This is the minor-project that is to be submitted to my University, during the 7th semester.We build a BiLSTM model and train the model on textual data - Twitter data. This project includes developing in two stages.
In this stage we develop a Deep Learning model, to train it on the data for sentiment classification. This stage includes:
The data is prepared from Twitter Scraping. The dataset includes thousands of tweets which are segregated into 3 domains, namely, positive, negative and neutral sentiments respecively.
-> We First tokenize the data, which consists of sentences or tweets. While tokenizing we also clean the data, cleaning the html codes, removing urls and hashtags, etc. -> Next we lemmatize the data, to shrink down the vocabulary of the dataset and reduce complexity of the data. -> Lastly we remove numbers or special characters from the data.
Lastly, we intantiate the BilSTM model, using the sequential wrapper class. Once the model is trained, then we save the weights of the models to be used later onto the 2nd stage.
In this stage we develop a web application on the Django framework. This stage includes:
So first we developed the django application and tested it on the development server. The application has a model-views-template architectural system. For our project
we only work upon the template-views sytem only as we are not storing data from the website itself.
Once developed, we deploy the web application on Heroku platform.
Link to site: https://sentiment-analysis-bilstm.herokuapp.com/ NOTE: The site takes time to load, this is due to the fact that BiLSTM is a heavy model for the free version of heroku hardware.