Nanopublications LDF server

Matches in Nanopublications for { ?s <http://purl.org/spar/c4o/hasContent> ?o ?g. }

• paragraph hasContent "Table 5 describes the analysis of signal (a) over the 3 frames that are mapped to DBPO properties. For each property and dataset, we computed the amount of available assertions and reported the gain relative to the fact extraction datasets. Although we considered the whole Italian DBpedia KB in these calculations, we observe that it has a generally low coverage with respect to the analysed properties, probably due to missing ontology mappings. For instance, the amount of assertions is always zero if we analyse the use case subset only, as no specific relevant mappings (e.g., Carriera_sportivo 29 to careerStation) currently exist. We view this as a major achievement, since our automatic approach also serves as a substitute for the manual mapping procedure." assertion.
• paragraph hasContent "Table 6 shows the results for signal (b). To obtain them, we proceeded as follows." assertion.
• paragraph hasContent "The A-Box enrichment is clearly visible from the results, given the low overlap and high gain in all approaches, despite the rather large size of the DBpedia use case subset, namely 6, 167, 678 assertions." assertion.
• section-11.3-title hasContent "A-Box Population" assertion.
• section-11-title hasContent "Evaluation" assertion.
• paragraph hasContent "First, we acknowledge that the use case frame repository is not exhaustive: LUs may have a higher ambiguity. For instance, giocare (to play) may trigger an additional frame depending on the context (as in the sentence to play as a defender); esordire (to start out) may also trigger the frame P ARTITA (match). Second, if a sentence is not in the gold standard, the supervised classifier should discard it (abstention). Third, the baseline approach may contain rules that are more harmful than beneficial, depending on the target KB reliability: for instance, the SportsEvent DBPO class leads to wrongly typed instances, due to the misuse of the template by Wikipedia editors. Finally, both the input corpus and the target KB originate from a relatively small Wikipedia chapter (i.e., Italian, with 1.23 million articles) if compared to the largest one (i.e., English, with almost 5 million articles). Therefore, we recognize that the T-Box and A-Box evaluation results may be proportionally different if obtained with English data." assertion.
• paragraph hasContent "Our approach has been tested on the Italian language, a specific domain, and with a small frame repository. Hence, we may consider the use case implementation as a monolingual closed-domain information extraction system. We outline below the points that need to be ad- dressed for scaling up to multilingual open information extraction:" assertion.
• section-12.1-title hasContent "Scaling Up" assertion.
• paragraph hasContent "We report below a list of technical improvements left for planned implementation:" assertion.
• section-12.2-title hasContent "Technical Future Work" assertion.
• section-12-title hasContent "Observations" assertion.
• paragraph hasContent "We locate our effort at the intersection of the following research areas:" assertion.
• paragraph hasContent "Although the borders are blurred, nowadays we can distinguish two principal fields in Information Extraction, namely Relation Extraction (RE) and Open In- formation Extraction (OIE). While both aim at structuring information in the form of relations between items, their difference relies in the relations set size, either fixed or potentially infinite. It is commonly argued that the main OIE drawback is the generation of noisy data [11,32], while RE is usually more accurate, but requires expensive supervision in terms of language resources [2,30,32]." assertion.
• paragraph hasContent "RE traditionally takes as input a finite set R of relations and a document d, and induces assertions in the form rel(subj, obj), where rel represent binary relations between a subject entity subj and an object entity obj mentioned in d. Hence, it may be viewed as a closed-domain paradigm. Recent efforts [3,2,30] have focused on alleviating the cost of full supervision via distant supervision. Distant supervision leverages available KBs to automatically annotate training data in the input documents. This is in contrast to our work, since we aim at enriching the target KB with external data, rather than using it as a source. Furthermore, our relatively cheap crowdsourcing technique serves as a substitute to distant supervision, while ensuring full supervision. Other approaches such as [7,33] instead leverage text that is not covered by the target KB, like we do." assertion.
• section-13.1.1-title hasContent "Relation Extraction" assertion.
• paragraph hasContent "OIE is defined as a function f (d) over a document d, yielding a set of triples (np 1 , rel, np 2 ), where nps are noun phrases and rel is a relation between them. Known complete systems include OLLIE [25], REVERB [14], and NELL [9]. Recently, it has been discussed that cross-utterance processing can improve the performance through logical entailments [1]. This paradigm is called “open” since it is not constrained by any schemata, but rather attempts to learn them from unstructured data. In addition, it takes as input heterogeneous sources of information, typically from the Web." assertion.
• paragraph hasContent "In general, most efforts have focused on English, due to the high availability of language resources. Approaches such as [15] explore multilingual directions, by leveraging English as a source and applying statistical machine translation (SMT) for scaling up to target languages. Although the authors claim that their approach do not directly depend on language resources, we argue that SMT still heavily relies on them. Furthermore, all the above efforts concentrate on binary relations, while we generate n-ary ones: under this perspective, EXEMPLAR [10] is a rule-based system which is closely related to ours." assertion.
• section-13.1.2-title hasContent "Open Information Extraction" assertion.
• section-13.1-title hasContent "Information Extraction" assertion.
• paragraph hasContent "DBPEDIA [23], FREEBASE [8] and YAGO [21] represent the most mature approaches for automatically building KBs from Wikipedia. Despite its crowd- sourced nature (i.e., fully manual), WIKIDATA [31] benefits from a rapidly growing community of active users, who have developed several robots for automatic imports of Wikipedia and third-party data. The KNOWLEDGE VAULT [11] is an example of KB construction combining Web-scale textual corpora, as well as additional semi-structured Web data such as HTML tables. Although our system may potentially create a KB from scratch from an input corpus, we prefer to improve the quality of existing resources and integrate into them, rather than developing a standalone one." assertion.
• section-13.2-title hasContent "Knowledge Base Construction" assertion.
• paragraph hasContent "OIE output can indeed be considered structured data compared to free text, but it still lacks of a disambiguation facility: extracted facts generally do not employ unique identifiers (i.e., URIs), thus suffering from intrinsic natural language polysemy (e.g., Jaguar may correspond to the animal or a known car brand). To tackle the issue, [12] propose a framework that clusters OIE facts and maps them to elements of a target KB. Similarly to us, they leverage EL techniques for disambiguation and choose DBpedia as the target KB. Nevertheless, the authors focus on A-Box population, while we also cater for the T-Box part. Moreover, OIE systems are used as a black boxes, in contrast to our full implementation of the extraction pipeline. Finally, relations are still binary, instead of our n-ary ones. Taking as input Wikipedia articles, L EGALO [28] exploits page links manually inserted by editors and attempts to induce the relations between them via NLP. Again, the extracted relations are binary and are not mapped to a target KB for enrichment purposes." assertion.
• section-13.3-title hasContent "Open Information Semantification" assertion.
• section-13-title hasContent "Related Work" assertion.
• paragraph hasContent "In a Web where the profusion of unstructured data limits its automatic interpretation, the necessity of Intelligent Web-reading Agents turns more and more evident. These agents should preferably be conceived to browse an extensive and variegated amount of Web sources corpora, harvest structured assertions out of them, and finally cater for target KBs, which can attenuate the problem of information overload. As a support to such vision, we have outlined two real-world scenarios involving general-purpose KBs:" assertion.
• paragraph hasContent "In this article, we presented a system that puts into practice our fourfold research contribution: first, we perform (1) N-ary relation extraction thanks to the implementation of Frame Semantics, in contrast to traditional binary approaches; second, we (2) jointly enrich both the T-Box and the A-Box parts of our target KB, through the discovery of candidate relations and the extraction of facts respectively. We achieve this with a (3) shallow layer of NLP technology only, namely grammatical analysis, instead of more sophisticated ones, such as syntactic parsing. Finally, we ensure a (4) fully supervised learning paradigm via an affordable crowdsourcing methodology. Our work concurrently bears the advantages and leaves out the weaknesses of RE and OIE: although we assess it in a closed-domain fashion via a use case (Section 3), the corpus analysis module (Section 5) allows to discover an exhaustive set of relations in an open-domain way. In addition, we overcome the supervision cost bottleneck trough crowdsourcing. Therefore, we believe our approach can represent a trade-off between open-domain high noise and closed-domain high cost." assertion.
• paragraph hasContent "The FACT EXTRACTOR is a full-fledged Information Extraction NLP pipeline that analyses a natural language textual corpus and generates structured machine-readable assertions. Such assertions are disambiguated by linking text fragments to entity URIs of the target KB, namely DBpedia, and are assigned a confidence score. For instance, given the sentence Buffon plays for Serie A club Juventus since 2001, our system produces the following dataset: @prefix dbpedia: <http://it.dbpedia.org/resource/> . @prefix dbpo: <http://dbpedia.org/ontology/> . @prefix fact: <http://fact.extraction.org/> . @prefix xsd: <http://www.w3.org/2001/XMLSchema#> . dbpedia:Gianluigi_Buffon dbpo:careerStation dbpedia:CareerStation_01 . dbpedia:CareerStation_01 dbpo:team dbpedia:Juventus_Football_Club ; fact:competition dbpedia:Serie_A ; dbpo:startYear "2001"^^xsd:gYear ; fact:confidence "0.906549"^^xsd:float ." assertion.
• paragraph hasContent "We estimate the validity of our approach by means of a use case in a specific domain and language, i.e., soccer and Italian. Out of roughly 52,000 Italian Wikipedia articles describing soccer players, we output more than 213, 000 triples with an average 78.5% F1. Since our focus is the improvement of existing resources rather than the development of a standalone one, we integrated these results into the ITALIAN DBPEDIA CHAPTER 30 and made them accessible through its SPARQL endpoint. Moreover, the codebase is publicly available as part of the DBPEDIA ASSOCIATION repository. 31" assertion.
• paragraph hasContent "We have started to expand our approach under the Wikidata umbrella, where we feed the primary sources tool. The community is currently concerned by the trustworthiness of Wikidata assertions: in order to authenticate them, they should be validated against references to external Web sources. Under this perspective, the FACT EXTRACTOR can serve as a reference suggestion mechanism for statement validation. To achieve this, we have successfully managed to switch the input corpus from Wikipedia to third-party corpora and translated our output to fit the Wikidata data model. The soccer use case has already been partially implemented: we have ran the baseline classifier and generated a small demonstra- tive dataset, named FBK-STREPHIT-SOCCER, which has been uploaded to the primary sources tool back-end. We invite the reader to play with it, by following the instructions in the project page. 32" assertion.
• paragraph hasContent "For future work, we foresee to scale up the imple- mentation towards multilingual open information extraction, thus paving the way to (a) its full deployment into the DBpedia Extraction Framework, and to (b) a thorough referencing system for Wikidata." assertion.
• section-14-title hasContent "Conclusion" assertion.
• paragraph hasContent "The FACT EXTRACTOR has been developed within the DBPEDIA ASSOCIATION and was partially funded by GOOGLE under the SUMMER OF CODE 2015 program." assertion.
• paragraph hasContent "Table 5" assertion.
• paragraph hasContent "Table 6" assertion.
• paragraph hasContent "Figure 1" assertion.
• paragraph hasContent "Figure 2" assertion.
• paragraph hasContent "Figure 3" assertion.
• paragraph hasContent "Figure 4" assertion.
• paragraph hasContent "Figure 5" assertion.
• paragraph hasContent "Figure 6" assertion.
• paragraph hasContent "Figure 7" assertion.
• paragraph hasContent "Figure 8" assertion.
• paragraph hasContent "Table 2" assertion.
• paragraph hasContent "Table 4" assertion.
• abstract hasContent "The contextual information in the built environment is highly heterogeneous, it goes from static information (e.g., information about the building structure) to dynamic information (e.g., user’s space-time information, sensors detections and events that occurred). This paper proposes to semantically fuse the building contextual information with data coming from a smart camera network by using ontologies and semantic web technologies. The ontology developed allows interoperability between the different contextual data and enables, without human interaction, real-time event detections to be performed and system reconfigurations. The use of semantic knowledge in multi-camera monitoring systems guarantees the protection of the user’s privacy by not sending nor saving any image, just extracting the knowledge from them. This paper presents a new approach to develop a "all-seeing" smart building, where the global system is the first step to attempt to provide Artificial Intelligence (AI) to a building. More details of the system and future works can be found at the following website: http://wisenet.checksem.fr/ ." assertion.
• paragraph hasContent "In Greek mythology, Argus Panoptes was a giant with a hundred eyes. It was impossible to deceive his vigilance, for only some of his eyes slept while the rest were awake. Argus was the servant of Hera. At his death, Hera rewarded the giant’s fidelity by trans- ferring his eyes to the feathers of the peacock, his favorite animal. "To have the eyes of Argus" is a popular expression which means to be lucid and vigilant. The term Panoptes means "all-seeing". Within the built en- vironment the term "all-seeing" is a quest in terms of access control, flow control and activities. In that context, a Panoptes building would characterize a smart building equipped with a network of cameras which could, in real-time, combine the different information seen and deduce the triggering of actions. In classical multi-camera based systems there is a monitor room with a central processing server where all the information is collected and analyzed in real-time by a human operator (or a set of them). However, as the size of the network increases, it becomes more difficult (or even impossible) for the human operator to monitor all the video streams at the same time and to identify events. Furthermore, having a large amount of infor- mation makes it infeasible to create a relation between actions that happened in the past and current actions." assertion.
• paragraph hasContent "Based on our experience, some issues and limitations of multi-camera based system deployed in built environments have been identified, such as:" assertion.
• paragraph hasContent "Many efforts have being devoted to deal with the aforesaid limitations of the multi-camera based sys- tem. The most prominent one is to rely on smart cameras to perform visual tasks semi-autonomously (with minimal human interaction). Smart cameras are specialized cameras that contain not only the image sen- sor but also a processing unit and some communication interfaces. In a few words, smart cameras are a self-contained vision systems [24,42]. The use of smart cameras in the built environment has become a growing trend due to the rich contextual data provided." assertion.
• paragraph hasContent "In the built environment, context is an essential factor since it provides information about the current status of users, places, objects, sensors and events. We assume that a smart building is a context-aware system because it extracts, interprets and uses the contextual information to automatically adapt its functionality according to the contextual changes." assertion.
• paragraph hasContent "A Panoptes building is a type of smart building that uses only smart cameras sensors and its main purpose is monitoring the different activities that occur in the built environment; in contrast to the smart building which uses different types of sensors and which mainly focuses on managing/monitoring the energy consumption. The creation of a Panoptes building is a complicate task due to the integration of data coming from different domains around the knowledge of the building composition. Many works have been done using semantic web standards such as Resource Description Framework (RDF) and Web Ontology Language (OWL) to represent contextual data [16]. On those systems, the ontology plays a crucial role in enabling the processing and sharing of information and knowledge, i.e., the use of an ontology allows interoperability between different domains." assertion.
• paragraph hasContent "This paper proposes an ontology for a Panoptes building that repurposes and integrates information about different domains composing the built context. The proposed ontology is the kernel of the WiseNET (Wise NETwork) system, which is a context-aware system whose main function is to perform reasoning about heterogeneous sources of information. Explicitly, the WiseNET system enhances the information of a smart camera network (SCN) with contextual information to allow autonomously real-time event/anomalies detection and system reconfiguration. The main contribution of this paper is the semantic fusion of Industry Foundations Classes (IFC) data with sensor information and other domain information in the Panoptes context, by using semantic web technologies. A semantic-based system is presented as well, which allows to overcome the multi-camera based system limitations and some computer vision problems, specially the privacy protection which nowadays is an important factor to consider.." assertion.
• paragraph hasContent "Given these insights, RQ3 was formulated: How can we design better crowdsourcing workflows using lay users or experts for curating LD sets, beyond one-step solutions for pointing out quality flaws? To do so, we adapted the crowdsourcing pattern known as Find-Fix- Verify, which has been originally proposed by Bern- stein et al. in [3]. Specifically, we wanted to know: can (i) we enhance the results of the LD quality issue de- tection through lay users by adding a subsequent step of cross-checking (Verify) to the initial Find stage? Or is it (ii) even more promising to combine experts and lay workers by letting the latter Verify the results of the experts’ Find step, hence drawing on the crowds’ complementary skills for deficiency identification we recognized before?”" assertion.
• section-introduction-title hasContent "Introduction" assertion.
• paragraph hasContent "Nowadays, multi-camera based systems have become a part of our daily life. They can be found in cities, commercial centers, supermarkets, offices, and even in houses. The advances in image sensor technology allow us to have smart cameras, which are low-cost and low-power systems that capture high-level description of a scene and analyze it in real-time [42]. These smart cameras can extract necessary/pertinent information from different images/video by employing different image processing algorithms such as face detection [47], person detection [8], people tracking [13], fall detection [35], object detection [33], etc." assertion.
• paragraph hasContent "Smart camera networks have been used in the built environment for a long time. The main applications focus on the following problematics:" assertion.
• paragraph hasContent "Most of the previous applications use a SCN de-ployed in a built environment to obtain and analyze different type of information. Therefore, they might be considered as Panoptes building applications. The main function of a Panoptes building is to combine the different information obtained by the SCN and to deduce the triggering of actions/events in real-time. In that context, Panoptes building applications should un- derstand the static building information as well as perceive (accurately) the dynamic and evolving data, i.e., they should be aware of their context. A context-aware system in the built environment is a complex task; it requires information from different domains such as environment data, sensing devices, spatio-temporal facts and details about the different events that may occur. For example, the required event information could be a set of concepts and relations concerning the different events that may occur in a built environment, their location, the time they occurred, the agents involved, the relation to other events and their consequences. In the case of the sensor information, the required data could be the description of the different sensing devices, the process implemented on them and their results. Regarding the environment, the required data could be the structure of the building, its topology and the different elements contained in a space." assertion.
• paragraph hasContent "The built environment data can be obtained using the Building Information Modeling (BIM) of a building. BIM becomes a general term designing the set of numerical data, objects and processes appended during the life-cycle of a building [11]. From the designing, construction and facility management steps, the BIM allows practitioners and managers to exchange data in a uniform way using the IFC standard [40]. The IFC standard gives the base of description, both semantic and graphic of all elements making the building [20]. This allows to aggregate all heterogeneous software dedicated to the built environment on an interoperability way. In the domain of interoperability three levels are described: technical, organizational and semantics [21]. The IFC aims the technical interoperability level [9]. The organizational level is in charge of the practitioners according to the law of each country and the rules of each enterprise. The semantics level aims to clearly specify the meaning of each element making the BIM." assertion.
• paragraph hasContent "An important work was made to bring the IFC EXPRESS schema into the semantic web world using OWL as the schema modeling language. Dibley et al. compared different frameworks that tried to achieve this goal [10]. The result is the ifcowl ontology whose main objective is to convert the IFC concepts and instances data into equivalent RDF data. The conversion procedure from EXPRESS to OWL can be found in [30]." assertion.
• paragraph hasContent "According to Studer, an ontology is a formal, explicit specification of a shared conceptualization [37]. In other words, an ontology is a set of concepts and relations used to describe and represent an area of concern. Currently, ontologies are represented using OWL-2 language, which is the recommendation of the World Wide Web Consortium (W3C) [43]. Other recommended technologies/languages in the semantic web field are: RDF, used for representing information in the form of a graph composed of triples [45]; RDF Schema (RDFS), which provides a vocabulary for creating a hierarchy of classes and properties [3]; SPARQL 1 , used to query RDF data [44]; and the Semantic Web Rule Language (SWRL 2), which is used for extending the OWL model with rule axioms [18]." assertion.
• paragraph hasContent "One important application of ontology is for semantic fusion, which consists on integrating and organiz- ing data and knowledge coming from multiple heterogeneous sources and to unified them into a consistent representation. There have being many works that combine semantic information coming from different sources in an ontology. Hong et al. presented some context-aware systems were the ontology plays a central role for enabling interoperability between devices and agents which are not designed to work together [16]. Dibley et al. developed an ontology framework that combines a sensor ontology with a building ontology and others supporting ontologies [10]. Other works have focused on using ontologies for combining computer vision with different kinds of information such as SanMiguel et al. which created an ontology composed mainly of knowledge about objects, events and image processing algorithms [34]; Chaochaisit et al. presented a semantic connection between sensor specification, localization methods and contextual information [5]; and Town which presented an ontology that fusion multiple computer vision stages with context information for image retrieval and event detections [39]. The use of ontologies as an interoperability agent warranties the information fusion. Consequently, we propose the creation of an ontology to combine and re-purpose the different types of information required by a Panoptes building." assertion.
• paragraph hasContent "Our approach differs from classical computer vision which deals with algorithm improvements [46] and signal processing problems [35], by dealing with a meaning problem in computer vision [39], where the observation of the smart camera is improved by semantically fusing it with contextual information (e.g., position of the sensors, position of the users, spaces in the environment and events that have occurred)." assertion.
• section-2-title hasContent "Background and related work" assertion.
• paragraph hasContent "The WiseNET ontology is an OWL-2 ontology that incorporates a vast corpus of concepts in the domain of a Panoptes building. The ontology provides a vocabulary for combining, analysing and re-purposing the information coming from the smart camera network (SCN) deployed in a built environment. The main function of the WiseNET ontology is to perform real-time event/anomalies detection and initiate system reconfiguration." assertion.
• paragraph hasContent "The goal for developing the WiseNET ontology was to create a shared understanding of the structure of information for a Panoptes building. The WiseNET developing process follows the Noy and McGuinness methodology for ontology development [27]. This methodology consists of seven steps which are: determine scope of the ontology, consider reuse, enumerate classes, define classes and properties, define constrains and create instances (ontology population shown on Sections 5 and 6)." assertion.
• paragraph hasContent "To determine the scope of the ontology we need to think about the kind of knowledge that should be covered by the ontology and its use, i.e., its domain." assertion.
• paragraph hasContent "Some competency questions were formulated to determine the focus of the ontology (Table 1). Those competency questions should be answered by the ontology and from them it can be extracted the different kind of knowledge that should be contained in the WiseNET ontology. Roughly, is knowledge about the environment, events, person, sensors and time.." assertion.
• section-3.1.1-title hasContent "Scope of the ontology" assertion.
• paragraph hasContent "When developing a new ontology it is recommended to reuse existing ontologies as much as possible. In this way, one can focus on defining the specific knowledge of the application. The reuse of external ontologies, not only saves time but also gives the advantage of using mature and proved ontological resources that have been validated by their applications and (some) by the W3C." assertion.
• paragraph hasContent "The WiseNET ontology reuses resources from many different ontologies (see Table 2). However, there are six key ontologies that cover most of the required concepts of the different domains, those are:" assertion.
• paragraph hasContent "Figure 1 shows the primary classes and properties reused by the WiseNET ontology. The external ontologies were not imported, from most of them only some concepts are reused, except from the ifcowl from which some instances are also considered. Not importing the external ontologies gives two main benefits: easing the ontology maintenance and improving the performance. The performance improvement is a very important factor due to the goal of the WiseNET ontology is to perform real-time reasoning." assertion.
• section-3.1.3-title hasContent "Links to existing ontologies" assertion.
• paragraph hasContent "Many of the competency questions involves more than one type of knowledge. Hence, the WiseNET ontology should be able to collect and combine the information from those different domains. In that context, it is necessary to define new concepts (classes and properties) that will allow us to: complete the information from the different domains, to describe attributes of instances according to our needs and, more importantly, to relate (i.e., link) the different domains." assertion.
• paragraph hasContent "The Tables 3 and 4 present some selected classes and properties with emphasis on built environment information. We propose to enhance the IFC information by adding functional facts to the spaces, i.e., information about the space usage (if it is a corridor, a reception, a coworking space, etc). Additionally, we propose to add information about the different types of alarm and security systems present in the built environment, specifically the security systems of the doors (e.g., key-lock system, card reader system and biometric system). This can allow the deduction of knowledge regarding the security restrictions of a space." assertion.
• paragraph hasContent "After having the complete terminology of the ontology, some constraints and characteristics of the class expressions and the property axioms need to be defined. Axioms are a set of formulas taken to be true and that every assignment of values should satisfied. Those constraints and characteristics will determine the expressiveness and decidability of the ontology, and their definition will depend on the description logic used." assertion.
• section-3.1.3-title hasContent "Classes and properties of WiseNET" assertion.
• section-3.1-title hasContent "Ontology development" assertion.
• paragraph hasContent "Description logics (DLs) are a family of formalism for representing knowledge [1]. The most no- table applications for the DLs is to provide the logical formalism for ontologies languages such as OWL. OWL-2, the current W3C ontology language recom-mendation, is based on the expressive description logic SROIQ(D) [17]. SROIQ(D) extends the well known description logic SHOIN (D) by including: a role constructor by general inclusion (R) and a qualified number restriction (Q). SROIQ(D) provides high expressive power with high computational cost of reasoning. Hence, to meet a more suitable compromise between the expressive power and the computational cost, the WiseNET ontology was defined using the SHOIQ(D) language which is more expressive than SHOIN (D) yet less expressive than SROIQ(D) and less computational complex [22]. A definition of the SHOIQ(D) constructors and some examples referencing the WiseNET ontology can be found in the Table 5." assertion.
• paragraph hasContent "Horrocks and Sattler presented a tableau decision procedure for SHOIQ(D) that solves the ontology consistency problem and allows the use of reasoning services, thus demonstrating the decidability of SHOIQ(D) [19]. One of the few requirements to preserve the decidability in SHOIQ(D) is to restrict the application of the qualified number restriction to simple roles, i.e., roles that are neither transitive nor have a transitive subrole [19]. This restriction is satisfy in the WiseNET ontology." assertion.
• paragraph hasContent "However, knowledge representation formalisms of the semantic web (such as DLs) have expressive limitations, for example composition of complex classes from classes and properties. Those limitations can be overcome by rule-based knowledge, specifically by using SWRL (Semantic Web Rule Language) rules [18]. SWRL rules are represented as implication of an antecedent (Body) and a consequent (Head):" assertion.
• paragraph hasContent "Reasoning becomes undecidable for the combination of OWL + SWRL, therefore the expressivity of SWRL needs to be reduced in order to assure decidability. Although many procedures exists to guarantee decidability of SWRL, the DL-safe rules was adapted [25]. This procedure consists on restricting the number of possible variables assignments, i.e., restricting the application of rules only to known OWL individuals (named individuals)." assertion.
• paragraph hasContent "Examples of DL-safe rules implemented in the WiseNET ontology are presented in Listing 1 and Listing 2. The first one states that if there are two spaces ’x’ and ’y’, and both contain the door ’d’, then those spaces are connected to each other. The second one states that if there are two spaces ’x’ and ’y’, and two smart cameras ’s1’ and ’s2’, and ’x’ is connected to ’y’, and ’s1’ is located in ’x’ and ’s2’ is located in ’y’, then those smart cameras are nearby each other." assertion.
• paragraph hasContent "To recapitulate, the WiseNET ontology decidability was insured by restricting the application of the qualified number restriction and by using DL-safe rules. As a result, a semantic reasoner can be employed for inferring logical consequences from a set of asserted facts or axioms. After finishing the formalization of the ontology the next step is to connect it to an operational architecture that enables the insertion of the required facts." assertion.
• section-3.2-title hasContent "Ontology decidability" assertion.
• section-3-title hasContent "Formal modeling" assertion.
• paragraph hasContent "The WiseNET ontology is the kernel of the WiseNET system. WiseNET is a semantic-based system that fuses heterogeneous sources of data such as data coming from sensors and the different contextual information. Due to the application of the paper (Panoptes building), we will focus in a specific type of sensor: smart cameras; however, the system is defined to include other type of sensors such as temperature, humidity, depth sensor, etc. The main goal of WiseNET is to improve classical computer vision and deep learning systems by considering the contextual information of the environment and by performing real-time reasoning. As a result, WiseNET may overcome some limitations of computer vision (e.g., false detections and missed detections), some drawbacks of deep learning (e.g., the need of large amount of training and testing data) and limitations of multi-camera based system (presented in Section 1) while allowing real-time event/anomalies detection and system reconfiguration." assertion.
• paragraph hasContent "Figure 2 illustrates the architecture of the WiseNET system. The system is articulated in three sections: the smart camera network, the central unit and the monitor unit. The SCN is a set of smart cameras distributed in an environment. The main functions of the SCN, in the WiseNET system, is to extract low-level features from a scene (such as person detection as shown in blue and green in the right side of the Figure 2), to convert the extracted data into knowledge and then to send it to the central unit. More information regarding the type of smart cameras used can be found in [23]. The central unit is composed of two elements the central API and the WiseNET ontology. The central API is in charge of the management of the ontology, for example: capturing the knowledge coming from the SCN and insert it to the ontology, retrieving inferred knowledge from the ontology, transferring data to the monitor unit, sending new configurations to the smart cameras, and other services. The WiseNET ontology is responsible of enabling interoperability between the incoming knowledge streams and the contextual data (e.g., environment information, previous knowledge and sensor in- formation) in order to deduce new knowledge and detect events/anomalies based on the history of activities. The central unit is also in charge of re-identifying people in the system by using their visual features extracted by the SCN. Eventually, the central unit could request extra information to the SCN. The monitor unit has as main function the visualization of the static and dynamic information; this unit will automatically retrieve information and will present it in a graphical manner, for example an occupancy map (as shown at the center of Figure 2). The monitor unit implements some queries to answer questions such as: how many people is in a room? what is the location of a person? and many others (see Table 1)." assertion.
• paragraph hasContent "The proposed system is context-aware and combines the information extracted by the SCN with logic rules and knowledge of what the camera observes, building information and events that may occurred. Our system differs from other computer vision systems mainly by three factors. First, no images are sent, the smart cameras extract the knowledge from the images and then this knowledge is sent to the central unit (for more details see Section 6.1). Second, the WiseNET system combines context information with the camera information to improve the detections, in this way, it can overcome missed detections or non-detectable in- formation (e.g., people out of sight of a camera). Third, the system uses an ontology to fusion the different kinds of information presented in a Panoptes building, such as: information of the environment, time, smart cameras, events, detectable objects, etc." assertion.
• paragraph hasContent "Notice that the architecture shown in Figure 2 presents the WiseNET system as a centralize system where there is no communication between the smart cameras, however the system could also be deployed in a distributed manner as presented in [2]." assertion.
• paragraph hasContent "Once the ontology is formally defined and implemented in a triplestore (a database for the storage of triples), the last step of the ontology development (as stated in Section 3) is to populate it. The next two sections will present the population from the IFC file and from the sensors respectively. Note that the second one consists of two parts. Firstly, the initialization of the sensors is populated, consisting on the description of the sensors and their relation to the built environment. Secondly, an ongoing population is performed each time the smart cameras detect a person in the building." assertion.
• section-4-title hasContent "Operational modeling" assertion.
• paragraph hasContent "After inserting the WiseNET ontology in the system, the a priori information about the built environment needs to be populated (inserted). The required information can be extracted from the IFC file of the environment. This population will be performed only once at the initialization of the system, therefore is considered as a static population." assertion.
• paragraph hasContent "The I3M (Institut Marey et Maison de la Métallurgie) building located in Dijon (France), will be used as an example for this section. The I3M building is composed of three building storeys from which we will focus on the third storey where a SCN has been deployed." assertion.
• paragraph hasContent "An IFC2x3 file, describing all the elements compos- ing the I3M building, was used in this project. It was obtained from the company in charge of the construction of this building and it was generated using the Revit CAD software 3 . Only a small portion of the IFC file is needed in the WiseNET system. Therefore, to improve the ontology performance, only the required information will be extracted from the IFC file and populated in the ontology." assertion.
• paragraph hasContent "Figure 3 shows, in the form of a graph, the main classes and properties required to be extracted from an IFC file and populated on the ontology. The extracted/populated data consists on information about the building, building storeys, spaces and elements contained on those spaces (such as doors, windows, walls). Furthermore, the topology of the building is also required, i.e., the relation between the building and the building storeys, between the building storeys and the spaces, between the spaces and the different elements (doors, windows and walls) and the relations between two spaces. We assume that two spaces are connected if and only if they share a door." assertion.
• paragraph hasContent "A framework was developed for extracting and populating the required IFC data into the WiseNET ontology (see Figure 4). The extraction/population framework employs semantic web technologies and it consists mainly of four processes: a compliance check of the IFC, a conversion of the IFC into ifcowl, the extraction of the pertinent instances from the ifcowl and finally, the population of the extracted instances and their relating properties into the WiseNET ontology." assertion.
• paragraph hasContent "The framework starts with an IFC file of a built environment, in this case the I3M building. The requirements of the IFC is to contain the following entities: IfcBuildingStorey, IfcRelSpaceBoundary, IfcRelDecomposes, IfcBuilding, IfcDoor, IfcWindow, IfcWall, and IfcSpace. The compliance check will verify the fulfillment of those requirements. Afterwards, the IFC is converted to RDF by using the IFC-to-RDF converter by Pauwels and Oraskari [29]. The result of the conversion is the ifcowl ontology with instances of the I3M building." assertion.
• paragraph hasContent "The ifcowl will be queried using SPARQL in order to extract the pertinent instances. The Listing 3 shows the SPARQL code used for the extraction. Line 4 obtains the building instance by using its class. Line 7 acquires the array of building storeys that decompose the building; and line 8 obtains the storeys inside that array. The same is done for the spaces that decompose the storeys on lines 11-12. Lines 15-16 obtain the ele- ments that are contained in a space. Lines 19-22 filter out the undesired elements just leaving the doors, windows and walls. Finally, line 25 saves all the classes of the elements. The result of the extraction query is a table where the columns corresponds to the variables used with the SELECT operator (line 1). This table is called the extracted table." assertion.
• section-5.1-title hasContent "Extraction query" assertion.
• paragraph hasContent "The extracted table contains a set of instances without any relations between them. That is why, the population query will create those relations using the WiseNET properties and then will insert them to the WiseNET ontology. To accomplish this, the population query will process row by row the extracted table. For exemplification, lets assume that the first row of the extracted table has the following values:" assertion.
• paragraph hasContent "As aforesaid, we assumed that two spaces are connected if and only if they share a door. This property spaceConnectedTo could be obtained from queries but its quite complex, therefore it was decided to formulate a rule to obtain this property (see Listing 1)." assertion.

• first
• previous
• next