• Journal of KIISE:Databases
• /
• v.37 no.4
• /
• pp.177-184
• /
• 2010

In these day, many data including sensor, delivery, credit and stock data are generated continuously in massive quantity. It is difficult to learn from these data because they are large in volume and changing fast in their concepts. To handle these problems, learning methods based in sliding window methods over time have been used. But these approaches have a problem of rebuilding models every time new data arrive, which requires a lot of time and cost. Therefore we need very simple incremental learning methods. Bayesian method is an example of these methods but it has a disadvantage which it requries the prior knowledge(probabiltiy) of data. In this study, we propose a learning method based on attribute values. In the proposed method, even though we don't know the prior knowledge(probability) of data, we can apply our new method to data. The main concept of this method is that each attribute value is regarded as an expert learner, summing up the expert learners lead to better results. Experimental results show our learning method learns from data very fast and performs well when compared to current learning methods(decision tree and bayesian).

• Journal of KIISE:Databases
• /
• v.31 no.5
• /
• pp.468-478
• /
• 2004

It is essential in various application areas of data mining and bioinformatics to effectively retrieve the occurrences of interesting patterns from sequence databases. For example, let's consider a network event management system that records the types and timestamp values of events occurred in a specific network component(ex. router). The typical query to find out the temporal casual relationships among the network events is as fellows: 'Find all occurrences of CiscoDCDLinkUp that are fellowed by MLMStatusUP that are subsequently followed by TCPConnectionClose, under the constraint that the interval between the first two events is not larger than 20 seconds, and the interval between the first and third events is not larger than 40 secondsTCPConnectionClose. This paper proposes an indexing method that enables to efficiently answer such a query. Unlike the previous methods that rely on inefficient sequential scan methods or data structures not easily supported by DBMSs, the proposed method uses a multi-dimensional spatial index, which is proven to be efficient both in storage and search, to find the answers quickly without false dismissals. Given a sliding window W, the input to a multi-dimensional spatial index is a n-dimensional vector whose i-th element is the interval between the first event of W and the first occurrence of the event type Ei in W. Here, n is the number of event types that can be occurred in the system of interest. The problem of‘dimensionality curse’may happen when n is large. Therefore, we use the dimension selection or event type grouping to avoid this problem. The experimental results reveal that our proposed technique can be a few orders of magnitude faster than the sequential scan and ISO-Depth index methods.hods.

• Journal of KIISE:Databases
• /
• v.33 no.1
• /
• pp.42-55
• /
• 2006

We propose a single Index approach for subsequence matching that supports moving average transform of arbitrary order in time-series databases. Using the single index approach, we can reduce both storage space overhead and index maintenance overhead. Moving average transform is known to reduce the effect of noise and has been used in many areas such as econometrics since it is useful in finding overall trends. However, the previous research results have a problem of occurring index overhead both in storage space and in update maintenance since tile methods build several indexes to support arbitrary orders. In this paper, we first propose the concept of poly-order moving average transform, which uses a set of order values rather than one order value, by extending the original definition of moving average transform. That is, the poly-order transform makes a set of transformed windows from each original window since it transforms each window not for just one order value but for a set of order values. We then present theorems to formally prove the correctness of the poly-order transform based subsequence matching methods. Moreover, we propose two different subsequence matching methods supporting moving average transform of arbitrary order by applying the poly-order transform to the previous subsequence matching methods. Experimental results show that, for all the cases, the proposed methods improve performance significantly over the sequential scan. For real stock data, the proposed methods improve average performance by 22.4${\~}$33.8 times over the sequential scan. And, when comparing with the cases of building each index for all moving average orders, the proposed methods reduce the storage space required for indexes significantly by sacrificing only a little performance degradation(when we use 7 orders, the methods reduce the space by up to 1/7.0 while the performance degradation is only $9\%{\~}42\%$ on the average). In addition to the superiority in performance, index space, and index maintenance, the proposed methods have an advantage of being generalized to many sorts of other transforms including moving average transform. Therefore, we believe that our work can be widely and practically used in many sort of transform based subsequence matching methods.

• Journal of KIISE:Software and Applications
• /
• v.31 no.5
• /
• pp.586-594
• /
• 2004

Because web documents become creation and disappearance rapidly, users require the recommend system that offers users to browse the web document conveniently and correctly. One largely untapped source of knowledge about large data collections is contained in the cumulative experiences of individuals finding useful information in the collection. Recommendation systems attempt to extract such useful information by capturing and mining one or more measures of the usefulness of the data. The existing Information Filtering system has the shortcoming that it must have user's profile. And Collaborative Filtering system has the shortcoming that users have to rate each web document first and in high-quantity, low-quality environments, users may cover only a tiny percentage of documents available. And dynamic recommendation system using the user browsing pattern also provides users with unrelated web documents. This paper classifies these web documents using the similarity between the web documents under the web document type and extracts the user browsing sequential pattern DB using the users' session information based on the web server log file. When user approaches the web document, the proposed Dynamic recommendation system recommends Top N-associated web documents set that has high similarity between current web document and other web documents and recommends set that has sequential specificity using the extracted informations and users' session information.

• Journal of KIISE:Databases
• /
• v.33 no.7
• /
• pp.693-707
• /
• 2006

To improve performance using a multidimensional index in similar sequence matching, we transform a high-dimensional sequence to a low-dimensional sequence, and then construct a low-dimensional MBR that contains multiple transformed sequences. In this paper we propose a formal method that transforms a high-dimensional MBR itself to a low-dimensional MBR, and show that this method significantly reduces the number of lower-dimensional transformations. To achieve this goal, we first formally define the new notion of MBR-safe. We say that a transform is MBR-safe if a low-dimensional MBR to which a high-dimensional MBR is transformed by the transform contains every individual low-dimensional sequence to which a high-dimensional sequence is transformed. We then propose two MBR-safe transforms based on DFT and DCT, the most representative lower-dimensional transformations. For this, we prove the traditional DFT and DCT are not MBR-safe, and define new transforms, called mbrDFT and mbrDCT, by extending DFT and DCT, respectively. We also formally prove these mbrDFT and mbrDCT are MBR-safe. Moreover, we show that mbrDFT(or mbrDCT) is optimal among the DFT-based(or DCT-based) MBR-safe transforms that directly convert a high-dimensional MBR itself into a low-dimensional MBR. Analytical and experimental results show that the proposed mbrDFT and mbrDCT reduce the number of lower-dimensional transformations drastically, and improve performance significantly compared with the $na\"{\i}ve$ transforms. These results indicate that our MBR- safe transforms provides a useful framework for a variety of applications that require the lower-dimensional transformation of high-dimensional MBRs.

• Journal of KIISE:Software and Applications
• /
• v.35 no.6
• /
• pp.392-403
• /
• 2008

User reviews are valuable information that can be used for various purposes. In particular, the product reviews on online shopping sites are important information which can directly affect the purchasing decision of the customers. In this paper, we present our design and implementation of a system for summarizing the customer's opinion and the features of each product by analyzing reviews on a commercial shopping site. During the analysis process, several natural language processing(NLP) techniques and the semantic dictionary were used. The semantic dictionary contains vocabularies that are used to express product features and customer's opinions. And it was constructed in semi-automatic way with the help of the tool we implemented. Furthermore, we discuss how to handle the vocabularies that have different meanings according to the context. We analyzed 1796 reviews about 20 products of 2 categories collected from an actual shopping site and implemented a novel ranking system. We obtained 88.94% for precision and 47.92% for recall on extracting opinion expression, which means our system can be applicable for real use.

• Journal of KIISE:Databases
• /
• v.37 no.1
• /
• pp.56-69
• /
• 2010

In blogspace, there are posts that deal with a common topic and bloggers that are interested in these posts. In this paper, we define a blog community as a group of these bloggers and posts. With a blog community, we can establish various business policies for target marketing, sharing high quality data, and mobilizing the activities in the blogspace. Unlike internet cafes, bloggers participate in blog communities without explicit membership. So, it is not easy to identify the members of a community. In this paper, we propose an effective approach for extracting a blog community that is related to a given topic. First, we choose seed posts that is highly related to a given topic, and select bloggers that are related to the topic with the seed posts. Then, we select posts that are related to the topic with the selected bloggers. By repeating this, we find all the posts and bloggers that are members of the community related to a given topic in blogspace. We verify the superiority of the proposed approach by analyzing extracted blog communities.

• Journal of Korea Multimedia Society
• /
• v.6 no.5
• /
• pp.909-920
• /
• 2003

Nowadays, IT for CRM has been growing and developed rapidly. Typical techniques are statistical analysis tools, on-line multidimensional analytical processing (OLAP) tools, and data mining algorithms (such neural networks, decision trees, and association rules). Among customer data, web log data is very important and to use these data efficiently, applying OLAP technology to analyze multi-dimensionally. To make OLAP cube, we have to precalculate multidimensional summary results in order to get fast response. But as the number of dimensions and sparse cells increases, data explosion occurs seriously and the performance of OLAP decreases. In this paper, we presented why the web log data sparsity occurs and then what kinds of sparsity patterns generate in the two and t.he three dimensions for OLAP. Based on this research, we set up the multidimensional data models and query models for benchmark with each sparsity patterns. Finally, we evaluated the performance of three OLAP systems (MS SQL 2000 Analysis Service, Oracle Express and C-MOLAP).

• The Journal of the Korea Contents Association
• /
• v.13 no.4
• /
• pp.427-439
• /
• 2013

With the fast development of complex science, lots of social networks are studied. We know that the social network is widely applied in analyzing issues in human culture, economics and web sciences. Recently we witness that some researchers began to compare the social network constructed from fiction literatures(literature social network) and the real social network obtained from practice. But we point that previous approaches for literature social network have some drawbacks since they completely depend on the biographical dictionary constructed for a designated literature. So since the previous approach focus on the few important characters and peoples around them, we can not understand the global structure of all characters appeared in the literature at least once. We propose one method to extract all characters appeared in the literature and how to make the social network from that information. Also we newly propose K-critical network by applying frequency of the named characters and the strength of relationship among all textual characters. Our experiment shows that the K-critical measure could be one crucial quantitative measure to compute the relationship strength among characters appeared in the object literature.

• Journal of KIISE:Databases
• /
• v.36 no.3
• /
• pp.226-234
• /
• 2009

A data stream is a massive unbounded sequence of data elements continuously generated at a rapid rate. Due to the characteristics of a data stream, it is impossible to save all the data elements of a data stream. Therefore it is necessary to define a new synopsis structure to store the summary information of a data stream. For this purpose, this paper proposes a cuboid prefix tree that can be effectively employed in evaluating an iceberg query over data streams. A cuboid prefix tree only stores those itemsets that consist of grouping attributes used in GROUP BY query. In addition, a cuboid prefix tree can compute multiple iceberg queries simultaneously by sharing their common sub-expressions. A cuboid prefix tree evaluates an iceberg query over an infinitely generated data stream while efficiently reducing memory usage and processing time, which is verified by a series of experiments.