Brazil

In this work, the technical feasibility of working with audio transcriptions from Youtube is analyzed, as well as presenting a method that allows data acquisition, pre-processing, and post-processing to work with this type of data. A topic modeling approach with the latent dirichlet allocation algorithm is used. An approach is also presented to dynamically determine the ideal number of topics that make up a given corpus. In the experiments, a database of 250 audio transcriptions was used, obtaining a model with coherence in the range of 40%.

BRACIS 2020

From audio to information: Learning topics from audiotranscripts

audio transcription

data mining

machine learning

topic modeling

The 9th Brazilian Conference on Intelligent Systems (BRACIS), the 16h Encontro Nacional de Inteligência Artificial e Computacional (ENIAC), and the 12th Concurso de Teses e Dissertações em Inteligência Artificial e Computacional (CTDIAC) was held from 20 to 23 October, 2020.

BRACIS originated from the combination of the two most important scientific events in Brazil in Artificial Intelligence (AI) and Computational Intelligence (CI): the Brazilian Symposium on Artificial Intelligence - SBIA (22 editions), and the Brazilian Symposium on Neural Networks - SBRN (13 editions). ENIAC is a national forum for researchers, practitioners, educators and students to present and discuss innovations, trends, experiences, developments and work in progress in the fields of Artificial and Computational Intelligence. CTDIAC is a satellite event whose goals are to disseminate and award the best dissertations and theses (MSc and PhDs concluded in Brazilian universities from June 2018 to May 2020) in Artificial and Computational Intelligence areas.

In this year, the Symposium on Knowledge Discovery, Mining and Learning (KDMiLe) will be held in conjunction with BRACIS 2020. This sysposium aims at integrating researchers, practitioners, developers, students and users to present their research results, to discuss ideas, and to exchange techniques, tools, and practical experiences – related to the Data Mining and Machine Learning areas.

technical paper

The Coronavirus disease 2019 (COVID-19) was first detected in China in December 2019. In a few months, the disease got pandemic proportions, overloading health systems all around the world. Risk factors related to the progression and outcome of the disease are still unclear. Moreover, clinical aspects of patients can differ between societies, and other demographic elements may impact survival responses. A better characterisation of local manifestation of COVID-19 is crucial to a better general understanding of the disease, and thus to improve treatment decisions and health systems’ management. In this article, we performed an initial analysis of clinical factors related to admission in ICU or death of SARS-CoV-2 confirmed Brazilian patients, based on 1,138,690 medical records from the Brazilian government. To our knowledge, this study is the first to assess clinical risk factors for disease progression in Brazil. We provide a concise data set of medical registers related to COVID-19 in the whole Brazilian territory, and we describe the baseline comorbidities and symptoms observed in the data collection. Then, we assess the correlation between the manifestation of symptoms/comorbidities and the patients’ survival response through Kaplan-Meier survival estimates. The results here reported are mostly in accordance with findings reported in previous works.

Clinical risk factors of ICU & fatal COVID-19 cases in Brazil

Fake news (FN) have affected people's lives in unimaginable ways. The automatic classification of FN is a vital tool to prevent their dissemination and support fact-checking. Related work has shown that FN spread faster, deeper, and more broadly than the truth on social media. Besides, deep learning has produced state-of-the-art solutions in this field, mainly based on textual attributes. In this paper, we propose initial experiments to combine compact representations of the textual news properties generated using DistilBERT, with topological metrics extracted from the social propagation network. Using a dataset related to politics and five distinct classification algorithms, our results are encouraging. Regarding the textual attributes, we reached results comparable to state-of-the-art solutions using only the news title and contents, which is useful for FN early detection. The topological attributes were not as effective, but the promising results encourage the investigation of alternative architectures for their combination.

Combining compact news representations generated using DistilBERT and topological features to classify fake news

Twitter has been one of the main sources of information and discussion during the COVID-19 pandemics. This paper characterizes a set of more than 56 million tweets written in Portuguese and collected over a period of 70 days. Our analysis includes the volume of messages, text of tweets, location of tweets, the main elements of tweets (e.g. hashtags and URLs) and the user profiles, including gender. The analyses showed the most discussed topics in the period were quarantine, hydroxychloroquine, agglomeration and social distance, and that the discussions were centered in political issues (e.g., most common hashtags include “fechadocombolsonaro" and “forabolsonaro").

A Characterization of Portugueses Tweets Regarding the Covid-19 Pandemic

The use of sensors in environments where they require constant monitoring has been increasing in recent years. The main goal is to guarantee the effectiveness, safety, and smooth functioning of the system. To identify the occurrence of abnormal events, we propose a methodology that aims to detect patterns that can lead to abrupt changes in the behavior of the sensor signals. To achieve this objective, we provide a strategy to characterize the time series, and we use a clustering technique to analyze the temporal evolution of the sensor system. To validate our methodology, we propose the clusters' stability index by windowing. Also, we have developed a parameterizable time series generator, which allows us to represent different operational scenarios for a sensor system where extreme anomalies may arise.

Extreme Events Characterization on Time Series

This paper describes a real application of graphical neural networks (GNNs) in the dynamic estimation of spatially distributed buoys that are of central importance in maritime navigation. We describe the techniques we used to process both data and background knowledge about the domain, indicating why GNNs are particularly well suited for this sort of task. We report our empirical results, demonstrating that GNNs profitably use the avaible relational structure.

Prediction of Environmental Conditions for Maritime Navigation using a Network of Sensors: A Practical Application of Graph Neural Networks

Sentiment analysis on social media data can be a challenging task, among other reasons, because labeled data for training is not always available. Transfer learning approaches address this problem by leveraging a labeled source domain to obtain a model for a target domain that is different but related to the source domain. However, the question that arises is how to choose proper source data for training the target classifier, which can be made considering the similarity between source and target data using distance metrics. This article investigates the relation between these distance metrics and the classifiers’ performance. For this purpose, we propose to evaluate four metrics combined with distinct dataset representations. Computational experiments, conducted in the Twitter sentiment analysis scenario, showed that the cosine similarity metric combined with bag-of-words normalized with term frequency-inverse document frequency presented the best results in terms of predictive power, outperforming even the classifiers trained with the target dataset in many cases.

Transfer learning for Twitter sentiment analysis: Choosing an effective source dataset

As demonstrações contábeis são parte essencial do processo de análise de risco de crédito de empresas. Tais documentos são apresentados em alta variedade de modelos, fator que limita a extração automática de seus dados. Para superar essa limitação, analisou-se agrupamentos desses documentos gerados com o algoritmo estado da arte RNG-HDBSCAN*. A pesquisa concluiu que os agrupamentos encontrados são aderentes ao processo de extração automática de dados e podem superar as limitações conhecidas, ou seja, têm potencial para melhorar a qualidade desse processo.

Improving automatic data extraction from financial statements with clustering analysis

Text simplification (TS) is a natural language transformation process that reduces linguistic complexity while preserving semantics and retaining its original meaning. This work aims to present a research proposal for automatic simplification of texts, precisely a split-and-rephrase approach based on an encoder-decoder neural network model. The proposed method was trained against the WikiSplit English corpus with the help of a part-of-speech tagger and obtained a BLEU score of 74.72%. We also experimented with this trained model to split-and-rephrase sentences written in Portuguese with relative success, showing the method’s potential.

Experimenting Sentence Split-and-Rephrase Using Part-of-Speech Labels

Clinical decision support systems is a research area in which Machine Learning (ML) techniques can be applied. Nevertheless, specifically in assisting pneumonia decision making, the use of ML has not been so expressive. To help matters, this work aims to contribute to the evolution of the intersection of such areas by presenting a Systematic Review of the Literature. It provides results which may help to identify, interpret and evaluate how ML techniques have been applied and some research enhancements yet to be done.

Machine Learning to Assist in Pneumonia Decision Making: A Systematic Review of the Literature

Brain-Computer Interfaces (ICM) allow the control of devices by modulating brain activity. Commonly, when based on motor imagery (IM) these systems use the energy (de)synchronization in the electroencephalogram signal (EEG), voluntarily caused by the individual, to identify and classify their motor intention. Therefore, the EEG segment used in the training of the learning algorithms plays a fundamental role in the description of the characteristics and, consequently, in the recognition of patterns in the signal. In this context, the objective of this work is to demonstrate the correlation between the temporal properties of the input EEG segment and the classification performance of a ICM-IM system. An auxiliary sliding window was used in order to obtain the variation of performance in function of the variation in the time and to support the decision making about the appropriate window. Simulations based on public EEG data point to significant variability in the location and width of the ideal window and suggest the need for individualized selection according to the cognitive patterns of each subject.

From audio to information: Learning topics from audiotranscripts

Downloads

Next from BRACIS 2020

Clinical risk factors of ICU & fatal COVID-19 cases in Brazil

Similar lecture

Machine Learning to Assist in Pneumonia Decision Making: A Systematic Review of the Literature

From audio to information: Learning topics from audiotranscripts

.css-70qvj9{display:-webkit-box;display:-webkit-flex;display:-ms-flexbox;display:flex;-webkit-align-items:center;-webkit-box-align:center;-ms-flex-align:center;align-items:center;}Downloads

Next from BRACIS 2020

Clinical risk factors of ICU & fatal COVID-19 cases in Brazil

Similar lecture

Machine Learning to Assist in Pneumonia Decision Making: A Systematic Review of the Literature

Downloads