Search

Regarding the spoken language understanding (SLU) pilot task of the Dialog State Tracking Challenge 5 (DSTC5), it is required to classify label sets of speech acts on human-to-human dialogues. In this paper, we propose a multi-label classification model with the assistance of algorithm adaptation method. To be specific, a Convolutional Neural Network (CNN) model on top of pre-trained word vectors is adapted for the multi-label classification task by utilizing a threshold learning mechanism. In order to evaluate the performance of our proposed model, comparative experiments on the DSTC5 dialogue datasets are conducted. Experimental results show that the proposed model outperforms most of the submitted model in the DSTC5 in terms of F1-score. Without any manually designed features, our model has advantage of handling the multi-label SLU task, using only publicly available pre-trained word vectors.

The proposed eHealth Recommendation Service System (eHeaRSS) is to recommend health service information to patients whenever and wherever. Nowadays, depending on online networks popularity, there are a lot of research starting to focus on patient-context to apply health care fields like tele-health, tele-care, medical traveling, and so on. The proposed eHeaRSS is like a family doctor to care patients, especially to use generated patients-context. eHeaRSS is a recommendation system for immediate and appropriate medical services using crowd sourcing in cloud computing environment. To do this, it is necessary to classify the data into symptoms, diseases, departments and doctors related data types. To consider hierarchies or subsume relationships among them, eHeaRSS Ontology (eHeaRSS-Ont) is developed through integration of 4-static ontology (4-Ont) which is comprised of symptom-, disease-, department-and doctor-Ont to create tailored recommendations depending on patient-context. For the recommendation service, Case-based reasoning (CBR) is applied. The extracted cases using 4-Ont are integrated by eHeaRSS-Ont depending on the patient-context. For customization, it is necessary to reconfigure the case using eHeaRSS Constraints Value Compatibility Map (eHeaRSS-CVCM). To prove the significance and efficiency of the eHeaRSS, we experimented using ontology-based data processing and proved the superiority of eHeaRSS with the provision of better recommendation than other DB-based systems.

As the number of RDF triples on the Linking Open Data (LOD) Cloud has been exponentially increased, the difficulties of information query have been increased. To query information on the LOD Cloud, the users have to have some capabilities for developing SPARQL and/or RDQL query statements and exact knowledge for web resources with RDF such as URI, DB title, and name of things. However, it is almost impossible to require the capabilities and/or knowledge to the users who are familiar with keywords search. So, we propose fully automated keyword-based SPARQL query generation system. In our system, the users can query information on the LOD Cloud without having capabilities about structured query language and having prior knowledge of web resources with RDF. The users should type a set of keywords into our system. To do so, we developed a propertybased path finding algorithm and an automated SPARQL query generation that can be used to provide query recommendations for the users. In the experimental section, we illustrate an example to validate the effectiveness of our system, and perform a simulation to show the superiority of the algorithms. The experiment results are not bad for a first attempt. If we improve the algorithms, we can expect a better result.

For effective searching of appropriate information, it is necessary to well organize data to access and store in data-base. So, we adopt a case structure as a formalized data form. Web resources are transformed into cases which help information processing and accessing. In addition, we define common-context which are shared concepts by comparatives and propose a common-context-based conceptual similarity through arc compression on ontology. Arc-based conceptual distance between comparative nodes is calculated under consideration of common-context. One of comparatives comes from the user requirements and another from indexes of a case. The distance is contingent upon consideration of common-context. The 'Node Compression (NC)' and 'Arc Compression (AC)' are proposed to support the dynamicity of similarity. NC is conducted between adjacent common-context nodes and leads calculation of conceptual distance between comparatives. AC is processed between non-adjacent common-context nodes. The conceptual arc compression is conducted by Weighted Partial Ontology (WPO) based on weights of arcs under consideration of common-context. The proposed NC and AC support to return conceptual distance between comparatives because it increases the concept-based reliability of search result. To verify the effectiveness, the proposed conceptual similarity is compared with that of edge-counting similarity method. We show that the proposed conceptual similarity calculation leads a higher similarity value for conceptually close classes compared with other methods.

Providing context-aware personalized service in smart home environment is very important to improve the quality of the disabled life. To reaching the goal, it needs to challenge due to diversity of needs: type of disability, degree of disability, preferences, tasks, and context. We propose a framework to recommend context-aware personalized services for the disabled in smart home environment called SOLVE-D. Main module of SOLVE-D is ontology repository that is consisted of three ontologies: generic service ontology, personalized service ontology, and service context ontology. The ontologies interact with each other to recommend context-aware personalized services, and to increase the quality of personalized services for disabled person in smart home environment. In addition, to support self-evolved ontologies and to provide the optimal service, SOLVE-D has follow modules such as Standard Service Recommendation Module (SSRM), Personalized Service Recommendation Module (PSRM), Personalized Service Ontology Generation Module (PSOGM), and Context and PSO Mapping Module (CPMM). The proposed ontologies and various modules of SOLVE-D are in perfect harmony to provide context-aware personalized services for` disabled users.

In this research, we propose a methodology for ontology construction and evolution using tags from tag-cloud to improve availability of analysis and utilization of large scale of data. It is possible to construct semantic relationships among tags. To do this, one way of building of relationships among tags is use of a popularized tag as a thing of ontology. However, it is not easy to select the popularized tag from tag cloud and to build a complete ontology with all of tags in the tag-cloud because tags are timely and dynamically updated added with tremendous numbers of tags. Therefore, we construct Primitive Ontologies (POs). A general tag is used to as a thing of a PO. The general tag is matched into one of requested keywords by users. In addition, we propose a methodology to evolve ontology using tags. To construct and evolve a PO, it defines classes and relationships between given tags by a user. Finally, the constructed several POs are integrated into a Tag-Based integrated Semantic ontology (TBiSont) by building relationships between classes that are included in each different PO. The construction of relationships among tags depends on linked resources into tags. The properties of tags are usually characterized by linked resources. It causes to extract semantic relationships among tags, because tags are comprised of meta-information that represents properties of resources. Thereby, the proposed PO is constructed by a general tag as a best-referring tag and relationships among tags including the general tag. The TBiSont as an integration of POs was applied into data analysis and a search process to verify efficiency of as in application systems. When we applied TBiSont into searching, it shows that TBiSont is more efficient than the other statistical-based search systems to increase accuracy or diversity of search results.

In this research, the proposed Tag-based Dynamic Data Catalog (TaDDCat) supports dynamic search of resources depending on user's requirements using tags from social web driven resources. It is possible for tags as meta data to match into some resources. Consequently, semantic relationships between tags can be extracted by dependency of relationship between tags and resources. TaDDCat using tags constructs hierarchical and associative relationships among tags for effective search of solution. Information template (ITM) is defined by extracted tags according to user's requirements and appropriately searched resources by the tags. In addition, primitive tree is extracted by hierarchical relationships among tags which are defined on semantic-tag cloud. The associative relationship is defined by shared resources between tags. Finally, new class which is generated by merged tags highly associative. The class creates new hierarchy with Hyper/hypo relationship between the class and former tags which are not merged. At last, integrated ontology (IO) is developed with relationships among tags. resources are matched into the IO are named by integrated information resource (IIR). The TaDDCat is completed by ITM with combination of Tags(T) and integrated information resource, and reasoning by relationships(R). The TaDDCats supports improvement of correctness and agility of search and decreasing of search effort by reduction of quantity of search data.

In this paper, we propose a context-based hybrid semantic matching framework for e-mentoring system. The framework adopts ontology technologies to define the attributes of mentors and mentees, the matching results, and satisfaction values of the mentoring, and to recommend adequate mentors via semantic reasoning. In the developed ontology named mentoring-matching ontology (M2O), the pairs of mentor-mentee and the satisfaction values of mentoring are managed as cases. To increase the satisfaction of matching, the framework performs rule-based filtering and two kinds of semantic matching to find appropriate mentees and the most similar mentor with the new mentee. We implemented a mobile e-mentoring system named Mint Story with the proposed matching framework. Also, we show the superiority of our framework in the evaluation section.

We propose a framework to support collaborative decision-making among multi-mobile users. This framework is comprised of an Agent's Case-based Reasoning, Integrated Ontology Generation and a Value Unification Module. The case-based reasoning module used to semantic similarity calculation and is applied to the extraction of the most similar case from each decision maker's case-base. Each the most similar case can be an alternative solution for the new requirement which makes the collaborative decision among the decision-makers. The number of alternative solutions depends on the number of the collaborative decision-makers. Each extracted case is mapped on a personalized personal ontology by projecting each of the cases onto the meta-ontology for deriving an appropriate solution for the new requirement. The personal ontology consists of the classes and/or instances and their relationships that are common in all the personalized personal ontologies. It is named an integrated ontology. As the next step, to find a collaboratively satisfactory solution, we need to determine the degree of unification of classes or instances on the integrated ontology. The weight of unification means the extent to which the classes and/or instances are shared among the decision-makers. The propagation of unification is performed by the value that has maximum weight of unification using the integrated ontology. The unification is executed recursively until it reaches a collaborative decision. To do so, we propose the ontology-based value unification algorithm. However, if we cannot find the solution after the unification is finished, the real-time interactive negotiation algorithm that is newly proposed by us is applied.

The main objective of this paper is to present and evaluate methods that can be used to extract valuable information from traffic data history for use in travel information and guidance systems. Through two different analyses, we identify the roads which the system should pay closer attention to. First, we introduce an approach that finds the patterns of the traffic speed which lie within the data history. Second, we discover the relations between traffic speeds and various features of roads. The result of our study let travel guidance systems know how heavy the traffic will be at any given road. We extend these approaches to tackle the problem of incident management, and designed a system that returns, the expected range of effect of an incident using road features.

This paper suggests a miniature internal chip antenna with quad band for operating at multi-band handset applications, including global system for mobile communication (GSM), digital communication system (DCS), personal communication system (PCS), and universal mobile telecommunication system (UMTS) band. The proposed antenna can be easily placed in practical handsets because of the small size of 28.3 times 5.1 times 3.2 mm3 . The electrical and structural characteristics of the proposed antenna make it attractive for use in mobile handset applications.