Bio-QA-GNN: Reasoning with Language Models and Knowledge Graphs for Interpretable Biomedical Question Answering

Nowadays the idea of injecting world or domain-specific structured knowledge into pre-trained language models (PLMs) is becoming an increasingly popular approach for solving problems such as biases, hallucinations, huge architectural sizes, and explainability lack—critical for real-world natural language processing applications in sensitive fields like bioinformatics. One recent work that has garnered much attention in Neuro-symbolic AI is QA-GNN, an end-to-end model for multiple-choice open-domain question answering (MCOQA) tasks via interpretable text-graph reasoning. Unlike previous publications, QA-GNN mutually informs PLMs and graph neural networks (GNNs) on top of relevant facts retrieved from knowledge graphs (KGs). However, taking a more holistic view, existing PLM+KG contributions mainly consider commonsense benchmarks and ignore or shallowly analyze performances on biomedical datasets. This thesis start from a propose of a deep investigation of QA-GNN for biomedicine, comparing existing or brand-new PLMs, KGs, edge-aware GNNs, preprocessing techniques, and initialization strategies. By combining the insights emerged in DISI's research, we introduce Bio-QA-GNN that include a KG. Working with this part has led to an improvement in state-of-the-art of MCOQA model on biomedical/clinical text, largely outperforming the original one (+3.63\% accuracy on MedQA). Our findings also contribute to a better understanding of the explanation degree allowed by joint text-graph reasoning architectures and their effectiveness on different medical subjects and reasoning types. Codes, models, datasets, and demos to reproduce the results are freely available at: \url{https://github.com/disi-unibo-nlp/bio-qagnn}.

Sensitivity and specificity of machine learning classifiers for glaucoma diagnosis using Spectral Domain OCT and standard automated perimetry

PURPOSE: To evaluate the sensitivity and specificity of machine learning classifiers (MLCs) for glaucoma diagnosis using Spectral Domain OCT (SD-OCT) and standard automated perimetry (SAP). METHODS: Observational cross-sectional study. Sixty two glaucoma patients and 48 healthy individuals were included. All patients underwent a complete ophthalmologic examination, achromatic standard automated perimetry (SAP) and retinal nerve fiber layer (RNFL) imaging with SD-OCT (Cirrus HD-OCT; Carl Zeiss Meditec Inc., Dublin, California). Receiver operating characteristic (ROC) curves were obtained for all SD-OCT parameters and global indices of SAP. Subsequently, the following MLCs were tested using parameters from the SD-OCT and SAP: Bagging (BAG), Naive-Bayes (NB), Multilayer Perceptron (MLP), Radial Basis Function (RBF), Random Forest (RAN), Ensemble Selection (ENS), Classification Tree (CTREE), Ada Boost M1(ADA),Support Vector Machine Linear (SVML) and Support Vector Machine Gaussian (SVMG). Areas under the receiver operating characteristic curves (aROC) obtained for isolated SAP and OCT parameters were compared with MLCs using OCT+SAP data. RESULTS: Combining OCT and SAP data, MLCs' aROCs varied from 0.777(CTREE) to 0.946 (RAN).The best OCT+SAP aROC obtained with RAN (0.946) was significantly larger the best single OCT parameter (p<0.05), but was not significantly different from the aROC obtained with the best single SAP parameter (p=0.19). CONCLUSION: Machine learning classifiers trained on OCT and SAP data can successfully discriminate between healthy and glaucomatous eyes. The combination of OCT and SAP measurements improved the diagnostic accuracy compared with OCT data alone.

Data mining applications in power systems - case-studies and future trends

Presently power system operation produces huge volumes of data that is still treated in a very limited way. Knowledge discovery and machine learning can make use of these data resulting in relevant knowledge with very positive impact. In the context of competitive electricity markets these data is of even higher value making clear the trend to make data mining techniques application in power systems more relevant. This paper presents two cases based on real data, showing the importance of the use of data mining for supporting demand response and for supporting player strategic behavior.

Correcção automática de respostas de texto livre para a aplicação UNI_NET-Classroom

Ao longo dos tempos foi possível constatar que uma grande parte do tempo dos professores é gasta na componente de avaliação. Por esse facto, há já algumas décadas que a correcção automática de texto livre é alvo de investigação. Sendo a correcção de exercícios efectuada pelo computador permite que o professor dedique o seu tempo em tarefas que melhorem a aprendizagem dos alunos. Para além disso, cada vez mais as novas tecnologias permitem o uso de ferramentas com bastante utilidade no ensino, pois para além de facilitarem a exposição do conhecimento também permitem uma maior retenção da informação. Logo, associar ferramentas de gestão de sala de aula à correcção automática de respostas de texto livre é um desafio bastante interessante. O objectivo desta dissertação foi a realização de um estudo relativamente à área de avaliação assistida por computador em que este trabalho se insere. Inicialmente, foram analisados alguns correctores ortográficos para seleccionar aquele que seria integrado no módulo proposto. De seguida, foram estudadas as técnicas mais relevantes e as ferramentas que mais se enquadram no âmbito deste trabalho. Neste contexto, a ideia foi partir da existência de uma ferramenta de gestão de sala de aula e desenvolver um módulo para a correcção de exercícios. A aplicação UNI_NET-Classroom, que foi a ferramenta para a qual o módulo foi desenvolvido, já continha um componente de gestão de exercícios que apenas efectuava a correcção para as respostas de escolha múltipla. Com este trabalho pretendeu-se acrescentar mais uma funcionalidade a esse componente, cujo intuito é dar apoio ao professor através da correcção de exercícios e sugestão da cotação a atribuir. Por último, foram realizadas várias experiências sobre o módulo desenvolvido, de forma a ser possível retirar algumas conclusões para o presente trabalho. A conclusão mais importante foi que as ferramentas de correcção automática são uma mais-valia para os professores e escolas.

Extraction automatique d´associations lexicales à partir de corpora

The automatic acquisition of lexical associations from corpora is a crucial issue for Natural Language Processing. A lexical association is a recurrent combination of words that co-occur together more often than expected by chance in a given domain. In fact, lexical associations define linguistic phenomena such as idiomes, collocations or compound words. Due to the fact that the sense of a lexical association is not compositionnal, their identification is fundamental for the realization of analysis and synthesis that take into account all the subtilities of the language. In this report, we introduce a new statistically-based architecture that extracts from naturally occurring texts contiguous and non contiguous. For that purpose, three new concepts have been defined : the positional N-gram models, the Mutual Expectation and the GenLocalMaxs algorithm. Thus, the initial text is fisrtly transformed in a set of positionnal N-grams i.e ordered vectors of simple lexical units. Then, an association measure, the Mutual Expectation, evaluates the degree of cohesion of each positional N-grams based on the identification of local maximum values of Mutual Expectation. Great efforts have also been carried out to evaluate our metodology. For that purpose, we have proposed the normalisation of five well-known association measures and shown that both the Mutual Expectation and the GenLocalMaxs algorithm evidence significant improvements comparing to existent metodologies.

Machine learning Vasicek model calibration with Gaussian processes

In this article, we calibrate the Vasicek interest rate model under the risk neutral measure by learning the model parameters using Gaussian processes for machine learning regression. The calibration is done by maximizing the likelihood of zero coupon bond log prices, using mean and covariance functions computed analytically, as well as likelihood derivatives with respect to the parameters. The maximization method used is the conjugate gradients. The only prices needed for calibration are zero coupon bond prices and the parameters are directly obtained in the arbitrage free risk neutral measure.

Terminology and knowledge engineering conference. New frontiers in the constructive symbiosis of terminology and knowledge engineering

To meet the increasing demands of the complex inter-organizational processes and the demand for continuous innovation and internationalization, it is evident that new forms of organisation are being adopted, fostering more intensive collaboration processes and sharing of resources, in what can be called collaborative networks (Camarinha-Matos, 2006:03). Information and knowledge are crucial resources in collaborative networks, being their management fundamental processes to optimize. Knowledge organisation and collaboration systems are thus important instruments for the success of collaborative networks of organisations having been researched in the last decade in the areas of computer science, information science, management sciences, terminology and linguistics. Nevertheless, research in this area didn’t give much attention to multilingual contexts of collaboration, which pose specific and challenging problems. It is then clear that access to and representation of knowledge will happen more and more on a multilingual setting which implies the overcoming of difficulties inherent to the presence of multiple languages, through the use of processes like localization of ontologies. Although localization, like other processes that involve multilingualism, is a rather well-developed practice and its methodologies and tools fruitfully employed by the language industry in the development and adaptation of multilingual content, it has not yet been sufficiently explored as an element of support to the development of knowledge representations - in particular ontologies - expressed in more than one language. Multilingual knowledge representation is then an open research area calling for cross-contributions from knowledge engineering, terminology, ontology engineering, cognitive sciences, computational linguistics, natural language processing, and management sciences. This workshop joined researchers interested in multilingual knowledge representation, in a multidisciplinary environment to debate the possibilities of cross-fertilization between knowledge engineering, terminology, ontology engineering, cognitive sciences, computational linguistics, natural language processing, and management sciences applied to contexts where multilingualism continuously creates new and demanding challenges to current knowledge representation methods and techniques. In this workshop six papers dealing with different approaches to multilingual knowledge representation are presented, most of them describing tools, approaches and results obtained in the development of ongoing projects. In the first case, Andrés Domínguez Burgos, Koen Kerremansa and Rita Temmerman present a software module that is part of a workbench for terminological and ontological mining, Termontospider, a wiki crawler that aims at optimally traverse Wikipedia in search of domainspecific texts for extracting terminological and ontological information. The crawler is part of a tool suite for automatically developing multilingual termontological databases, i.e. ontologicallyunderpinned multilingual terminological databases. In this paper the authors describe the basic principles behind the crawler and summarized the research setting in which the tool is currently tested. In the second paper, Fumiko Kano presents a work comparing four feature-based similarity measures derived from cognitive sciences. The purpose of the comparative analysis presented by the author is to verify the potentially most effective model that can be applied for mapping independent ontologies in a culturally influenced domain. For that, datasets based on standardized pre-defined feature dimensions and values, which are obtainable from the UNESCO Institute for Statistics (UIS) have been used for the comparative analysis of the similarity measures. The purpose of the comparison is to verify the similarity measures based on the objectively developed datasets. According to the author the results demonstrate that the Bayesian Model of Generalization provides for the most effective cognitive model for identifying the most similar corresponding concepts existing for a targeted socio-cultural community. In another presentation, Thierry Declerck, Hans-Ulrich Krieger and Dagmar Gromann present an ongoing work and propose an approach to automatic extraction of information from multilingual financial Web resources, to provide candidate terms for building ontology elements or instances of ontology concepts. The authors present a complementary approach to the direct localization/translation of ontology labels, by acquiring terminologies through the access and harvesting of multilingual Web presences of structured information providers in the field of finance, leading to both the detection of candidate terms in various multilingual sources in the financial domain that can be used not only as labels of ontology classes and properties but also for the possible generation of (multilingual) domain ontologies themselves. In the next paper, Manuel Silva, António Lucas Soares and Rute Costa claim that despite the availability of tools, resources and techniques aimed at the construction of ontological artifacts, developing a shared conceptualization of a given reality still raises questions about the principles and methods that support the initial phases of conceptualization. These questions become, according to the authors, more complex when the conceptualization occurs in a multilingual setting. To tackle these issues the authors present a collaborative platform – conceptME - where terminological and knowledge representation processes support domain experts throughout a conceptualization framework, allowing the inclusion of multilingual data as a way to promote knowledge sharing and enhance conceptualization and support a multilingual ontology specification. In another presentation Frieda Steurs and Hendrik J. Kockaert present us TermWise, a large project dealing with legal terminology and phraseology for the Belgian public services, i.e. the translation office of the ministry of justice, a project which aims at developing an advanced tool including expert knowledge in the algorithms that extract specialized language from textual data (legal documents) and whose outcome is a knowledge database including Dutch/French equivalents for legal concepts, enriched with the phraseology related to the terms under discussion. Finally, Deborah Grbac, Luca Losito, Andrea Sada and Paolo Sirito report on the preliminary results of a pilot project currently ongoing at UCSC Central Library, where they propose to adapt to subject librarians, employed in large and multilingual Academic Institutions, the model used by translators working within European Union Institutions. The authors are using User Experience (UX) Analysis in order to provide subject librarians with a visual support, by means of “ontology tables” depicting conceptual linking and connections of words with concepts presented according to their semantic and linguistic meaning. The organizers hope that the selection of papers presented here will be of interest to a broad audience, and will be a starting point for further discussion and cooperation.

NovaIntell - projecto de text-Mining para a língua portuguesa numa empresa de Gestão de Informação e Conhecimento

Dissertação apresentada como requisito parcial para obtenção do grau de Mestre em Estatística e Gestão de Informação

Recommending media content based on machine learning methods

Dissertação para obtenção do Grau de Mestre em Engenharia Informática

Automatic information retrieval through text-mining

The dissertation presented for obtaining the Master’s Degree in Electrical Engineering and Computer Science, at Universidade Nova de Lisboa, Faculdade de Ciências e Tecnologia

Machine learning Gaussian short rate

Dissertação para obtenção do Grau de Doutor em Estatística e Gestão do Risco