Sentiment Analysis of Rental Property Reviews

Rui Jorge Rodrigues de Castilho Breda Perdigoto

Key information

Authors:

Rui Jorge Rodrigues de Castilho Breda Perdigoto (Rui Jorge Rodrigues de Castilho Breda Perdigoto)

Supervisors:

Giovani Loiola da Silva (Giovani Loiola da Silva)

Published in

December 4, 2025

Abstract

The large amount of text data found in user reviews requires automated techniques to be processed and analyzed. These techniques must leverage different approaches from the usual analysis of numerical data, due to the inherently different structure of text. This work explores the application of Natural Language Processing techniques to user reviews in rental property listings. A foundation of the fundamental concepts in this area is provided. In addition, the state of the art is presented, namely Transformers and Bidirectional Encoders. Text Mining techniques were employed to complement the exploratory data analysis of the property listings. This allowed for the use of descriptive statistics and visualizations applied not only to regular numerical and categorical data, but also text within user comments. The primary objective was to analyze the textual data comprising user comments left on rental propriety reviews, so as to obtain a numerical measurement of user sentiment towards properties. Sentiment Analysis models were used to generate numerical values from user comments on reviews in order to compute listing ratings. These were Valence Aware Dictionary for sEntiment Reasoning, a lexicon model, and a model based on Bidirectional Encoder Representations from Transformers. The results were evaluated and performance differences were tested with statistical significance tests. Models had similar performance when restricting to only English comments. The transformer model had the benefit of supporting other languages however, and the inclusion of more comments in languages other than English led to a significant improvement in performance.