Enhanced Sentiment Analysis Toward Specific Locations and Neighborhoods with Advanced Machine Learning Techniques

Abstract

Sentiment analysis has become an important field of study in recent years because it enables the evaluation of public opinions collected from multiple data sources. This study highlights the importance of understanding public perceptions regarding specific areas and communities, which is essential for urban planning, tourism, real estate, and community engagement. By using diverse sources such as social media platforms and online reviews, the study applies sentiment analysis techniques to identify shared attitudes and emotional reactions toward geographical locations. The resulting analysis provides detailed insights that support decision-making processes in areas such as city planning, tourism development, and public service improvement. These sentiments are classified into three categories: positive, negative, and neutral. This study applies comparative machine learning approaches to a QA-based geospatial aspect-based sentiment analysis (ABSA) dataset in order to examine probabilistic and sequential modeling behavior. The research specifically focuses on four major characteristics: “price,” “safety,” “transit-location,” and “general,” which were identified as the most common aspects within the dataset. The methodology involved dividing the dataset, containing both single and multiple place mentions, into train, development (dev), and test sets. Specifically, 70% of the data was allocated for training, 10% for development, and 20% for testing. The evaluated models included logistic regression, gradient boosting, Bayesian network, long short-term memory, and GRU. Among all models, the Bayesian network achieved the highest accuracy of 88%, demonstrating strong potential for urban sentiment analysis and informed decision-making in city planning and tourism