فهرست نشریات علمی دسترسی آزاد ایران

This study aimed to predict and categorize accommodation prices on the “Shab” website, a leading accommodation booking platform in Iran. In the first phase, we extracted complete information of 7448 accommodations through web scraping and formed the IRAPP dataset which is publicly available for future research. IRAPP encompasses a variety of features: textual descriptions of accommodations, numerical attributes such as prices and ratings, and binary indicators of facility availability for individuals with disabilities and the elderly. In the second phase, data preprocessing involved transforming textual features into feature vectors using advanced natural language processing models, including AraBERT, FARBERT, RoBERTa, and Shiraz. These models effectively extract deep features from Persian texts, significantly enhancing prediction accuracy. The third phase focused on training various machine learning and deep learning models to predict accommodation prices, with performance assessed using standard error metrics. Three specific tasks were defined: price prediction, 5-class classification, and binary classification. A range of machine learning models, such as Random Forest, Logistic Regression, SVC, Naïve Bayes, Decision Tree, MLP, XGBoost, and AdaBoost, were employed in this analysis. Results from the model implementations highlighted the superiority of newer large language models in extracting textual features and demonstrated the beneficial effects of integrating textual and numerical features on prediction accuracy and price classification. Specifically, the voting regressor model outperformed others in the price prediction task, the XGBoost model excelled in the 5-class price classification task, and the Random Forest model achieved the highest efficiency in binary classification compared to the other models.