Publication

Abstract : Sentiment analysis is a useful method for figuring out what people think in areas like tourism, where customer feedback is a key element of making business decisions. Many people who travel wish to share their experiences at different places by writing reviews on social media. People who work in the tourism business can improve service delivery by understanding more about this unstructured data. Some of the most common machine learning approaches for sentiment analysis are Support Vector Machines (SVM), Naive Bayes, and Random Forest. But these strategies aren't enough to understand the complete meaning of a text in its context or the long-term connections between words. It's impossible to determine what tourists really think when they leave evaluations because of this. To overcome these challenges, we provide the Deep Ensemble Learning Model, which puts CNNs, RNNs, and LSTMs together into one model. This method removes stop words, breaks up and lemmatizes social media reviews, and then processes the reviews. Deep learning algorithms are utilized to uncover spatial, sequential, and temporal patterns after acquiring the features from GloVe word embeddings. First, each model tries to figure out if the emotion is good or bad. After then, the final guess is established by either weighted averaging or a vote from the majority. The ensemble in issue is both robust and able to cover a wide range of tourist sensations. A lot of traveler reviews from Twitter and TripAdvisor were used to test the system. The suggested strategy had an average accuracy and F1-score that was 6% to 12% higher than typical machine learning methods. The ensemble might combine the best aspects of CNN, RNN, and LSTM to swiftly and effectively handle sentence-level patterns, uncover long-term connections, and record local phrases.