Hot Topic detection in Tweets using CNN-LSTM Classifier and Modified Bald Eagle Search Algorithm

Abstract

A good feature engineering produces high accuracy in hot topic detection but its essential to
have an excessive degree of expertise and is also prolonged process. The neural network
shows the better results in less time by replacing the need for high degree of expertise. In this
article, CNN-LSTM-MBES hybrid model is built to produce promising results in identifying
the hot topics in twitter. A single CNN learns the features deeply but cannot remember the
longer sequence of data while LSTM can remember the longer sequence of data along with
important data and at same time discard the unimportant data to infer the meaningful pattern,
but LSTM takes longer time for training the model. This challenge can be resolved by
introducing CNN along with LSTM and weight updation using the modified bald eagle
search (MBES) algorithm. Proposing the modified bald eagle search (MBES) algorithm by
introducing the weight factor in choosing the search space instead of a random selection in
the BES, eliminates the local optimal problem which is prevailing problem in BES. This
modified bald eagle search (MBES) algorithm is used to optimize and improve the
performance of CNN-LSTM model. Twitter data is considered in our experimental study.
The proposed model performance is compared current classifiers. The proposed model
reported an accuracy of 99.71% which is 13.6% increase in accuracy compared to other
existing classifiers and drop-in computation time when compared with CNN-LSTM-BES.