THE ROLE OF NLP IN CHATBOT SYSTEMS FOR REAL ESTATE BROKERAGE SUPPORT

Abstract

This paper presents the results of a comprehensive study on the effectiveness of natural language processing (NLP) methods in chatbot systems designed to automate the search for real estate properties. The relevance of this research arises from the rapid growth of unstructured textual data generated by users through natural language queries and contained within property listings on various online platforms. Traditional keyword-based search mechanisms are unable to provide sufficient accuracy, as they neglect semantic relationships, contextual dependencies, and the linguistic variability inherent in natural language.

In this study, a prototype chatbot with an integrated NLP module was developed to analyze user queries, extract essential property attributes (such as type, price, location, number of rooms, and area), and perform semantic search based on the extracted criteria. For experimental evaluation, a specialized dataset was created, consisting of 520 real estate listings and 75 natural-language queries of varying complexity. Three approaches to textual representation and search were compared: the statistical TF-IDF method, the neural Word2Vec model, and the transformer-based BERT model.

The results demonstrate that BERT significantly outperforms alternative approaches in terms of semantic search accuracy. The achieved Precision@5 score reached 0.83, which is 39% higher compared to TF-IDF and 24% higher than Word2Vec. The developed chatbot prototype was shown to provide highly relevant property selections while reducing the number of irrelevant results and improving overall query processing efficiency.

The findings confirm the feasibility and effectiveness of integrating modern NLP models into brokerage-support systems and highlight promising directions for future development, including the enhancement of intelligent chatbots and the expansion of semantic search technologies within the real estate domain.