Bayesian Approach to News Recommendation Systems
Palavras-chave:Bayesian network, Clustering, Online news, Recommender system.
ResumoThis research was responsible for the development of a method for recommending news in online newspapers. This study takes into consideration that each reader has specific needs and interests when reading online newspapers, and it is a challenge to bring personalized and individualized information, in order to meet each reader's needs. The main goal here was solving or minimizing this problem when there is a new reader, because the system has little or no information over the reader’s preferences. This descriptive research used as a subject a new reader from a news portal and all data collected from the web browsing experience was performed without that user’s knowledge. The research may be characterized as applied, since it generated knowledge enough for solving the problem of online newspaper readers. A quantitative approach was adopted, because the news recommended by the system were classified and the system’s accuracy was quantified comparing the system`s suggestions and the decisions made by the readers. The solution adopted involved accessing three different methods. The Bayesian network was adopted as the main method when generating news suggestions to the new reader and the excess of variables was clustered using the K-means algorithm. The probabilities missing on this network were captured through the EM algorithm (Expectation Maximization). This algorithm uses cases in which variables were used to learn how to predict their values when they are not being observed.
Silva, J. B., Schreiber, J. N. C., & Nara, E. O. B. (2017). Bayesian Approach to News Recommendation Systems. Ciência Da Informação, 44(3). Recuperado de http://revista.ibict.br/ciinf/article/view/1902
- A publicação se reserva o direito de efetuar, nos originais, alterações de ordem normativa, ortográfica e gramatical, com vistas a manter o padrão culto da língua, respeitando, porém, o estilo dos autores;
- As provas finais não serão enviadas aos autores;
- Os originais não serão devolvidos aos autores;
- Os autores mantém os direitos totais sobre seus trabalhos publicados na revista Ciência da Informação, ficando sua reimpressão total ou parcial, depósito ou republicação sujeita à indicação de primeira publicação na revista, por meio da licensa CC-By;
- Deve ser consignada a fonte de publicação original;
- As opiniões emitidas pelos autores dos artigos são de sua exclusiva responsabilidade;
- Cada autor receberá dois exemplares da revista, caso esteja disponível no formato impresso.