Financial news about brazilian companies listed on B3 and source-codes to perform sentiment analysis

This package contains a dataset of financial news (written in Portuguese) and the source codes (in Python) to perform sentiment analysis on these news, according to two approaches: (i) based on three lexicons (also in Portuguese), being two of then proposed by the authors and specifically developed for the financial market; and (ii) based on machine learning, particularly with Naive Bayes and Multilayer Perceptrons. The dataset (file "NewsDatabase.zip") contains 828 news, downloaded from Brazilian newspapers through a web scrapper and manually labeled as positive or negative, according to an investor's sentiment.

This dataset contains two sets of files, with and without the application of stemming. All documents were preprocessed with steps of tokenization, normalization, and removal of special characters and stop words. In the source codes (file "Source-Codes.zip"), the two proposed dictionaries can be found in the file "financial_dictionary.py".