Search

BACKGROUND: Cladribine is a nucleoside analogue interfering with synthesis and repair of DNA. Treatment with cladribine leads to a preferential reduction in lymphocytes, resulting in profound depletion of B-cells with a rapid recovery of naïve B-cells, while T-cell show a lesser but long-lasting depletion It is approved for treatment of relapsing multiple sclerosis (MS). Cladribine tablets 3.5 mg/kg bodyweight are administered in two yearly treatment courses, each including two treatment series lasting 4 or 5 days, one at the start of the first month and the other at the start of the second month.OBJECTIVE: To describe treatment patterns of cladribine in a real-world setting.METHODS: Registry based observational cohort study with prospectively enrolled cases from December 2017 through June 2021. The data source is The Danish Multiple Sclerosis Registry, which is a near complete nationwide population-based registry. Outcomes were length of the treatment, preceding and following treatments, treatment response, and safety data.RESULTS: In total 268 patients had started therapy with cladribine tablets, 89 men and 179 women, with a median age of 40 years (interquartile range (IQR) 32-48. The disease course was relapsing-remitting MS in 97.8% of the patients, and at treatment start the median time from disease onset was 8.1 years (IQR 4.2-14.5) and EDSS 2.5 (IQR 1.5-3.5). Thirty-four patients (12.7%) were treatment naïve while 56 (20.9%) had received one previous disease-modifying therapy (DMT), 67 (25.0%) two, and 111 (41.4%) three or more previous DMTs. In total, 214 (80.0%) patients had completed the full treatment of two courses of cladribine, while 54 (20.0%) had received only one course of cladribine tablets. The median follow-up time after cladribine initiation was 34.7 months (IQR 23.3-43.7). Compared with an annualized relapse rate (ARR) of 0.67 (95% CI [0.56, 0.79]) in the year prior to start of cladribine, ARR was reduced to 0.11 (95% CI [0.08, 0.15]) in year 0-2 after 3-month re-baseline with cladribine (84.8% reduction). Adverse events, reported in 44 (16.4%) of the patients, were mild or moderate, and herpes zoster was only reported in 2 patients. In total, 30 (11.2%) patients discontinued cladribine treatment, of whom 7 (2.6%) discontinued because of adverse effects and 12 (4.5%) discontinued because of disease activity.CONCLUSION: In this nationwide review of all Danish patients starting therapy with cladribine tablets in a real-world setting, cladribine treatment was safe, and the therapeutic response was as expected from previous clinical trials. A prolonged observation period is necessary to assess the long-term benefit and risk of cladribine.

Background and purpose: Real-world evidence regarding the effectiveness and safety of ocrelizumab for the treatment of multiple sclerosis (MS) is limited. The aim was to evaluate the effectiveness and safety of ocrelizumab treatment for MS in a real-world setting. Methods: A nationwide population-based cohort study was conducted where clinical and magnetic resonance imaging data of MS patients enrolled prospectively in the Danish Multiple Sclerosis Registry who initiated ocrelizumab treatment between January 2018 and November 2020 were analyzed. Results: A total of 1104 patients (85.7% relapsing–remitting MS [RRMS], 8.8% secondary progressive MS [SPMS], 5.5% primary progressive MS [PPMS]) were included, with a median follow-up period of 1.3 years. At baseline, the mean age was 41.4 years in the RRMS group, 44.5 years in the PPMS group and 50.3 years in the SPMS group. Median Expanded Disability Status Scale score was 2.5, 3.5 and 5.5, respectively. Most RRMS and SPMS patients had received previous disease-modifying therapies (87.5% and 91.8%, respectively), whereas PPMS patients were mostly treatment naïve (78.7%). After ocrelizumab initiation, 9.3% of the patients experienced a relapse and 8.7% a 24 weeks confirmed disability worsening. Conversely, 16.7% showed a 24 weeks confirmed disability improvement. After ~1 year of treatment, most patients (94.5%) were free of magnetic resonance imaging activity. Ocrelizumab was generally well tolerated, as side effects were only reported for 10% of patients, mostly consisting of infusion-related reactions and infections. Conclusions: It is shown that most MS patients treated with ocrelizumab are clinically stabilized and with an adverse event profile consistent with the experience from the pivotal clinical trials.

Predicting the final outcome of an ongoing process instance is a key problem in many real-life contexts. This problem has been addressed mainly by discovering a prediction model by using traditional machine learning methods and, more recently, deep learning methods, exploiting the supervision coming from outcome-class labels associated with historical log traces. However, a supervised learning strategy is unsuitable for important application scenarios where the outcome labels are known only for a small fraction of log traces. In order to address these challenging scenarios, a semi-supervised learning approach is proposed here, which leverages a multi-target DNN model supporting both outcome prediction and the additional auxiliary task of next-activity prediction. The latter task helps the DNN model avoid spurious trace embeddings and overfitting behaviors. In extensive experimentation, this approach is shown to outperform both fully-supervised and semi-supervised discovery methods using similar DNN architectures across different real-life datasets and label-scarce settings.

The ever-increasing attention of process mining (PM) research to the logs of low structured processes and of non-process-aware systems (e.g., ERP, IoT systems) poses a number of challenges. Indeed, in such cases, the risk of obtaining low-quality results is rather high, and great effort is needed to carry out a PM project, most of which is usually spent in trying different ways to select and prepare the input data for PM tasks. Two general AI-based strategies are discussed in this paper, which can improve and ease the execution of PM tasks in such settings: (a) using explicit domain knowledge and (b) exploiting auxiliary AI tasks. After introducing some specific data quality issues that complicate the application of PM techniques in the above-mentioned settings, the paper illustrates these two strategies and the results of a systematic review of relevant literature on the topic. Finally, the paper presents a taxonomical scheme of the works reviewed and discusses some major trends, open issues and opportunities in this field of research.

Ticket Management Systems are widespread in disparate kinds of companies and organizations, as they represent a fundamental tool for handling customer requests and issues in an efficient and effective manner. In particular, accurately categorizing incoming tickets is a key task in real-life application settings (e.g., helpdesk/CRM systems and bug tracking systems), in order to improve ticket processing efficiency and effectiveness (e.g., in terms of customer satisfaction). In this work, we propose a comprehensive ticket-categorization analysis that relies on inducing and exploiting a heterogeneous ensemble of deep learning architectures, in addition to a range of functionalities for acquiring, integrating and pre-processing ticket-related information coming from different channels (e.g. mail, chat, web form, etc.). Experimental results conducted on the specific application scenario concerning the data of a publicly available ticket-mining dataset have proven the effectiveness of the framework in different ticket categorization tasks.

Traditionally, Expert Systems have found a natural application in the behavioral analysis of processes. In fact, they have proved effective in the tasks of interpreting the data collected during the process executions and of analyzing these data with the aim of diagnosing/detecting anomalies. In this context, we focus on log data generated by executions of business processes, and consider the issue of detecting “insecure” process instances, involving some kind of security breach (e.g. attacks, frauds). We propose a hybrid framework for accomplishing a security-oriented classification of activity-unawaretraces, i.e., traces consisting of “low-level” events with no explicit reference to the “high-level” activities the analysts are typically familiar with. The framework integrates two classification approaches traditionally used as alternative ways to decide on the “secureness” of process traces: (i) a model-driven approach, using knowledge of behavioral models expressed at the abstraction level of the activities, and (ii) an example driven approach, exploiting the availability of event sequences labeled by experts as symptomatic of “secure” or “insecure” behavior. The core of our solution is a meta-classifier combining (i) and (ii) thanks to a probabilistic Montecarlo mechanism that allows the traces to be simultaneously viewed as sequences of low-level events and of high-level activities. The framework has been empirically proved effective in jointly exploiting the two aforementioned forms of knowledge/expertise, typically coming from different experts, and in acting as a sort of “super-expert” classification tool. Its accuracy and efficiency make it a solid basis for implementing a novel kind of expert system for the security-oriented monitoring/analysis of business processes.

Process Mining (PM) is meant to extract knowledge on the behavior of business processes from historical log data. Lately, an increasing attention has been gained by the Predictive Process Monitor- ing, a field of PM that tries to extend process monitoring systems with prediction capabilities and, in particular. Several current proposals in literature to this problem rely on Deep Neural Networks, mostly on recurrent architectures (in particular LSTM) owing to their capability to capture the inherent sequential nature of process data. Very recently, however, an alternative solution based on a convolutional architecture (CNN) has been proposed in the literature, which was shown to achieve compelling results. Inspired by this line of research, we here propose a novel convolution-based deep learning approach to the prediction of the next activity which relies on: (i) extracting high-level features (at different levels of abstraction) through the computation of time-oriented dilated convolutions over traces, and (ii) exploiting residual-like connections to make the training of the predictive model more robust and faster. Preliminary results on real-life datasets confirm the validity of our proposal, compared with an LSTM-based and a CNN-based approach in the literature.

In social insects, cuticular hydrocarbons function in nest-mate recognition and also provide a waxy barrier against desiccation, but basic evolutionary features, including the heritability of hydrocarbon profiles and how they are shaped by natural selection are largely unknown. We used a new pharaoh ant ( Monomorium pharaonis ) laboratory mapping population to estimate the heritability of individual cuticular hydrocarbons, genetic correlations between hydrocarbons, and fitness consequences of phenotypic variation in the hydrocarbons. Individual hydrocarbons had low to moderate estimated heritability, indicating that some compounds provide more information about genetic relatedness and can also better respond to natural selection. Strong genetic correlations between compounds are likely to constrain independent evolutionary trajectories, which is expected, given that many hydrocarbons share biosynthetic pathways. Variation in cuticular hydrocarbons was associated with variation in colony productivity, with some hydrocarbons experiencing strong directional selection. Altogether, this study builds on our knowledge of the genetic architecture of the social insect hydrocarbon profile and indicates that hydrocarbon variation is shaped by natural selection.

Computer Science is a relatively young discipline, but in the last two decades the advances in hardware technology and software engineering has induced notable changes in the way users interact with computers. In particular, several processes involving data have changed in a radical manner. As a matter of fact, the amount of data stored in repositories has grown at impressive rates due to the rise of data sources, such as sensor networks, social networks or operational processes. Moreover, the heterogeneity of data has dramatically increased. In a word, data and their management have became more and more complex. Besides the data issues, new challenges have emerged from an application viewpoint, enabled by the informatization and digitalization of even the most trivial processes in the society. New applications span from real-time monitoring of dependable systems by means of cyber-physical sensors, to information diffusion at all levels by means of persistently connected devices, to the capability of tracking and delivering adhoc services to individuals.We're living in a connected world and the new technologies allow us with unforeseen possibilities. Along the recent years, new strategies (specifically, data loading, transformation and extraction, data management, and data mining approaches) have been proposed in order to deal with these unprecedented source of information. The advent of Big Data technologies has empowered the whole discipline of data analytics, by eventually creating even new professional figures which summarize skills in computer science and databases, statistics, machine learning and economics. However, the explosion of this new discipline is not new, and finds its path in the several contributions throughout the last 20 years. The Italian community has been particularly active in this scenario, by providing cutting-edge scientific results and prototypes in the field. As a proof of concept of this, this chapter tries to summarize some important contributions to the analysis of complex data both from the technological and the application point of view. In the following, we briefly recall the basic data types and methods that we have explored in the last years. In order to give the reader a more appealing view of the mentioned approaches, we also describe some applications that we have carried out with respect to some key important topics in the framework of the H2020 societal challenges.

Increasing amounts of data are becoming available in the form of "asynchronous" sequences of event records, associated each with a content and a temporal mark, e.g. sequences of activities on social media, clickstream data, user interaction logs, point- of-interest trajectories, business process logs, application logs and IoT logs, to name a few. Such a kind of data are more general than classic time series, as the lapse of time between consecutive events in a sequence may be an arbitrary continuous value. Usually, events in the same sequence exhibit hidden correlations (e.g., an event can cause or prevent the occurrence of certain kinds of events in the future). Generative models constitute a powerful and versatile means for analysing such data (e.g., by supporting variegate key tasks, such as data completion, data denoising, simulation analyses), as well as for enabling the very generation of new data instances (e.g., for preventing information leakage). In particular, if devised in a conditional fashion, these models can be exploited to predict which events will happen in the remainder of a given (unfinished) sequence and when, based on the sequence's history. Different kinds of neural generative models have been used in the last years for analysing sequence data, which range from Recurrent Neural Networks (RNNs), to Self-attention models, to more sophisticated frameworks like Variational Auto- encoders (VAEs) and Generative Adversarial Networks (GANs). In particular, roughly speaking, basic GAN frameworks implement a sort min-max game, where a "discrim- inator" sub-net is trained to distinguish real data instances from those produced by a "generator" sub-net, which is trained instead to fool the former sub-net. In principle, compared to previous solutions, GANs could yield models that are more general (owing to the GANs' capability to learn implicit data distributions) and more robust to "exposure bias" issues (i.e., to the risk of accumulating errors at inference time). However, they are notoriously difficult to train and tune optimally --in particular, they may converge slowly, and eventually reach an equilibrium that does not ensure a good enough generator (e.g., the latter may suffer from severe mode collapse issues). Moreover, the discrete nature of temporally-marked event sequences calls for extending traditional GAN schemes, as to prevent the problem of breaking the differentiability of the discriminators output w.r.t. the generators parameters (that would arise when sampling one event sequence from the generator's distribution).

The problem of inducing a model for forecasting the outcome of an ongoing process instance from historical log traces has attracted notable attention in the field of Process Mining. Approaches based on deep neural networks have become popular in this context, as a more effective alternative to previous feature- based outcome-prediction methods. However, these approaches rely on a pure supervised learning scheme, and unfit many real- life scenarios where the outcome of (fully unfolded) training traces must be provided by experts. Indeed, since in such a scenario only a small amount of labeled traces are usually given, there is a risk that an inaccurate or overfitting model is discovered. To overcome these issues, a novel outcome-discovery approach is proposed here, which leverages a fine-tuning strategy that learns general-enough trace representations from unlabelled log traces, which are then reused (and adapted) in the discovery of the outcome predictor. Results on real-life data confirmed that our proposal makes a more effective and robust solution for label- scarcity scenarios than current outcome-prediction methods

Intrusion detection systems have to cope with many challenging problems, such as unbalanced datasets, fast data streams and frequent changes in the nature of the attacks (concept drift). To this aim, here, a distributed genetic programming (GP) tool is used to generate the combiner function of an ensemble; this tool does not need a heavy additional training phase, once the classifiers composing the ensemble have been trained, and it can hence answer quickly to concept drifts, also in the case of fast-changing data streams. The above-described approach is integrated into a novel cybersecurity framework for classifying non stationary and unbalanced data streams. The framework provides mechanisms for detecting drifts and for replacing classifiers, which permits to build the ensemble in an incremental way. Tests conducted on real data have shown that the framework is effective in both detecting attacks and reacting quickly to concept drifts.

Modern intrusion detection systems must be able to discover new types of attacks in real-time. To this aim, automatic or semi-automatic techniques can be used; outlier detection algorithms are particularly apt to this task, as they can work in an unsupervised way. However, due to the different nature and behavior of the attacks, the performance of different outlier detection algorithms varies largely. In this ongoing work, we describe an approach aimed at understanding whether an ensemble of outlier algorithms can be used to detect effectively new types of attacks in intrusion detection systems. In particular, Genetic Programming (GP) is adopted to build the combining function of an ensemble of local and global outlier detection algorithms, which are used to detect different types of attack. Preliminary experiments, conducted on the well-known NSL-KDD dataset, are encouraging and confirm that, depending on the type of attacks, it would be better to use only local or only global detection algorithms and that the GP-based ensemble improves the performance in comparison with commonly used combining functions.

In many application contexts, a business process' executions are subject to performance constraints expressed in an aggregated form, usually over predefined time windows, and detecting a likely violation to such a constraint in advance could help undertake corrective measures for preventing it. This paper illustrates a prediction-aware event processing framework that addresses the problem of estimating whether the process instances of a given (unfinished) window w will violate an aggregate performance constraint, based on the continuous learning and application of an ensemble of models, capable each of making and integrating two kinds of predictions: single-instance predictions concerning the ongoing process instances of w, and time-series predictions concerning the "future" process instances of w (i.e. those that have not started yet, but will start by the end of w). Notably, the framework can continuously update the ensemble, fully exploiting the raw event data produced by the process under monitoring, suitably lifted to an adequate level of abstraction. The framework has been validated against historical event data coming from real-life business processes, showing promising results in terms of both accuracy and efficiency.

The discovery of a control-flow model for a process is here faced in a challenging scenario where each trace in the given log \(L_E\) encodes a sequence of low-level events without referring to the process’ activities. To this end, we define a framework for inducing a process model that describes the process’ behavior in terms of both activities and events, in order to effectively support the analysts (who typically would find more convenient to reason at the abstraction level of the activities than at that of low-level events). The proposed framework is based on modeling the generation of \(L_E\) with a suitable Hidden Markov Model (HMM), from which statistics on precedence relationships between the hidden activities that triggered the events reported in \(L_E\) are retrieved. These statistics are passed to the well-known Heuristics Miner algorithm, in order to produce a model of the process at the abstraction level of activities. The process model is eventually augmented with probabilistic information on the mapping between activities and events, encoded in the discovered HMM. The framework is formalized and experimentally validated in the case that activities are “atomic” (i.e., an activity instance triggers a unique event), and several variants and extensions (including the case of “composite” activities) are discussed.

Log analysis and querying recently received a renewed interest from the research community, as the effective understanding of process behavior is crucial for improving business process management. Indeed, currently available log querying tools are not completely satisfactory, especially from the viewpoint of easiness of use. As a matter of fact, there is no framework which meets the requirements of easiness of use, flexibility and efficiency of query evaluation. In this paper, we propose a framework for graphical querying of (process) log data that makes the log analysis task quite easy and efficient, adopting a very general model of process log data which guarantees a high level of flexibility. We implemented our framework by using a flexible storage architecture and a user-friendly data analysis interface, based on an intuitive and yet expressive graph-based query language. Moreover, we exploit our results as a basis for a better and dynamic definition of process cube structures. Experiments performed on real data confirm the validity of the approach.

We consider the scenario where the executions of different business processes are traced into a log, where each trace describes a process instance as a sequence of low-level events representing basic kinds of operations. In this context, we address a novel problem: given a description of the processes' behaviors in terms of high-level activities instead of low-level events, and in the presence of uncertainty in the mapping between events and activities, find all the interpretations of each trace $$\Phi $$. Specifically, an interpretation is a pair $$\langle \sigma , W \rangle $$ that provides a two-level "explanation" for $$\Phi $$: $$\sigma $$ is a sequence of activities that may have triggered the events in $$\Phi $$, and W is a process whose model admits $$\sigma $$. To solve this problem, we propose a probabilistic framework representing "consistent" $$\Phi $$'s interpretations, where each interpretation is associated with a probability score.

The issue of devising efficient and effective solutions for supporting the analysis of process logs has recently received great attention from the research community, as effectively accomplishing any business process management task requires understanding the behavior of the processes. In this paper, we propose a new framework supporting the analysis of process logs, exhibiting two main features: a flexible data model (enabling an exhaustive representation of the facets of the business processes that are typically of interest for the analysis) and a graphical query language, providing a user-friendly tool for easily expressing both selection and aggregate queries over the business processes and the activities they are composed of. The framework can be easily and efficiently implemented by leveraging either “traditional” relational DBMSs or “innovative” NoSQL DBMSs, such as Neo4J.

Despite performance-oriented process mining techniques have been successfully applied in numerous application contexts, when applied to processes featuring complex and heterogeneous behaviors, they hardly produce models with a satisfactory level of accuracy, generality, and readability. In particular, the presence of deviant (i.e. anomalous/exceptional) traces often lead to cumbersome models with misleading performance statistics. Noise/outlier filtering solutions allows to alleviate this problem, and to discover a better model for "normal" executions, but do not provide insight on the nature and impact of deviant ones. The process discovery approach proposed here tries to recognize and describe both a normal execution scenario and a number of deviant ones for the analyzed log, by inducing two different kinds of models: (i) a list of readable clustering rules defining the deviance scenarios; (ii) a performance model for each discovered deviance scenario, and a "distilled" one for the "nor- mal" cases that do not fall in any deviant scenario. Technically, these models are discovered by mainly exploiting a conceptual clustering method, that greedily tries to separate groups of traces that maximally deviate from the current normality model. Tests on real-life logs confirmed the validity of the approach, and its ability to both find good performance models and support the analysis of deviant process instances.

Process mining methods have been proven effective in turning historical log data into actionable process knowledge. However, most of them work under the assumption that the events reported in the logs can be easily mapped to well-defined process activities, that are the terms in which analysts are used to reason on the processes’ behaviors. We here consider the challenging scenario where this assumption does not hold: the log traces are sequences of low-level operations with no explicit reference to the corresponding high-level process activities. In this setting, we face the fundamental problem of bringing the log traces to the abstraction level of the analyst’s perspective. Formally, given a trace Φ, and on the basis of a high-level behavioral description of the processes, we search for every possible interpretation⟨σ, W⟩ of Φ, where σ is a sequence of high-level activities whose execution may have generated the sequence of low-level operations Φ, and, in turn, W is a process that may have triggered the execution of σ. We address this problem probabilistically, and propose a framework that builds a compact representation of Φ’s interpretations, each associated with a probability score. This probability measures how likely the associated interpretation is the correct one, and it is evaluated adopting a revision paradigm guided by the background knowledge provided by the processes’ models. Notably, our approach can deal with “complex” activities (i.e., each generating a sequence of low-level operations, rather than a single one), and with the case that the traces encode process instances exhibiting some deviation from the expected behaviors encoded in the process models.

Increasing attention has been paid to the detection and analysis of “deviant” instances of a business process that are connected with some kind of “hidden” undesired behavior (e.g. frauds and faults). In particular, several recent works faced the problem of inducing a binary classification model (here named deviance detection model ) that can discriminate between deviant traces and normal ones, based on a set of historical log traces (labeled as either deviant or normal). Current solutions rely on applying standard classifier-induction methods to a feature-based representation of the given traces, where the features include sequence-based patterns extracted from the corresponding sequences of activities. However, there is no consensus on which kinds of patterns are the most suitable for such a task. On the other hand, mixing multiple pattern families together may produce a heterogenous, redundant and sparse representation of the traces that likely leads to poor deviance detection models. In this paper, we propose an ensemble-learning method for solving this problem, where multiple base classifiers are trained on different feature-based views of the log (each obtained by mapping the traces onto a distinguished collection of patterns). A stacking procedure is used to combine the discovered base models into an overall probabilistic model that associates any new trace with an estimate of the probability that it reflects a deviant process instance. This helps the analyst prioritize the inspection of the cases that are more likely to be deviant. The method also takes advantage of all nonstructural data available in the log, and employs a resampling mechanism to deal with the rarity of deviances in the training log. It has been conceived as the core of a comprehensive framework for detecting and analyzing business process deviances. The framework supports the analyst to investigate suspect deviances, and provides some feedback to the learning method for improving the accuracy of the discovered deviance detection models. Tests on several real-life datasets proved the validity of the approach, as concerns its capability to discover an accurate deviance detection model, and to effectively exploit new (originally unlabeled) traces via active learning and self-training mechanisms.

The execution logs of a business process have been recently exploited to extract classification models for discriminating “deviant” instances of the process - i.e. instances diverging from normal/desired outcomes (e.g., frauds, faults, SLA violations). Regarding all log traces as sequences of task labels, current solutions essentially map each trace onto a vector space where the features correspond to sequence-oriented patterns, and any standard classifier-induction method can be applied to separate the two classes of instances. An ensemble-learning approach was also recently proposed to combine multiple base learners trained on heterogenous pattern-based log views. However, as these approaches simply abstract each event into an activity symbol, they disregard all the non structural event data that are typically stored in real-life logs, and which may well help improve the detection of deviances. Moreover, the usefulness of deviance models could be enhanced by equipping each prediction with a confidence measure, allowing the analyst to focus on (or prioritize) more suspicious cases. To overcome these limitations, we propose a multi-view ensemble learning approach, which: (i) fully exploits the multi-dimensional nature of log events, with the help of a clustering-based trace abstraction method; and (ii) implements a context- and probability-aware stacking method for combining base models' predictions. Tests on a real-life log confirmed the validity of the approach, and its capability to achieve compelling performances w.r.t. state-of-the-art methods.

Process discovery has emerged as a powerful approach to support the analysis and the design of complex processes. It consists of analyzing a set of traces registering the sequence of tasks performed along several enactments of a transactional system, in order to build a process model that can explain all the episodes recorded over them. An approach to accomplish this task is presented that can benefit from the background knowledge that, in many cases, is available to the analysts taking care of the process (re-)design. The approach is based on encoding the information gathered from the log and the (possibly) given background knowledge in terms of precedence constraints , that is, of constraints over the topology of the resulting process models. Mining algorithms are eventually formulated in terms of reasoning problems over precedence constraints, and the computational complexity of such problems is thoroughly analyzed by tracing their tractability frontier. Solution algorithms are proposed and their properties analyzed. These algorithms have been implemented in a prototype system, and results of a thorough experimental activity are discussed.

Monitoring the performances of a business process is a key issue in many organizations, especially when predefined constraints exist on them, due to contracts or internal require- ments. Several approaches were defined recently in the literature for predicting the performances of a single process instance. However, in many real situations, process-oriented performance metrics and associated constraints are defined in an aggregated form, on a time-window basis. This work right addresses the problem of predicting whether (the process instances in) each time window will infringe an aggregate performance constraint, at a series of checkpoints within the window. To this end, at each checkpoint, three kinds of measures are to be estimated: what performance outcome each ongoing process instance will yield, how many process instances will start in the rest of the window, and what their aggregate performance outcomes will be. The approach proposed is general (it can reuse a wide range of regression methods), and it can be embedded in a continuous monitoring-and-learning scheme. Tests on real-life logs showed its validity in terms of prediction accuracy.

The ability of social insects to discriminate nestmates (NMs) from non-nestmates (nNMs) is mainly achieved through chemical communication. To ultimately understand this recognition and its decision rules, identification of the recognition cues is essential. Although recognition cues are most likely cuticular hydrocarbons (CHCs), identifying the exact cues for specific species has remained a daunting task, partly due to the sheer number of odor compounds. Perhaps unsurprisingly, one of the few species where the recognition cues have been identified, Formica exsecta, has only around ten major hydrocarbons on its cuticle. In this study we use previous results of this species to search for nestmate recognition cues (NMR cues) in two other species of ants, Camponotus aethiops, and Monomorium pharaonis. Employing chemical distances and observed aggression between colonies, we first ask which type of data normalization, centroid, and distance calculation is most diagnostic to discriminate between NMR cues and other compounds. We find that using a “global centroid” instead of a “colony centroid” significantly improves the analysis. One reason may be that this new approach, unlike previous ones, provides a biologically meaningful way to quantify the chemical distances between NMs, allowing for within-colony variation in recognition cues. Next, we ask which subset of hydrocarbons most likely represents the cues that the ants use for nestmate recognition, which shows less clear results for C. aethiops and M. pharaonis than for F. exsecta, possibly due to less than ideal datasets. Nonetheless, some compound sets performed better than others, showing that this approach can be used to identify candidate compounds to be tested in bio-assays, and eventually crack the sophisticated code that governs nestmate recognition. ispartof: Frontiers in Ecology and Evolution vol:2 status: published

Process Mining techniques have been gaining attention, especially as concerns the discovery of predictive process models. Traditionally focused on workflows, they usually assume that process tasks are clearly specified, and referred to in the logs. This limits however their application to many real-life BPM environments (e.g. issue tracking systems) where the traced events do not match any predefined task, but yet keep lots of context data. In order to make the usage of predictive process mining to such logs more effective and easier, we devise a new approach, combining the discovery of different execution scenarios with the automatic abstraction of log events. The approach has been integrated in a prototype system, supporting the discovery, evaluation and reuse of predictive process models. Tests on real-life data show that the approach achieves compelling prediction accuracy w.r.t. state-of-the-art methods, and finds interesting activities' and process variants' descriptions. © 2014 Springer International Publishing.

Because of the widespread diffusion of semistructured data in XML format, much research effort is currently devoted to support the storage and retrieval of large collections of such documents. XML documents can be compared as to their structural similarity, in order to group them into clusters so that different storage, retrieval, and processing techniques can be effectively exploited. In this scenario, an efficient and effective similarity function is the key of a successful data management process. We present an approach for detecting structural similarity between XML documents which significantly differs from standard methods based on graph-matching algorithms, and allows a significant reduction of the required computation costs. Our proposal roughly consists of linearizing the structure of each XML document, by representing it as a numerical sequence and, then, comparing such sequences through the analysis of their frequencies. First, some basic strategies for encoding a document are proposed, which can focus on diverse structural facets. Moreover, the theory of discrete Fourier transform is exploited to effectively and efficiently compare the encoded documents (i.e., signals) in the domain of frequencies. Experimental results reveal the effectiveness of the approach, also in comparison with standard methods.

Predicting run-time performances is a hot issue in ticket resolution processes. Recent efforts to take account for the sequence of resolution steps, suggest that predictive Process Mining (PM) techniques could be applied in this field, if suitably adapted to the peculiarities of ticket systems. In particular, the performances of a ticket instance usually depend on which kinds of experts worked on it (more than on the mere sequence of resolution tasks), while relevant information about ticket cases is stored in the form of text fields, which are usually disregarded by PM approaches. Instead of relying on a-priori experts groups, we devise an ad-hoc method for clustering experts according to their real working patterns, based on log data. Regarding the discovered groups as abstractions for log events, we also perform a predictive clustering of ticket cases, while using context data as input attributes for splitting the tickets. In this way, different (context-dependent) execution scenarios are recognized for the process, and equipped with more accurate performance predictors. The approach was validated on a real application scenario, where it showed better results than state-of-the-art solutions.

A key task in process mining consists of building a graph of causal dependencies over process activities, which can then help derive more expressive models in some high-level modeling language. An approach to accomplishing this task is presented, where the learning process can exploit background knowledge available to the analyst. The method is based on encoding the information gathered from the log and the (possibly) given background knowledge in terms of precedence constraints, i.e., constraints over the topology of the graphs. Learning algorithms are formulated in terms of reasoning problems over precedence constraints. The whole approach has been implemented in a prototype system and results of experimental activity are reported.

This paper presents a novel approach to the discovery of predictive process models, which are meant to support the run-time prediction of some performance indicator (e.g., the remaining processing time) on new ongoing process instances. To this purpose, we combine a series of data mining techniques (ranging from pattern mining, to non-parametric regression and to predictive clustering) with ad-hoc data transformation and abstraction mechanisms. As a result, a modular representation of the process is obtained, where different performancerelevant variants of it are provided with separate regression models, and discriminated on the basis of context information. Notably, the approach is capable to look at the given log traces at a proper level of abstraction, in a pretty automatic and transparent fashion, which reduces the need for heavy intervention by the analyst (which is, indeed, a major drawback of previous solutions in the literature). The approach has been validated on a real application scenario, with satisfactory results, in terms of both prediction accuracy and robustness.

A key task in process mining consists of building a graph of causal dependencies over process activities, which can then be used to derive more expressive models in some high-level modeling language. An approach to accomplish this task is presented where the learning process can exploit the background knowledge that, in many cases, is available to the analysts taking care of the process (re-)design. The method is based on encoding the information gathered from the log and the (possibly) given background knowledge in terms of precedence constraints, i.e., constraints over the topology of the graphs. Learning algorithms are eventually formulated in terms of reasoning problems over precedence constraints, and the computational complexity of such problems is thoroughly analyzed by tracing their tractability frontier. The whole approach has been implemented in a prototype system leveraging a solid constraint programming platform, and results of experimental activity are reported. © 2012 The Author(s).

Process-oriented systems have been increasingly attracting data mining researchers, mainly due to the advantages that the application of inductive process mining techniques to log data could open to both the analysis of complex processes and the design of new process models. However, the actual impact of process mining in the industry is endangered by some simplifying assumptions these techniques relies on. In fact, current approaches have still problems to mine models over languages that allow for complex constructs, e.g., duplicate tasks, hidden tasks, non-free-choice constructs, and/or when noise is admitted in the log. In this paper, some advances to facing these problems are made, by proposing an algorithm which can deal with duplicate and hidden tasks, as well as with the presence of noise and non-free choice relationships among process activities. Importantly, due to the local nature of the search strategy exploited by the algorithm, the proposed approach seems suited to scale in real-world application scenarios.

Process Mining techniques exploit the information stored in the execution log of a process to extract some high-level process model, useful for analysis or design tasks. Most of these techniques focus on “structural” aspects of the process, in that they only consider what elementary activities were executed and in which ordering. Hence, any other “non-structural” data, usually kept in real log systems (e.g., activity executors, parameter values), are disregarded, yet being a potential source of knowledge. In this paper, we overcome this limitation by proposing a novel approach to the discovery of process models, where the behavior of a process is characterized from both structural and non-structural viewpoints. Basically, we recognize different executions’ classes via a structural clustering approach, and model them with a collection of specific workflows. Relevant correlations between these classes and non-structural properties are captured by a rule-based classification model, which can be used for both explanation and prediction. In order to empower the versatility of our approach, we also combine it with a pre-processing method, which allows to restructure the log events according to different analysis perspectives, and to study them at the right abstraction level. Interestingly, such an approach reduces the risk of obtaining knotty, ”spaghetti-like”, process models when analyzing the logs of loosely-structured processes consisting of low-level operations that are performed in a more autonomous way than in traditional BPM platforms. Preliminary results on real-life application scenario confirm the validity of the approach.

Process mining techniques have been receiving great attention in the literature for their ability to automatically support process (re)design. Typically, these techniques discover a concrete workflow schema modelling all possible execution patterns registered in a given log, which can be exploited subsequently to support further-coming enactments. In this paper, an approach to process mining is introduced that extends classical discovery mechanisms by means of an abstraction method aimed at producing a taxonomy of workflow models. The taxonomy is built to capture the process behavior at different levels of detail. Indeed, the most-detailed mined models, i.e., the leafs of the taxonomy, are meant to support the design of concrete workflows, as it happens with existing techniques in the literature. The other models, i.e., non-leaf nodes of the taxonomy, represent instead abstract views over the process behavior that can be used to support advanced monitoring and analysis tasks. All the techniques discussed in the paper have been implemented, tested, and made available as a plugin for a popular process mining framework (ProM). A series of tests, performed on different synthesized and real datasets, evidenced the capability of the approach to characterize the behavior encoded in input logs in a precise and complete way, achieving compelling conformance results even in the presence of complex behavior and noisy data. Moreover, encouraging results have been obtained in a real-life application scenario, where it is shown how the taxonomical view of the process can effectively support an explorative ex-post analysis, hinged on the different kinds of process execution discovered from the logs.

In this paper we propose an architecture that exploit web pages stuctural information for the extraction of relevant information from them. In this architecture, a primary role played by a distance-based classification methodology is devised. Such a methodology is based on an efficient and effective technique for detecting structural similarities among semistructured documents, which significantly differs from standard methods based on graph-matching algorithms. The technique is based on the idea of representing the structure of a document as a time series in which each occurrence of a tag corresponds to a given impulse. By analyzing the frequencies of the corresponding Fourier transform, we can hence state the degree of similarity between documents. Experiments on real data show the effectiveness of the proposed technique.

The high-order co-clustering problem, i.e., the problem of simultaneously clustering several heterogeneous types of domains, is usually faced by minimizing a linear combination of some optimization functions evaluated over pairs of correlated domains, where each weight expresses the reliability/relevance of the associated contingency table. Clearly enough, accurately choosing these weights is crucial to the effectiveness of the co-clustering, and techniques for their automatic tuning are particularly desirable, which are instead missing in the literature. This paper faces this issue by proposing an information-theoretic framework where the co-clustering problem does not need any explicit weighting scheme for combining pairwise objective functions, while a suitable notion of agreement among these functions is exploited. Based on this notion, an algorithm for co-clustering a “star-structured” collection of domains is defined.

Constrained graphs are directed graphs describing the control flow of processes models. In such graphs, nodes represent activities involved in the process, and edges the precedence relationship among such activities. Typically, nodes and edges can specify some constraints, which control the interaction among the activities. Faced with the above features constrained graphs are widely used in the modelling and analysis of Workflow processes. In this paper we overview two mining problems related to the analysis of constrained graphs, namely the analysis of frequent patterns of execution, and the induction of a constrained graph from a set of execution traces. We discuss some complexity aspects related to the problem of reasoning and mining on constrained graphs, and overview two algorithms for the mentioned problems.

We propose an incremental algorithm for discovering clusters of duplicate tuples in large databases. The core of the approach is the usage of an indexing technique which, for any newly arrived tuple mu, allows to efficiently retrieve a set of tuples in the database which are mostly similar to P, and which are likely to refer to the same real-world entity which is associated with mu. The proposed index is based on a hashing approach which tends to assign similar objects to the same buckets. Empirical and analytical evaluation demonstrates that the proposed approach achieves satisfactory efficiency results, at the cost of low accuracy loss.

In this paper we investigate on the exploitation of data mining techniques to analyze data coming from the enactment of workflow-based processes in a service-oriented Grid infrastructure. The extracted knowledge allows users to better comprehend the behavior of the enacted processes, and can be profitably exploited to provide advanced support to several phases in the life-cycle of workflow processes, including (re-)design, matchmaking, scheduling and performance monitoring. To this purpose, we focus on recent data mining techniques specifically aimed at enabling re- fined analyzes of workflow executions. Moreover, we introduce a comprehensive system architecture that supports the management of Grid workflows by fully taking advantage of such mining techniques.