• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.25 no.12
• /
• pp.1784-1789
• /
• 2021

Korean sentiment classification task is used in real-world services such as chatbots and analysis of user's purchase reviews. And due to the development of deep learning technology, neural network models with high performance are being applied. However, the neural network model is not easy to interpret what the input sentences are predicting due to which words, and recently, model interpretation methods for interpreting these neural network models have been popularly proposed. In this paper, we used the LIME algorithm among the model interpretation methods to interpret which of the words in the input sentences of the models learned with the korean sentiment classification dataset. As a result, the interpretation of the Bi-LSTM model with 85.24% performance included 25,283 words, but 84.20% of the transformer model with relatively low performance showed that the transformer model was more reliable than the Bi-LSTM model because it contains 26,447 words.

• Korean Security Journal
• /
• no.48
• /
• pp.177-206
• /
• 2016

The purpose of this study is to verify the factors that affect the fear of crime by applying the risk interpretation model. Especially, whereas previous studies have not proven micro individual factor that the risk interpretation model had presented, This study includes micro individual elements such as neighborhood factor, perceived risk of crime, fears of crime as main variables. This study utilized secondary data of the National Crime Victimization Survey 2012, conducted by the Korean Institute of Criminology. In this study, multiple regression analysis of two stages and Sobel Test were conducted for verifying the individual influence of each independent variables and identifying the causal relationship between the variables set out in the risk analysis model. As the result, it appeared that the higher level of perceived risk of crime, neighborhood factor, crime experience, education, income cause the higher degree of the fear of crime. On the other hand, the lower degree of age was found to induce the higher level of the fear of crime. In addition, female showed the higher degree of the fear of crime than man. The causal relationship between the variables set out in the risk interpretation model was presented significantly in all variables, except for education.

• Korean Management Science Review
• /
• v.15 no.2
• /
• pp.177-200
• /
• 1998

One of organization's generic functions is the interpretation of events to carry out decision-making activities. In intelligent Data Interpretation System(IDIS), Interpreting is computationally modeled as classification of new data into categories having similar features. We define the Extensional Object Model(ExOM) as a formalism for IDIS. In ExOM, objects and categories are loosely coupled to provide flexibility for both object description and category definition in data gathering and interpretation process. Objects are classified inductively based on exemplars of categories as well as deductively based on category structures.

• Economic and Environmental Geology
• /
• v.25 no.1
• /
• pp.79-85
• /
• 1992

An Interpretation technique is presented to determine strike, dip, velocity and depth of 3-D planar layers from refraction or reflection traveltime curve. This interpretation technique determines the direction of emerging ray from the slope of the traveltime curve and traces the emerging ray to the refractor or reflector. The ray direction in the last layer is used to decide the normal vector to the refractor or reflector from whick its dip, strike and velocity are calculated. The vertical depth to the refractor or reflector is computed by using the intercept or zero-offset time and the ray direction in each layer. Some tests on the interpretation method are performed for the sysnthetic traveltimes generated in 3-D model layers and show that the paramerters of the model layers are accurately determined.

• Journal of the Korean Association of Geographic Information Studies
• /
• v.13 no.4
• /
• pp.170-180
• /
• 2010

With development of the earth's subsurface space, the need for a reliable subsurface spatial model such as a cross-section, boring log is increasing. However, the ground mass was essentially uncertain. To generate model was uncertain because of the shortage of data and the absence of geotechnical interpretation standard(non-statistical uncertainty) as well as field environment variables(statistical uncertainty). Therefore, the current interpretation of the data and the generation of the model were accomplished by a highly trained experts. In this study, a geotechnical ontology model was developed using the current expert experience and knowledge, and the information content was calculated in the ontology hierarchy. After the relative distance between the information contents in the ontology model was combined with the distance between cluster centers, a cluster analysis that considered the geotechnical semantics was performed. In a comparative test of the proposed method, k-means method, and expert's interpretation, the proposed method is most similar to expert's interpretation, and can be 3D-GIS visualization through easily handling massive data. We expect that the proposed method is able to generate the more reasonable subsurface spatial information model without geotechnical experts' help.

• Journal of Internet Computing and Services
• /
• v.7 no.5
• /
• pp.123-134
• /
• 2006

In this research paper, We have proposed an interpretation structure concerned with context-awareness which is essential to provide intelligent service. The proposed model has an interpretation structure that integrated information of both external and internal context, and been designed to keep independency between context information and application service. To verify the interpretation structure that proposed, we established a test bed (Ubi-Class) in which the concept of context awareness has been introduced to the existing class room, and constituted a scenario, and tested whether intelligent service is performed according to context interpretation.

• Proceedings of the Korean Society of Propulsion Engineers Conference
• /
• 2004.03a
• /
• pp.532-543
• /
• 2004

The present work have been developed the interpretation processor including the behavior of material failure and the separation phenomena under transient dynamic loading (the operation of explosive bolt) using AUTODYN V4.3, SoildWork 2003 and TrueGrid V2.1 programs. It has been demonstrated that the interpretation in ridge-cut explosive bolt under dynamic loading condition should be necessary to the appropriate failure model and the basic stress of bolt failure is the principal stress. The use of this interpretation processor developing the present work could be extensively helped to design the shape and the amount of explosives in the explosive bolt having a complex geometry. It is also proved that the interpretation processor approach is an accurate and effective analysis technique to evaluate the separation mechanism in explosive bolts.

• Journal of the Korean Society for Nondestructive Testing
• /
• v.24 no.1
• /
• pp.8-14
• /
• 2004

Model-based inversion tools for eddy current signals have been developed by combining neural networks and finite element modeling, for quantitative flaw characterization in steam generator tubes. In the present work, interpretation of experimental eddy current signals was carried out in order to validate the developed inversion tools. A database was constructed using the synthetic flaw signals generated by the finite element model. The hybrid neural networks composed of a PNN classifier and BPNN size estimators were trained using the synthetic signals. Experimental eddy current signals were obtained from axisymmetric artificial flaws. Interpretation of flaw signals was conducted by feeding the experimental signals into the neural networks. The interpretation was excellent, which shows that the developed inversion tools would be applicable to the Interpretation of real eddy current signals.

• Proceedings of the Korea Society for Industrial Systems Conference
• /
• 2002.06a
• /
• pp.166-169
• /
• 2002

In this paper, implements the abstract interpretation of the imperative language While in SMV model-checker and explain how to apply the logic of CTL which example the security of information flow. And show the way to translate the abstract program of While into SMV program and explain the derive process of CTL logic to test the security of the information flow. For the various security test, it is suitable to use the model-checking than to implements the abstract interpretation.

• Communications for Statistical Applications and Methods
• /
• v.6 no.2
• /
• pp.601-610
• /
• 1999

Crossover design is officially except for special occasions recommended by KFDA and FDA for assessing Bioequivalence between two drugs one for reference and the other for innovator. Such design is regarded as a special case of latin square split-polt or repeated measurement design and its main difference with other designing methods is that each subject in an experiment is exposed two drugs in sequence. Therefore general statistical analysis is not suitable since the model for this experiment includes carryover effect in addition to period and sequence effect. in this paper analysis for crossover model with two drugs and its interpretation are mainly discussed and an example is given for illustration.