• Proceedings of the Acoustical Society of Korea Conference
• /
• 1995.06a
• /
• pp.83-87
• /
• 1995

Keyword Spotting 이란 음성인식의 한 분야로서 입력된 음성에서 미리 정해진 특정단어 또는 복수 개의 단어들 중 어느 것이 포함되어 있는지의 여부를 찾아내고 이 단어를 식별해 내는 작업을 의미한다. 음소모델을 이용하여 Keyword Spotting 시스템을 구성할 경우 새로운 keyword의 추가 또는 변경이 필요할 때 단순히 그 발음사전에 따라 음소모델들을 연결시킴으로써 keyword 모델을 구성할 수 있으므로 단어모델에 의한 방법에 비해 장점이 있다. 본 논문에서는 triphone을 기본단위로 하는 HMM 에 의해 keyword 모델을 구성하고, non-keyword 모델 및 silence 모델을 함께 사용하는 keyword spotting 시스템을 구성하였다. 이러한 시스템에서 non-keyword 모델은 keyword와 keyword가 아닌 음성을 구분 지어주는 역할을 하므로 인식성능의 향상을 위해서는 적절한 non-keyword 모델의 선택이 필요하다. 본 논문에서는 10개의 state를 갖는 단일모델, 조음방법에 의해 음소들을 clustering 한 모델, 그리고 통계적 방법에 의해 음소들을 clustering 한 모델들을 각각 non-keyword 모델로 사용하여 그 성능을 비교하였다. 6개의 keyword를 대상으로 한 화자독립 keyword spotting 실험결과, 통계적 방법에 의해 음소들을 6 또는 7개의 그룹으로 clustering 한 방법이 가장 우수한 인식성능을 나타냈다.

• Speech Sciences
• /
• v.10 no.3
• /
• pp.225-239
• /
• 2003

In this paper, we develop a keyword spotting system using vocabulary-independent speech recognition technique, and investigate several non-keyword modeling and post-processing methods to improve its performance. In order to model non-keyword speech segments, monophone clustering and Gaussian Mixture Model (GMM) are considered. We employ likelihood ratio scoring method for the post-processing schemes to verify the recognition results, and filler models, anti-subword models and N-best decoding results are considered as an alternative hypothesis for likelihood ratio scoring. We also examine different methods to construct anti-subword models. We evaluate the performance of our system on the automatic telephone exchange service task. The results show that GMM-based non-keyword modeling yields better performance than that using monophone clustering. According to the post-processing experiment, the method using anti-keyword model based on Kullback-Leibler distance and N-best decoding method show better performance than other methods, and we could reduce more than 50% of keyword recognition errors with keyword rejection rate of 5%.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.19 no.11
• /
• pp.758-764
• /
• 2018

Digital documents such as patents, scholarly papers and research reports have author keywords which summarize the topics of documents. Different documents are likely to describe the same topic if they share the same keywords. Document clustering aims at clustering documents to similar topics with an unsupervised learning method. However, it is difficult to apply to a large amount of documents event though the document clustering is utilized to in various data analysis due to computational complexity. In this case, we can cluster and connect massive documents using keywords efficiently. Existing bottom-up hierarchical clustering requires huge computation and time complexity for clustering a large number of keywords. This paper proposes an inverted index based bottom-up clustering for keywords and analyzes the results of clustering with massive keywords extracted from scholarly papers and research reports.

• The Journal of the Acoustical Society of Korea
• /
• v.25 no.7
• /
• pp.319-324
• /
• 2006

We Propose two new methods for non-keyword modeling to improve the performance of speaker- and vocabulary-independent keyword spotting system. The first method is decision tree clustering of monophone at the state level instead of monophone clustering method based on K-means algorithm. The second method is multi-state multiple mixture modeling at the syllable level rather than single state multiple mixture model for the non-keyword. To evaluate our method, we used the ETRI speech DB for training and keyword spotting test (closed test) . We also conduct an open test to spot 100 keywords with 400 sentences uttered by 4 speakers in an of fce environment. The experimental results showed that the decision tree-based state clustering method improve 28%/29% (closed/open test) than the monophone clustering method based K-means algorithm in keyword spotting. And multi-state non-keyword modeling at the syllable level improve 22%/2% (closed/open test) than single state model for the non-keyword. These results show that two proposed methods achieve the improvement of keyword spotting performance.

• Journal of the Korean Operations Research and Management Science Society
• /
• v.40 no.4
• /
• pp.49-63
• /
• 2015

Keyword indexed in multidisciplinary journals show trends about science and technology innovation. Nature and Science were selected as multidisciplinary journals for our analysis. In order to reduce the effect of plurality of keyword, stemming algorithm were implemented. After this process, we fitted growth curve of keyword (stem) following bass model, which is a well-known model in diffusion process. Bass model is useful for expressing growth pattern by assuming innovative and imitative activities in innovation spreading. In addition, we construct keyword co-occurrence network and calculate network measures such as centrality indices and local clustering coefficient. Based on network metrics and yearly frequency of keyword, time series analysis was conducted for obtaining statistical causality between these measures. For some cases, local clustering coefficient seems to Granger-cause yearly frequency of keyword. We expect that local clustering coefficient could be a supportive indicator of emerging science and technology.

• Journal of Internet Computing and Services
• /
• v.14 no.6
• /
• pp.107-115
• /
• 2013

Various networks can be observed in the world. Knowledge networks which are closely related with technology and research are especially important because these networks help us understand how knowledge is produced. Therefore, many studies regarding knowledge networks have been conducted. The assortativity coefficient represents the tendency of connections between nodes having a similar property as figures. The relevant characteristics of the assortativity coefficient help us understand how corresponding technologies have evolved in the keyword-based patent network which is considered to be a knowledge network. The relationships of keywords in a knowledge network where a node is depicted as a keyword show the structure of the technology development process. In this paper, we suggest two hypotheses basedon the previous research indicating that there exist core nodes in the keyword network and we conduct assortativity analysis to verify the hypotheses. First, the patents network based on the keyword represents disassortativity over time. Through our assortativity analysis, it is confirmed that the knowledge network shows disassortativity as the network evolves. Second, as the keyword-based patents network becomes disassortavie, clustering coefficients become lower. As the result of this hypothesis, weconfirm the clustering coefficient also becomes lower as the assortative coefficient of the network gets lower. Another interesting result concerning the second hypothesis is that, when the knowledge network is disassorativie, the tendency of decreasing of the clustering coefficient is much higher than when the network is assortative.

• Journal of the Korean Society for Library and Information Science
• /
• v.44 no.3
• /
• pp.155-173
• /
• 2010

In this study, a new keyword extraction algorithm is applied to a single document with term clustering. A single document is divided by multiple passages, and two ways of calculating similarities between two terms are investigated; the first-order similarity and the second-order distributional similarity. In this experiment, the best cluster performance is achieved with a 50-term passage from the second-order distributional similarity. From the results of first experiment, the second-order distribution similarity was also applied to various keyword extraction methods using statistic information of terms. In the second experiment, pf(paragraph frequency) and $tf{\times}ipf$(term frequency by inverse paragraph frequency) were found to improve the overall performance of keyword extraction. Therefore, it showed that the algorithm fulfills the necessary conditions which good keywords should have.

• The Journal of the Korea Contents Association
• /
• v.10 no.5
• /
• pp.29-36
• /
• 2010

Past clustering researches are focused on extraction of keyword for word similarity grouping. However, too many candidates to compare and compute bring high complexity, low speed and low accuracy. To overcome these weaknesses, this paper proposed a topical web document clustering model using not only keyword but also named entities such as person name, organization, location, and so on. By several experiments, we prove effects of our model compared with traditional model based on only keyword and analyze how different effects show according to characteristics of document collection.

• The KIPS Transactions:PartD
• /
• v.16D no.6
• /
• pp.859-870
• /
• 2009

Despite the surge in the research for keyword search method over relational database, only little attention has been devoted to studying on relational data stream.The research for keyword search over relational data stream is intense interest because streaming data is recently a major research topic of growing interest in the data management. In this regard we first analyze the researches related to keyword search methodover relational data stream, and then this paper focuses on the method of minimizing the join cost occurred while processing keyword search queries. As a result, we propose an advanced keyword search method that can yield more meaningful results for users on relational data streams. We also propose a query optimization method using layered-clustering for efficient query processing.

• The KIPS Transactions:PartB
• /
• v.15B no.1
• /
• pp.73-80
• /
• 2008

In this paper, an advertisement-keyword finding method using document clustering is proposed to solve problems by ambiguous words and incorrect identification of main keywords. News articles that have similar contents and the same advertisement-keywords are clustered to construct the contextual information of advertisement-keywords. In addition to news articles, the web page and summary of a product are also used to construct the contextual information. The given document is classified as one of the news article clusters, and then cluster-relevant advertisement-keywords are used to identify keywords in the document. We could achieve 21% precision improvement by our proposed method.