Memorial de projetos : detector de spam com redes neurais recorrentes
Resumo
Resumo: O spam por SMS deixou de ser apenas um incômodo e passou a representar uma séria ameaça à segurança digital, envolvendo fraudes, phishing e malware. Para enfrentar esse desafio crescente, técnicas de Inteligência Artificial, especialmente do Processamento de Linguagem Natural (PLN), têm se mostrado eficazes no combate a essas ameaças, interpretando as nuances da linguagem e classificando a veracidade do conteúdo analisado. As Redes Neurais Recorrentes (RNNs) permitem analisar padrões em mensagens, mas enfrentam limitações em sequências longas devido ao desaparecimento do gradiente, que as impede de recordar informações cruciais do início de textos mais longos. Como uma alternativa, a arquitetura Long Short-Term Memory (LSTM), projetada para superar essa limitação, consegue manter informações por mais tempo, garantindo uma detecção de spam mais apurada. Dessa forma, o uso de RNNs e LSTMs é uma solução promissora para proteger os usuários, podendo ser aplicado em smartphones, sistemas de filtragem em operadoras ou plataformas de segurança corporativa Abstract: SMS spam is no longer just an annoyance; it has become a serious threat to digital security, acting as a vector for fraud, phishing, and malware. To address this growing challenge, Artificial Intelligence techniques, particularly Natural Language Processing (NLP), have proven to be highly effective in combating these threats. These methods work by interpreting the nuances of language to classify the authenticity of the content. While Recurrent Neural Networks (RNNs) allow for the analysis of patterns in messages, they are limited by the vanishing gradient problem in long sequences, which hinders their ability to remember crucial information from the start of a text. The Long Short-Term Memory (LSTM) architecture was designed as an alternative to overcome this, retaining information over longer periods to guarantee more precise spam detection. Consequently, the use of RNNs and LSTMs represents a promising solution for user protection, suitable for implementation on smartphones, in carrier filtering systems, or on corporate security platforms