Busca em catálogo de produtos: uma comparação entre banco de dados relacional e motor de busca.

Abstract

The objective of TCE-AC is to oversee the expenses and revenues of municipalities and the state of Acre. In recent years, it has modernized its working methods. In particular, fast access to the prices being practiced is crucial for both the oversight process and the general population. To achieve this, the Price Database is employed, being continuously updated and currently containing tens of millions of invoice records. Given the scenario, as the system relies on relational databases for conducting queries, it often experiences delays in producing results in various situations and occasionally yields less relevant outcomes. To address this issue, the proposal is to implement Elasticsearch as the search engine for the system. Elasticsearch employs indexing techniques and features tools that optimize query execution and results. Additionally, strategies for continuous data loading will be implemented, along with documenting the challenges encountered during the implementation. To evaluate the proposed solution, statistics related to response times and query quality were measured before and after the implementation of Elasticsearch. Result quality was assessed using techniques such as NDCG (Normalized Discounted Cumulative Gain) and F1-score, based on the determination of relevant and non-relevant documents for each query. As a result, it was observed that Elasticsearch reduced query response times by a factor of 10 when compared to results involving SQL Server. Furthermore, there was an improvement in result relevance of approximately 2%, leading to an average NDCG of 95.3% for queries with 10 results, which are the default in the system.