독후감 텍스트의 토픽모델링 적용에 관한 탐색적 연구

이수상 한국도서관·정보학회 2016 한국도서관정보학회지 Vol.47 No.4

The purpose of this study is to explore application of topic modeling for topic analysis of book report. Topic modeling can be understood as one method of topic analysis. This analysis was conducted with texts in 23 book reports using LDA function of the “topicmodels” package provided by R. According to the result of topic modeling, 16 topics were extracted. The topic network was constructed by the relation between the topics and keywords, and the book report network was constructed by the relation between book report cases and topics. Next, Centrality analysis was conducted targeting the topic network and book report network. The result of this study is following these. First, 16 topics are shown as network which has one component. In other words, 16 topics are interrelated. Second, book report was divided into 2 groups, book reports with high centrality and book reports with low centrality. The former group has similarities with others, the latter group has differences with others in aspect of the topics of book reports. The result of topic modeling is useful to identify book reports’ topics combining with network analysis. 이 연구는 독후감 텍스트의 주제분석에 토픽모델링의 활용방안을 탐색하는 것을 목적으로 하고 있다. 텍스트의 주제분석 방안으로서 토픽모델링 분석방법을 이해하고, R에서 제공하는 “topicmodels” 패키지의 LDA 함수를 사용하여 23건의 사례 독후감 텍스트들을 대상으로 실제의 분석작업을 수행하였다. 토픽모델링 분석결과 16개의 토픽들을 추출하였고, 토픽과 구성 단어들의 관계에서 토픽 네트워크, 사례 독후감과 토픽들의 관계에서 독후감 네트워크를 구성하였다. 이후 토픽 네트워크와 독후감 네트워크를 대상으로 중심성 분석을 수행하였으며, 분석결과는 다음과 같다. 첫째, 16개의 토픽들이 1개의 컴포넌트를 가지는 네트워크로 나타났다. 이것은 16개 토픽들이 상호 연관되어 있다는 것을 의미한다. 둘째, 독후감 네트워크에서는 연결정도 중심성이 높은 독후감들과 낮은 독후감들로 구분이 되었다. 전자의 독후감들은 다른 독후감들과 주제적으로 유사성을 가지며, 후자의 독후감들은 다른 독후감들과 주제적으로 상이성을 가지는 것으로 해석하였다. 토픽모델링의 결과를 네트워크 분석과 결합함으로써 독후감의 주제파악에 유용한 결과들을 얻게 되었다.

사회선택 통계모형의 방법론적 특성과 p2와 p*모형을 활용한 남자고등학생의 영어도움 네트워크 분석

김종민(Kim, Chong Min) 한국교육평가학회 2015 교육평가연구 Vol.28 No.3

본 연구의 목적은 사회 네트워크 분석 중 사회선택 통계모형의 방법론적 특성을 탐색하고 이를 활용하여 남자고등학생의 영어도움 네트워크 구조를 분석하는 것이다. 이에 첫째, 사회선택 통계모형인 p2 와 p* 모형을 중심으로 방법론적 특성을 탐색하였고, 둘째, 경기도 소재 남자 고등학생의 영어도움네트워크 자료를 토대로, p2 와 p* 모형을 활용하여 네트워크의 구조(호혜성과 전이성)와 고등학생의 특성(성취목표, 영어 학습동기 및 영어 학업성취도)이 영어도움 네트워크에 어떤 영향을 주는지 분석하였다. 먼저 방법론적 특성으로 p* 모형은 p2 모형에서 모형화하기 힘들었던 전이성을 포함한 복잡한 네트워크 구조를 모형 명세화 할 수 있는 반면, p2 모형은 전이성 등의 복잡한 네트워크 구조가 존재하지않는 네트워크나 네트워크 자체가 별로 없는 경우 네트워크의 다변량 분석과 다층분석 그리고 다양한모형 명세화가 가능하고 또한 추정의 수렴성과 추정의 표준오차의 효율성 측면에서 유리하다. 그리고 남자고등학생의 영어도움 네트워크를 p2 와 p* 모형으로 분석결과, 남자고등학생의 특성인 사회적 수행접근목표, 영어 학습 내재적 동기 그리고 영어 학업성취도 점수가 높을수록 영어도움 네트워크가 증가 하였고, 네트워크 구조인 호혜성과 전이성은 같은 반 남자 고등학생의 영어도움 네트워크에 긍정적인영향을 주었다. The purpose of this study is to explore methodological characteristics of social network analysis with a focus on social selection models and to analyze male high schools students’ help networks for English learning using p2 and p* models. Among social selection models, p2 and p* models were presented as statistical models and methodological characteristics of both models were introduced. Each social selection statistical mdoel has its own methodological characteristics. p* models can specify complex network structure like transitivity which could hardly be specified in p2 models. On the other hand, p2 models can analyze multivariate models and multilevel models and specify various models when social networks are sparse or simple without complex network structure-like transitivity, with an advantage in the convergence of estimation and efficiency of estimation errors. The application of p2 and p* models was demonstrated using male high school students’ social networks for English learning in Korea. The research question was to examine which attributes(achievement goals, English-learning motivation, and English achievement scores) and network structure(reciprocity and transitivity) affect English-learning help networks. The results of p2 and p* models indicated that the transitive triplets(transitivity), social performance-approach goals, the intrinsic motivation to learn English, and English achievement scores were statistically significant and positively related to providing and seeking help for English learning among students in the same classroom.

Prediction Model of Sports Performance Based on Grey BP Neural Network

Deng Kui,Xiao Liu,Xu Liang,Song Haiyan 보안공학연구지원센터 2016 International Journal of u- and e- Service, Scienc Vol.9 No.8

The best annual performances of the world women’s pentathlons during 2005~2013 are statistically collected in this article, and the prediction of the best performance of the world women’s heptathlon in 2013 is taken as the research object. According to the best annual performances of the world women’s heptathlons during 2005~2012, the sports performance prediction model composed of GM(1,1) grey prediction model and BP neural network prediction model in serial connection is established in this article, and this model is applied to predict the best annual performance of the world women’s heptathlon in 2013. Through the comparison of the actual value of the best annual performance of the world women’s heptathlon in 2013 and the predicted value of the model, the application of the grey BP neural network prediction model in sports performance prediction is researched and analyzed in this article. The research result shows that for the sports performance prediction problem, the grey BP neural network prediction model has the features of high prediction accuracy, simple application and strong generalization performance, and this model is also superior to single GM(1,1) grey prediction model and BP neural network model.

동적 소셜네트워크 구조 변수를 적용한 가상 재화 구매 모형 연구

이희태,배정호 한국유통과학회 2019 유통과학연구 Vol.17 No.3

Purpose – The existing marketing studies using Social Network Analysis have assumed that network structure variables are time-invariant. However, a node’s network position can fluctuate considerably over time and the node’s network structure can be changed dynamically. Hence, if such a dynamic structural network characteristics are not specified for virtual goods purchase model, estimated parameters can be biased. In this paper, by comparing a time–invariant network structure specification model(base model) and time-varying network specification model(proposed model), the authors intend to prove whether the proposed model is superior to the base model. In addition, the authors also intend to investigate whether coefficients of network structure variables are random over time. Research design, data, and methodology – The data of this study are obtained from a Korean social network provider. The authors construct a monthly panel data by calculating the raw data. To fit the panel data, the authors derive random effects panel tobit model and multi-level mixed effects model. Results – First, the proposed model is better than that of the base model in terms of performance. Second, except for constraint, multi-level mixed effects models with random coefficient of every network structure variable(in-degree, out-degree, in-closeness centrality, out-closeness centrality, clustering coefficient) perform better than not random coefficient specification model. Conclusion – The size and importance of virtual goods market has been dramatically increasing. Notwithstanding such a strategic importance of virtual goods, there is little research on social influential factors which impact the intention of virtual good purchase. Even studies which investigated social influence factors have assumed that social network structure variables are time-invariant. However, the authors show that network structure variables are time-variant and coefficients of network structure variables are random over time. Thus, virtual goods purchase model with dynamic network structure variables performs better than that with static network structure model. Hence, if marketing practitioners intend to use social influences to sell virtual goods in social media, they had better consider time-varying social influences of network members. In addition, this study can be also differentiated from other related researches using survey data in that this study deals with actual field data.

Application of Neural Network Model to Vehicle Emissions

김대현,이정 서울시립대학교 도시과학연구원 2010 도시과학국제저널 Vol.14 No.3

The issue of air quality is now a major concern around the world and the vehicle emissions model is very important. Most of the current vehicle emission models are multiple regression techniques. In this study, a neural network-based model has been proposed to achieve better estimation accuracy. The estimation performance of two models, the proposed neural network-based model and a general regression model, has been compared using mean absolute error (MAE). A comparative study between two models to estimate vehicle emissions, the proposed neural network-based model and a general regression model, has been conducted to assess the estimation performance of the proposed model in terms of mean absolute percentage error. Experimental results in this study revealed that the neural network model performed better as it was able to decrease the error for emission estimation comparing with the multiple regression models. More importantly, in this study a lookup table (LUT) method has been proposed to overcome the black-box problem, which is a disadvantage of the neural network models. It could be useful for any other researches to estimate emissions without developing and training the neural network model which can be a time-consuming task. The issue of air quality is now a major concern around the world and the vehicle emissions model is very important. Most of the current vehicle emission models are multiple regression techniques. In this study, a neural network-based model has been proposed to achieve better estimation accuracy. The estimation performance of two models, the proposed neural network-based model and a general regression model, has been compared using mean absolute error (MAE). A comparative study between two models to estimate vehicle emissions, the proposed neural network-based model and a general regression model, has been conducted to assess the estimation performance of the proposed model in terms of mean absolute percentage error. Experimental results in this study revealed that the neural network model performed better as it was able to decrease the error for emission estimation comparing with the multiple regression models. More importantly, in this study a lookup table (LUT) method has been proposed to overcome the black-box problem, which is a disadvantage of the neural network models. It could be useful for any other researches to estimate emissions without developing and training the neural network model which can be a time-consuming task.

Performance Prediction Model of University Students Based on the Grey BP Neural Network

Liao Yu,Liu Zongxin 보안공학연구지원센터 2016 International Journal of u- and e- Service, Scienc Vol.9 No.10

This article counted the best performance of students entrepreneurship courses from 2005 to 2014, and took the best performance prediction of 2014 entrepreneurship course as the research object. According to the best annual performance of entrepreneurship courses from 2005 to 2014, this article established the grade prediction model of series combination of GM (1, 1) grey prediction model and BP neural network prediction model, and the established model was used to predict the best annual performance of students entrepreneurship course. Through comparing the actual value of the best annual performance of 2014 entrepreneurship course and the predicted value c by the model, this article analyzed the application of grey BP neural network prediction model in the students entrepreneurship performance prediction. The research results showed that for entrepreneurship performance prediction problem, the grey BP neural network prediction model had high prediction precision , simple application, and it can be widely used, and had more advantages than single GM (1, 1) grey prediction model and BP neural network model.

Lightweight object detection network model suitable for indoor mobile robots

Lin Jiang,Wenkang Nie,Jianyang Zhu,Xumin Gao,Bin Lei 대한기계학회 2022 JOURNAL OF MECHANICAL SCIENCE AND TECHNOLOGY Vol.36 No.2

This work proposes a lightweight object detection network model ShuffleNetSSD (S-SSD) to solve the problem of single shot multibox detector (SSD) network model where it cannot meet the real-time performance requirement in the task of object detection and recognition of indoor mobile robot. This model is suitable for indoor mobile robot by improving the SSD network model based on ShuffleNet network. The main idea of the improvement is that SSSD replaces VGG-16 network as the basic feature extraction network of SSD network model with ShuffleNet network. The proposed model is based on the design of deep separable convolution, point-by-point grouping convolution, and channel rearrangement. It retains the design idea of multiscale feature graph detection of SSD network model. This model ensures a slight decline in detection accuracy while greatly reduces the amount of computation generated by the network operation, thereby greatly improving the detection rate. A data set for the task of object detection and recognition of indoor mobile robot is made. The S-SSD lightweight network model is superior to the original SSD network model and tiny-YOLO lightweight network model in terms of detection accuracy and detection rate, and can simultaneously meet the requirement of detection accuracy and real-time performance in the task of indoor object detection and recognition of mobile robot. These findings are verified through the comparative experiments of object detection accuracy and detection rate and real-time object detection and recognition of mobile robot under the actual indoor scene.

네트워크 중심성 척도가 추천 성능에 미치는 영향에 대한 연구

이동원(Dongwon Lee) 한국지능정보시스템학회 2021 지능정보연구 Vol.27 No.1

Collaborative filtering, which is often used in personalization recommendations, is recognized as a very useful technique to find similar customers and recommend products to them based on their purchase history. However, the traditional collaborative filtering technique has raised the question of having difficulty calculating the similarity for new customers or products due to the method of calculating similaritiesbased on direct connections and common features among customers. For this reason, a hybrid technique was designed to use content-based filtering techniques together. On the one hand, efforts have been made to solve these problems by applying the structural characteristics of social networks. This applies a method of indirectly calculating similarities through their similar customers placed between them. This means creating a customer’s network based on purchasing data and calculating the similarity between the two based on the features of the network that indirectly connects the two customers within this network. Such similarity can be used as a measure to predict whether the target customer accepts recommendations. The centrality metrics of networks can be utilized for the calculation of these similarities. Different centrality metrics have important implications in that they may have different effects on recommended performance. In this study, furthermore, the effect of these centrality metrics on the performance of recommendation may vary depending on recommender algorithms. In addition, recommendation techniques using network analysis can be expected to contribute to increasing recommendation performance even if they apply not only to new customers or products but also to entire customers or products. By considering a customer’s purchase of an item as a link generated between the customer and the item on the network, the prediction of user acceptance of recommendation is solved as a prediction of whether a new link will be created between them. As the classification models fit the purpose of solving the binary problem of whether the link is engaged or not, decision tree, k-nearest neighbors (KNN), logistic regression, artificial neural network, and support vector machine (SVM) are selected in the research. The data for performance evaluation used order data collected from an online shopping mall over four years and two months. Among them, the previous three years and eight months constitute social networks composed of and the experiment was conducted by organizing the data collected into the social network. The next four months’ records were used to train and evaluate recommender models. Experiments with the centrality metrics applied to each model show that the recommendation acceptance rates of the centrality metrics are different for each algorithm at a meaningful level. In this work, we analyzed only four commonly used centrality metrics: degree centrality, betweenness centrality, closeness centrality, and eigenvector centrality. Eigenvector centrality records the lowest performance in all models except support vector machines. Closeness centrality and betweenness centrality show similar performance across all models. Degree centrality ranking moderate across overall models while betweenness centrality always ranking higher than degree centrality. Finally, closeness centrality is characterized by distinct differences in performance according to the model. It ranks first in logistic regression, artificial neural network, and decision tree withnumerically high performance. However, it only records very low rankings in support vector machine and K-neighborhood with low-performance levels. As the experiment results reveal, in a classification model, network centrality metrics over a subnetwork that connects the two nodes can effectively predict the connectivity between two nodes in a social network. Furthermore, each metric has a different performance depending on the classification model type. This result implie

공급네트워크 구조와 재고 비용 성과의 관계

박철순(Chulsoon Park),김성학(Sunghak Kim) 한국생산관리학회 2017 韓國生産管理學會誌 Vol.28 No.1

최근 공급 사슬을 네트워크 관점에서 바라보는 연구가 활발히 진행되고 있음에도 불구하고 아직까지 네트워크 구조와 성 과 간의 관계를 명확히 밝히는데 한계가 있었다. 우리는 이론적 배경을 바탕으로 공급네트워크의 성과가 공급네트워크의 구조에 따라서 달라질 수 있음을 가설로 제시하였다. 이를 검정하기 위해 공급사슬의 대표적인 모형인 비어게임을 네트워 크 형태로 확장한 행위자 기반 모형을 세우고 모의실험을 진행하였다. 구조화된 실험 설계를 통한 모의실험으로 얻은 방 대한 데이터를 분석한 결과 공급네트워크의 구조를 나타내는 지표 중 하나인 평균 경로 길이(average path length)와 성과 간에 역U자 관계가 있음을 밝혔다. 다시 말해, 공급네트워크의 참여 기업 간 논리적 거리가 가깝거나 먼 경우가 논 리적 거리가 중간 정도인 경우에 비해 공급네트워크 재고 비용이 낮게 나타났다. 나아가 이러한 구조와 성과의 역U자 관 계는 리드타임이 길수록, 단위당 재고 유지비용이 클수록 강화됨을 확인했다. 또한, 주문 정책에 따라 역U자 관계가 다르 게 나타남을 보였다. Although supply chain research with a network perspective has been growing fast recently, there are some limitations to verifying a relationship between supply network structure and performance due to the lack of data availability and the complexity of mathematical modeling. Based on a theoretical background, we propose a hypothesis that suggests a curvilinear relationship between supply network structure and performance. In order to test the hypothesis, we develop an agent-based model that extends the Beer Distribution Game from a serial chain to a complex network structure. The model considers different factors that may affect the supply network performance such as lead time, ordering policy, demand pattern, and product cost characteristics. Through deliberately structured simulation experiments, we develop a huge data set at the supply network level. The results show that there is an inverted-U relationship between the average path length (APL) of a supply network and its performance. In other words, supply network performance increases and then decreases as its APL increases. This means that we may achieve minimum supply network costs when the supply network structure is either centralized such as with a hub and spoke structure or decentralized as with a tree structure. Furthermore, the inverted-U relationship between structure and performance is moderated by lead time, ordering policy and unit inventory holding cost. As the overall lead time of supply network members increases or the unit inventory holding cost rises, the slope of the inverted-U shape steepens.

토픽모델링과 네트워크 분석을 활용한 〈亂中日記〉 텍스트 연구

정성훈(Jung, Sung-hoon) 국어국문학회 2021 국어국문학 Vol.- No.197

텍스트 마이닝은 자연언어처리(NLP)와 형태소분석 기술에 기반하여 비정형화된 다량의 텍스트에서 유의미한 단어를 추출하고, 텍스트와 단어의 빈도를 고려하여 문맥(context) 수준의 의미를 찾아내는 방법이다. 이러한 텍스트 마이닝 방법 중에서 최근 각광받고 있는 방법 중의 하나가 토픽모델링(topic modeling)이다. 이에 본 연구에서는 먼저 토픽모델링의 알고리즘에 대한 소개를 하고, 한문 고전문헌의 일기텍스트 중에서 대표적인 〈난중일기〉를 대상으로 토픽모델링을 적용하였다. 연도별 ‧ 계절별로 〈난중일기〉에 나타나는 주제(topic)와 그 특징을 파악하고, 이를 네트워크 구조로 전환하여 중심성과 경향성을 파악해 보았다. 그 결과, 〈난중일기〉에는 10개의 숨겨진 주제들이 있었고, 대부분은 〈난중일기〉의 내용들과 밀접한 관련이 있는 내용들이었다. 특히 토픽모델링으로 추출된 10개의 주제 중에서 주제 2, 3, 4, 6 등 4개의 주제는 해석가능성이 아주 높았다. 또한 네트워크 분석 결과, 주제 3이 〈난중일기〉의 텍스트의 핵심을 이루는데, 통상적인 공무 활동의 일, 병사를 관리 ‧ 감독하는 일, 군수품 준비, 여가생활(음주, 바둑, 활쏘기 시합) 등의 내용이 중심으로 밝혀졌다. 본 연구는 한문 고전문헌의 하나인 〈난중일기〉를 텍스트 마이닝의 하나인 토픽모델링으로 분석해 보고자 한 점에서 의의가 있다고 할 수 있겠다. 나아가 이러한 토픽모델링 분석은 디지털화된 대량의 한문 고전문헌을 분석하는 데 유용한 방법이 될 수 있을 것이다. Text mining is a method of extracting meaningful words from a large amount of atypical texts based on natural language processing(NLP) and morpheme analysis. We can find latent word meanings in the context by analyzing the frequency of meaningful words. One of these text mining methods that has recently been in the spotlight is topic modeling. First of all, in this study, we will introduce the algorithm of topic modeling and try applying topic modeling to 〈nanjungilgi〉, which is a representative diary text in the classical Sino-Korean text. The purpose of this study is to examine the topics and the characteristics in 〈nanjungilgi〉, and to convert them into a network structure to understand its centrality and tendency. As a result, we can find 10 latent topics in 〈nanjungilgi〉, most of which were closely related to the contents of 〈nanjungilgi〉. In particular, among the 10 topics extracted from topic modeling analysis, 4 topics, such as topic 2, 3, 4, and 6, had very high interpretability. In addition, as a result of network analysis, topic 3 forms the core of the context of 〈nanjungilgi〉, which includes the work of normal public service, management and supervision of soldiers, preparation of munitions, leisure life(drinking, go game, archery matches, etc.). This study is meaningful in that it intends to analyze 〈nanjungilgi〉, which is one of the classical Sino-Korean text, using topic modeling. Furthermore, topic modeling analysis like this can be a useful method to analyze a large amount of digitized classical Sino-Korean text.