Paper Title
A Novel Federated LSTM Model with Conventional LSTM Model for Sentiment Analysis of Twitter Datasets

Abstract
The purpose of tweet sentiment analysis is to determine the feelings or thoughts behind the tweet. For the tweet sentiment classification, there are many classical classifiers available that are proposed by researchers. Here in this paper, we are using a federated approach to train our Long Short-Term Memory (LSTM) neural network for sentiment analysis purposes. Federated learning is the mechanism for on-device and distributed learning in machine learning. The federated learning approach works differently fundamentally because, as training happens on different devices, it is not possible to use the classic gradient descent approach. Continual learning is also possible with a federated approach. In this paper, for the federated training, we used the "Federated Averaging" algorithm. We trained our LSTM neural network on 40,000 tweets in a federated way. After the experiment, it was discovered that the federated learning method slightly performed better than server-based learning, with the best 73.5% precision, 73.5% recall, 73.5% F1 score, 74.0% accuracy, and the lowest 9.11 loss rate on the "Sentiment 140" dataset. This experiment was designed to carry out LSTM training in a centralized and federated manner to compare various aspects of federated learning. Keywords - Federated learning, Long Short-Term Memory (LSTM), Server based learning, Tweet sentiment Analysis.