Search

Purpose: The work presented in this article focusses on improving the interpretability of probabilistic topic models created from a large collection of scientific documents that evolve over time. Methods: Several time-dependent approaches based on topic models were compared to analyse the annual evolution of latent concepts in the CORD-19 corpus: Dynamic Topic Model, Dynamic Embedded Topic Model, and BERTopic. Then COVID-19 period (December 2019 - present) has been analysed in greater depth, month by month, to explore the evolution of what is written about the disease. Results: The evaluations suggest that the Dynamic Topic Model is the best choice to analyse the CORD-19 corpus. A novel topic labelling strategy is proposed for dynamic topic models to analyse the evolution of latent concepts. It incorporates content changes in both the annual evolution of the corpus and the monthly evolution of the COVID-19 disease. The generated labels are manually validated using two approaches: through the most relevant documents on the topic, and through the documents that share the most semantically similar label topics. Conclusions: The labelling enables the interpretation of topics. The novel method for dynamic topic labelling fits the content of each topic and supports the semantics of the topics.

Final master project

Searching for similar documents and exploring major themes covered across groups of documents are common activities when browsing collections of scientific papers. This manual knowledge-intensive task can become less tedious and even lead to unexpected relevant findings if unsupervised algorithms are applied to help researchers. Most text mining algorithms represent documents in a common feature space that abstract them away from the specific sequence of words used in them. Probabilistic Topic Models reduce that feature space by annotating documents with thematic information. Over this low-dimensional latent space some locality-sensitive hashing algorithms have been proposed to perform document similarity search. However, thematic information gets hidden behind hash codes, preventing thematic exploration and limiting the explanatory capability of topics to justify content-based similarities. This paper presents a novel hashing algorithm based on approximate nearest-neighbor techniques that uses hierarchical sets of topics as hash codes. It not only performs efficient similarity searches, but also allows extending those queries with thematic restrictions explaining the similarity score from the most relevant topics. Extensive evaluations on both scientific and industrial text datasets validate the proposed algorithm in terms of accuracy and efficiency.

Public procurement is a large market affecting almost every organisation and individual; therefore, governments need to ensure its efficiency, transparency, and accountability, while creating healthy, competitive, and vibrant economies. In this context, open data initiatives and integration of data from multiple sources across national borders could transform the procurement market by such as lowering the barriers of entry for smaller suppliers and encouraging healthier competition, in particular by enabling cross-border bids. Increasingly more open data is published in the public sector; however, these are created and maintained in siloes and are not straightforward to reuse or maintain because of technical heterogeneity, lack of quality, insufficient metadata, or missing links to related domains. To this end, we developed an open linked data platform, called TheyBuyForYou, consisting of a set of modular APIs and ontologies to publish, curate, integrate, analyse, and visualise an EU-wide, cross-border, and cross-lingual procurement knowledge graph. We developed advanced tools and services on top of the knowledge graph for anomaly detection, cross-lingual document search, and data storytelling. This article describes the TheyBuyForYou platform and knowledge graph, reports their adoption by different stakeholders and challenges and experiences we went through while creating them, and demonstrates the usefulness of Semantic Web and Linked Data technologies for enhancing public procurement. The work presented in this article was funded by the EC H2020 project TheyBuyForYou (grant 780247).

The ICT infrastructures of medium and large organisations that offer ICT services (infrastructure, platforms, software, applications, etc.) are becoming increasingly complex. Nowadays, these environments combine all sorts of hardware (e.g., CPUs, GPUs, storage elements, network equipment) and software (e.g., virtual machines, servers, microservices, services, products, AI models). Tracking, understanding and acting upon all the data produced in the context of such environments is hence challenging. Configuration management databases have been so far widely used to store and provide access to relevant information and views on these components and on their relationships. However, different databases are organised according to different schemas. Despite existing efforts in standardising the main entities relevant for configuration management, there is not yet a core set of ontologies that describes these environments homogeneously, and which can be easily extended when new types of items appear. This paper presents an ontology network created with the purpose of serving as an initial step towards an homogeneous representation of this domain, and which has been already used to produce a knowledge graph for a large ICT company.

Searching for similar documents and exploring the major themes are common activities when browsing document collections. With the ongoing growth in the number of digital documents in multiple languages, we need better tools to browse large multilingual corpora. Manual document annotation has been traditionally used to facilitate such document browsing. However, manual annotation is knowledge-intensive and tedious task and can be alleviated by using automatic document annotation algorithms. Most algorithms represent documents in a common feature space that abstracts them away from the specific sequence of words used in them. Probabilistic Topic Models reduce that feature space by annotating documents with thematic information. Over this low-dimensional latent space some algorithms have been proposed to perform document similarity search, including collections of texts in multiple languages. However, dictionaries or comparable corpora are required to create multilingual topics and thematic information is usually hidden behind specific representations that limits the explanatory capability of topics to justify content-based similarities. In this thesis we address the challenge of automatically relating documents from large multilingual corpora based on the knowledge offered by the topics covered in the collection, and without the need for theme-aligned data. In order to do so, we have created a framework where probabilistic topic models can be created and reused, a hierarchical model for describing documents with thematic annotations and an unsupervised algorithm that relates multilingual documents from their most relevant themes. Evaluations on classifying and sorting documents by similar content reveal good results on multiple domains. ----------RESUMEN---------- La búsqueda de documentos similares y la exploración de los principales temas tratados son actividades comunes cuando se examinan colecciones de documentos. Con el continuo crecimiento del número de documentos digitales en múltiples idiomas, se necesitan mejores herramientas que permitan la navegación de corpus multilingües de gran tamaño. Tradicionalmente se ha utilizado anotaciones manuales para facilitar esa exploración. Sin embargo, es una tarea tediosa que requiere conocimiento del dominio, y puede aliviarse mediante algoritmos automáticos de anotación de documentos. La mayoría de los algoritmos representan documentos en un espacio de características comunes que los abstraen de la secuencia específica de palabras utilizadas en ellos. Los modelos probabilísticos de tópicos reducen ese espacio de características anotando los documentos con información temática. Sobre este espacio latente de reducidas dimensiones se han propuesto algoritmos que realizan búsquedas de documentos semejantes, incluso en colecciones de textos en múltiples idiomas. Sin embargo, para crear temas multilingües se necesitan datos o diccionarios que permitan alinear los temas y la información temática queda oculta tras representaciones que limitan su capacidad explicativa para justificar las relaciones basadas en el contenido. En esta tesis abordamos el desafío de relacionar automáticamente documentos multilingües a gran escala sin perder el conocimiento que ofrecen los temas para explicar las relaciones y sin necesitar corpus paralelos o comparables. Para ello, hemos creado un marco de trabajo donde se pueden crear y reutilizar modelos probabilísticos de tópicos, un modelo jerárquico para describir documentos con anotaciones temáticas y un algoritmo no supervisado que relaciona documentos multilingües a partir de sus principales temas. Las evaluaciones exhaustivas en múltiples dominios han mostrado buenos resultados en tareas de clasificación y recuperación de documentos por contenido similar.

Background Drug–drug interactions (DDIs) that involve antiretrovirals (ARVs) tend to cause harm if unrecognized, especially in the context of comorbidity and polypharmacy. Methods A linkage was established between the drug dispensing registry of Madrid and the Liverpool human immunodeficiency virus (HIV) DDI database (January 2017–June 2017). Polypharmacy was defined as the use of ≥5 non-HIV medications, and DDIs were classified by a traffic-light ranking for severity. Results A total of 22 945 people living with HIV (PLWH) and 6 613 506 individuals without HIV had received medications. ARV regimens were predominantly based on integrase inhibitors (51.96%). Polypharmacy was higher in PLWH (32.94%) than individuals without HIV (22.16%; P < .001); this difference was consistently observed across all age strata except for individuals ≥75 years. Polypharmacy was more common in women than men in both PLWH and individuals without HIV. The prevalence of contraindicated combinations involving ARVs was 3.18%. Comedications containing corticosteroids, quetiapine, or antithrombotic agents were associated with the highest risk for red-flag DDI, and the use of raltegravir- or dolutegravir-based antiretroviral therapy was associated with an adjusted odds ratio of 0.72 (95% confidence interval, .60–.88; P = .001) for red-flag DDI. Conclusions Polypharmacy was more frequent among PLWH across all age groups except those aged ≥75 years and was more common in women. The detection of contraindicated medications in PLWH suggests a likely disconnect between hospital and community prescriptions. Switching to alternative unboosted integrase regimens should be considered for patients with risk of harm from DDIs.

Public procurement is a large market affecting almost every organisation and individual. Governments need to ensure efficiency, transparency, and accountability, while creating healthy, competitive, and vibrant economies. In this context, we built a platform, consisting of a set of modular APIs and ontologies to publish, curate, integrate, analyse, and visualise an EU-wide, cross-border, and cross-lingual procurement knowledge graph. We developed end-user tools on top of the knowledge graph for anomaly detection and cross-lingual document search. This paper describes our experiences and challenges faced in creating such a platform and knowledge graph and demonstrates the usefulness of Semantic Web technologies for enhancing public procurement.

This paper reports on cross-evaluating the two software tools for automated term extraction (ATE) from English texts: NaCTeM TerMine and UPM Term Extractor. The objective was to find the most fitting software for extracting the bags of terms to be the part of our instrumental pipeline for exploring terminological saturation in text document collections in a domain of interest. The choice of these particular tools from the bunch of the other available is explained in our review of the related work in ATE. The approach to measure terminological saturation is based on the use of the THD algorithm developed in frame of our OntoElect methodology for ontology refinement. The paper presents the suite of instrumental software modules, experimental workflow, 2 synthetic and 3 real document collections, generated datasets, and set-up of our experiments. Next, the results of the cross-evaluation experiments are presented, analyzed, and discussed. Finally the paper offers some conclusions and recommendations on the use of ATE software for measuring terminological saturation in retrospective text document collections.

There are many scenarios where we may want to find pairs of textually similar documents in a large corpus (e.g. a researcher doing literature review, or an R&D project manager analyzing project proposals). To programmatically discover those connections can help experts to achieve those goals, but brute-force pairwise comparisons are not computationally adequate when the size of the document corpus is too large. Some algorithms in the literature divide the search space into regions containing potentially similar documents, which are later processed separately from the rest in order to reduce the number of pairs compared. However, this kind of unsupervised methods still incur in high temporal costs. In this paper, we present an approach that relies on the results of a topic modeling algorithm over the documents in a collection, as a means to identify smaller subsets of documents where the similarity function can then be computed. This approach has proved to obtain promising results when identifying similar documents in the domain of scientific publications. We have compared our approach against state of the art clustering techniques and with different configurations for the topic modeling algorithm. Results suggest that our approach outperforms (> 0.5) the other analyzed techniques in terms of efficiency., Accepted at the 9th International Conference on Knowledge Capture (K-CAP 2017)

We present librAIry, a novel architecture to store, process and analyze large collections of textual resources, integrating existing algorithms and tools into a common, distributed, high-performance workflow. Available text mining techniques can be incorporated as independent plug&play modules working in a collaborative manner into the framework. In the absence of a pre-defined flow, librAIry leverages on the aggregation of operations executed by different components in response to an emergent chain of events. Extensive use of Linked Data (LD) and Representational State Transfer (REST) principles are made to provide individually addressable resources from textual documents. We have described the architecture design and its implementation and tested its effectiveness in real-world scenarios such as collections of research papers, patents or ICT aids, with the objective of providing solutions for decision makers and experts in those domains. Major advantages of the framework and lessons-learned from these experiments are reported.

Manuscript of the master's thesis: Generation of a spanish annotated corpus with biomedical entities Abstract: The great advances in the technological world have led to an exponential increase in the data and information produced, especially in Biomedicine. However, the format in which this knowledge is structured is highly heterogeneous and, in the worst-case scenario, non-machine readable. Consequently, it is more difficult for biomedical professionals to access it. The field of Natural Language Processing provides models/tools that extract the information and transform it in a format that is machine-readable so the retrieval and access to the knowledge can be automated and easy. To carry out this information extraction task, these models need to be provided with data already structured to employ them as a reference, such as an annotated corpus. However, most of the biomedical annotated corpus are in English and therefore there is a lack of these kinds of NLP resources in other languages such as Spanish. In addition, there is also a large percentage of corpus that is not openly available, which further hinders its usage to create more resources in NLP. That is why this project aims to annotate a Spanish corpus with biomedical entities and with semantic relationships and publish it openly. Firstly, we performed a systematic literature review of English and Spanish corpora annotated with biomedical entities with the purpose of identifying which is the methodology and format employed by most of the researchers to perform that task. Then we designed a work environment to annotate a Spanish corpus with eleven types of biomedical entities and eighth types of semantic relations. Finally, we have carried out a preliminary annotation to a set of 18 clinical cases resulting in a Spanish corpus with 324 biomedical entities and 170 semantic relations.