The widespread deployment of machine learning systems in critical real-world decision-making applications has highlighted the urgent need for counterfactual explainability methods that operate effectively. Global counterfactual explanations, expressed as actions to offer recourse, aim to provide succinct explanations and insights applicable to large population subgroups. High effectiveness, measured by the fraction of the population that is provided recourse, ensures that the actions benefit as many individuals as possible. Keeping the cost of actions low ensures the proposed recourse actions remain practical and actionable. Limiting the number of actions that provide global counterfactuals is essential to maximize interpretability. The primary challenge, therefore, is to balance these trade-offs—maximizing effectiveness, minimizing cost, while maintaining a small number of actions. We introduce GLANCE, a versatile and adaptive algorithm that employs a novel agglomerative approach, jointly considering both the feature space and the space of counterfactual actions, thereby accounting for the distribution of points in a way that aligns with the model's structure. This design enables the careful balancing of the trade-offs among the three key objectives, with the size objective functioning as a tunable parameter to keep the actions few and easy to interpret. Our extensive experimental evaluation demonstrates that GLANCE consistently shows greater robustness and performance compared to existing methods across various datasets and models.

The rapid integration of machine learning (ML) into critical decisionmaking systems has heightened concerns over fairness, particularly regarding spatial biases often tied to sensitive socioeconomic factors. In response, we propose a model-agnostic post-processing method for spatial bias mitigation that operates without access to the original training data. Our approach formulates an optimization problem that minimizes a fairness measure robust to gerrymandering, subject to a constraint specifying the allowable deviation from the original model's performance ensuring spatial fairness while preserving accuracy. This measure has a 0–1 scale, offering an intuitive way to quantify spatial bias. Comprehensive evaluations on real-world datasets show that our framework effectively reduces spatial bias and achieves fairer outcomes with minimal performance loss, outperforming other state-of-the-art post-processing methods. This work advances spatial fairness methodologies, offering practitioners an efficient, interpretable, and adaptable post-processing solution to mitigate location-based discrimination in ML applications.

In most time-dependent real world phenomena, the distribution of data changes through time. This effect poses challenges to the training of time series Machine Learning models and their prediction results. Algorithms that address this problem, train the machine learning model on a subsequence of the input time series containing recent samples of the distribution, in order to capture the latest state of the distribution. However, estimating the optimal size for the training subsequence is not straightforward. State-of-the-art methods use change point detection algorithms to detect the change points of the distribution and set the most recent change point as the start of the training subsequence. In this paper, we show that this approach is not optimal. Instead, we present an algorithm that retrieves the training subsequence which minimizes the prediction error of the machine learning model. We evaluate our algorithm on two synthetic and four real-world datasets and show that it significantly increases the accuracy of the models.

Over the last decade, the use of Machine/Deep learning algorithms and methodologies has found widespread application across various domains in wildfire science, with fire occurrence risk prediction being one of the extensively covered areas. Many algorithms and architectures have been explored to solve the problem as a binary classification task focusing mainly on the model’s classification performance through the standard classification metrics. However, in the context of an application for predicting the fire occurrence risk we have to enhance the trust in the model’s output by deriving a well defined fire susceptibility index, as well as interpretable insights for the model’s predictions when applied to real-world datasets. In this manuscript we propose model agnostic solutions towards (a) the production of a probability-based binning of the model’s prediction distribution for generating a scalar wildfire occurrence risk and (b) of a representative sampling for producing local prediction explanations spanning the whole dataset, to tackle the slow processing times of explainable AI frameworks, when generating vast amounts of local interpretations. The objective is to contribute the development of a reliable application predicting next day’s fire risk, facilitating the adoption of ML-based solutions by users who are not experts in data science.

In this work, we present Fairness Aware Counterfactuals for Subgroups (FACTS), a framework for auditing subgroup fairness through counterfactual explanations. We start with revisiting (and generalizing) existing notions and introducing new, more refined notions of subgroup fairness. We aim to (a) formulate different aspects of the difficulty of individuals in certain subgroups to achieve recourse, i.e. receive the desired outcome, either at the micro level, considering members of the subgroup individually, or at the macro level, considering the subgroup as a whole, and (b) introduce notions of subgroup fairness that are robust, if not totally oblivious, to the cost of achieving recourse. We accompany these notions with an efficient, model-agnostic, highly parameterizable, and explainable framework for evaluating subgroup fairness. We demonstrate the advantages, the wide applicability, and the efficiency of our approach through a thorough experimental evaluation on different benchmark datasets.

With the lowering costs of sensors, high-volume and high-velocity data are increasingly being generated and analyzed, especially in IoT domains like energy and smart homes. Consequently, applications that require accurate short-term forecasts and predictions are also steadily increasing. In this paper, we provide an overview of a novel end-to-end platform that provides efficient ingestion, compression, transfer, query processing, and machine learning-based analytics for high-frequency and high-volume time series from IoT. The performance of the platform is evaluated using real-world dataset from RES installations. The results show the importance of high-frequency analytics and the surprisingly positive impact of error bounded lossy compression on machine learning in the form of AutoML. For example, when detecting yaw misalignments in wind turbines, an improvement of 9% in accuracy was observed for AutoML models on lossy compressed data compared to the current industry standard of 10-minute aggregated data. Thus, these small-scale experiments show the potential of the platform, and larger pilots are planned.

The paper gives an account of an infrastructure that will be integrated into a platform aimed at providing a multi-faceted experience to visitors of Northern Greece using mythology as a starting point. This infrastructure comprises a multi-lingual and multi-modal corpus (i.e., a corpus of textual data supplemented with images, and video) that belongs to the humanities domain along with a dedicated database (content management system) with advanced indexing, linking and search functionalities. We will present the corpus itself focusing on the content, the methodology adopted for its development, and the steps taken towards rendering it accessible via the database in a way that also facilitates useful visualizations. In this context, we tried to address three main challenges: (a) to add a novel annotation layer, namely geotagging, (b) to ensure the long-term maintenance of and accessibility to the highly heterogeneous primary data – even after the life cycle of the current project – by adopting a metadata schema that is compatible to existing standards; and (c) to render the corpus a useful resource to scholarly research in the digital humanities by adding a minimum set of linguistic annotations.

In this paper, we handle the problem of next day wildfire prediction via the use of machine learning. In contrast to most works in the relevant literature, we set the problem to its realistic basis, with respect to its large scale, the extreme imbalance in the data distribution, the required high spatial granularity of the predictions and the consideration of the strong spatial correlations inherent in the data. We implement a machine learning workflow that exploits Tree Ensemble and Neural Network algorithms, upon which an extensive hyperparameter search procedure is performed, via cross-validation, in order to select a set of effective models that are expected to generalize well on new data. Our experiments on the whole Greek territory demonstrate the effectiveness of the proposed methodology, rendering it directly applicable to real-world scenarios. Finally, several insights towards further improving the effectiveness of current models are discussed.

Explanations for algorithmically generated recommendations is an important requirement for transparent and trustworthy recommender systems. When the internal recommendation model is not inherently interpretable (e.g., most contemporary systems are complex and opaque), or when access to the system is not available (e.g., recommendation as a service), explanations have to be generated post-hoc, i.e., after the system is trained. In this common setting, the standard approach is to provide plausible interpretations of the observed outputs of the system, e.g., by building a simple surrogate model that is inherently interpretable, and explaining that model. This however has several drawbacks. First, such explanations are not truthful, as they are rationalizations of the observed inputs and outputs constructed by another system. Second, there are privacy concerns, as to train a surrogate model, one has to know the interactions from users other than the one who seeks an explanation. Third, such explanations may not be scrutable and actionable, as they typically return weights for items or other users that are difficult to comprehend, and hard to act upon so to improve the quality of one’s recommendations.

In this paper, we introduce an approach for improving the quality of the geocoding process. Geocoding refers to the procedure of mapping an address of textual form to a pair of accurate spatial coordinates. While there is a variety of available geocoders, both open source and commercial, that curate this mapping in either a semi-automated or fully-automated way, there is no one-size-fits-all system. Depending on the underlying algorithm of each geocoder, its output may be very accurate for some addresses, districts or countries, while failing to properly locate some others. Given that, our setup can be thought of as a meta-geocoding pipeline, built on top of the available geocoders. We propose a machine learning approach, which, given an address and a sequence of coordinate pairs suggested by standalone geocoders, it is able to identify the most accurate one. In order to achieve this, we formulate the task as a multi-class classification problem and introduce a series of domain specific training features, capturing essential information about each coordinate pair suggestion, as well as computing comparative metrics among different suggestions. These features are fed into several classification algorithms and are evaluated on a proprietary address dataset of a geo-marketing company. Furthermore, we present LGM-GC, a QGIS plugin, which provides the functionality of our approach through a user-friendly interface.

An important task on land registration processes is to be able to determine the prevalent data provenance for a finalized polygon that represents a cadastral parcel, since the finalized polygon is derived by the examination of a set of initial polygons, drawn from several individual registers (databases). These registers might contain different, partially similar or conflicting information regarding the ownership, usage and polygon geometry of a cadastral parcel. In such cases, the cadastration expert either select one of of the initial geometries, or (in cases none of the initial accurately represents the finalized land parcel) creates a new geometry. Maintaining this provenance information is of high importance for further cadastration and validation/quality assessment processes; however, due to the gradual and long lasting nature of cadastration procedures, this information is absent from large parts of cadastral databases. In this paper, we present an approach for effectively classifying such land parcel polygons with respect to their provenance information. We propose a method that can produce highly accurate provenance recommendations based only on attributes derived from the geometry of a land parcel. In particular, we implement a set of spatial training features, capturing polygon properties and relations. These features are fed into several classification algorithms and are evaluated on a proprietary dataset of a cadastration company.

Interlinking spatio-textual data comprises a core problem within the research literature, as well as a task of high practical importance in a plethora of industrial applications involving GIS systems. In its general form, it consists in identifying, between two sources of spatio-texual entities, pairs of entities that match, i.e. correspond to the same real-world entities. In this paper, we focus on interlinking spatio-textual entities based solely on their name, that is we handle the problem of toponym interlinking. To solve the problem, works in the literature exploit generic string similarity measures and either apply them as is, or integrate them as training features in classification models, without adapting/extending them based on the specific characteristics of toponyms. In this work, we showcase that domain knowledge can significantly improve the accuracy of toponym interlinking, by proposing domain specific similarity measures that take into account specificities of toponyms. We assess the implemented measures on Geonames and demonstrate significant increases in interlinking accuracy compared to baseline methods.

In this demonstration, we present LGM-PC, a QGIS plugin for automatic recommendation of categories for new POIs. LGM-PC allows users to train classification models on individual areas of POIs and, then, use these models in order to classify new POIs into categories. The tool produces ranked category recommendations, based solely on the name of the POI, its coordinates and properties of its surrounding POIs, which are already annotated with categories. The user is then required to validate the produced recommendations, by selecting the most fitting category. Being implemented as a QGIS plugin, LGM-PC allows the visualization of POIs on map layers, for further assisting the user in the final category selection task.

State of the art works for Point-Of-Interest (POI) classification use either traditional feature extraction methods, or, more recently, deep learning (DL) methods, in order to train classification models on historical data, i.e. POIs already annotated with categories, and then deploy these models to classify new, unannotated POIs. These methods are either inherently limited to learning simple, rather than complex, relationships, or learn embeddings trivially and without taking into account domain knowledge or semantic information, thus yielding disappointing classification results. In this vision paper, we discuss the problem of POI classification, identify limitations in the state of the art and propose novel research directions. We propose a framework for learning meaningful context embeddings for POIs, that incorporate domain knowledge and semantic information extracted from external data sources, and we discuss how the proposed approach can overcome inherent limitations of the state of the art. Specifically, we prescribe the construction of fine-grained embeddings, by refining the consideration of spatial neighborhoods of POIs, assessing and selecting the proper attributes to be used as context and target information in the embedding learning process, as well as by properly enriching these embeddings with external, semantic knowledge. Additionally, we advocate the joint utilization of traditional training features and embedding-derived features for POI classification, and argue towards the usefulness of the proposed approach.

Point of Interest (POI) data is widely used in many modern applications and services related to navigation, tourism, social networking, logistics, and many more. In this paper, we propose a comprehensive and vendor-agnostic data model to represent multi-faceted and enriched POI profiles. Harnessing the versatility of Linked Data technologies, this semantically rich ontology accommodates and extends existing POI formats for assembling and managing POI data from heterogeneous sources. Furthermore, we have developed the open-source software TripleGeo, which can effectively transform POI data from diverse sources and formats (geographical files, databases, and semi-structured data) to their RDF representations and vice versa. Thus, it is possible to import POI data from various existing systems and products, transfer and address the data integration challenges in the Linked Data domain, and export back the results. Our empirical study confirms the validity and efficiency of this framework for a variety of real-world POI assets and formats, underscoring its robustness to cope with scalable data volumes.

An immense wealth of data is already accessible through the Semantic Web and an increasing part of it also has geospatial context or relevance. Although existing technology is mature enough to integrate a variety of information from heterogeneous sources into interlinked features, it still falls behind when it comes to representation and reasoning on spatial characteristics. It is only lately that several RDF stores have begun to accommodate geospatial entities and to enable some kind of processing on them. To address interoperability, the OGC has recently adopted the GeoSPARQL standard, which defines a vocabulary for representing geometric types in RDF and an extension to the SPARQL language for formulating queries. In this paper, we provide a comprehensive review of the current state-of-the-art in geospatially-enabled semantic data management. Apart from an insightful analysis of the available architectures in industry and academia, we conduct an evaluation study on prominent RDF stores with geospatial support. We also compare their performance and attested capabilities to renowned DBMSs widely used in geospatial applications. We introduce a methodology suitable to assess RDF stores for robustness against large geospatial datasets, and also for expressiveness on a variety of queries involving both spatial and thematic criteria. As our findings demonstrate, the potential for query optimization, advanced indexing schemes, and spatio-semantic extensions is significant. Towards this goal, we point out several challenging issues for joint research by the GIS and Semantic Web communities.

Personalized retrieval models aim at capturing user interests to provide personalized results that are tailored to the respective information needs. User interests are however widely spread, subject to change, and cannot always be captured well, thus rendering the deployment of personalized models challenging. We take a different approach and study ranking models for user intent. We exploit user feedback in terms of click data to cluster ranking models for historic queries according to user behavior and intent. Each cluster is finally represented by a single ranking model that captures the contained search interests expressed by users. Once new queries are issued, these are mapped to the clustering and the retrieval process diversifies possible intents by combining relevant ranking functions. Empirical evidence shows that our approach significantly outperforms baseline approaches on a large corporate query log.

BACKGROUND: MicroRNAs are small endogenously expressed non-coding RNA molecules that regulate target gene expression through translation repression or messenger RNA degradation. MicroRNA regulation is performed through pairing of the microRNA to sites in the messenger RNA of protein coding genes. Since experimental identification of miRNA target genes poses difficulties, computational microRNA target prediction is one of the key means in deciphering the role of microRNAs in development and disease. RESULTS: DIANA-microT 3.0 is an algorithm for microRNA target prediction which is based on several parameters calculated individually for each microRNA and combines conserved and non-conserved microRNA recognition elements into a final prediction score, which correlates with protein production fold change. Specifically, for each predicted interaction the program reports a signal to noise ratio and a precision score which can be used as an indication of the false positive rate of the prediction. CONCLUSION: Recently, several computational target prediction programs were benchmarked based on a set of microRNA target genes identified by the pSILAC method. In this assessment DIANA-microT 3.0 was found to achieve the highest precision among the most widely used microRNA target prediction programs reaching approximately 66%. The DIANA-microT 3.0 prediction results are available online in a user friendly web server at http://www.microrna.gr/microT.