• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제18권3호
• /
• pp.591-609
• /
• 2024

In this study, preprocessings with all combinations were examined in terms of the effects on decreasing word number, shortening the duration of the process and the classification success in balanced and imbalanced datasets which were unbalanced in different ratios. The decreases in the word number and the processing time provided by preprocessings were interrelated. It was seen that more successful classifications were made with Turkish datasets and English datasets were affected more from the situation of whether the dataset is balanced or not. It was found out that the incorrect classifications, which are in the classes having few documents in highly imbalanced datasets, were made by assigning to the class close to the related class in terms of topic in Turkish datasets and to the class which have many documents in English datasets. In terms of average scores, the highest classification was obtained in Turkish datasets as follows: with not applying lowercase, applying stemming and removing stop words, and in English datasets as follows: with applying lowercase and stemming, removing stop words. Applying stemming was the most important preprocessing method which increases the success in Turkish datasets, whereas removing stop words in English datasets. The maximum scores revealed that feature selection, feature size and classifier are more effective than preprocessing in classification success. It was concluded that preprocessing is necessary for text classification because it shortens the processing time and can achieve high classification success, a preprocessing method does not have the same effect in all languages, and different preprocessing methods are more successful for different languages.

• International Journal of Internet, Broadcasting and Communication
• /
• 제16권1호
• /
• pp.99-105
• /
• 2024

Runway released an updated version, Gen-2, in March 2023, which introduced new features that are different from Gen-1: it can convert text and images into videos, or convert text and images together into video images based on text instructions. This update will be officially open to the public in June 2023, so more people can enjoy and use their creativity. With this new feature, users can easily transform text and images into impressive video creations. However, as with all new technologies, comes the instability of AI, which also affects the results generated by Runway. This article verifies the feasibility of using Runway to generate the desired video from several aspects through personal practice. In practice, I discovered Runway generation problems and propose improvement methods to find ways to improve the accuracy of Runway generation. And found that although the instability of AI is a factor that needs attention, through careful adjustment and testing, users can still make full use of this feature and create stunning video works. This update marks the beginning of a more innovative and diverse future for the digital creative field.

• International Journal of Contents
• /
• 제14권1호
• /
• pp.1-6
• /
• 2018

Text, as one of the most influential inventions of humanity, has played an important role in human life since ancient times. The rich and precise information embodied in text is very useful in a wide range of vision-based applications such as the text data extracted from images that can provide information for automatic annotation, indexing, language translation, and the assistance systems for impaired persons. Therefore, natural-scene text detection with active research topics regarding computer vision and document analysis is very important. Previous methods have poor performances due to numerous false-positive and true-negative regions. In this paper, a fully-convolutional-network (FCN)-based method that uses supervised architecture is used to localize textual regions. The model was trained directly using images wherein pixel values were used as inputs and binary ground truth was used as label. The method was evaluated using ICDAR-2013 dataset and proved to be comparable to other feature-based methods. It could expedite research on text detection using deep-learning based approach in the future.

• 스마트미디어저널
• /
• 제2권1호
• /
• pp.31-36
• /
• 2013

A robust and efficient text extraction is very important for an accuracy of Optical Character Recognition (OCR) systems. Natural scene images with degradations such as uneven illumination, perspective distortion, complex background and multi color text give many challenges to computer vision task, especially in text extraction. In this paper, we propose a method for extraction of the text in signboard images based on a combination of mean shift algorithm and weighting scheme of hue and saturation in HSI color space for clustering algorithm. The number of clusters is determined automatically by mean shift-based density estimation, in which local clusters are estimated by repeatedly searching for higher density points in feature vector space. Weighting scheme of hue and saturation is used for formulation a new distance measure in cylindrical coordinate for text extraction. The obtained experimental results through various natural scene images are presented to demonstrate the effectiveness of our approach.

• 정보관리학회지
• /
• 제22권3호
• /
• pp.261-287
• /
• 2005

이 논문의 목적은 SVM(지지벡터기계) 분류기의 성능을 문헌간 유사도를 이용해서 향상시키는 것이다. SVM은 효과적인 기계학습 시스템으로서 최고 수준의 문헌자동분류 기술로 인정받고 있다. 이 연구에서는 문헌 벡터 자질 표현에 기반한 SVM 문헌자동분류를 제안하였다. 제안한 방식은 분류 자질로 색인어 대신 문헌 벡터를, 자질 값으로 가중치 대신 벡터유사도를 사용한다. 제안한 방식에 대한 실험 결과, SVM 분류기의 성능을 향상시킬 수 있었다. 실행 효율 향상을 위해서 문헌 벡터 자질 선정 방안과 범주 센트로이드 벡터를 사용하는 방안을 제안하였다. 실험 결과 소규모의 벡터 자질 집합만으로도 색인어 자질을 사용하는 기존 방식보다 나은 성능을 얻을 수 있었다.

• 정보관리학회지
• /
• 제29권2호
• /
• pp.155-171
• /
• 2012

텍스트 범주화에 있어서 일반적인 문제는 문헌을 표현하는 핵심적인 용어라도 학습문헌 집합에 나타나지 않으면 이 용어는 분류자질로 선정되지 않는다는 것과 형태가 다른 동의어들은 서로 다른 자질로 사용된다는 점이다. 이 연구에서는 위키피디아를 활용하여 문헌에 나타나는 동의어들을 하나의 분류자질로 변환하고, 학습문헌 집합에 출현하지 않은 입력문헌의 용어를 가장 유사한 학습문헌의 용어로 대체함으로써 범주화 성능을 향상시키고자 하였다. 분류자질 선정 실험에서는 (1) 비학습용어 추출 시 범주 정보의 사용여부, (2) 용어의 유사도 측정 방법(위키피디아 문서의 제목과 본문, 카테고리 정보, 링크 정보), (3) 유사도 척도(단순 공기빈도, 정규화된 공기빈도) 등 세 가지 조건을 결합하여 실험을 수행하였다. 비학습용어를 유사도 임계치 이상의 최고 유사도를 갖는 학습용어로 대체하여 kNN 분류기로 분류할 경우 모든 조건 결합에서 범주화 성능이 0.35%~1.85% 향상되었다. 실험 결과 범주화 성능이 크게 향상되지는 못하였지만 위키피디아를 활용하여 분류자질을 선정하는 방법이 효과적인 것으로 확인되었다.

• 한국정보처리학회논문지
• /
• 제3권5호
• /
• pp.1294-1306
• /
• 1996

본 논문에서는 한글 문서에 대해 높은 압축률을 얻기 위해 한글의 형태적 특징인 조사와 어말어미의 출현 빈도를 이용한 효율적인 한글 문서 압축 기법들을 제안하였으며 제안된 기법들의 성능 분석을 위하여 기존의 압축 기법들과 압축률을 비교 분석하였다. 한글 문서에서 조사와 어말어미가 반복적으로 출현한다는 형태적인 특성으로부터 높은 압축률을 얻기 위해 출현 빈도가 상대적으로 높은 64개의 조사 및 어말어미를 선정 하여 고정 사전을 구성하고, 이를 이용하여 한글 문서를 압축하도록 기존의 LZ77기법과 LZW기법을 수정하여 각각 HLZ77기법과 HLZW기법을 제안하였다. 또한, 본 연구에서는 수정 제안된 HLZ77기법과 HLZW기법의 성능을 분석하기 위하여 4가지 기법을 실 제 재현하여 여러 형태의 한글 문서를 대상으로 압축률을 비교하였다. 성능 결과로 부터 일반적인 한글 문서에 대해 한글의 형태적인 특성을 이용하는 HLZ77기법과 HLZW 기법이 각각 LZ77기법과 LZW기법 보다 우수한 압축률을 나타냄을 알 수 있었다.

• Journal of Information Processing Systems
• /
• 제8권4호
• /
• pp.575-588
• /
• 2012

Exploiting unlabeled text data with a relatively small labeled corpus has been an active and challenging research topic in text mining, due to the recent growth of the amount of biomedical literature. Biomedical named-entity recognition is an essential prerequisite task before effective text mining of biomedical literature can begin. This paper proposes an Active Co-Training (ACT) algorithm for biomedical named-entity recognition. ACT is a semi-supervised learning method in which two classifiers based on two different feature sets iteratively learn from informative examples that have been queried from the unlabeled data. We design a new classification problem to measure the informativeness of an example in unlabeled data. In this classification problem, the examples are classified based on a joint view of a feature set to be informative/non-informative to both classifiers. To form the training data for the classification problem, we adopt a query-by-committee method. Therefore, in the ACT, both classifiers are considered to be one committee, which is used on the labeled data to give the informativeness label to each example. The ACT method outperforms the traditional co-training algorithm in terms of f-measure as well as the number of training iterations performed to build a good classification model. The proposed method tends to efficiently exploit a large amount of unlabeled data by selecting a small number of examples having not only useful information but also a comprehensive pattern.

• 인터넷정보학회논문지
• /
• 제20권1호
• /
• pp.1-10
• /
• 2019

빅 데이터에서 텍스트 마이닝은 많은 수의 데이터로부터 많은 특징 추출하기 때문에, 클러스터링 및 분류 과정의 계산 복잡도가 높고 분석결과의 신뢰성이 낮아질 수 있다. 특히 텍스트마이닝 과정을 통해 얻는 Term document matrix는 term과 문서간의 특징들을 표현하고 있지만, 희소행렬 형태를 보이게 된다. 본 논문에서는 탐지모델을 위해 텍스트마이닝에서 개선된 GA(Genetic Algorithm)을 이용한 특징 추출 방법을 설계하였다. TF-IDF는 특징 추출에서 문서와 용어간의 관계를 반영하는데 사용된다. 반복과정을 통해 사전에 미리 결정된 만큼의 특징을 선택한다. 또한 탐지모델의 성능 향상을 위해 sparsity score(희소성 점수)를 사용하였다. 스팸메일 세트의 희소성이 높으면 탐지모델의 성능이 낮아져 최적화된 탐지 모델을 찾기가 어렵다. 우리는 fitness function에서 s(F)를 사용하여 희소성이 낮고 TF-IDF 점수가 높은 탐지모델을 찾았다. 또한 제안된 알고리즘을 텍스트 분류 실험에 적용하여 성능을 검증하였다. 결과적으로, 제안한 알고리즘은 공격 메일 분류에서 좋은 성능(속도와 정확도)을 보여주었다.

• Journal of Information Processing Systems
• /
• 제19권3호
• /
• pp.289-301
• /
• 2023

Due to the development and dissemination of modern technology, anyone can easily communicate using services such as social network service (SNS) through a personal computer (PC) or smartphone. The development of these technologies has caused many beneficial effects. At the same time, bad effects also occurred, one of which was the spam problem. Spam refers to unwanted or rejected information received by unspecified users. The continuous exposure of such information to service users creates inconvenience in the user's use of the service, and if filtering is not performed correctly, the quality of service deteriorates. Recently, spammers are creating more malicious spam by distorting the image of spam text so that optical character recognition (OCR)-based spam filters cannot easily detect it. Fortunately, the level of transformation of image spam circulated on social media is not serious yet. However, in the mail system, spammers (the person who sends spam) showed various modifications to the spam image for neutralizing OCR, and therefore, the same situation can happen with spam images on social media. Spammers have been shown to interfere with OCR reading through geometric transformations such as image distortion, noise addition, and blurring. Various techniques have been studied to filter image spam, but at the same time, methods of interfering with image spam identification using obfuscated images are also continuously developing. In this paper, we propose a deep learning-based spam image detection model to improve the existing OCR-based spam image detection performance and compensate for vulnerabilities. The proposed model extracts text features and image features from the image using four sub-models. First, the OCR-based text model extracts the text-related features, whether the image contains spam words, and the word embedding vector from the input image. Then, the convolution neural network-based image model extracts image obfuscation and image feature vectors from the input image. The extracted feature is determined whether it is a spam image by the final spam image classifier. As a result of evaluating the F1-score of the proposed model, the performance was about 14 points higher than the OCR-based spam image detection performance.