, Embedding, GlobalMaxPooling1D, Dense, Bidirectional,BatchNormalization, GlobalMaxPool1D, Dropout, Flatten Layers
  • And its padding Sequence = 20 (randomly taken avg length of tweets text in this dataset is 15)

Procedure

  • 'Here, is an sample text for the input #TextFile' (raw input tweet text content)
  • 'Here sample text input' (after textCleaning and removing the stopwords)
  • ['Here sample text input'] (converted to python list)
  • ['Here','sample','text','input'] (Sentense is splitted int words)
  • [10,250,381,58] (each number that maps to position of the word in the VOCABULARY built using taking all tweet text in the dataset)
  • [10,250,381,58, 0,0,0,0,0 ,0,0,0] (padded sequence for length = 12 this is used make all sentense to same length = input layer size)
  • Now it can be feeded into Nueral Network.

Tips:

  • If you Reduce the padding sequence Model accuracy increases but it will ignore words that comes after seq length

!!! Training these model takes lot of time if have nvidia gpu install cuda drivers that reduces the training time significantly but, it is completely optional !!!

About

Sentiment Analysis on Covid 19 Tweets using Deep Neural Network i.e. LSTM and BiLSTM

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published