• ETRI Journal
• /
• 제44권5호
• /
• pp.769-779
• /
• 2022

Spectral clustering has become a typical and efficient clustering method used in a variety of applications. The critical step of spectral clustering is the similarity measurement, which largely determines the performance of the spectral clustering method. In this paper, we propose a novel spectral clustering algorithm based on the local similarity measure of shared neighbors. This similarity measurement exploits the local density information between data points based on the weight of the shared neighbors in a directed k-nearest neighbor graph with only one parameter k, that is, the number of nearest neighbors. Numerical experiments on synthetic and real-world datasets demonstrate that our proposed algorithm outperforms other existing spectral clustering algorithms in terms of the clustering performance measured via the normalized mutual information, clustering accuracy, and F-measure. As an example, the proposed method can provide an improvement of 15.82% in the clustering performance for the Soybean dataset.

• 한국음향학회지
• /
• 제34권4호
• /
• pp.303-309
• /
• 2015

화자 분할은 사전에 분류되지 않은 데이터를 각각의 화자로 분류하는 연구이며 DBSCAN(Density-Based Spatial Clustering of Applications with Noise)은 간결함과 계산의 효율성으로 인해 화자분할 분야에 널리 사용되어 왔다. 그러나 클러스터의 데이터들이 공간적이지 않으며 서로 다른 클러스터가 근접하여 경계를 공유할 때 오버클러스터링 문제가 발생하여 DBSCAN의 성능이 하락한다. 본 논문에서는 DBSCAN과 문제점을 설명하고, 개체의 지역적 특성에 기반한 밀도 기반 클러스터링 알고리즘을 제안한다. 제안하는 알고리즘은 개체의 지역적 밀도와 분산의 정도에 따라 가변적인 판단 기준을 탐색에 이용한다. DBSCAN과 제안 기법의 실험을 통해 성능을 비교하고 제안 기법의 효용을 보인다. 실험 결과 제안한 방법은 오버클러스터링이 발생하지 않으며 DBSCAN에 비해 보다 높은 정확도를 보여 지역적 특성을 이용한 접근 방법이 효과적임을 증명한다.

• 한국멀티미디어학회논문지
• /
• 제23권1호
• /
• pp.1-7
• /
• 2020

Clustering methods essentially take a two-step approach; extracting feature vectors for dimensionality reduction and then employing clustering algorithm on the extracted feature vectors. However, for clustering images, the traditional clustering methods such as stacked auto-encoder based k-means are not effective since they tend to ignore the local information. In this paper, we propose a method first to effectively reduce data dimensionality using convolutional auto-encoder to capture and reflect the local information and then to accurately cluster similar data samples by using a hierarchical clustering approach. The experimental results confirm that the clustering results are improved by using the proposed model in terms of clustering accuracy and normalized mutual information.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제14권10호
• /
• pp.3972-3988
• /
• 2020

Existing methods always rely on statistical features to extract local words for microblog user geolocation. There are many non-local words in extracted words, which makes geolocation accuracy lower. Considering the statistical and semantic features of local words, this paper proposes a microblog user geolocation method by extracting local words based on word clustering and wrapper feature selection. First, ordinary words without positional indications are initially filtered based on statistical features. Second, a word clustering algorithm based on word vectors is proposed. The remaining semantically similar words are clustered together based on the distance of word vectors with semantic meanings. Next, a wrapper feature selection algorithm based on sequential backward subset search is proposed. The cluster subset with the best geolocation effect is selected. Words in selected cluster subset are extracted as local words. Finally, the Naive Bayes classifier is trained based on local words to geolocate the microblog user. The proposed method is validated based on two different types of microblog data - Twitter and Weibo. The results show that the proposed method outperforms existing two typical methods based on statistical features in terms of accuracy, precision, recall, and F1-score.

• 디지털산업정보학회논문지
• /
• 제10권2호
• /
• pp.47-59
• /
• 2014

In this paper, I propose a new energy efficient clustering scheme to prolong the network lifetime by reducing energy consumption at the sensor node. It is possible that a node determines whether to participate in clustering with certain probability based on local density. This scheme is useful under the environment that sensor nodes are deployed unevenly within the sensing area. By adjusting the probability of participating in clustering dynamically with local density of nodes, the energy consumption of the network is reduced. So, the lifetime of the network is extended. In the region where nodes are densely deployed, it is possible to reduce the energy consumption of the network by limiting the number of node which is participated in clustering with probability which can be adjusted dynamically based on local density of the node. Through computer simulation, it is verified that the proposed scheme is more energy efficient than LEACH protocol under the environment where node are densely located in a specific area.

• Communications for Statistical Applications and Methods
• /
• 제19권2호
• /
• pp.213-217
• /
• 2012

Fuzzy k-means clustering is an attractive alternative to the ordinary k-means clustering in analyzing multivariate data. Fuzzy versions yield more natural output by allowing overlapped k groups. In this study, we modify a fuzzy k-means clustering algorithm to be used for undirected social networks, apply the algorithm to both real and simulated cases, and report the results.

• 한국지능시스템학회논문지
• /
• 제22권1호
• /
• pp.81-86
• /
• 2012

본 논문에서는 전형적인 Linear Discriminant Analysis을 확장시켜 전체 입력공간을 다수의 지역공간으로 분할하고 분할된 공간에 Local Linear Discriminant Analysis 기반으로 하여 패턴 분류 규칙을 설계하는 새로운 방법을 제안한다. 전체 입력공간을 여러 개의 지역공간으로 분할하기 위한 방법으로 unsupervised clustering의 대표적인 방법인 k-Means 클러스터링 기법과 최적화 알고리즘인 차분 진화 연산 알고리즘을 사용한다. 제안된 알고리즘의 성능 평가를 위해 기존의 패턴 분류기와 비교 결과를 제시한다.

• 제어로봇시스템학회논문지
• /
• 제13권1호
• /
• pp.46-51
• /
• 2007

In this paper, we propose a new data clustering method using local probability and hypothesis theory. To cluster the test data set we analyze the local area of the test data set using local probability distribution and decide the candidate class of the data set using mean standard deviation and variance etc. To decide each class of the test data, statistical hypothesis theory is applied to the decided candidate class of the test data set. For evaluating, the proposed classification method is compared to the conventional fuzzy c-mean method, k-means algorithm and Discriminator analysis algorithm. The simulation results show more accuracy than results of fuzzy c-mean method, k-means algorithm and Discriminator analysis algorithm.

• 전기학회논문지
• /
• 제58권11호
• /
• pp.2269-2274
• /
• 2009

Customer clustering method is used to make a TDLP (typical daily load profile) to estimate the quater hourly load profile of non-AMR (Automatic Meter Reading) customer. In this paper, repeated small-sized clustering method is supposed to improve the classifying ability of TDLP. K-means algorithm is well-known clustering technology of data mining. To reduce the local maxima of k-means algorithm, proposed method clusters average load profiles to small-sized clusters and selects the highest error rated cluster and clusters this to small-sized clusters repeatedly to minimize the local maxima.

• 대한산업공학회지
• /
• 제29권1호
• /
• pp.90-99
• /
• 2003

Object clustering, which makes classification for a set of objects into a number of groups such that objects included in a group have similar characteristic and objects in different groups have dissimilar characteristic each other, has been exploited in diverse area such as information retrieval, data mining, group technology, etc. In this study, an object-clustering problem with similarity coefficients between objects is considered. At first, an evaluation function for the optimization problem is defined. Then, a genetic algorithm and local optimization technique based on heuristic method are proposed and used in order to obtain near optimal solutions. Solutions from the genetic algorithm are improved by local optimization techniques based on object relocation and cluster merging. Throughout extensive experiments, the validity and effectiveness of the proposed algorithms are tested.