“Bestaurentz” – Traveler Decision Support System Using Natural Language Processing

Abstract

Picking a suitable hotel investigating the accessible information is one of the most complicated undertakings for travelers when planning their journeys. Around thousands of reviews with a range of information existing in social media web sites these days. With the growth of decision support system, explorers can assess the existing selections easily. Consequently, the travelers can make their pronouncements easily. Intention of this development is to develop traveler decision support system using online reviews. Deep learning techniques were used to develop this system and for Sentiment analysis of the hotel reviews. By this system, the emotions and the related information can be gathered from online reviews and capable of summarizing the final result. Nevertheless, this project was to label hotel reviews as positive, negative, or neutral and finally allow users to search them by keywords like food, cleanliness, etc. This research was done using the hotel reviews about the Sri Lankan hotels. Reviews were collected from the booking.com website. This website is a very popular website among local and foreign travelers in recent years. Nearly 66000 reviews were used as the data set for this research. All these reviews were written in the English language. In this project, a very basic machine learning algorithm which is Naive Bayes was used to build the classifier first. Due to the low accuracy of the classifier, wanted to use a new machine-learning algorithm to create the classifier. Considering the recent results of the NLP researches and the internal implementation, to create the sentiment analysis model, the convolutional neural network was used which is a class of artificial neural networks. This algorithm is most commonly used to analyze visual imageries but recently it has given promising results with text classification also. In order to directly classify the hotel reviews as positive, neutral, and negative, an emotional feature extraction mechanism was used to create the CNN classifier. This feature contains the list of emotional categories that are widely used and accepted by the Word Emotion Association Lexicon. It extracts the frequencies of emotional categories from the given textual data. The significance of using this feature can be measured in terms of the concept that it is necessary to obtain emotions from the texts as they convey significant information to identify hatred speeches, overexcited texts, encouraging comments, or such strong emotions, that are useful to classify them on the emotional basis. Ultimately incorporating the emotion extraction feature with the classification model, accuracy could be improved as expectedly.