• Journal of Korean Library and Information Science Society
• /
• v.45 no.3
• /
• pp.5-36
• /
• 2014

The Web is rich with various sources of information that go beyond the contents of documents, such as hyperlinks and manually classified directories of Web documents such as Yahoo. This research extends past fusion IR studies, which have repeatedly shown that combining multiple sources of evidence (i.e. fusion) can improve retrieval performance, by investigating the effects of combining three distinct retrieval approaches for Web IR: the text-based approach that leverages document texts, the link-based approach that leverages hyperlinks, and the classification-based approach that leverages Yahoo categories. Retrieval results of text-, link-, and classification-based methods were combined using variations of the linear combination formula to produce fusion results, which were compared to individual retrieval results using traditional retrieval evaluation metrics. Fusion results were also examined to ascertain the significance of overlap (i.e. the number of systems that retrieve a document) in fusion. The analysis of results suggests that the solution spaces of text-, link-, and classification-based retrieval methods are diverse enough for fusion to be beneficial while revealing important characteristics of the fusion environment, such as effects of system parameters and relationship between overlap, document ranking and relevance.

• The Journal of Society for e-Business Studies
• /
• v.25 no.1
• /
• pp.203-214
• /
• 2020

It is necessary to classify technical documents such as patents, R&D project reports in order to understand the trends of technology convergence and interdisciplinary joint research, technology development and so on. Text mining techniques have been mainly used to classify these technical documents. However, in the case of classifying technical documents by text mining algorithms, there is a disadvantage that the features representing technical documents must be directly extracted. In this study, we propose a BERT-based document classification model to automatically extract document features from text information of national R&D projects and to classify them. Then, we verify the applicability and performance of the proposed model for classifying documents.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.22 no.8
• /
• pp.1049-1054
• /
• 2018

Keyword searching used in the past as a method of finding similar patents, and automated classification by machine learning is using in recently. Keyword searching is a method of analyzing data that is formalized through data refinement. While the accuracy for short text is high, long one consisted of several words like as document that is not able to analyze the meaning contained in sentences. In semantic analysis level, the method of automatic classification is used to classify sentences composed of several words by unstructured data analysis. There was an attempt to find similar documents by combining the two methods. However, it have a problem in the algorithm w the methods of analysis are different ways to use simultaneous unstructured data and regular data. In this paper, we study the method of extracting keywords implied in the document and using the LDA(Latent Semantic Analysis) method to classify documents efficiently without human intervention and finding similar patents.

• Journal of the Korean Society for information Management
• /
• v.24 no.1 s.63
• /
• pp.251-271
• /
• 2007

This paper studies the problem of classifying documents with labeled and unlabeled learning data, especially with regards to using document similarity features. The problem of using unlabeled data is practically important because in many information systems obtaining training labels is expensive, while large quantities of unlabeled documents are readily available. There are two steps In general semi-supervised learning algorithm. First, it trains a classifier using the available labeled documents, and classifies the unlabeled documents. Then, it trains a new classifier using all the training documents which were labeled either manually or automatically. We suggested two types of semi-supervised learning algorithm with regards to using document similarity features. The one is one step semi-supervised learning which is using unlabeled documents only to generate document similarity features. And the other is two step semi-supervised learning which is using unlabeled documents as learning examples as well as similarity features. Experimental results, obtained using support vector machines and naive Bayes classifier, show that we can get improved performance with small labeled and large unlabeled documents then the performance of supervised learning which uses labeled-only data. When considering the efficiency of a classifier system, the one step semi-supervised learning algorithm which is suggested in this study could be a good solution for improving classification performance with unlabeled documents.

• Management & Information Systems Review
• /
• v.37 no.4
• /
• pp.41-62
• /
• 2018

As the volume of unstructured data increases through various social media, Internet news articles, and blogs, the importance of text analysis and the studies are increasing. Since text analysis is mostly performed on a specific domain or topic, the importance of constructing and applying a domain-specific dictionary has been increased. The quality of dictionary has a direct impact on the results of the unstructured data analysis and it is much more important since it present a perspective of analysis. In the literature, most studies on text analysis has emphasized the importance of dictionaries to acquire clean and high quality results. However, unfortunately, a rigorous verification of the effects of dictionaries has not been studied, even if it is already known as the most essential factor of text analysis. In this paper, we generate three dictionaries in various ways from 39,800 news articles and analyze and verify the effect each dictionary on the accuracy of document classification by defining the concept of Intrinsic Rate. 1) A batch construction method which is building a dictionary based on the frequency of terms in the entire documents 2) A method of extracting the terms by category and integrating the terms 3) A method of extracting the features according to each category and integrating them. We compared accuracy of three artificial neural network-based document classifiers to evaluate the quality of dictionaries. As a result of the experiment, the accuracy tend to increase when the "Intrinsic Rate" is high and we found the possibility to improve accuracy of document classification by increasing the intrinsic rate of the dictionary.

• Journal of the Institute of Convergence Signal Processing
• /
• v.11 no.3
• /
• pp.209-214
• /
• 2010

This paper proposes an algorithm of segmentation and title detection for document image. The automated title detection method that we have developed is composed of two phases, segmentation and title area detection. In the first phase, we extract and segment the document image. To perform this operation, the binary map is segmented by combination of morphological operation and CCA(connected component algorithm). The first phase provides segmented regions that would be detected as title area for the second stage. Candidate title areas are detected using geometric information, then we can extract the title region that is performed by removing non-title regions. After classification step that removes non-text regions, projection is performed to detect a title region. From the fact that usually the largest font is used for the title in the document, horizontal projection is performed within text areas. In this paper, we proposed a method of segmentation and title detection for various forms of document images using geometric features and projection profile analysis. The proposed system is expected to have various applications, such as document title recognition, multimedia data searching, real-time image processing and so on.

• Korean Journal of Oriental Medicine
• /
• v.11 no.2
• /
• pp.67-96
• /
• 2005

This Study focused on IPC (International Patent Classification) for TKM (Traditional Korea Medicine) Paper. The results processed for 9,000 TKM paper by using 8th in IPC Classification. The name of Herbal Medicine assigned to IPC Classification, we assigned to two part for main-Classification(A61K) and sub-Classification (A61P). The results obtained about 77% for A61K and about 96% for A61K36 among them. And also analysed about 23% for sub-Classification(A61P) additionally. Main-Classification is distributed A61K > A61H37 > A61B5 > A61N > A61M1. Detailed Main-Classification for A61K is distributed A61K36 > A61K35 > A61K33 among Main-Classification. TKM Paper mainly analysed A61K36 and A61H37 in Main-Classification. According to the results. 'The Korean Journal of Herbology' has high-valued for Utilization as a Non Patent Document. we should constructed Database system for protection of intellectual property rights. And after We will registered minimum documentation of PCT.

• The KIPS Transactions:PartB
• /
• v.12B no.6 s.102
• /
• pp.703-714
• /
• 2005

Researches in text categorization have been confined to whole-document-level classification, probably due to lacks of full-text test collections. However, full-length documents availably today in large quantities pose renewed interests in text classification. A document is usually written in an organized structure to present its main topic(s). This structure can be expressed as a sequence of sub-topic text blocks, or passages. In order to reflect the sub-topic structure of a document, we propose a new passage-level or passage-based text categorization model, which segments a test document into several Passages, assigns categories to each passage, and merges passage categories to document categories. Compared with traditional document-level categorization, two additional steps, passage splitting and category merging, are required in this model. By using four subsets of Routers text categorization test collection and a full-text test collection of which documents are varying from tens of kilobytes to hundreds, we evaluated the proposed model, especially the effectiveness of various passage types and the importance of passage location in category merging. Our results show simple windows are best for all test collections tested in these experiments. We also found that passages have different degrees of contribution to main topic(s), depending on their location in the test document.

• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 2001.05a
• /
• pp.187-190
• /
• 2001

Due to the recent proliferation of the internet, it is broadly granted that the necessity of the automatic document categorization has been on the rise. Since it is a heavy time-consuming work and takes too much manpower to process and classify manually, we need a system that categorizes them automatically as their contents. In this paper, we propose the automatic E-mail response system that is based on 2 hierarchical document clustering methods. One is to get the final result from the classifier trained seperatly within each class, after clustering the whole documents into 3 groups so that the first classifier categorize the input documents as the corresponding group. The other method is that the system classifies the most distinct classes first as their similarity, successively. Neural networks have been adopted as classifiers, we have used dendrograms to show the hierarchical aspect of similarities between classes. The comparison among the performances of hierarchical and non-hierarchical classifiers tells us clustering methods have provided the classification efficiency.

• ETRI Journal
• /
• v.37 no.4
• /
• pp.813-823
• /
• 2015

Because protein is a primary element responsible for biological or biochemical roles in living bodies, protein function is the core and basis information for biomedical studies. However, recent advances in bio technologies have created an explosive increase in the amount of published literature; therefore, biomedical researchers have a hard time finding needed protein function information. In this paper, a classification system for biomedical literature providing protein function evidence is proposed. Note that, despite our best efforts, we have been unable to find previous studies on the proposed issue. To classify papers based on protein function evidence, we should consider whether the main claim of a paper is to assert a protein function. We, therefore, propose two novel features - protein and assertion. Our experimental results show a classification performance with 71.89% precision, 90.0% recall, and a 79.94% F-measure. In addition, to verify the usefulness of the proposed classification system, two case study applications are investigated - information retrieval for protein function and automatic summarization for protein function text. It is shown that the proposed classification system can be successfully applied to these applications.