Estudo de caso de um sistema de gerenciamento de banco de dados vetorial para manipulação de dados de redes sociais online

Abstract

The growing volume of unstructured data, driven by social media platforms, poses chal lenges for traditional relational Database Management Systems (DBMS), which are lim ited to lexical searches. Vector DBMSs emerge as a solution, yet they introduce archi tectural complexity. An alternative is the integration of vector capabilities into existing DBMSs. This work aims to evaluate the viability of the PostgreSQL extension PGVec tor as an integrated solution for the analysis of unstructured data. To this end, a case study was developed using a dataset of tweets from the X platform (formerly Twitter). The workflow covers text preprocessing and the generation of vector representations (em beddings) using the “paraphrase-multilingual-MiniLM-L12-v2” model from the sentence transformers library. The generated embeddings were stored and indexed in PostgreSQL using PGVector. Subsequently, the K-means clustering algorithm was applied for unsu pervised topic discovery, with the optimal number of clusters (K) determined by the Elbow Method. The results demonstrate the applicability of the integrated approach, enabling the execution of a complete Machine Learning pipeline within the relational ecosystem. This work contributes by presenting a practical and validated roadmap, serving as a resource for future applications in the field.