Dong-Hyuk Im

Linked tag: image annotation using semantic relationships between image tags

State of the art image tagging systems are limited because they allow users to annotate image tags in noun form, which cannot fully express the semantics of image content. In this paper, we propose Linked Tag, a semi-automatic image annotation system that inserts semantic relationships between tags. The proposed annotation method connects image tags using predicate words that can capture the contexts in which the image tags are used. In particular, we exploit Linked Data such as DBPedia in order to connect the image tags with a property value. Compared with tag-based annotation and ontology-based annotation systems, Linked Tag eliminates a large amount of manual labor and enhances the semantic expression of image content. We also introduce two annotation-based applications on Linked Tag. First, we propose SPARQL query processing for image retrieval, which enables us to express visual appearance as well as semantic information. Second, we propose a novel tag-ranking algorithm based on the link analysis in the RDF annotation graph. Finally, we demonstrate the operation of our proposed system and analyze its efficacy.

SigMR: MapReduce-based SPARQL query processing by signature encoding and multi-way join

Large numbers of Resource Description Framework triples are available in Linked Data which can grow exponentially. It makes SPARQL query processing engines infeasible on a single machine. To address this scalability issue, MapReduce framework-based SPARQL engines have been proposed, but we note that these methods are limited in terms of join evaluations. The two-way join-based approach evaluates joins via a sequence of binary multiplications that require multiple MapReduce jobs, which involves costly disk accesses between MapReduce jobs. The multi-way join-based approach combines multiple two-way join operations, which allows the simultaneous evaluation of joins during one MapReduce job. However, the size of data for the MapReduce job might increase exponentially if a complex query is given. In this study, we propose SigMR, a pruning method for multi-way join-based SPARQL query processing in MapReduce. In the proposed approach, a SPARQL query can be evaluated in a single MapReduce job, where the size of data is reduced dramatically by pruning based on our signature encoding technique, thereby overcoming the weaknesses of the previous approaches. In experiments, we showed that the query processing time required was lower with our approach than existing MapReduce-based methods.

xStore: Federated temporal query processing for large scale RDF triples on a cloud environment

Abstract Temporal information retrieval tasks have a long history in information retrieval field and also have attracted neuroscientists working on memory system. It becomes more important in Semantic Web where structured data in RDF triples, often with temporal information, are rapidly accumulated over time. Existing triple stores already support loading RDF triples and answering a given SPARQL query with time interval constraints. However, few triple stores has been optimized for processing time interval queries which are important for temporal information retrieval tasks. In this paper, we propose xStore , a federated SPARQL engine running on a cloud environment, which supports a fast processing of temporal queries. xStore is built on top of heterogeneous storages such as key-value stores and conventional triple stores. Experiments over real-world temporal datasets showed that our approach is faster than a conventional SPARQL engine for processing temporal queries.

G-Diff: A Grouping Algorithm for RDF Change Detection on MapReduce

Linked Data is a collection of RDF data that can grow exponentially and change over time. Detecting changes in RDF data is important to support Linked Data consuming applications with version management. Traditional approaches for change detection are not scalable. This has led researchers to devise algorithms on the MapReduce framework. Most works simply take a URI as a Map key. We observed that it is not efficient to handle RDF data with a large number of distinct URIs since many Reduce tasks have to be created. Even though the Reduce tasks are scheduled to run simultaneously, too many small Reduce tasks would increase the overall running time. In this paper, we propose G-Diff, an efficient MapReduce algorithm for RDF change detection. G-Diff groups triples by URIs during Map phase and sends the triples to a particular Reduce task rather than multiple Reduce tasks. Experiments on real datasets showed that the proposed approach takes less running time than previous works.

STAG: Semantic Image Annotation Using Relationships between Tags

The state-of-art image tagging system has a limitation in that it allows users to annotate image tags in noun form that cannot fully express the semantics of image. In this paper, we propose a STAG, semi-automatic annotation system for image using semantic relationships between social tags. By connecting the image tags using predicate word, we can capture the contexts in which image tags are used. Compared with ontology-based annotation system, this reduces a large amount of manual jobs and enhances the expression of image content. In addition, we are able to use SPARQL-like query for image retrieval.

A dynamic and parallel approach for repetitive prime labeling of XML with MapReduce

A massive amount of extensible markup language (XML) data from various areas is available on the Web. Answering structural queries against XML data is important, as it is the core of information retrieval systems for XML data. Labeling scheme has been suggested for rapid query processing of massive XML data. Interval-based, prefix-based, and prime number labeling scheme exist. Of these, the prime number labeling scheme has the advantage of query processing by arithmetic operations. Recently, the repetitive prime number labeling scheme was proposed; this scheme produces a smaller label size than conventional prime number labeling using prime numbers repetitively. However, a parallel algorithm for the repetitive prime number labeling scheme does not exist; therefore, this scheme is difficult to apply to massive XML data. In this paper, a dynamic and parallel approach of XML labeling algorithm that works with MapReduce is proposed for, particularly, the repetitive prime number labeling scheme. Two optimization techniques are devised: the label assignment order adjustment to further reduce the label size and the upper tree compressing technique to reduce the memory requirements during the labeling process. Experiments over real-world XML data confirmed that the techniques are effective than the previous works.

Design and Implementation of NFC-Based Mobile Coupon for Small Traders and Enterprisers

Mobile coupon is an electronic ticket and is currently used for a financial discount and rebate. NFC is a short range wireless standard and has an advantage that the communication is very simple. In this paper, we propose a NFC-based mobile coupon for small traders and enterprisers. Since it is expensive for small shops to create and distribute their own coupons, NFC communication is only used to send the customers identification. Users can receive and check coupon information from their mobile phones. In addition, our platform provides some community service that the small merchants work together.

A MapReduce-Based Approach for Prefix-Based Labeling of Large XML Data

A massive amount of XML (Extensible Markup Language) data is available on the web, which can be viewed as tree data. One of the fundamental building blocks of information retrieval from tree data is answering structural queries. Various labeling schemes have been suggested for rapid structural query processing. We focus on the prefix-based labeling scheme that labels each node with a concatenation of its parent’s label and its child order. This scheme has been adapted in RDF (Resource Description Framework) data management systems that index RDF data in tree by grouping subjects. Recently, a MapReduce-based algorithm for the prefix-based labeling scheme was suggested. We observe that this algorithm fails to keep label size minimized, which makes the prefix-based labeling scheme difficult for massive real-world XML datasets. To address this issue, we propose a MapReduce-based algorithm for prefix-based labeling of XML data that reduces label size by adjusting the order of label assignments based on the structural information of the XML data. Experiments with real-world XML datasets show that the proposed approach is more effective than previous works.

Wireless Security System Incorporating Tilt Sensors and Web Cameras

The need for research and development of the security and alarm devices has been expanding in order to prevent growing crimes. In this paper, we propose the security system using both sensors and cameras to complement the drawbacks of the existing system. The sensors are used to recognize emergency situations more correctly, and cameras to overcome the difficulty of on-site check. Also, the user can directly check the situation of the scene from the server or users smart phone.

Design of U-Health System with Statistical Disease-Identification Technique

In this paper, we propose U-health system that enhances reliability in order to possibly diagnose patients condition and judge abnormal sign or emergency situation by applying the disease-identification algorithm. In addition, the proposed system enhances speed by allowing family and medical-institution practitioners to possibly confirm the identified result through the disease identification technique algorithm, by using smart phone anytime and anywhere.