Algorithmic bias gained notoriety in over the years due to impacts caused by Artificial Intelligence (AI) systems in historically discriminated vulnerable groups [1] [2] [3]. Hate speech detection models are used to identify unwanted contents in social media platforms like Instagram, Threads and X. These platforms are often used as data source during data collection process for dataset constructions utilized in AI models training for hate speech detection. However, the acquisition of bias can occur in many steps of development cycle of AI models. Therefore, this work contributes to diminish algorithmic racism by a methodology that evaluates the presence of racial bias in databases and trained classifiers for hate speech detection in Brazilian Portuguese. The models Logistic Regression (LR), Multilayer Perceptron (MLP), BERTimbaubaseand Tucano-1b1 were trained with OFFCOMBR-2, HateBR and TUPY-E databases for hate speech detection and their respective models were used to predict hate speech over BR-RAPData, a database constructed through data collection of Brazilian RAP lyrics. Our results show that at least one model trained with OFFCOMBR-2, HateBR or TUPY-E achieved F1 score over 70%, a substantial result according to the literature [4] [5] [6]. Nonetheless, analyses of the BR-RAPData prediction for hate speech shows that, in some cases, over 50% of the content in the dataset was targeted as hate speech by the trained classifiers.

Current scientific applications have been producing large amounts of data. The processing, handling and analysis of such data require large-scale computing infrastructures such as clusters and grids. In this area, studies aim at improving the performance of data-intensive applications by optimizing data accesses. In order to achieve this goal, distributed storage systems have been considering techniques of data replication, migration, distribution, and access parallelism. However, the main drawback of those studies is that they do not take into account application behavior to perform data access optimization. This limitation motivated this paper which applies strategies to support the online prediction of application behavior in order to optimize data access operations on distributed systems, without requiring any information on past executions. In order to accomplish such a goal, this approach organizes application behaviors as time series and, then, analyzes and classifies those series according to their properties. By knowing properties, the approach selects modeling techniques to represent series and perform predictions, which are, later on, used to optimize data access operations. This new approach was implemented and evaluated using the OptorSim simulator, sponsored by the LHC-CERN project and widely employed by the scientific community. Experiments confirm this new approach reduces application execution time in about 50 percent, specially when handling large amounts of data.