Search

While humans are used to reason about other humans’ behavior, they are not readily able to understand the decision processes of artificial agents. This can be harmful in human-robot interaction (HRI) settings where a user may suspect erroneous or, even worse, intentionally non-cooperative behavior, resulting in reduced acceptance of the robot. In order to mitigate such negative effects, autonomous robots may be equipped with the ability to adequately explain their behavior. To that end, a robot is required to have the ability to (1) robustly detect a user’s need for explanation and (2) identify the situation-specific nature of the explanation need. Further it needs to be endowed with (3) communicative capabilities in order to deliver suitable explanations and ensure sufficient understanding. This extended abstract presents recent work towards endowing a social robot with such qualities and discusses how robots can meet users’ explanation needs more adequately.

Humans constantly create narratives to provide explanations for how and why something happens. Designing systems able to build such narratives would therefore contribute to building more human-centric systems, and to support uses like decision-making processes. Here, a narrative is seen as a sequence of events. My thesis investigates how a narrative can be built computationally. Four research questions are identified: representation, construction, link prediction and evaluation. A case study on the French Revolution, based upon Wikidata and Wikipedia is presented. This prototype helps identifying the first challenges such as dynamic representation and evaluation of a narrative.

Ontology Alignment (OA) is generally performed by requesting two parties to provide their complete knowledge to a third party that suggests potential schema alignments. This might however not always be possible or helpful, as for example, when two organisations want to query each other’s knowledge, and none of them is willing to share their schema due to information privacy considerations. This Ph.D. explores how to allow multi-agent communication in cases where agents operate using different ontologies that cannot be fully exposed or shared. Our preliminary experiments focus on the case where agents’ knowledge is describing a common set of entities and has the form of Knowledge Graphs (KGs). The suggested methodology is based on the grounded naming game, where agents are forced to develop their own language in order to refer to corresponding schema concepts of different ontologies. This way, agents that use different ontologies can still communicate successfully for a task at hand, without revealing any private information. We have performed some proof of concept experiments applying our suggested method on artificial cases and we are on the process of extending our methodology so that it can be applied in real-world KGs.

A meta-analysis is a Science of Science method widely used in the medical and social sciences to review, aggregate and quantitatively synthesise a body of studies that address the same research question. With the volume of research growing exponentially every year, conducting meta-analyses can be costly and inefficient, as a significant amount of time and human efforts needs to be spent in finding studies meeting research criteria, annotating them, and properly performing the statistical analyses to summarise the findings. In this work, we show these issues can be tackled with semantic representations and technologies, using a social science scenario as case-study. We show how the domain-specific content of research outputs can be represented and used to facilitate their search, analysis and synthesis. We present the very first representation of the domain of human cooperation, and the application we built on top of this to help experts in performing meta-analyses semi-automatically. Using few application scenarios, we show how our approach supports the various phases meta-analyses, and more in general contributes towards research replication and automated hypotheses generation.

Health care professionals rely on treatment search engines to efficiently find adequate clinical trials and early access programs for their patients. However, doctors lose trust in the system if its underlying processes are unclear and unexplained. In this paper, a model-agnostic explainable method is developed to provide users with further information regarding the reasons why a clinical trial is retrieved in response to a query. To accomplish this, the engine generates features from clinical trials using by using a knowledge graph, clinical trial data and additional medical resources. and a crowd-sourcing methodology is used to determine their importance. Grounded on the proposed methodology, the rationale behind retrieving the clinical trials is explained in layman's terms so that healthcare processionals can effortlessly perceive them. In addition, we compute an explainability score for each of the retrieved items, according to which the items can be ranked. The experiments validated by medical professionals suggest that the proposed methodology induces trust in targeted as well as in non-targeted users, and provide them with reliable explanations and ranking of retrieved items.

Privacy assistants aim to provide meaningful privacy recommendations to users. Here, we describe a web-based testing environment for smartphone privacy assistants called the “Mock App Store” (MAS). The MAS was developed to test a particular privacy assistant – the value-centered privacy assistant (VcPA) – which assists users in selecting applications based on their value profile. While the MAS was designed with the VcPA in mind, it could also be utilized to test other state-of-the-art privacy assistant technology.

Shared vocabularies and ontologies are essential for many applications. Although standards and recommendations already cover many areas, adaptations are usually necessary to represent concrete use-cases properly. Domain experts are unfamiliar with ontology engineering, which creates special requirements for needed tool support. Simple sketch applications are usually too imprecise, while comprehensive ontology editors are often too complicated for non-experts. We present Neologism 2.0 – an open-source tool for quick vocabulary creation through domain experts. Its guided vocabulary creation and its collaborative graph editor enable the quick creation of proper vocabularies, even for non-experts, and dramatically reduces the time and effort to draft vocabularies collaboratively. An RDF export allows quick bootstrapping of any other Semantic Web tool.

Internet of Things (IoT) brings together heterogeneous data from smart devices in smart homes. Smart devices operate within different platforms, but ontologies can be used to create a common middle ground that allows communications between these smart devices outside of those platforms. The data communicated by the smart devices can be used to train the prediction algorithms used in forecasters. This research will first focus on the creation of a mapping to transform IoT data into a knowledge graph than can be used in the common middle ground and investigate the effect of using that IoT knowledge graph data as input for prediction algorithms. Experiments to determine the impact of incorporating other related information in the training of the prediction algorithms will be performed by using external datasources that can be linked to the knowledge graph and by using federated learning over IoT data from other smart homes. Initial results on the transformation mapping of IoT data to an ontology is presented.

This position paper presents an attempt to improve the scalability of existing object recognition methods, which largely rely on supervision and imply a huge availability of manually-labelled data points. Moreover, in the context of mobile robotics, data sets and experimental settings are often handcrafted based on the specific task the object recognition is aimed at, e.g. object grasping. In this work, we argue instead that publicly available open data such as ShapeNet can be used for object classification first, and then to link objects to their related concepts, leading to task-agnostic knowledge acquisition practices. To this aim, we evaluated five pipelines for object recognition, where target classes were all entities collected from ShapeNet and matching was based on: (i) shape-only features, (ii) RGB histogram comparison, (iii) a combination of shape and colour matching, (iv) image feature descriptors, and (v) inexact, normalised cross-correlation, resembling the Deep, Siamese-like NN architecture of Submariam et al. (2016). We discussed the relative impact of shape-derived and colour-derived features, as well as suitability of all tested solutions for future application to real-life use cases.

Query popularity is a main feature in web-search auto-completion. Several personalization features have been proposed to support specific users' searches, but often do not meet the privacy requirements of a medical environment (e.g. clinical trial search). Furthermore, in such specialized domains, the differences in user expertise and the domain-specific language users employ are far more widespread than in web-search. We propose a query auto-completion method based on different relevancy and diversity features, which can appropriately meet different user needs. Our method incorporates indirect popularity measures, along with graph topology and semantic features. An evolutionary algorithm optimizes relevance, diversity, and coverage to return a top-k list of query completions to the user. We evaluated our approach quantitatively and qualitatively using query log data from a clinical trial search engine, comparing the effects of different relevancy and diversity settings using domain experts. We found that syntax-based diversity has more impact on effectiveness and efficiency, graph-based diversity shows a more compact list of results, and relevancy the most effect on indicated preferences.

This volume contains the proceedings of the Poster and Demo Track of the 17th International Conference on Semantic Systems, SEMANTiCS 2021, and the Workshop on Digital Twins, which took place from September 6 - 9, 2021 in Amsterdam, The Netherlands. SEMANTiCS is the annual meeting place for professionals who make semantic computing work, who understand its benefits and encounter its limitations. Every year, SEMANTiCS attracts information managers, IT architects, software engineers and researchers from organisations ranging from research facilities, NPOs, through public administrations to the largest and/or most innovative companies in the world.

Household robots need semantics to understand that a detergent is a cleaning product that can be used to clean physical objects like a table, but laundry detergent is only used to clean/wash laundry. A safely acting autonomous robot should also know that both will not be used as ingredients for meal preparation. We propose a new approach to connect robot sensor data to Linked Data in order to give robotic agents semantic product information about objects that can be found in their environment so that the action to be performed with a given object can be inferred. For this, we use the robot’s belief state when recognizing a product and link it to a product ontology that follows Semantic Web standards. We then use the product class information to fetch further information from external sources like Wikidata or ConceptNet that contain action information (e.g. laundry detergent is used for laundering). At last, the action results are mapped to internally known actions of the robotic agent so that it knows which action can be performed with the perceived object.

In this paper, we tackle the problem of lack of understandability of deep learning systems by integrating heterogeneous knowledge sources, and in the specific we present how we used FrameNet to guarantee the correct learning for an LSTM-based semantic parser in the task of Spoken Language Understanding for robots. The problem of the explainability of Artificial Intelligence (AI) systems, i.e. their ability to explain decisions to both experts and end users, has attracted growing attention in the latest years, affecting their credibility and trustworthiness. Trusting these systems is fundamental in the context of AI-based robotic companions interacting in natural language, as the users’ acceptance of the robot also relies on the ability to explain the reasons behind its actions. Following similar approaches, we first use the values of the neural attention layers employed in the semantic parser as a clue to analyze and interpret the model’s behavior and reveal the intrinsic bias induced by the training data. We then show how the integration of knowledge from external resources such as FrameNet can help minimizing, or mitigating, such bias, and consequently guarantee the model to provide the correct interpretations. Our preliminary, but promising results suggest that (i) attention layers can improve the model understandability; (ii) the integration of different knowledge bases can help overcoming the limitations of machine learning models; and (iii) an approach combining the strengths of both knowledge engineering and machine learning can foster the development of more transparent, understandable intelligent systems.

In this paper, we tackle the problem of lack of understandability of deep learning systems by integrating heterogeneous knowledge sources, and in the specific we present how we used FrameNet to guarantee the correct learning for an LSTM-based semantic parser in the task of Spoken Language Understanding for robots. The problem of the explainability of Artificial Intelligence (AI) systems, i.e. their ability to explain decisions to both experts and end users, has attracted growing attention in the latest years, affecting their credibility and trustworthiness. Trusting these systems is fundamental in the context of AI-based robotic companions interacting in natural language, as the users’ acceptance of the robot also relies on the ability to explain the reasons behind its actions. Following similar approaches, we first use the values of the neural attention layers employed in the semantic parser as a clue to analyze and interpret the model’s behavior and reveal the intrinsic bias induced by the training data. We then show how the integration of knowledge from external resources such as FrameNet can help minimizing, or mitigating, such bias, and consequently guarantee the model to provide the correct interpretations. Our preliminary, but promising results suggest that (i) attention layers can improve the model understandability; (ii) the integration of different knowledge bases can help overcoming the limitations of machine learning models; and (iii) an approach combining the strengths of both knowledge engineering and machine learning can foster the development of more transparent, understandable intelligent systems.

To support the decision-making process in various industrial applications, many companies use knowledge management and Information Retrieval (IR). In an industrial setting, knowledge is extracted from data that is often stored in a semi-structured or unstructured format. As a result, Natural Language Processing (NLP) methods have been applied to a number of IR steps. In this work, we explore how NLP and particularly topic modelling can be used to improve the relevance of spare part retrieval in the context of maintenance services. A proposed methodology extracts topics from short maintenance service reports that also include part replacement data. An intuition behind the proposed methodology is that every topic should represent a specific root cause. Experimental were conducted for an ad-hoc retrieval system of service case descriptions and spare parts. The results have shown that our modification improves a baseline system thus boosting the performance of maintenance service solution recommendation.

A meta-analysis is a Science of Science method widely used in the medical and social sciences to review, aggregate and quantitatively synthesise a body of studies that address the same research question. With the volume of research growing exponentially every year, conducting meta-analyses can be costly and inefficient, as a significant amount of time and human efforts needs to be spent in finding studies meeting research criteria, annotating them, and properly performing the statistical analyses to summarise the findings. In this work, we show these issues can be tackled with semantic representations and technologies, using a social science scenario as case-study. We show how the domain-specific content of research outputs can be represented and used to facilitate their search, analysis and synthesis. We present the very first representation of the domain of human cooperation, and the application we built on top of this to help experts in performing meta-analyses semi-automatically. Using few application scenarios, we show how our approach supports the various phases meta-analyses, and more in general contributes towards research replication and automated hypotheses generation.

In the past year we have seen the release of a huge volume of open bibliographic citation data, thanks primarily to the efforts of the Initiative for Open Citations (I4OC, https://i4oc.org). However, the incomplete coverage of these data is one of the most important issues that the open scholarship community is currently facing. In this paper we present an approach for creating open citation data while supporting journal editors in their task of curating the reference lists contained within articles submitted to them for publication. Our contributions are twofold: (a) a basic workflow that supports editors in the management and correction of bibliographic references, and (b) a tool, called BCite, that creates open citation data compliant with an existing RDF-based citation repository: the Open Citations Corpus.

Design and Specification of the First Smart City Competition

We present our work-in-progress on handling temporal RDF graph data using the Ethereum distributed ledger. The motivation for this work are scenarios where multiple distributed consumers of streamed data may need or wish to verify that data has not been tampered with since it was generated – for example, if the data describes something which can be or has been sold, such as domestically-generated electricity. We describe a system in which temporal annotations, and information suitable to validate a given dataset, are stored on a distributed ledger, alongside the results of fixed SPARQL queries executed at the time of data storage. The model adopted implements a graph-based form of temporal RDF, in which time intervals are represented by named graphs corresponding to ledger entries. We conclude by discussing evaluation, what remains to be implemented, and future directions.

In this paper we exploit knowledge from Linked Data to ease the process of analysing scholarly data. In the last years, many techniques have been presented with the aim of analysing such data and revealing new, unrevealed knowledge, generally presented in the form of “patterns”. How-ever, the discovered patterns often still require human interpretation to be further exploited, which might be a time and energy consuming process. Our idea is that the knowledge shared within Linked Data can actuality help and ease the process of interpreting these patterns. In practice, we show how research communities obtained through standard network analytics techniques can be made more understand- able through exploiting the knowledge contained in Linked Data. To this end, we apply our system Dedalo that, by performing a simple Linked Data traversal, is able to automatically label clusters of words, corresponding to topics of the different communities.

Knowledge Discovery consists in discovering hidden regularities in large amounts of data using data mining techniques. The obtained patterns require an interpretation that is usually achieved using some background knowledge given by experts from several domains. On the other hand, the rise of Linked Data has increased the number of connected cross-disciplinary knowledge, in the form of RDF datasets, classes and relationships. Here we show how Linked Data can be used in an Inductive Logic Programming process, where they provide background knowledge for finding hypotheses regarding the unrevealed connections between items of a cluster. By using an example with clusters of books, we show how different Linked Data sources can be used to automatically generate rules giving an underlying explanation to such clusters.

Link traversal is one of the biggest advantages of Linked Data, as it allows the serendipitous discovery of new knowledge thanks to the natural connections between data of different sources. Our general problem is to understand how such a property can benefit the Knowledge Discovery process: in particular, we aim at using Linked Data to explain the patterns of data that have been extracted from a typical data min- ing process such as clustering. The strategy we propose here is Linked Data traversal, in which we explore and build on-the-fly an unknown Linked Data graph by simply deferencing entities’ URIs until we find, by following the links between entities, a valid explanation to our clusters. The experiments section gives an insight into the performance of such an approach, in terms of time and scalability, and show how the links easily gather knowledge from different data sources.