Short text classification applied to item description: Some methods evaluation

The increasing demand for information classification based on content in the age of social media and e-commerce has led to need automated product using their descriptions. This study aims evaluate various techniques this task, with a focus descriptions written Portuguese. A pipeline is implemented preprocess data, including lowercasing, accent removal, unigram tokenization. bag words method then used convert text into numerical five are applied: argmaxtf, argmaxtfnorm, argmaxtfidf from retrieval, two machine learning methods logistic regression support vector machines. performance each technique evaluated simple accuracy via thirty-fold cross validation. results show that achieves highest mean among techniques.