Ensuring the privacy of users whose data are used to train Natural Language Processing (NLP) models is necessary to build and maintain customer trust. Differential Privacy (DP) has emerged as the most successful method to protect the privacy of individuals. However, applying DP to the NLP domain comes with unique challenges. The most successful previous methods use a generalization of DP for metric spaces, and apply the privatization by adding noise to inputs in the metric space of word embeddings. However, these methods assume that one specific distance measure is being used, ignore the density of the space around the input, and assume the embeddings used have been trained on public data. In this work we propose Truncated Exponential Mechanism (TEM), a general method that allows the privatization of words using any distance metric, on embeddings that can be trained on sensitive data. Our method makes use of the exponential mechanism to turn the privatization step into a selection problem. This allows the noise applied to be calibrated to the density of the embedding space around the input, and makes domain adaptation possible for the embeddings. In our experiments, we demonstrate that our method outperforms the state-of-the-art in terms of utility for the same level of privacy, while providing more flexibility in the metric selection.

We study the problem of releasing the set union of users' items subject to differential privacy. Previous approaches consider only the set of items for each user as the input. We propose incorporating the item frequency, which is typically available in set union problems, to boost the utility of private mechanisms. However, using the global item frequency over all users would largely increase privacy loss. We propose to use the local item frequency of each user to approximate the global item frequency without incurring additional privacy loss. Local item frequency allows us to design greedy set union mechanisms that are differentially private, which is impossible for previous greedy proposals. Moreover, while all previous works have to use uniform sampling to limit the number of items each user would contribute to, our construction eliminates the sampling step completely and allows our mechanisms to consider all of the users' items. Finally, we propose to transfer the knowledge of the global item frequency from a public dataset into our mechanism, which further boosts utility even when the public and private datasets are from different domains. We evaluate the proposed methods on multiple real-life datasets.

Learning from continuous data streams via classification/regression is prevalent in many domains. Adapting to evolving data characteristics (concept drift) while protecting data owners' private information is an open challenge. We present a differentially private ensemble solution to this problem with two distinguishing features: it allows anunbounded number of ensemble updates to deal with the potentially never-ending data streams under a fixed privacy budget, and it ismodel agnostic, in that it treats any pre-trained differentially private classification/regression model as a black-box. Our method outperforms competitors on real-world and simulated datasets for varying settings of privacy, concept drift, and data distribution.

We propose a new method that satisfies approximate differential privacy for top-$k$ selection with unordered output in the unknown data domain setting, not relying on the full knowledge of the domain universe. Our algorithm only requires looking at the top-$\bar{k}$ elements for any given $\bar{k} \geq k$, thus, enforcing the principle of minimal privilege. Unlike previous methods, our privacy parameter $\varepsilon$ does not scale with $k$, giving improved applicability for scenarios of very large $k$. Moreover, our novel construction, which combines the sparse vector technique and stability efficiently, can be applied as a general framework to any type of query, thus being of independent interest. We extensively compare our algorithm to previous work of top-$k$ selection on the unknown domain, and show, both analytically and on experiments, settings where we outperform the current state-of-the-art.

This study proposes a predictive model to detect the delay in bank teller queues. Since there are penalties and fines applied to the branches that leave their clients waiting for a long time, detecting these cases as early as possible is essential. Four models were tested: one using a Queuing Theory's formula and the other three using Data Mining algorithms - Deep Learning (DL), Gradient Boost Machine (GBM), and Random Forest (RF). The results indicated the GBM model as the most efficient, with an accuracy of 97% and a F1-measure of 75%.

Each Brazilian Deputy receives a quota of money quota to cover the politician activity expenses, besides their salary. The amount of money reserved for that quota can sum up to almost 1 billion of Brazilian currency (approximately 300 million US Dollars) in a 4 year legislature. Civic society is using that data to perform independent auditing to verify expenses that are against the rules. This article presents the application of deep Autoencoders to identify anomalies in that data. The anomalies found indicate new suspicious expenses and several data quality problems in the data opened to the society.

This article presents a study using Brazilian publicly listed companies' financial statements footnotes, in order to verify if these documents can provide information to predict variations in the debt of the corresponding firms. From a text mining perspective, we built classification models by assigning a class to each company in a period of time based on the variation of the debt to equity ratio. We conducted experiments using two different classifiers: random forest and support vector machine. The caret package in R language was used for this work. When we employed the random forest classifier, we got accuracies 4 percent points greater than the baseline accuracy, suggesting that the textual content of financial statements has some potential use for debt prediction.

The Department of Research and Strategic Information (DIE), from the Brazilian Office of the Comptroller General (CGU), is responsible for investigating potential problems related to federal expenditures. To pursue this goal, DIE regularly has to analyze large volumes of data to search for anomalies that can reveal suspicious activities. With the growing demand from the citizens for transparency and corruption prevention, DIE is constantly looking for new methods to automate these processes. In this work, we investigate IT purchases anomalies in the Federal Government Procurement System by using a deep learning algorithm to generate a predictive model. This model will be used to prioritize actions carried out by the office in its pursuit of problems related to this kind of purchases. The data mining process followed the CRISP-DM methodology and the modeling phase tested the parallel resources of the H2O tool. We evaluated the performance of twelve deep learning with auto-encoder models, each one generated under a different set of parameters, in order to find the best input data reconstruction model. The best model achieved a mean squared error (MSE) of 0.0012775 and was used to predict the anomalies over the test file samples.

This article presents a study conducted in a Brazilian bank, in order to assist the institution account managers in the approach to customers with loans in arrears. This approach is carried out to propose alternatives to customers return to timely payments situation, but the efficiency of this approach is small, accounting for only about 6.8% of customers. A predictive model, using classification was used to help identify customers with the most potential to return to a normal situation, reaching a 85.5% accuracy rate with the winning algorithm, Gradient Boosting Method. It was implemented in the integrated Platform H2O with R language, exploring the grid mode and parallel processing models advantages.