Brazil

Tax evasion is a problem that affects our society, costing billions of Brazilian reais of public funds every year. Stopping this practice is a complex challenge that involves analyzing a large and diverse volume of data. In this work, we propose an approach to analyze invoices and extract information about measures and units from product descriptions using a neural network with the BiLSTM-CRF architecture. Our method can validate product quantity information to, for instance, check whether any product was bought or sold by a business without issuing an invoice. The results were evaluated according to precision, recall, and f-score. The proposed approach can correctly detect more than 90% of cases of each type of information, showing its feasibility to process invoice data.

PROPOR 2022

Automatic Recognition of Units of Measurement in Product Descriptions from Tax Invoices Using Neural Networks

bilstm-crf.

named-entity recognition

neural networks

The purpose of the AAAI conference is to promote research in artificial intelligence (AI) and scientific exchange among AI researchers, practitioners, scientists, and engineers in affiliated disciplines. 

AAAI 2022

poster

The International Conference on Computational Processing of Portuguese (PROPOR) is the main event in the area of human language processing that is focused on theoretical and technological issues of written and spoken Portuguese and Galician. The meeting has been a very rich forum for the exchange of ideas and partnerships for the research and industry communities dedicated to the automated processing of Portuguese, promoting the development of methodologies, resources and projects that can be shared among researchers and practitioners in the field.

Find more information and pay the registration fee on [event organizer’s website](https://sites.universidadedefortaleza.com/propor2022/?page_id=401)

The International Conference on Computational Processing of Portuguese (PROPOR) is the main event in the area of human language processing that is focused on theoretical and technological issues of written and spoken Portuguese and Galician.

Deep learning techniques have been shown to be efficient in various tasks, especially in the development of speech recognition systems, that is, systems that aim to transcribe an audio sentence in a sequence of written words. Despite the progress in the area, speech recognition can still be considered difficult, especially for languages lacking available data, such as Brazilian Portuguese (BP). In this sense, this work presents the development of an public Automatic Speech Recognition (ASR) system using only open available audio data, from the fine-tuning of the Wav2vec 2.0 XLSR-53 model pre-trained in many languages, over BP data. The final model presents an average word error rate of 12.4% over 7 different datasets (10.5% when applying a language model). According to our knowledge, the obtained error is the lowest among open end-to-end (E2E) ASR models for BP.

Brazilian Portuguese Speech Recognition Using Wav2vec 2.0

BERT has revolutionized the NLP field by enabling transfer learning with large language models that can capture complex textual patterns, reaching the state-of-the-art for an expressive number of NLP applications. For text classification tasks, BERT has already been extensively explored. However, aspects like how to better cope with the different embeddings provided by the BERT output layer and the usage of language-specific instead of multilingual models are not well studied in the literature, especially for the Brazilian Portuguese language. The purpose of this article is to conduct an extensive experimental study regarding different strategies for aggregating the features produced in the BERT output layer, with a focus on the sentiment analysis task. The experiments include BERT models trained with Brazilian Portuguese corpora and the multilingual version, contemplating multiple aggregation strategies and open-source datasets with predefined training, validation, and test partitions to facilitate the reproducibility of the results. BERT achieved the highest ROC-AUC values for the majority of cases as compared to TF-IDF. Nonetheless, TF-IDF represents a good trade-off between the predictive performance and computational cost.

BERT for Sentiment Analysis: Pre-trained and Fine-Tuned Alternatives

This paper proposes an interface between linguistic concepts and computational interventions to support hate speech detection.
We applied a Natural Language Processing pipeline involving topic modeling and semantic role labeling, allowing a semi-automatic identification of hate speech. We also discuss how such speech qualifies as a type of verbal violence widespread on social networks to reinforce a sexist stereotype. Finally, we use Twitter data to analyze information that resulted in virtual attacks against a specific person.
As an achievement, this work validates the use of linguistic features to annotate data either as hate speech or not. It also proposes using fallacies as a potential additional feature to identify potential intolerant discourses.

Sexist Hate Speech: Identifying Potential Online Verbal Violence Instances

Artificial Intelligence applied to Law is getting attention in the community, both from the Judiciary and the lawyers, due to possible gains in procedural celerity and the automation of repetitive tasks, among other use cases. Many of its bene-fits can be derived from Natural Language Processing since legal proceedings are textual document-based. Named Entity Recognition is the NLP task of identifying and classifying named entities in unstructured text to help achieve these goals. In recent years, transformers emerged as the new state-of-the-art architecture for some tasks in NLP systems. NLP resources in Portuguese, such as models and datasets, are needed to allow Portuguese speaking countries to benefit from this new NLP level. This paper presents the first fine-tuned BERT models trained ex-clusively on Brazilian Portuguese for Legal NER. The models achieved new state-of-the-art on LeNER-Br dataset, a Portuguese NER corpus for the legal domain. We also built a prototype application for Judiciary users to evaluate how the model performed with authentic law documents. The results showed that the models were able to extract information with good quality. Both the models and the prototype application are publicly available for the community.

Fostering Judiciary applications with new fine-tuned models for Legal Named Entity Recognition in Portuguese

Most approaches to role-filler entity extraction (REE) rely on large labeled training corpora in which entity mentions are directly annotated in the input document. In this work, we leverage an existing knowledge base (KB) of entities to perform document-level REE from drug seizure petitions. We propose a system that learns to extract entities from petitions to fill 29 roles of a drug seizure event. Although we have access to a KB covering more than 170 thousand entities and six thousand petitions, such that each entity in the KB is linked to a specific petition, the mentions to an entity within a petition's text are not annotated. The lack of these annotations brings challenges related to mismatches between entity values in the KB and entity mentions in the documents. Additionally, there are entities with same type or same value. Thus, we propose a distant annotation method to overcome these challenges and automatically label petition documents using the available KB. This annotation method includes a parameter that controls the balance between precision and recall. We also propose a strategy to effectively tune this parameter in order to optimize a given metric. We then train a BERT-based sequence labeling model that learns to identify entity mentions and label them. Our system achieves an F1-score of 78.59 with precision over 82%. We also report ablation studies regarding the distant annotation method.

Entity Extraction from Portuguese Legal Documents Using Distant Supervision

We present some results of applying natural language processing (NLP) techniques in the domain of History, having as object of investigation the Brazilian Historical-Biographical Dictionary (Dicionário Histórico-Biográfico Brasileiro, DHBB). After improving or adding annotation of specific fields, information extraction techniques based on manually derived patterns were applied to three relevant problems: the age of entrance in Brazilian politics, the academic background of Brazilian politicians, and family ties among the political elites.

Automatic information extraction: a distant reading of the Brazilian Historical-Biographical Dictionary

Products, services, among many other things in life have a quality standard, are inclusive, or do not harm customers. Regulations required from their manufacturers or providers make it possible. This type of requirement also exists in the finance sector. Governments, international agencies, or civil institutions are responsible for creating, applying, and inspecting these regulations. Regulators from all spheres (federal, state, and municipal) constantly demand changes in the finance sector to meet current needs adequately. This paper presents the constant evolution of a banking compliance application in Brazil. It aims to classify the relevance or irrelevance of regulatory documents published by more than 100 Brazilian regulators, affecting the businesses of more than 40 departments of Banco do Brasil. The application uses a hybrid strategy, combining machine learning and rules for a binary classification challenge involving each company department. This work also presents a particular type of corpus imbalance called The Imbalance Within Class.

Banking Regulation Classification in Portuguese

Deep learning transformers have drastically improved systems that automatically answer questions in natural language. However, different questions demand different answering techniques; here we propose, build and validate an architecture that integrates different modules to answer two distinct kinds of queries. Our architecture takes a free-form natural language text and classifies it to send it either to a Neural Question Answering Reasoner or a Natural Language parser to SQL. We implemented a complete system for the Portuguese language, using some of the main tools available for the language and translating training and testing datasets. Experiments show that our system selects the appropriate answering method with high accuracy (over 99%), thus validating a modular question answering strategy.


Integrating Question Answering and Text-to-SQL in Portuguese

Named entity recognition (NER) tasks imply token-level labels. Annotating documents can be time-consuming, costly, and prone
to human error. In many real-life scenarios, the lack of labeled data has
become the biggest bottleneck preventing NER being effectively used in
some domains and with some natural languages, with negative impacts
on the quality of some tasks. To overcome the barrier of the lack of annotated data for new application domains in some natural languages,
we propose a method that uses the output of an ensemble of NER’s to
automatically annotate the data needed to train a Bidirectional Encoder
Representations from Transformers (BERT) based NER for Portuguese.
The performance was assessed using MiniHAREM dataset with promising results. For domain relevant classes such as LOCAL, F1, Precision
and Recall above 50% were obtained when training only with automatically annotated data.

Named Entity Extractors for New Domains by Transfer Learning with Automatically Annotated Data

Open Information Extraction (OpenIE) is the task of extracting structured information from text. Recent advances in applying Deep Learning to OpenIE tasks have improved the state of the art for the task, although few works have been produced for languages other than English. In this work, we propose PortNOIE, a neural framework for open information extraction for the Portuguese language. We evaluate our method on a manually annotated corpus of Open IE extractions, obtaining better performance than the current state of the art for OpenIE for Portuguese, both based on rule-based approaches or neural methods.

Premium content

Downloads

Next from PROPOR 2022

Brazilian Portuguese Speech Recognition Using Wav2vec 2.0

Similar lecture

AI Driven Accounts Payable Transformation