One of the crucial indicators of the heart's functioning, is the so-called left ventricular ejection fraction (LVEF), which measures the heart's ability to pump blood, and corresponds to the relative change in volume within the heart's left ventricle between it's most expanded (end-diastole) and most contracted state (end-systole) during a cardiac cycle. A reduced LVEF is a key indicator of heart failure, and as such, its accurate measurement plays a prominent role in cardiology. This work proposes a machine learning approach for estimating the LVEF from short echocardiogram videos. Our model, based on gradient-boosted trees, is significantly simpler than the state of the art, but is competitive in terms of accuracy and has a higher degree of explainability. The proposed model operates on a set of geometric features of the heart's left ventricle, tracking its evolution during the cardiac cycle; some of these features are novel and are proposed here for the first time. We discuss the performance of our model on a dataset of over 10,000 samples, including the relative importance of our proposed features, and show that the model's estimation error is well within the margin of variation that occurs when the same LVEF is measured by different experts.

The purpose of citation indexes and metrics is intended to be a measure for scientific innovation and quality for researchers, journals, and institutions. However, those metrics are often prone to abuse and manipulation by excessive and unethical self-citations induced by authors, reviewers, editors, or journals. Identifying whether there are or not legitimate reasons for self-citations is normally determined during the review process, where the participating parts may have intrinsic incentives, rendering the legitimacy of self-citations, after publication, questionable. In this paper, we conduct a large-scale analysis of journal self-citations while taking into consideration the similarity between a publication and its references. Specifically, we look into PubMed Central articles published since 1990 and compute similarities of article-reference pairs using sentence embeddings. We examine journal self-citations with an aim to distinguish between justifiable and unethical self-citations.

In the medical domain, a Systematic Literature Review (SLR) attempts to collect all empirical evidence, that fit pre-specified eligibility criteria, in order to answer a specific research question. The process of preparing an SLR consists of multiple tasks that are labor-intensive and time-consuming, involving large monetary costs. Technology-assisted review (TAR) methods automate the different processes of creating an SLR and they are particularly focused on reducing the burden of screening for reviewers. We present a novel method for TAR that implements a full pipeline from the research protocol to the screening of the relevant papers. Our pipeline overcomes the need of a Boolean query constructed by specialists and consists of three different components: the primary retrieval engine, the inter-review ranker and the intra-review ranker, combining learning-to-rank techniques with a relevance feedback method. In addition, we contribute an updated version of the Task 2 of the CLEF 2019 eHealth Lab dataset, which we make publicly available. Empirical results on this dataset show that our approach can achieve state-of-the-art results.

Web crawlers account for more than a third of the total web traffic and they are threatening the security, privacy and veracity of web applications and their users. Businesses in finance, ticketing, and publishing, as well as websites with rich and unique content are the ones mostly affected by their actions. To deal with this problem, we present a novel web robot detection approach that takes advantage of the content of a website based on the assumption that human web users are interested in specific topics, while web robots crawl the web randomly. Our approach extends the typical user session representation of log-based features with a novel set of features that capture the semantics of the content of the requested resources. In addition, we contribute a new real-world dataset, which we make publicly available, towards alleviating the scarcity of open data in this field. Empirical results on this dataset validate our assumption and show that our approach outranks state-of-the-art methods for web robot detection.

The figures found in biomedical literature are a vital part of biomedical research, education, and clinical decision. The multitude of their modalities and the lack of corresponding metadata constitute search and information, retrieval a difficult task. In this paper, we introduce novel multi-label modality classification approaches for biomedical figures without segmenting the compound figures. In particular, we investigate using both simple and compound figures for training a multi-label model to be used for annotating either all figures or only those predicted as compound by a compound figure detection model. Using data from the medical task of ImageCLEF 2016, we train our approaches with visual features and compare them with the approach involving compound figure separation into sub-figures. Furthermore, we study how multimodal learning, from both visual and textual features affects the tasks of classifying biomedical figures by modality and detecting compound figures. Finally, we present a web application for medical figure retrieval, which is based on one of our classification approaches and allows users to search for figures of PubMed Central from any device and provide feedback about the modality of a figure classified by the system.

The constantly expanding medical libraries contain immense amounts of information, including evidence from healthcare research. Gathering and interpreting this evidence can be both challenging and time-consuming for researchers conducting systematic reviews. Technologically assisted review (TAR) aims to assist this process by finding as much relevant information as possible with the least effort. Toward this, we present an incremental learning method that ranks documents, previously retrieved, by automating the process of title and abstract screening. Our approach combines a learning-to-rank model trained across multiple reviews with a model focused on the given review, incrementally trained based on relevance feedback. The classifiers use as features several similarity metrics between the documents and the research topic, such as Levenshtein distance, cosine similarity and BM25, and vectors derived from word embedding methods such as Word2Vec and Doc2Vec. We test our approach using the dataset provided by the Task II of CLEF eHealth 2017 and we empirically compare it with other approaches participated in the task.

Certain business environments, like health-care or customer service, host complex and highly variable business processes. In such situations, we expect fluctuating process behavior, which is difficult to attribute to specific causes, at least automatically. This work aims to provide process analysts with an additional tool to discover factors that affect the process flow. To this end, we propose a three-stage methodology to deal with the several challenges of this goal. Adhering to the process mining paradigm that suggests for evidence-based process analysis and improvement, we introduce a horizontal partitioning approach to identify elements of process behavior during the first stage. Then, during the second stage, we discuss how log manipulations can yield characteristics that reflect various perspectives of the process. Finally, we propose a multi-target feature evaluation step to deliver insights about the associations between characteristics and process behavior. The proposed methodology is designed to tackle challenges related to the general correlation problem of process mining, like dealing with general process behavior (not just local decisions) and relaxing the independence assumption among the elements of behavior. We demonstrate our approach step by step through a case study on a real-world, open dataset.

Web robots constitute nowadays more than half of the total web traffic. Malicious robots threaten the security, privacy and performance of the web, while non-malicious ones are involved in analytics skewing. The latter constitutes an important problem for large websites with unique content, as it can lead to false impressions about the popularity and impact of a piece of information. To deal with this problem, we present a novel web robot detection approach for content-rich websites, based on the assumption that human web users are interested in specific topics, while web robots crawl the web randomly. Our approach extends the typical representation of user sessions with a novel set of features that capture the semantics of the content of the requested resources. Empirical results on real-world data from the web portal of an academic publisher, show that the proposed semantic features lead to improved web robot detection accuracy.

Systematic reviews are literature reviewing processes that aim to retrieve all relevant content based on a specific topic, in an exhaustive manner. Such reviews are particularly useful in healthcare, where decision making must take into account all possible evidence, and are usually done by constructing a boolean query and submitting it to a database, and then screening the retrieved documents for relevant ones. Task 2 of CLEF 2018 eHealth lab focuses on automating this process on two fronts: Sub-Task 1 is about bypassing the construction of the boolean query, retrieving relevant documents and ranking them by relevance based on a protocol that describes a topic, and Sub-Task 2 is about ranking the documents retrieved by an already constructed query by Cochrane experts. We present our approaches for both sub-tasks, which combine a learning-to-rank model trained on multiple reviews with a model incrementally trained on each individual review using relevance feedback.

We describe the approach we employed for Task II of CLEF eHealth 2017, concerning title and abstract screening in diagnostic test accuracy reviews. Our approach combines a learning-to-rank model trained across multiple reviews with a model focused on the given review, incrementally trained based on relevance feedback. Our learning-to-rank model is built using extreme gradient boosting on features computed by considering the similarity of different fields of the documents (title, abstract), with different fields of the topics (title, query). Our incrementally trained model is a support vector machine trained on a TF-IDF representation of title and abstract of the documents. The results of our approach are promising, reaching 0.658 normalized cumulative gain in the top 10 ranked documents in the simple evaluation setting and 0.846 in the cost-effective evaluation setting, the latter assuming feedback can be obtained from an intermediate user/oracle instead of the end-user.

Recent industry reports assure the rise of web robots which comprise more than half of the total web traffic. They not only threaten the security, privacy and efficiency of the web but they also distort analytics and metrics, doubting the veracity of the information being promoted. In the academic publishing domain, this can cause articles to be faulty presented as prominent and influential. In this paper, we present our approach on detecting web robots in academic publishing websites. We use different supervised learning algorithms with a variety of characteristics deriving from both the log files of the server and the content served by the website. Our approach relies on the assumption that human users will be interested in specific domains or articles, while web robots crawl a web library incoherently. We experiment with features adopted in previous studies with the addition of novel semantic characteristics which derive after performing a semantic analysis using the Latent Dirichlet Allocation (LDA) algorithm. Our real-world case study shows promising results, pinpointing the significance of semantic features in the web robot detection problem.

The figures found in biomedical literature are a vital part of biomedical research, education and clinical decision. The multitude of their modalities and the lack of corresponding meta-data, constitute search and information retrieval a difficult task. We present multi-label modality classification approaches for biomedical figures. In particular, we investigate using both simple and compound figures for training a multi-label model to be used for annotating either all figures, or only those predicted as compound by an initial compound figure detection model. Using data from the medical task of ImageCLEF 2016, we train our approaches with visual features and compare them with the standard approach involving compound figure separation into sub-figures. Furthermore, we present a web application for medical figure retrieval, which is based on one of our classification approaches and allows users to search for figures of PubMed Central.

Αυτή η διατριβή παρουσιάζει πρωτότυπη έρευνα στους τομείς των επιστημών της πληροφορικής και της βιβλιοθηκονομίας, και συγκεκριμένα στον τομέα της ανακάλυψης γνώσης από ακαδημαϊκές δημοσιεύσεις. Η ανακάλυψη γνώσης από ακαδημαϊκές δημοσιεύσεις συγκεντρώνει πολλαπλά ερωτήματα από διαφορετικά ερευνητικά πεδία, όπως ανάκτηση πληροφοριών, μηχανική μάθηση και επεξεργασία φυσικής γλώσσας, και στοχεύει στην κατανόηση, προώθηση και χρήση του δημοσιευμένου επιστημονικού υλικού για την αποτελεσματικότερη αντιμετώπιση των προβλημάτων της κοινωνίας μας. Τα προϊόντα αυτού του ερευνητικού πεδίου είναι νέοι αλγόριθμοι και μέθοδοι που διερευνούν, αναλύουν και χρησιμοποιούν τις πληροφορίες που βρίσκονται στις επιστημονικές δημοσιεύσεις πιο αποτελεσματικά, γρηγορότερα και με ευκολία. Η συμβολή μας αφορά την ανακάλυψη γνώσης από ακαδημαϊκές εκδόσεις σε τέσσερις διαφορετικά προβλήματα: ανάκτηση ολικής ανάκλησης, σημασιολογική δεικτοδότηση, πρόσβαση σε βιβλιογραφικές πηγές και ανάλυση αυτό-παραπομπών. Τα κοινά στοιχεία σε αυτές τις εργασίες είναι η χρήση ακατέργαστων δεδομένων που προέρχονται από δημοσιεύσεις και η ανάπτυξη νέων μεθόδων μηχανικής μάθησης. Ο τομέας εφαρμογής της διατριβής είναι οι ακαδημαϊκές δημοσιεύσεις αλλά οι μέθοδοι που αναπτύχθηκαν μπορούν εύκολα να εφαρμοστούν σε άλλα πεδία όπου τα έγγραφα κειμένου είναι η κύρια πηγή δεδομένων. Πρώτον, παρουσιάζουμε μια νέα προσέγγιση για την αξιολόγηση εγγράφων που στοχεύει να βοηθήσει τους ερευνητές να δημιουργήσουν συστηματικές βιβλιογραφικές ανασκοπήσεις. Η προσέγγισή μας ανακτά και κατατάσσει αποτελεσματικά έγγραφα με βάση ένα ερώτημα, χρησιμοποιώντας τεχνικές learning-to-rank μαζί με μια επαναληπτική μέθοδο ανατροφοδότησης. Δεύτερον, παρουσιάζουμε μια προσέγγιση πολλαπλών ετικετών για την ταξινόμηση βιοϊατρικών εικόνων. Αυτή η μέθοδος δεν χρησιμοποιεί έναν αλγόριθμο διαχωρισμού εικόνων και χρησιμοποιεί οπτικά χαρακτηριστικά καθώς και χαρακτηριστικά κειμένου. Τρίτον, προτείνουμε μια καινοτόμο προσέγγιση εντοπισμού ρομπότ του διαδικτύου που λαμβάνει υπόψη το περιεχόμενο ενός ιστότοπου. Η κύρια συνεισφορά μας είναι μια νέα αναπαράσταση για διαδικτυακές συνεδρίες, βασισμένη στο LDA, που ποσοτικοποιεί τη σημασιολογική διακύμανση του περιεχομένου ιστού μιας συνεδρίας. Τέλος, παρουσιάζουμε έναν νέο τρόπο ανίχνευσης δυνητικά μη θεμιτών αυτό-παραπομπών με βάση τη σημασιολογική ομοιότητα μιας εργασίας και των παραπομπών αυτής. Παρουσιάζουμε τη βαθμολογία ReLy, η οποία βασίζεται σε sentence embeddings και ποσοτικοποιεί τη σημασιολογική ομοιότητα άρθρου-παραπομπής. Όλες οι παραπάνω προσεγγίσεις αξιολογούνται και συγκρίνονται με αντίστοιχες κορυφαίες μεθόδους, σε πολλαπλές πειραματικές ρυθμίσεις και με δεδομένα που προέρχονται από τον πραγματικό κόσμο. Τα αποτελέσματα καταδεικνύουν σημαντικές βελτιώσεις ή νέα ευρήματα σε όλες τις περιπτώσεις και δημοσιεύουμε, σε πολλές περιπτώσεις, τα σύνολα δεδομένων που χρησιμοποιούνται για ενθάρρυνση της διαφάνειας και της μελλοντικής έρευνας.