Ellen Souza

Modelos baseados em BERT vêm sendo largamente utilizados, tornando-se o estado da arte para muitas tarefas de Processamento de Linguagem Natural e também para Recuperação de Informação. A arquitetura Sentence-BERT permitiu que esses modelos fossem facilmente utilizados para a busca semântica de documentos, já que ela gera embeddings contextuais que podem ser comparados através de medidas de similaridade. Para melhor investigar a aplicação de modelos baseados em BERT para Recuperação de Informação, este trabalho avaliou 12 modelos Sentence-BERT, disponíveis publicamente, para a recuperação de documentos no cenário legislativo brasileiro. Duas variantes do algoritmo BM25 foram utilizadas como baseline: Okapi BM25 e BM25L. O BM25L alcançou melhores resultados, com significância estatística, mesmo no cenário em que os documentos não foram pré-processados, enquanto que apenas um dos modelos de linguagem, ajustado usando dados legislativos brasileiros, obteve um desempenho similar para uma das três bases de dados utilizadas

Legislative documents present substantial obstacles to summarization due to their complex argument structures and specialized terminology. This research investigates the application of Large Language Models (LLMs) in summarizing Brazilian legislative proposals from the Chamber of Deputies, examining a dataset of over 56 thousand texts from 2013 to 2023. The paper explores three main summarization methodologies: extractive, abstractive, and hybrid, with an emphasis on abstractive summarization using LLMs. The performance of the LLM LLAMA2-13b is assessed using metrics such as ROUGE, BLEU, METEOR, BERTScore, and BERTopic, compared against reference summaries. The results show that LLMs can generate coherent and informative summaries, with positive evaluation metric results. Notably, the study reveals that traditional summary evaluation metrics may not be adequate for evaluating LLMs in summarization tasks. On the other hand, metrics based on pre-trained models like BERT provide a more effective evaluation of this innovative automatic summarization approach.

This work investigates the application of Active Learning methodologies for data annotation in Named Entity Recognition (NER) tasks mainly when used in documents from the legal domain in Portuguese. Its aim is to determine an algorithm able to improve the efficiency of the annotation process and reduce the human cost involved, without compromising the quality of the classifiers trained in these corpora. Three sample selection methods were explored: (i) Multi-Criteria Active Learning, using informativeness, representativeness, and diversity as selection criteria, (ii) Dynamic Selection Guided by Entity Volume, and (iii) Random Sentence Selection (used as a baseline for evaluating the other two). The study was conducted using the BERT model for classification, employing different amounts of labeled data for each approach (annotation budgets). The results show that, although Multi-Criteria Active Learning performed better in some scenarios, Dynamic Selection Guided by Entity Volume consistently showed good performance, especially for low annotation budgets, in addition to being computationally more efficient. Thus, the analysis of the results suggests that the volume of named entities is a good predictor for selecting informative samples. This study contributes to the Active Learning field by applying these techniques to modern language models and providing efficient solutions for reducing costs in data annotation for Named Entity Recognition.

Legislative public spaces are important structures for participatory democracy, allowing citizens’ voices to get engaged with politic decisions. As a consequence of the popularization of information and communication technologies, internet based tools have played an important role to improve public participation in political decisions, known as e-Democracy. These tools are usually composed of a set of functionalities or small services, named microservices. The better the microservices, the higher the citizen participation. This work investigates how to extract useful knowledge from citizen participation in the microservices of the public portal of the Brazilian Chamber of Deputies. For such, it analyzes public comments incorporating Natural Language Processing and Artificial Intelligence techniques in a platform named Ulysses. The tasks developed on this paper focus on a temporal analysis of comments on bills in the portal through Stance Detection and dynamic Topic Modeling tasks. For the first task, OxêSD, a BERTimbau-based model, was trained on two different corpora, one of them translated into Portuguese, and its predictive performance was evaluated using the F1 and ROC-AUC metrics, achieving 73% for both on our proposed Political-BRSD a mixed dataset containing both translated content from a bigger multilingual dataset (adapted from x-Stance) and bill-specific content (adapted from Ulysses-SD); for the second, BERTopic, a Topic Modeling framework, was used. Visualization tools to analyze how the proposed approach addressed the task were also used to explore the knowledge extracted. They allow the user to understand over time how the comments relate to each other and how the comments relate to a given legislative bill.

Social-digital inclusion is a pertinent phenomenon that seeks, among other things, to ensure access and technological appropriation to individuals who are in a situation of exclusion. While research institutes have been generating valuable insights into social-digital inclusion among the general population in recent years, there remains a dearth of specific data concerning individuals with disabilities. To address this gap, this study developed a research instrument and collected data from individuals with intellectual and multiple disabilities, who represent a social group with significant e-exclusion. The research was conducted at the Association of Parents and Friends of the Exceptional in Serra Talhada. Two questionnaires were administered: one focused on household-related data and the other on specific information regarding individuals with disabilities. Fifty families participated in the first questionnaire, with twenty individuals assisted by the association completing the second. While many results align with those found in the general population, it became evident that there is a need to consider the manner in which technology is utilized—advocating for conscious and responsible use. Finally, the findings of this research can serve as a foundation for the development of institutional and public policies, as well as for further research with this population.

Apesar de existirem documentos legais a exemplo de Leis, Decretos e Declarações que tratam da educação como um direito de todo indivíduo, ainda é possível perceber que o acesso e permanência das pessoas com deficiência ou transtornos em instituições de ensino profissionalizante podem ser dificultados por diversos fatores, entre eles a falta de acessibilidade. O presente estudo objetivou investigar a inclusão de pessoas com deficiência ou transtorno nas instituições de ensino superior e técnico, públicas e privadas, em modalidade presencial de Serra Talhada – PE. Adotamos o método de pesquisa survey, caracterizado como um levantamento de dados, utilizando análise quali-quantitativa para podermos descrever o atual cenário. Os resultados indicam que as pessoas com deficiência ou transtornos estão conseguindo chegar ao ensino profissionalizante, mas o quantitativo de pessoas típicas ainda é muito superior. A pesquisa também oportunizou perceber que há um número de matrículas proporcionalmente maior nas instituições públicas; o quantitativo referente à condição de pessoa com deficiência sobressai o de pessoas com transtornos; há um equilíbrio entre homens e mulheres, a maioria desses vieram de escolas públicas e atualmente estão matriculados no período noturno, em cursos da área de humanas. Além disso, evidenciamos que as principais dificuldades enfrentadas no processo de inclusão em Serra Talhada são: a falta de profissional capacitado e a escassez de recursos de acessibilidade para atender as necessidades específicas dos alunos já matriculados, mas especialmente dos que vierem a ingressar.