Multi-relational networks capture intricate relationships in data and have diverse applications across fields such as biomedical, financial, and social sciences. As networks derived from increasingly large datasets become more common, identifying efficient methods for representing and analyzing them becomes crucial. This work extends the Prime Adjacency Matrices (PAMs) framework, which employs prime numbers to represent distinct relations within a network uniquely. This enables a compact representation of a complete multi-relational graph using a single adjacency matrix, which, in turn, facilitates quick computation of multi-hop adjacency matrices. In this work, we enhance the framework by introducing a lossless algorithm for calculating the multi-hop matrices and propose the Bag of Paths (BoP) representation, a versatile feature extraction methodology for various graph analytics tasks, at the node, edge, and graph level. We demonstrate the efficiency of the framework across various tasks and datasets, showing that simple BoP-based models perform comparably to or better than commonly used neural models while improving speed by orders of magnitude.

The outbreak of the COVID-19 pandemic led policy makers and public health officials around the world to implement non-pharmaceutical interventions to suppress the spread of the virus. The goal was to reduce human mobility and social contacts, considered as the main factors of virus diffusion. However, these containment measures needed to be revised on a frequent basis to avoid serious economic and social costs. Moreover, spatial disparities needed to be taken into consideration, since the behavior of the virus was different according to the geographical context. Therefore, a frequent update on the short-term forecasts of the epidemic course, which considered spatial heterogeneities, was crucial for planning appropriate mitigation strategies. In this article, we present a simple epidemiological model based on Cellular Automata, that takes into account human mobility and produces short-term forecasts of daily virus infections. Cellular Automata allow the discretization of time and space and thus, the spatio-temporal dynamics of the disease can be explored at the desired scale. We apply our model on real daily infection and mobility data from Spain and show that it is reliable in predicting the short-term daily infections trajectory both at the country level and at the regional level of Autonomous Communities. Furthermore, compared against four state-of-the-art methods, the proposed method achieves comparable forecasting performance with significantly lower computational resources.

Neurosymbolic (NeSy) AI has emerged as a promising direction to integrate neural and symbolic reasoning. Unfortunately, little effort has been given to developing NeSy systems tailored to sequential/temporal problems. We identify symbolic automata (which combine the power of automata for temporal reasoning with that of propositional logic for static reasoning) as a suitable formalism for expressing knowledge in temporal domains. Focusing on the task of sequence classification and tagging we show that symbolic automata can be integrated with neural-based perception, under probabilistic semantics towards an end-to-end differentiable model. Our proposed hybrid model, termed NeSyA (Neuro Symbolic Automata) is shown to either scale or perform more accurately than previous NeSy systems in a synthetic benchmark and to provide benefits in terms of generalization compared to purely neural systems in a real-world event recognition task. Code is available at: https://github.com/nmanginas/nesya

Complex Event Recognition (CER) systems receive as input a stream of time-stamped events and identify situations of interest that satisfy a given pattern. Streaming environments are characterized by the high rate and volume of input data, and thus, scalability is of crucial importance. At the same time, noise and uncertainty are ubiquitous in temporal data, and not considering them, leads to erroneous detections. To confront these challenges, we present a tensor-based formalization of the Event Calculus (EC) for probabilistic inference, and demonstrate the scalability of our approach with the use of CER datasets from two real-world application domains. Moreover, we demonstrate the benefits of our approach, in terms of processing time, by comparing it against a probabilistic logic programming implementation of EC.

Neuro-Symbolic Artificial Intelligence (NeSy AI) has emerged as a promising direction for integrating neural learning with symbolic reasoning. Typically, in the probabilistic variant of such systems, a neural network first extracts a set of symbols from sub-symbolic input, which are then used by a symbolic component to reason in a probabilistic manner towards answering a query. In this work, we address the problem of formally verifying the robustness of such NeSy probabilistic reasoning systems, therefore paving the way for their safe deployment in critical domains. We analyze the complexity of solving this problem exactly, and show that a decision version of the core computation is $\mathrm{NP}^{\mathrm{PP}}$-complete. In the face of this result, we propose the first approach for approximate, relaxation-based verification of probabilistic NeSy systems. We demonstrate experimentally on a standard NeSy benchmark that the proposed method scales exponentially better than solver-based solutions and apply our technique to a real-world autonomous driving domain, where we verify a safety property under large input dimensionalities.

Abstract In this work, we generalize the ideas of Kaiming initialization to Graph Neural Networks (GNNs) and propose a new scheme (G-Init) that reduces oversmoothing, leading to very good results in node and graph classification tasks. GNNs are commonly initialized using methods designed for other types of Neural Networks, overlooking the underlying graph topology. We analyze theoretically the variance of signals flowing forward and gradients flowing backward in the class of convolutional GNNs. We then simplify our analysis to the case of the GCN and propose a new initialization method. Results indicate that the new method (G-Init) reduces oversmoothing in deep GNNs, facilitating their effective use. Our approach achieves an accuracy of 61.60% on the CS dataset (32-layer GCN) and 69.24% on Cora (64-layer GCN), surpassing state-of-the-art initialization methods by 25.6 and 8.6 percentage points, respectively. Extensive experiments confirm the robustness of our method across multiple benchmark datasets, highlighting its effectiveness in diverse settings. Furthermore, our experimental results support the theoretical findings, demonstrating the advantages of deep networks in scenarios with no feature information for unlabeled nodes (i.e., “cold start” scenario).

Abstract The performance of Graph Neural Networks (GNNs) diminishes as their depth increases. That is mainly attributed to oversmoothing, which leads to similar node representations through repeated graph convolutions. To enable deep GNNs, several approaches have been proposed, among which the use of residual connections. Residual connections have proven effective in benchmark datasets, but the way in which they improve the performance of deep GNNs has not been fully studied. We show that residual connections force the model to focus on the local neighborhood of graph nodes, making the GNN equivalent to the sum of shallow GCNs. We explain theoretically why this is the case and verify the theoretical results experimentally. However, our findings raise the question of whether residual connections are helpful in cases where deep networks are necessary. We assess this experimentally, in two situations: (a) in the presence of the “cold start" problem, i.e. when there is no feature information about unlabeled nodes; and (b) in a new synthetic dataset of controllable long-interactions. These experiments highlight the drawbacks of GNNs using residual connections, while showing that simpler methods can be more effective.

Abstract In this work we investigate an observation made by Kipf and Welling (5th International Conference on Learning Representations, 2017), who suggested that untrained Graph Convolutional Networks (GCNs) can generate meaningful node embeddings. In particular, we investigate the effect of training only a single layer of a GCN or a GAT (Graph Attention Network), while keeping the rest of the layers frozen. We propose a basis on which the effect of the untrained layers and their contribution to the generation of embeddings can be predicted. Moreover, we show that network width influences the dissimilarity of node embeddings produced after the initial node features pass through the untrained part of the model. Additionally, we establish a connection between partially trained GCNs and oversmoothing, showing that they are capable of reducing it. We verify our theoretical results experimentally and show the benefits of using deep networks that resist oversmoothing, in a “cold start” scenario, where there is a lack of feature information for unlabeled nodes.

We address the need for focused retrieval and integration of biomedical literature, enabling improved subject annotation of documents with domain concepts for which no manual labels are available. To do so, we propose a novel zero-shot method, called PN Relabeler, that improves heuristic concept-level annotations by relabeling the documents Predicted as Negative (PN) by the concept occurrence (CO) heuristic. CO has proven to be a good heuristic for fine-grained semantic indexing (FGSI) of biomedical literature. In particular, it is quite precise but still misses some document labels, i.e. it suffers from lower recall. The PN Relabeler method addresses this problem, by introducing a novel approach to combine heuristic annotations for unseen labels with knowledge learned from past labels. To do so, it first tackles the intermediate task of zero-shot relabeling of documents labeled with CO-based FGSI annotations. Then, it builds upon the power of domain-specific deep pretrained language models to improve the recall of the heuristic annotation, i.e. reduce the rate of false negative cases. The results reveal that relabeling with PN Relabeler improves the micro-F1 by more than 6%, compared to CO annotations, and by 3%, compared to state-of-the-art CO-based FGSI methods. This highlights the potential of learning from the errors of a strong heuristic on concepts where manual labels are already available. This knowledge is useful for improving the heuristic FGSI labels on new unseen concepts without manual labels in a zero-shot scenario.

Large Language Models (LLMs) have been shown to effectively perform zero-shot document retrieval, a process that typically consists of two steps: i) retrieving relevant documents, and ii) re-ranking them based on their relevance to the query.This paper presents GENRA, a new approach to zero-shot document retrieval that incorporates rank aggregation to improve retrieval effectiveness.Given a query, GENRA first utilizes LLMs to generate informative passages that capture the query's intent.These passages are then employed to guide the retrieval process, selecting similar documents from the corpus.Next, we use LLMs again for a second refinement step.This step can be configured for either direct relevance assessment of each retrieved document or for re-ranking the retrieved documents.Ultimately, both approaches ensure that only the most relevant documents are kept.Upon this filtered set of documents, we perform multi-document retrieval, generating individual rankings for each document.As a final step, GENRA leverages rank aggregation, combining the individual rankings to produce a single refined ranking.Extensive experiments on benchmark datasets demonstrate that GENRA improves existing approaches, highlighting the effectiveness of the proposed methodology in zero-shot retrieval.

Abstract In this paper, we propose a novel framework, called STAL, which makes use of unlabeled graph data, through a combination of Active Learning and Self-Training, in order to improve node labeling by Graph Neural Networks (GNNs). GNNs have been shown to perform well on many tasks, when sufficient labeled data are available. Such data, however, is often scarce, leading to the need for methods that leverage unlabeled data that are abundant. Active Learning and Self-training are two common approaches towards this goal and we investigate here their combination, in the context of GNN training. Specifically, we propose a new framework that first uses active learning to select highly uncertain unlabeled nodes to be labeled and be included in the training set. In each iteration of active labeling, the proposed method expands also the label set through self-training. In particular, highly certain pseudo-labels are obtained and added automatically to the training set. This process is repeated, leading to good classifiers, with a limited amount of labeled data. Our experimental results on various datasets confirm the efficiency of the proposed approach.

OBJECTIVE: This article presents the novel BioASQ Synergy research process which aims to facilitate the interaction between biomedical experts and automated question-answering systems. MATERIALS AND METHODS: The proposed research allows systems to provide answers to emerging questions, which in turn are assessed by experts. The assessment of the experts is fed back to the systems, together with new questions. With this iteration, we aim to facilitate the incremental understanding of a developing problem and contribute to solution discovery. RESULTS: The results suggest that the proposed approach can assist researchers to navigate available resources. The experts seem to be very satisfied with the quality of the ideal answers provided by the systems, suggesting that such systems are already useful in answering open research questions. DISCUSSION: BioASQ Synergy aspires to provide a tool that gives the experts easy and personalized access to the latest findings in a fast-growing corpus of material. CONCLUSION: In this article, we envisioned BioASQ Synergy as a continuous dialogue between experts and systems to issue open questions. We ran an initial proof-of-concept of the approach, in order to evaluate its usefulness, both from the side of the experts, as well as from the side of the participating systems.

Abstract Symbolic event recognition systems detect event occurrences using first-order logic rules. Although existing online structure learning approaches ease the discovery of such rules in noisy data streams, they assume the existence of fully labelled training data. Splice is a recent online graph-based approach that estimates the labels of unlabelled data and makes it possible to learn such rules from semi-supervised training sequences of logical interpretations. However, Splice labelling depends significantly on the metric used to compute the distances of unlabelled examples to their labelled counterparts. Moreover, there is no guarantee about the quality of the labelling found in the local graphs that are built as the data stream in. In this paper, we propose a new online learning method, which includes an enhanced hybrid measure that combines an optimised structural distance, and a data-driven one. The former is guided by feature selection targeted to kNN classification, while the latter is a mass-based dissimilarity. Additionally, the enhanced Splice method, improves the graph construction process, by storing a synopsis of the past, in order to achieve more informed labelling on the local graphs. We evaluate our approach by learning Event Calculus theories for the tasks of human activity recognition, maritime monitoring, and fleet management. The evaluation suggests that our approach outperforms its predecessor, in terms of inferring the missing labels and improving the predictive accuracy of the underlying structure learning system.

We present a system for Complex Event Recognition (CER) based on automata. While multiple such systems have been described in the literature, they typically suffer from a lack of clear and denotational semantics, a limitation which often leads to confusion with respect to their expressive power. In order to address this issue, our system is based on an automaton model which is a combination of symbolic and register automata. We extend previous work on these types of automata, in order to construct a formalism with clear semantics and a corresponding automaton model whose properties can be formally investigated. We call such automata Symbolic Register Transducers (SRT). The distinctive feature of SRT , compared to previous automaton models used in CER, is that they can encode patterns relating multiple input events from an event stream, without sacrificing rigor and clarity. We show how SRT can be used in CER in order to detect patterns upon streams of events, using our framework that provides declarative and compositional semantics. We also compare our SRT -based CER engine against other state-of-the-art CER systems and show that it is both more expressive and more efficient.

In this work, we examine the auditor opinions that are provided in financial reports of public companies, when they express issues related to going concerns. Auditor opinions provide explicit insights regarding potential threats to the financial status of companies. We, therefore, provide methods for the automated classification of the auditor narratives to going concern issues, and we investigate which of those issues are related to bankruptcy events. We focus on annual reports of public US companies and publicly available bankruptcy labels to learn models that label these reports with probable going concern issues in an automated way. Our experimental results validate our approach and provide evidence that the analysis of these narratives can lead to the identification of specific issues related to bankruptcy concerns and thus alarm the interested parties.

Most phenomena related to biomedical tasks are inherently complex, and in many cases, are expressed as signals on biomedical Knowledge Graphs (KGs). In this work, we introduce the use of a new representation framework, the Prime Adjacency Matrix (PAM) for biomedical KGs, which allows for very efficient network analysis. PAM utilizes prime numbers to enable representing the whole KG with a single adjacency matrix and the fast computation of multiple properties of the network. We illustrate the applicability of the framework in the biomedical domain by working on different biomedical knowledge graphs and by providing two case studies: one on drug-repurposing for COVID-19 and one on important metapath extraction. We show that we achieve better results than the original proposed workflows, using very simple methods that require no training, in considerably less time.

Complex Event Recognition (CER) systems detect event occurrences in streaming input using predefined event patterns. Techniques that learn event patterns from data are highly desirable in CER. Since such patterns are typically represented by symbolic automata, we propose a family of such automata where the transition-enabling conditions are defined by Answer Set Programming (ASP) rules, and which, thanks to the strong connections of ASP to symbolic learning, are learnable from data. We present such a learning approach in ASP, capable of jointly learning the structure of an automaton, while synthesizing the transition guards from building-block predicates, and a scalable, incremental version thereof that progressively revises models learnt from mini-batches using Monte Carlo Tree Search. We evaluate our approach on three CER datasets and empirically demonstrate its efficacy.

The performance of Graph Neural Networks (GNNs) deteriorates as the depth of the network increases. That performance drop is mainly attributed to oversmoothing, which leads to similar node representations through repeated graph convolutions. We show that in deep GNNs the activation function plays a crucial role in oversmoothing. We explain theoretically why this is the case and propose a simple modification to the slope of ReLU to reduce oversmoothing. The proposed approach enables deep networks without the need to change the network architecture or to add residual connections. We verify the theoretical results experimentally and further show that deep networks, which do not suffer from oversmoothing, are beneficial in the presence of the “cold start” problem, i.e. when there is no feature information about unlabeled nodes.

We present a system for online, incremental composite event recognition. In streaming environments, the usual case is for data to arrive with a (variable) delay from, and to be revised by, the underlying sources. We propose RTEC_inc, an incremental version of RTEC, a composite event recognition engine with formal, declarative semantics, that has been shown to scale to several real-world data streams. RTEC deals with delayed arrival and revision of events by computing all queries from scratch. This is often inefficient since it results in redundant computations. Instead, RTEC_inc deals with delays and revisions in a more efficient way, by updating only the affected queries. We compare RTEC_inc and RTEC experimentally using real-world and synthetic datasets. The results are compatible with our complexity analysis and show that RTEC_inc outperforms RTEC in many practical cases.

This paper applies different link prediction methods on a knowledge graph generated from biomedical literature, with the aim to compare their ability to identify unknown drug-gene interactions and explain their predictions. Identifying novel drug-target interactions is a crucial step in drug discovery and repurposing. One approach to this problem is to predict missing links between drug and gene nodes, in a graph that contains relevant biomedical knowledge. Such a knowledge graph can be extracted from biomedical literature, using text mining tools. In this work, we compare state-of-the-art graph embedding approaches and contextual path analysis on the interaction prediction task. The comparison reveals a trade-off between predictive accuracy and explainability of predictions. Focusing on explainability, we train a decision tree on model predictions and show how it can aid the understanding of the prediction process. We further test the methods on a drug repurposing task and validate the predicted interactions against external databases, with very encouraging results.

Abstract Complex Event Recognition (CER) systems detect event occurrences in streaming time-stamped input using predefined event patterns. Logic-based approaches are of special interest in CER, since, via Statistical Relational AI, they combine uncertainty-resilient reasoning with time and change, with machine learning, thus alleviating the cost of manual event pattern authoring. We present a system based on Answer Set Programming (ASP), capable of probabilistic reasoning with complex event patterns in the form of weighted rules in the Event Calculus, whose structure and weights are learnt online. We compare our ASP-based implementation with a Markov Logic-based one and with a number of state-of-the-art batch learning algorithms on CER data sets for activity recognition, maritime surveillance and fleet management. Our results demonstrate the superiority of our novel approach, both in terms of efficiency and predictive performance. This paper is under consideration for publication in Theory and Practice of Logic Programming (TPLP).

Clinical trials offer a fundamental opportunity to discover new treatments and advance the medical knowledge.However, the uncertainty of the outcome of a trial can lead to unforeseen costs and setbacks.In this study, we propose a new method to predict the effectiveness of an intervention in a clinical trial.Our method relies on generating an informative summary from multiple documents available in the literature about the intervention under study.Specifically, our method first gathers all the abstracts of PubMed articles related to the intervention.Then, an evidence sentence, which conveys information about the effectiveness of the intervention, is extracted automatically from each abstract.Based on the set of evidence sentences extracted from the abstracts, a short summary about the intervention is constructed.Finally, the produced summaries are used to train a BERT-based classifier, in order to infer the effectiveness of an intervention.To evaluate our proposed method, we introduce a new dataset which is a collection of clinical trials together with their associated PubMed articles.Our experiments demonstrate the effectiveness of producing short informative summaries and using them to predict the effectiveness of an intervention.

Lefteris Loukas, Manos Fergadiotis, Ilias Chalkidis, Eirini Spyropoulou, Prodromos Malakasiotis, Ion Androutsopoulos, Georgios Paliouras. Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2022.

Abstract Computational systems and methods are often being used in biological research, including the understanding of cancer and the development of treatments. Simulations of tumor growth and its response to different drugs are of particular importance, but also challenging complexity. The main challenges are first to calibrate the simulators so as to reproduce real‐world cases, and second, to search for specific values of the parameter space concerning effective drug treatments. In this work, we combine a multi‐scale simulator for tumor cell growth and a genetic algorithm (GA) as a heuristic search method for finding good parameter configurations in reasonable time. The two modules are integrated into a single workflow that can be executed in parallel on high performance computing infrastructures. In effect, the GA is used to calibrate the simulator, and then to explore different drug delivery schemes. Among these schemes, we aim to find those that minimize tumor cell size and the probability of emergence of drug resistant cells in the future. Experimental results illustrate the effectiveness and computational efficiency of the approach.

Early and precise prognosis of dementia is a critical medical challenge. The design of an optimal computational model that addresses this issue, and at the same time explains the underlying mechanisms that lead to output decisions, is an ongoing challenge. In this study, we focus on assessing the risk of an individual converting to Dementia in the short (next year) and long (one to five years) term, given only a few early-stage observations. Our goal is to develop a machine learning model that could assist the prediction of dementia from regular clinical data. The results show that combining various machine learning techniques together can successfully define ways to identify the risks of developing dementia over the following five years with accuracies considerably above average rates. These findings suggest that accurately developed models can be considered as a promising tool to improve early dementia prognosis.

We present a system for online, incremental composite event recognition. In streaming environments, the usual case is for data to arrive with a (variable) delay from, and to be revised by, the underlying sources. We propose RTECinc, an incremental version of RTEC, a composite event recognition engine with formal, declarative semantics, that has been shown to scale to several real-world data streams. RTEC deals with delayed arrival and revision of events by computing all queries from scratch. This is often inefficient since it results in redundant computations. Instead, RTECinc deals with delays and revisions in a more efficient way, by updating only the affected queries. We examine RTECinc theoretically, presenting a complexity analysis, and show the conditions in which it outperforms RTEC. Moreover, we compare RTECinc and RTEC experimentally using real-world and synthetic datasets. The results are compatible with our theoretical analysis and show that RTECinc outperforms RTEC in many practical cases.

In biomedical research, unified access to up-to-date domain-specific knowledge is crucial, as such knowledge is continuously accumulated in scientific literature and structured resources. Identifying and extracting specific information is a challenging task and computational analysis of knowledge bases can be valuable in this direction. However, for disease-specific analyses researchers often need to compile their own datasets, integrating knowledge from different resources, or reuse existing datasets, that can be out-of-date. In this study, we propose a framework to automatically retrieve and integrate disease-specific knowledge into an up-to-date semantic graph, the iASiS Open Data Graph. This disease-specific semantic graph provides access to knowledge relevant to specific concepts and their individual aspects, in the form of concept relations and attributes. The proposed approach is implemented as an open-source framework and applied to three diseases (Lung Cancer, Dementia, and Duchenne Muscular Dystrophy). Exemplary queries are presented, investigating the potential of this automatically generated semantic graph as a basis for retrieval and analysis of disease-specific knowledge.

Cancer remains one of the major public health challenges worldwide. After cardiovascular diseases, cancer is one of the first causes of death and morbidity in Europe, with more than 4 million new cases and 1.9 million deaths per year. The suboptimal management of cancer patients during treatment and subsequent follows up are major obstacles in achieving better outcomes of the patients and especially regarding cost and quality of life In this paper, we present an initial data-driven approach to analyze the resources and services that are used more frequently by lung-cancer patients with the aim of identifying where the care process can be improved by paying a special attention on services before diagnosis to being able to identify possible lung-cancer patients before they are diagnosed and by reducing the length of stay in the hospital. Our approach has been built by analyzing the clinical notes of those oncological patients to extract this information and their relationships with other variables of the patient. Although the approach shown in this manuscript is very preliminary, it shows that quite interesting outcomes can be derived from further analysis.

In the maritime domain, vessels typically maintain straight, predictable routes at open sea, except in the rare cases of adverse weather conditions, accidents and traffic restrictions. Consequently, large amounts of streaming positional updates from vessels can hardly contribute additional knowledge about their actual motion patterns. We have been developing a system for vessel trajectory compression discarding a significant part of the original positional updates, with minimal trajectory reconstruction error. In this work, we present an extension of this system, that allows the user to fine-tune trajectory compression according to the requirements of a given application. The extended system avoids the issues of hyper-parameter tuning, supports incremental optimization and facilitates composite maritime event recognition. Finally, we report empirical results from a comprehensive empirical evaluation against two real-world datasets of vessel positions.

The vision of IASIS project is to turn the wave of big biomedical data heading our way into actionable knowledge for decision makers. This is achieved by integrating data from disparate sources, including genomics, electronic health records and bibliography, and applying advanced analytics methods to discover useful patterns. The goal is to turn large amounts of available data into actionable information to authorities for planning public health activities and policies. The integration and analysis of these heterogeneous sources of information will enable the best decisions to be made, allowing for diagnosis and treatment to be personalised to each individual. The project offers a common representation schema for the heterogeneous data sources. The iASiS infrastructure is able to convert clinical notes into usable data, combine them with genomic data, related bibliography, image data and more, and create a global knowledge base. This facilitates the use of intelligent methods in order to discover useful patterns across different resources. Using semantic integration of data gives the opportunity to generate information that is rich, auditable and reliable. This information can be used to provide better care, reduce errors and create more confidence in sharing data, thus providing more insights and opportunities. Data resources for two different disease categories are explored within the iASiS use cases, dementia and lung cancer.

Biomedical literature in MEDLINE/PubMed is semantically indexed with MeSH thesaurus entries (subject annotations) which may correspond to more than one related but distinct domain concepts. In such cases, the subject annotations do not follow the level of detail available in the domain and do not always suffice to meet the information needs of domain experts. In this work, we propose a method to automatically refine subject annotations at the level of concepts and employ it in the case of the MeSH descriptor for Alzheimer's Disease, which corresponds to six different concepts representing disease sub-types. The results indicate that the use of concept-occurrence as weak supervision can improve upon the predictive performance of literal string matching alone. The refined annotations can support more precise concept-based search, enable the integration of subject annotations with other semantic information and facilitate the maintenance of subject annotation consistency, as the MeSH thesaurus evolves with the addition of more detailed entries.

Complex Event Processing (CEP) systems have appeared in abundance during the last two decades. Their purpose is to detect in real–time interesting patterns upon a stream of events and to inform an analyst for the occurrence of such patterns in a timely manner. However, there is a lack of methods for forecasting when a pattern might occur before such an occurrence is actually detected by a CEP engine. We present Wayeb, a tool that attempts to address the issue of Complex Event Forecasting. Wayeb employs symbolic automata as a computational model for pattern detection and Markov chains for deriving a probabilistic description of a symbolic automaton.

In the recent years research on dynamic social network has increased, which is also due to the availability of data sets from streaming media. Modeling a network's dynamic behaviour can be performed at the level of communities, which represent their mesoscale structure. Communities arise as a result of user to user interaction. In the current work we aim to predict the evolution of communities, i.e. to predict their future form. While this problem has been studied in the past as a supervised learning problem with a variety of classifiers, the problem is that the "knowledge" of a classifier is opaque and consequently incomprehensible to a human. Thus we have employed first order logic, and in particular the event calculus to represent the communities and their evolution. We addressed the problem of predicting the evolution as an online Inductive Logic Programming problem (ILP), where the issue is to learn first order logical clauses that associate evolutionary events, and particular Growth, Shrinkage, Continuation and Dissolution to lower level events. The lower level events are features that represent the structural and temporal characteristics of communities. Experiments have been performed on a real life data set form the Mathematics StackExchange forum, with the OLED framework for ILP. In doing so we have produced clauses that model both short term and long term correlations.

Information shared on Twitter is ever increasing and users-recipients are overwhelmed by the number of tweets they receive, many of which of no interest. Filters that estimate the interest of each incoming post can alleviate this problem, for example by allowing users to sort incoming posts by predicted interest (e.g., "top stories" vs. "most recent" in Facebook). Global and personal filters have been used to detect interesting posts in social networks. Global filters are trained on large collections of posts and reactions to posts (e.g., retweets), aiming to predict how interesting a post is for a broad audience. In contrast, personal filters are trained on posts received by a particular user and the reactions of the particular user. Personal filters can provide recommendations tailored to a particular user's interests, which may not coincide with the interests of the majority of users that global filters are trained to predict. On the other hand, global filters are typically trained on much larger datasets compared to personal filters. Hence, global filters may work better in practice, especially with new users, for which personal filters may have very few training instances ("cold start" problem).

No abstract available.

Abstract Systems for symbolic event recognition infer occurrences of events in time using a set of event definitions in the form of first-order rules. The Event Calculus is a temporal logic that has been used as a basis in event recognition applications, providing among others, direct connections to machine learning, via Inductive Logic Programming (ILP). We present an ILP system for online learning of Event Calculus theories. To allow for a single-pass learning strategy, we use the Hoeffding bound for evaluating clauses on a subset of the input stream. We employ a decoupling scheme of the Event Calculus axioms during the learning process, that allows to learn each clause in isolation. Moreover, we use abductive-inductive logic programming techniques to handle unobserved target predicates. We evaluate our approach on an activity recognition application and compare it to a number of batch learning techniques. We obtain results of comparable predicative accuracy with significant speed-ups in training time. We also outperform hand-crafted rules and match the performance of a sound incremental learner that can only operate on noise-free datasets.

BACKGROUND: This article provides an overview of the first BIOASQ challenge, a competition on large-scale biomedical semantic indexing and question answering (QA), which took place between March and September 2013. BIOASQ assesses the ability of systems to semantically index very large numbers of biomedical scientific articles, and to return concise and user-understandable answers to given natural language questions by combining information from biomedical articles and ontologies. RESULTS: The 2013 BIOASQ competition comprised two tasks, Task 1a and Task 1b. In Task 1a participants were asked to automatically annotate new PUBMED documents with MESH headings. Twelve teams participated in Task 1a, with a total of 46 system runs submitted, and one of the teams performing consistently better than the MTI indexer used by NLM to suggest MESH headings to curators. Task 1b used benchmark datasets containing 29 development and 282 test English questions, along with gold standard (reference) answers, prepared by a team of biomedical experts from around Europe and participants had to automatically produce answers. Three teams participated in Task 1b, with 11 system runs. The BIOASQ infrastructure, including benchmark datasets, evaluation mechanisms, and the results of the participants and baseline methods, is publicly available. CONCLUSIONS: A publicly available evaluation infrastructure for biomedical semantic indexing and QA has been developed, which includes benchmark datasets, and can be used to evaluate systems that: assign MESH headings to published articles or to English questions; retrieve relevant RDF triples from ontologies, relevant articles and snippets from PUBMED Central; produce "exact" and paragraph-sized "ideal" answers (summaries). The results of the systems that participated in the 2013 BIOASQ competition are promising. In Task 1a one of the systems performed consistently better from the NLM's MTI indexer. In Task 1b the systems received high scores in the manual evaluation of the "ideal" answers; hence, they produced high quality summaries as answers. Overall, BIOASQ helped obtain a unified view of how techniques from text classification, semantic indexing, document and passage retrieval, question answering, and text summarization can be combined to allow biomedical experts to obtain concise, user-understandable answers to questions reflecting their real information needs.

Systems for symbolic event recognition accept as input a stream of time-stamped events from sensors and other computational devices, and seek to identify high-level composite events, collections of events that satisfy some pattern. RTEC is an Event Calculus dialect with novel implementation and `windowing' techniques that allow for efficient event recognition, scalable to large data streams. RTEC supports the expression of rather complex events, such as `two people are fighting', using simple primitives. It can operate in the absence of filtering modules, as it is only slightly affected by data that are irrelevant to the events we want to recognise. Furthermore, RTEC can deal with applications where event data arrive with a (variable) delay from, and are revised by, the underlying sources. RTEC can update already recognised events and recognise new events when data arrive with a delay or following data revision. We evaluate RTEC both theoretically, presenting a complexity analysis, and experimentally, using two real-world applications. The evaluation shows that RTEC can support real-time event recognition and is capable of meeting the performance requirements identified in a survey of event processing use cases.

Symbolic event recognition systems have been successfully applied to a variety of application domains, extracting useful information in the form of events, allowing experts or other systems to monitor and respond when significant events are recognised. In a typical event recognition application, however, these systems often have to deal with a significant amount of uncertainty. In this article, we address the issue of uncertainty in logic-based event recognition by extending the Event Calculus with probabilistic reasoning. Markov logic networks are a natural candidate for our logic-based formalism. However, the temporal semantics of the Event Calculus introduce a number of challenges for the proposed model. We show how and under what assumptions we can overcome these problems. Additionally, we study how probabilistic modelling changes the behaviour of the formalism, affecting its key property—the inertia of fluents. Furthermore, we demonstrate the advantages of the probabilistic Event Calculus through examples and experiments in the domain of activity recognition, using a publicly available dataset for video surveillance.

Abstract We present a system for recognising human activity given a symbolic representation of video content. The input of our system is a set of time-stamped short-term activities (STA) detected on video frames. The output is a set of recognised long-term activities (LTA), which are pre-defined temporal combinations of STA. The constraints on the STA that, if satisfied, lead to the recognition of an LTA, have been expressed using a dialect of the Event Calculus. In order to handle the uncertainty that naturally occurs in human activity recognition, we adapted this dialect to a state-of-the-art probabilistic logic programming framework. We present a detailed evaluation and comparison of the crisp and probabilistic approaches through experimentation on a benchmark dataset of human surveillance videos.

Organisations require techniques for automated transformation of the Big Data they collect into operational knowledge. This requirement may be addressed by employing event processing systems that detect activities/events of special significance within an organisation, given streams of low-level information that are difficult to be utilised by humans [4]. Systems for event processing and in particular event recognition (‘event pattern matching’) accept as input a stream of time-stamped, simple or low-level events. A low-level event is the result of applying a computational derivation process to some other event, such as an event coming from a sensor. Using low-level events as input, event processing systems identify composite or high-level events of interest — collections of events that satisfy some pattern. Consider, for example, the recognition of attacks on nodes of a computer network given the TCP/IP messages, the recognition of suspicious trader behaviour given the transactions in a financial market, and the recognition of whale songs given a symbolic representation of whale sounds. Numerous event processing systems have been proposed in the literature [3]. Systems with a logic-based representation of event structures, for example, have been attracting considerable attention. They exhibit a formal, declarative semantics, allowing for verification and a code maintenance, they have proven to be efficient and scalable, and they are supported by machine learning tools, minimising human effort in the system development. In this tutorial, we review formal event processing systems. High-level event ‘definitions’ impose temporal and, possibly, atemporal constraints on subevents, that is, low-level events or other high-level events. We will review a Chronicle Recognition System, the Event Calculus, ProbLog and Markov Logic Networks. The Chronicle Recognition System is a purely temporal reasoning system that allows for efficient event processing. It has been used in various domains, ranging from medical applications to computer network management. The Event Calculus allows for the representation of temporal, as well as atemporal constraints. Consequently, the Event Calculus may be used in applications requiring

This paper provides an overview of the workshop Web-Scale Classification: Web Classification in the Big Data Era which was held in New York City, on February 28th as a workshop of the seventh International Conference on Web Search and Data Mining. The goal of the workshop was to discuss and assess recent research focusing on classification and mining in Web-scale category systems. The workshop brought together members of several communities such web mining, machine learning, text classification and social media mining.

A new transfer learning method is presented in this paper, addressing a particularly hard transfer learning problem: the case where the target domain shares only a subset of its classes with the source domain and only unlabeled data are provided for the target domain. This is a situation that occurs frequently in real-world applications, such as the multiclass document classification problems that motivated our work. The proposed approach is a transfer learning variant of the Probabilistic Latent Semantic Analysis (PLSA) model that we name TL-PLSA. Unlike most approaches in the literature, TL-PLSA captures both the difference of the domains and the commonalities of the class sets, given no labelled data from the target domain. We perform experiments over three different datasets and show the difficulty of the task, as well as the promising results that we obtained with the new method.

Artificial Intelligence-based event recognition systems carry high potential for organisations to utilise their structured and unstructured data. The application of these systems as a backbone of decision support systems allows for effective and efficient information management. To sufficiently evaluate such kind of integrated systems recognising events for the benefit of decision makers, a holistic methodology is necessary. We propose a new methodology which complements existing approaches for technology-oriented verification and validation by user-oriented evaluation (user experience analysis). We illustrate the proposed methodology by evaluating EP-IRM, an event processing system for intelligent resource management. This case study shows that our methodology offers invaluable information about the performance and acceptance of an event based decision support system.

This paper presents a method along with a set of measures for evaluating learned ontologies against gold ontologies. The proposed method transforms the ontology concepts and their properties into a vector space representation to avoid the common string matching of concepts and properties at the lexical layer. The proposed evaluation measures exploit the vector space representation and calculate the similarity of the two ontologies (learned and gold) at the lexical and relational levels. Extensive evaluation experiments are provided, which show that these measures capture accurately the deviations from the gold ontology. The proposed method is tested using the Genia and the Lonely Planet gold ontologies, as well as the ontologies in the benchmark series of the Ontology Alignment Evaluation Initiative.

The paper is motivated by the need to handle robustly the uncertainty of temporal intervals, e.g. as it occurs in automated event detection in video streams. The paper introduces a two-dimensional mapping of Allen's relations, based on orthogonal characteristics of interval relations, namely relative position and relative size. The hourglass-shaped mapping also represents limit cases that correspond to durationless intervals. Based on this mapping, we define two sets of primitive interval relations in terms of the relative positioning and relative size of intervals. These primitives are then used to derive a probabilistic set of Allen's relations. A number of example cases are presented to illustrate how the proposed approach can improve the robustness of interval relations.

This paper presents a knowledge discovery framework for the construction of Community Web Directories, a concept that we introduced in our recent work, applying personalization to Web directories. In this context, the Web directory is viewed as a thematic hierarchy and personalization is realized by constructing user community models on the basis of usage data. In contrast to most of the work on Web usage mining, the usage data that are analyzed here correspond to user navigation throughout the Web, rather than a particular Web site, exhibiting as a result a high degree of thematic diversity. For modeling the user communities, we introduce a novel methodology that combines the users' browsing behavior with thematic information from the Web directories. Following this methodology, we enhance the clustering and probabilistic approaches presented in previous work and also present a new algorithm that combines these two approaches. The resulting community models take the form of Community Web Directories. The proposed personalization methodology is evaluated both on a specialized artificial and a general-purpose Web directory, indicating its potential value to the Web user. The experiments also assess the effectiveness of the different machine learning techniques on the task.

This paper proposes a method for learning ontologies given a corpus of text documents. The method identifies concepts in documents and organizes them into a subsumption hierarchy, without presupposing the existence of a seed ontology. The method unco

Focused crawlers are programs that wander in the Web, using its graph structure, and gather pages that belong to a specific topic. The most critical task in Focused Crawling is the scoring of the URLs as it designates the path that the crawler will follow, and thus its effectiveness. In this paper we propose a novel scheme for assigning scores to the URLs, based on the Reinforcement Learning (RL) framework. The proposed approach learns to select the best classifier for ordering the URLs. This formulation reduces the size of the search space for the RL method and makes the problem tractable. We evaluate the proposed approach on-line on a number of topics, which offers a realistic view of its performance, comparing it also with a RL method and a simple but effective classifier-based crawler. The results demonstrate the strength of the proposed approach.

In this paper we propose a novel relation extraction method, based on grammatical inference. Following a semisupervised learning approach, the text that connects named entities in an annotated corpus is used to infer a context free grammar. The grammar learning algorithm is able to infer grammars from positive examples only, controlling overgeneralisation through minimum description length. Evaluation results show that the proposed approach performs comparable to the state of the art, while exhibiting a bias towards precision, which is a sign of conservative generalisation.

Determining the size of an ontology that is automatically learned from texts is an open issue. In this paper, we study the similarity between ontology concepts at different levels of a taxonomy, quantifying in a natural manner the quality of the ontology attained. Our approach is integrated in a method for language-neutral learning of ontologies from texts, which relies on conditional independence tests over thematic topics that are discovered using LDA.

This paper proposes a method for learning ontologies given a corpus of text documents. The method identifies concepts in documents and organizes them into a subsumption hierarchy, without presupposing the existence of a seed ontology. The method uncovers latent topics in terms of which document text is being generated. These topics form the concepts of the new ontology. This is done in a language neutral way, using probabilistic space reduction techniques over the original term space of the corpus. Given multiple sets of concepts (latent topics) being discovered, the proposed method constructs a subsumption hierarchy by performing conditional independence tests among pairs of latent topics, given a third one. The paper provides experimental results over the GENIA corpus from the domain of biomedicine.

This article investigates the effectiveness of voting and stacked generalization -also known as stacking- in the context of information extraction (IE). A new stacking framework is proposed that accommodates well-known approaches for IE. The key idea is to perform cross-validation on the base-level data set, which consists of text documents annotated with relevant information, in order to create a meta-level data set that consists of feature vectors. A classifier is then trained using the new vectors. Therefore, base-level IE systems are combined with a common classifier at the meta-level. Various voting schemes are presented for comparing against stacking in various IE domains. Well known IE systems are employed at the base-level, together with a variety of classifiers at the meta-level. Results show that both voting and stacking work better when relying on probabilistic estimates by the base-level systems. Voting proved to be effective in most domains in the experiments. Stacking, on the other hand, proved to be consistently effective over all domains, doing comparably or better than voting and always better than the best base-level systems. Particular emphasis is also given to explaining the results obtained by voting and stacking at the meta-level, with respect to the varying degree of similarity in the output of the base-level systems.

Interest in the analysis of user behaviour on the Internet has been increasing rapidly, especially since the advent of electronic commerce. In this context, we argue here for the usefulness of constructing communities of users with common behaviour, making use of machine learning techniques. In particular, we assume that the users of any service on the Internet constitute a large community and we aim to construct smaller communities of users with common characteristics. The paper presents the results of three case studies for three different types of Internet service: a digital library, an information broker and a Web site. Particular attention is paid on the different types of information access involved in the three case studies: query-based information retrieval, profile-based information filtering and Web-site navigation. Each type of access imposes different constraints on the representation of the learning task. Two different unsupervised learning methods are evaluated: conceptual clustering and cluster mining. One of our main concerns is the construction of meaningful communities that can be used for improving information access on the Internet. Analysis of the results in the three case studies brings to surface some of the important properties of the task, suggesting the feasibility of a common methodology for the three different types of information access on the Internet.

This paper presents a method that assists in maintaining a rule-based named-entity recognition and classification system. The underlying idea is to use a separate system, constructed with the use of machine learning, to monitor the performance of the rule-based system. The training data for the second system is generated with the use of the rule-based system, thus avoiding the need for manual tagging. The disagreement of the two systems acts as a signal for updating the rule-based system. The generality of the approach is illustrated by applying it to large corpora in two different languages: Greek and French. The results are very encouraging, showing that this alternative use of machine learning can assist significantly in the maintenance of rule-based systems.

The recognition of Proper Nouns (PNs) is considered an important task in the area of Information Retrieval and Extraction. However the high performance of most existing PN classifiers heavily depends upon the availability of large dictionaries of domain-specific Proper Nouns, and a certain amount of manual work for rule writing or manual tagging. Though it is not a heavy requirement to rely on some existing PN dictionary (often these resources are available on the web), its coverage of a domain corpus may be rather low, in absence of manual updating. In this paper we propose a technique for the automatic updating of an PN Dictionary through the cooperation of an inductive and a probabilistic classifier. In our experiments we show that, whenever an existing PN Dictionary allows the identification of 50% of the proper nouns within a corpus, our technique allows, without additional manual effort, the successful recognition of about 90% of the remaining 50%.