Search

Scientific knowledge has been traditionally disseminated and preserved through research articles published in journals, conference proceedings, and online archives. However, this article-centric paradigm has been often criticized for not allowing to automatically process, categorize, and reason on this knowledge. An alternative vision is to generate a semantically rich and interlinked description of the content of research publications. In this paper, we present the Artificial Intelligence Knowledge Graph (AI-KG), a large-scale automatically generated knowledge graph that describes 820K research entities. AI-KG includes about 14M RDF triples and 1.2M reified statements extracted from 333K research publications in the field of AI, and describes 5 types of entities (tasks, methods, metrics, materials, others) linked by 27 relations. AI-KG has been designed to support a variety of intelligent services for analyzing and making sense of research dynamics, supporting researchers in their daily job, and helping to inform decision-making in funding bodies and research policymakers. AI-KG has been generated by applying an automatic pipeline that extracts entities and relationships using three tools: DyGIE++, Stanford CoreNLP, and the CSO Classifier. It then integrates and filters the resulting triples using a combination of deep learning and semantic technologies in order to produce a high-quality knowledge graph. This pipeline was evaluated on a manually crafted gold standard, yielding competitive results. AI-KG is available under CC BY 4.0 and can be downloaded as a dump or queried via a SPARQL endpoint.

Online educational platforms are enabling learners to consume a great variety of content and share opinions on their learning experience. The analysis of the sentiment behind such a collective intelligence represents a key element for supporting both instructors and learning institutions on shaping the offered educational experience. Combining Word Embedding representations and deep learning architectures has made possible to design sentiment analysis systems able to accurately measure the text polarity on several contexts. However, the application of such representations and architectures on educational data still appears limited. Therefore, considering the over-sensitiveness of the emerging models to the context where the training data is collected, conducting adaptation processes that target the e-learning context becomes crucial to unlock the full potential of a model. In this chapter, we describe a deep learning approach that, starting from Word Embedding representations, measures the sentiment polarity of textual reviews posted by learners after attending online courses. Then, we demonstrate how Word Embeddings trained on smaller e-learning-specific resources are more effective with respect to those trained on bigger general-purpose resources. Moreover, we show the benefits achieved by combining Word Embeddings representations with deep learning architectures instead of common machine learning models. We expect that this chapter will help stakeholders to get a clear view and shape the future research on this field.

Background Networks whose nodes have labels can seem complex. Fortunately, many have substructures that occur often (“motifs”). A societal example of a motif might be a household. Replacing such motifs by named supernodes reduces the complexity of the network and can bring out insightful features. Doing so repeatedly may give hints about higher level structures of the network. We call this recursive process Recursive Supernode Extraction. Results This paper describes algorithms and a tool to discover disjoint (i.e. non-overlapping) motifs in a network, replacing those motifs by new nodes, and then recursing. We show applications in food-web and protein-protein interaction (PPI) networks where our methods reduce the complexity of the network and yield insights. Conclusions SuperNoder is a web-based and standalone tool which enables the simplification of big graphs based on the reduction of high frequency motifs. It applies various strategies for identifying disjoint motifs with the goal of enhancing the understandability of networks.

Social media are providing the humus for the sharing of knowledge and experiences and the growth of community activities (e.g., debating about different topics). The analysis of the user-generated content in this area usually relies on Sentiment Analysis. Word embeddings and Deep Learning have attracted extensive attention in various sentiment detection tasks. In parallel, the literature exposed the drawbacks of traditional approaches when content belonging to specific contexts is processed with general techniques. Thus, ad-hoc solutions are needed to improve the effectiveness of such systems. In this paper, we focus on user-generated content coming from the e-learning context to demonstrate how distributional semantic approaches trained on smaller context-specific textual resources are more effective with respect to approaches trained on bigger general-purpose ones. To this end, we build context-trained embeddings from online course reviews using state-of-the-art generators. Then, those embeddings are integrated in a deep neural network we designed to solve a polarity detection task on reviews in the e-learning context, modeled as a regression. By applying our approach on embeddings trained using background corpora from different contexts, we show that the performance is better when the background context is aligned with the regression context.

In this paper we propose a Natural Language Processing engine that allows the NAO humanoid robot to execute natural language actions spoken by the user. To this aim we created an ontology that describes body parts, actions, and incompatibilities between actions. The system can work in two modes: stateless and stateful. In stateless mode, the robot returns to its default position after each action. In stateful mode, it performs the actions sequentially and may refuse a command if incompatible with the robot current state. Our system handles compound and multiple expressions that the robot understands and performs.

During the last decades, a huge amount of data have been collected in clinical databases in the form of medical reports, laboratory results, treatment plans, etc., representing patients health status. Hence, digital information available for patient-oriented decision making has increased drastically but it is often not mined and analyzed in depth since: (i) medical documents are often unstructured and therefore difficult to analyze automatically, (ii) doctors traditionally rely on their experience to recognize an illness, give a diagnosis, and prescribe medications. However doctors experience can be limited by the cases they are treated so far and medication errors can occur frequently. In addition, it is generally hard and time-consuming inferring information for comparing unstructured data and evaluating similarities between heterogeneous resources. Technologies as Data Mining, Natural Language Processing, and Machine Learning can provide possibilities to explore and exploit potential knowledge from diagnosis history records and help doctors to prescribe medication correctly to decrease medication error effectively. In this paper, we design and implement a medical recommender system that is able to cluster a collection of medical reports on features detected by IBM Watson and Framester, two emerging tools from, respectively, Cognitive Computing and Frame Semantics, and then, giving a medical report from a specific patient as input, to recommend similar other medical reports from patients who had analogues symptoms. Experiments and results have proved the quality of the resulting clustering and recommendations, and the key role that these innovative services can play on the biomedical sector. The proposed system is able to classify new medical cases thus supporting physicians to take more correct and reliable actions about specific diagnosis and cares.

With the proliferation in number and scale of online courses, several challenges have emerged in supporting stakeholders during their delivery and fruition. Machine Learning and Semantic Analysis can add value to the underlying online environments in order to overcome a subset of such challenges (e.g. classification, retrieval, and recommendation). However, conducting reproducible experiments in these applications is still an open problem due to the lack of available datasets in Technology-Enhanced Learning (TEL), mostly small and local. In this paper, we propose COCO, a novel semantic-enriched collection including over 43 K online courses at scale, 16 K instructors and 2,5 M learners who provided 4,5 M ratings and 1,2 M comments in total. This outruns existing TEL datasets in terms of scale, completeness, and comprehensiveness. Besides describing the collection procedure and the dataset structure, we depict and analyze two potential use cases as meaningful examples of the large variety of multi-disciplinary studies made possible by having COCO.

Complex network analysis is rising as an essential tool to understand properties of ecological landscape networks, and as an aid to land management. The most common methods to build graph models of ecological networks are based on representing functional connectivity with respect to a target species. This has provided good results, but the lack of a model able to capture general properties of the network may be seen as a shortcoming when the activity involves the proposal for modifications in land use. Similarity scores, calculated between nature protection areas, may act as a building block for a graph model intended to carry a higher degree of generality. The present work compares several design choices for similarity-based graphs, in order to determine which is most suitable for use in land management.

Complex network analysis is being applied on topological models of ecological networks, to extrapolate their advanced properties and as part of the activity of land management. Commonly employed methods tend to focus on single target species. This is satisfactory for cognitive analysis, but the limited view provided by these models results in a lack of general information needed for land planning. Similarity scores computed for pairs of nature protection areas are proposed as a building block of a general model to address this shortcoming.

International audience; The article discusses the process of versioning for ontologies and semantic artefacts developed using Semantic Web technologies. We describe methods for encoding versioning and other relevant information in metadata properties and we illustrate with examples from the MOD2.0 specification. Building on our experiences with the AgroPortal ontology repository and the Linked Open Vocabulary, we raise several questions, such as which metadata properties to use, how metadata values should be coordinated, and what stay the same over versions, and what should change. We propose recommendations for better versioning ontologies with clear semantics-identifiers, descriptions, status, dates, links-and suggests that the recommendations can be generalized to any digital objects that need to be versioned and semantically described.