Categorização e avaliação de algoritmos de sistemas de recomendação na base de dados IFG-Produz.

Abstract

With the growth of scientific academic productions over the years, finding relevant publications has become a difficult task. Considering scientific article recommendation systems, this difficulty is mitigated by recommending specific research items. Following the claims in the literature, algorithms for recommending scientific articles tend to have more effective results than researchers searching search engines. Due to the recommendation algorithms can use different data, such as the researcher’s background, area of expertise, and published articles, thus specifying the possible contexts in association with other articles. Thus, the work presented in this document provides a new public and annotated database for the recommendation system of scientific articles, with research in several areas. Besides the new database, we present a characterization of it, comparing it with other databases available in the literature. Among the analyses, we included the comparison of several recommendation system algorithms. According to the experimental results, the available database has better effectiveness when applying content-based filtering strategies, mainly due to the good textual description of articles and projects. The experiments show that databases with less sparsity tend to have better results when applying collaborative filtering strategies. In general, this work contributes to the area of recommendation of scientific articles by providing a new database for future works, while presenting a comparison of techniques in the state of the art of recommendation systems.