Search

Many students fail because of academic and financial difficulties and dissatisfaction with their instruction and academic environment. Educational data mining(EDM) community tries to find solutions for such problems by mining student's data. There are a wide variety of current methods popular within educational data mining but the topics of EDM research are changing. Discovery with models has recently gained widespread use in EDM and it was the second most common category of EDM research by 2008-2009. In this article an important sub method of discovery with models (model based cluster analysis) is addressed. We explain the rule of MBCA in EDM, MBCA algorithms, interests of MBCA, and an experimental comparison of MBCA algorithms, and attributes selection methods on student's data.

The judgment that whether an email is spam or non-spam may vary from person to person. Different individuals can have totally different responses to the same email based on their preferences. This paper presents an innovative approach that incorporates user preferences to construct an anti-spam mail system, which is different from the conventional content-based approaches. We build a user preference ontology to formally represent the important concepts and rules derived from a data mining process. Then we use an inference engine that utilizes the knowledge to predict the user's action on new incoming emails. We also suggest a new rule optimization procedure inspired from logic synthesis to improve comprehensibility and exclude redundant rules. Experimental results showed that our user preference based architecture achieved good performance and the rules derived from the architecture and the optimization method have better quality in terms of comprehensibility.

Dissemination techniques of data have been rapidly developed for wireless communication environment. Since the advent of ubiquitous computing, semantic-based researches have been sped-up between service provider and user. We have tried to maintain user information with a semantic cache scheme. In this paper, we propose a semantic prefetching (SP) strategy for efficient data management, considering semantic information of user. We concentrate on data prefetching methods to send mass data to high bandwidth channels. In the SP, the prepared semantic descriptions are prefetched from server to client according as the client locates in anywhere. The semantic descriptions are dynamically composed of user patterns, such as the request patterns of user, the size of querying area, the direction of user and the average speed of user. We also propose a new cache management scheme, called to semantic least recently used (S-LRU). The evaluation result shows that the system performance is significantly improved.

Over the last few years, the increase in spatial data has led to more research on spatial indexing. Most studies, however, are based on adding or changing various options in R-tree, and few studies have focused on increasing search performance via minimum bounding rectangle (MBR) compression. In a spatial index, a greater number of node entries lowers tree heights and decreases the number of node accesses, thereby shrinking disk I/O. This study proposes a new MBR compression scheme using semi-approximation (SA) and SAR-tree, which indexes spatial data using R-tree. Since SA decreases the size of MBR keys, halves QMBR enlargement, and increases node utilization, it improves the overall search performance. This study mathematically analyzes the number of node accesses and evaluates the performance of SAR-tree using real location data. The results show that the proposed index performs better than existing MBR compression schemes.

In mobile computing, a huge number of users expect various data services based on their location. Wireless data broadcast is suitable for meeting requests from lots of users owing to its scalability. Air indexing techniques have been developed for energy efficient query processing of mobile clients in the wireless data broadcast. Especially for spatial queries related to the user's location, the index based on Hilbert Curve is proposed. However, the index makes clients listen to a lot of data objects in order to process range queries. In this paper, we describe indexing technique for spatial range queries. And we propose the distributed air index based on space partition. For the performance evaluation, it is implemented using a discrete event-driven simulation package, SimJava. A Simulation is conducted with the real dataset that contains 5922 cities of Greece. The results show that the proposed index outperforms the index based on the Hilbert Curve in the energy efficiency and query processing time significantly.

The increased need for spatial data for location-based services or geographical information systems (GISs) in mobile computing has led to more research on spatial indexing, such as R-tree. The R-tree variants approximate spatial data to a minimal bounding rectangle (MBR). Most studies are based on adding or changing various options in R-tree, while a few studies have focused on increasing search performance via MBR compression. This study proposes a novel MBR compression scheme that uses semi-approximation (SA) MBRs and SAR-tree. Since SA decreases the size of MBR keys, halves QMBR enlargement, and increases node utilization, it improves the overall search performance. This scheme decreases quantized space more than existing quantization schemes do, and increases the utilization of each disk allocation unit. This study mathematically analyzes the number of node accesses and evaluates the performance of SAR-tree using real location data. The results show that the proposed index performs better than existing MBR compression schemes.

This paper addresses the problem of monitoring continuous k-nearest neighbor (k-NN) queries. In order to support highly dynamic environments, where objects and/or queries are frequently moving, monitoring continuous k-NN require real-time updated results when objects and/or queries change their locations. Thus, it is important to minimize time delay for maintaining up to date the results. In this paper, we present the monitoring method to shorten time delay for updating continuous k-NN queries based on the notion of result region and the minimum bounding rectangle enclosing all objects inside each cell, referred to as cMBR, in the main-memory grid index structure. Simulations are conducted to show the efficiency of the proposed method.

Studies on spatial index, which is used for location-based services in mobile computing or GIS have increased in proportion to the increase in the spatial data. However, these studies were on the indices based on R-tree, and there are a few studies on how to increase the search performance of the spatial data by compressing MBRs. This study was conducted in order to propose a new MBR compression scheme, SA (Semi-approximation), and a SAR-tree that indexes spatial data using R-tree. The basic idea of this paper is the compression of MBRs in a spatial index. Since SA decreases the size of MBR keys, halves QMBR enlargement, and increases node utilization. Therefore, the SAR-tree heightens the overall search performance. The experiments show that the proposed index has increased performance, higher than that of the pre-established schemes on compression of MBRs.

We present a feature selection method by fuzzy inference and its application to spam-mail filtering in this work. The proposed fuzzy inference method outperforms information gain and chi squared test methods as a feature selection method in terms of error rate. In the case of junk mails, since the mail body has little text information, it provides insufficient hints to distinguish spam mails from legitimate ones. To address this problem, we follow hyperlinks contained in the email body, fetch contents of a remote web page, and extract hints from both original email body and fetched web pages. A two-phase approach is applied to filter spam mails in which definite hint is used first, and then less definite textual information is used. In our experiment, the proposed two-phase method achieved an improvement of recall by 32.4% on the average over the 1st phase or the 2nd phase only works.

This paper gives a comparative study of feature selection methods in spam-mail filtering. In our experiment, the fuzzy inference method showed about 6% and 10% improvements over information gain and χ2-test as a feature selection method in terms of the average error rate which is more important than typical information retrieval measures. Since it is not easy to reduce error rate, our work can be regarded as a meaningful research for email users suffering from unsolicited emails flooding indiscriminately.

This paper presents a method to reduce features less contributing to the classification of user preferred news groups among several news groups by the use of fuzzy inference and coefficient of determination. To this end, we extract a number of representative keywords from example documents through fuzzy inference. From the observation of training patterns, we found that lots of keywords in training patterns are empty. Thus, a new method to train neural network through reduction of unnecessary dimensions by the statistical coefficient of determination is proposed in this paper. Experimental results show that the proposed method is superior to the method using lots of input attributes in terms of within-cluster variance and its standard deviation.

With the development of e-commerce and information access, a large amount of information can be found online, which makes a good recommendation service to be urgently necessary. While many collaborative recommender systems (CRS) have succeeded in capturing the similarity among users or items based on ratings, there are still some challenges for them to be a more efficient RS. In this paper, we address three problems in CRS, that is user bias, nontransitive association, and new item problem, and show that the ICHM suggested in our previous work is able to solve the addressed problems. A series of experiments are carried out to show that our approach is feasible.

In this work, we present a service determining user's preferred news groups among various ones. For this end, candidate terms from example documents of each news group are extracted and a number of representative keywords among them are chosen through fuzzy inference. They are then presented to Kohonen network for learning representative keywords of each news group. From the observation of training patterns, we could find the sparseness problem that lots of keywords in training patterns are empty. Thus, a method to train neural network through reduction of unnecessary dimensions by the statistical coefficient of determination is used in this paper. Experimental results show that the method is superior to the method using every input dimension in terms of cluster overlap defined by using within-cluster distance and between-clusters distance.