Design And Performance, Song Mao, Jong Woo Kim et al. · 2004
The Medical Article Records System or MARS has been developed at the U.S. National Library of Medicine (NLM) for automated data entry of bibliographical information from medical journals into MEDLINE, the premier bibliographic citation database at NLM. Currently, a rule-based algorithm (called ZoneCzar) is used for labeling important bibliographical fields (title, author, affiliation, and abstract) on medical journal article page images. While rules have been created for medical journals with regular layout types, new rules have to be manually created for any input journals with arbitrary or new layout types. Therefore, it is of interest to label any journal articles independent of their layout styles. In this paper, we first describe a system (called ZoneMatch) for automated generation of crucial geometric and non-geometric features of important bibliographical fields based on string-matching and clustering techniques. The rule-based algorithm is then modified to use these features to perform style-independent labeling. We then describe a performance evaluation method for quantitatively evaluating our algorithm and characterizing its error distributions. Experimental results show that the labeling performance of the rule-based algorithm is significantly improved when the generated features are used.
Cited by 0
Jongwoo Kim, Daniel X. Le, George R. Thoma · 2001
The National Library of Medicine (NLM) is developing an automated system to produce bibliographic records for its MEDLINE ® database. This system, named Medical Article Record System (MARS), employs document image analysis and understanding techniques and optical character recognition (OCR). This paper describes a key module in MARS called the Automated Labeling (AL) module, which labels all zones of interest (title, author, affiliation, and abstract) automatically. The AL algorithm is based on 120 rules that are derived from an analysis of journal page layouts and features extracted from OCR output. Experiments carried out on more than 11,000 articles in over 1,000 biomedical journals show the accuracy of this rule-based algorithm to exceed 96%.
Cited by 0
Jian Liang, David Doermann · 2002
Logical structure analysis of document images is an important problem in document image understanding. In this paper, we propose a graph matching approach to label logical components on a document page. Our system is able to learn a model for a document class, use this model to label document images through graph matching, and adaptively improve the model with error feed back. We tested our method on journal/proceeding article title pages. The
Cited by 0
Michail Gongolidis · LA Referencia (Red Federada de Repositorios Institucionales de Publicaciones Científicas) · 2020
Summarization is the notion of abstracting key content from information sources. The task of summarization has become vital in our everyday lives especially nowadays with the exponentially growing amounts of available information that is generated over the internet. In Natural Language processing, automatic text summarization is the problem of creating short, accurate and fluent text pieces from large textual data. In this thesis, I test state-of-the-art Deep Learning models for summarizing single documents and document clusters. Later, I compare the resulting cluster summaries with principal cluster labeling techniques to determine how accurate the Deep Learning models can be in labeling document clusters. For single-document summarization I chose a model consisted of multiple neural layers based on the encoder-decoder architecture with a novel Recurrent neural cell ( Rotational Unit of Memory-RUM ) in its core. I trained and evaluated the model on a new U.S. patent dataset. For document cluster summarization, I tested a framework that combines a fundamental extractive statistical natural language processing technique ( Maximal Marginal Relevance ) with the abstractive encoder-decoder model to summarize the document clusters. Finally, I evaluated the resulting summaries by comparing them with the top words of the clusters based on the Word Frequency and their TF*IDF scores. The RUM-based encoder-decoder model surpassed all the state-of-the-art models that were tested on the selected dataset and the MMR framework successfully summarized document clusters of up to 100 U.S. patent abstracts.
Cited by 0
Alexandrin Popescul, Lyle H. Ungar · 2000
Automatically labeling document clusters with words which indicate their topics is difficult to do well. The most commonly used method, labeling with the most frequent words in the clusters, ends up using many words that are virtually void of descriptive power even after traditional stop words are removed. Another method, labeling with the most predictive words, often includes rather obscure words. We present two methods of labeling document clusters motivated by the model that words are generated by a hierarchy of mixture components of varying generality. The first method assumes existence of a document hierarchy (manually constructed or resulting from a hierarchical clustering algorithm) and uses a 2 test of significance to detect different word usage across categories in the hierarchy. The second method selects words which both occur frequently in a cluster and effectively discriminate the given cluster from the other clusters. We compare these methods on abstracts of documents sel...
Cited by 0
ISO, Trusted CI · IUScholarWorks (Indiana University) · 2021
In response to incident '2019-10-02_01', a scheme for labeling documents was developed.The goal of this scheme (and policy to implement it) is to ensure that future documents produced by Trusted CI are published in the manner identified by the document's maintainer.The scheme and policy should be simple to employ by the maintainer, it should not be too obtrusive on a document's title page and it should be sufficiently tractable by the publications' liaison in both identifying and parsing the distribution label specified.Moreover, it should not conflict with existing distribution statements, thus, providing a query-able string component to allow batch processing of Trusted CI's numerous documents.
Cited by 0Open access
Österberg, Adam · KTH Publication Database DiVA (KTH Royal Institute of Technology) · 2022
Labeling is the process of giving a set of data a descriptive name. This thesis dealt with documents with no additional information and aimed at clustering them using topic modeling and labeling them using Wikipedia as a second source. Labeling documents is a new field with many potential solutions. This thesis examined one method in a practical setting. Unstructured data was preprocessed and clustered using a topic model. Frequent words from each cluster were used to generate a search query sent to Wikipedia, where titles and categories from the most relevant pages were stored as candidate labels. Each candidate label was evaluated based on the frequency of common cluster words among the candidate labels. The frequency was weighted proportional to the relevance of the original Wikipedia article. The relevance was based on the order of appearance in the search results. The five labels with the highest scores were chosen to describe the cluster. The clustered documents consisted of exam questions that students use to practice before a course exam. Each question in the cluster was scored by someone experienced in the relevant topic by evaluating if one of the five labels correctly described the content. The method proved unreliable, with only one course receiving labels considered descriptive for most of its questions. A significant problem was the closely related data with all documents belonging to one overarching category instead of a dataset containing independent topics. However, for one dataset, 80 % of the documents received a descriptive label, indicating that labeling using secondary sources has potential, but needs to be investigated further.
Cited by 0Open access
Chandranath Adak · 2014
A document image contains texts and non-texts, it may be printed, handwritten, or hybrid of both. In this paper we deal with printed document where textual region is of printed characters, and non-texts are mainly photo images. Here we propose a model which performs labeling of different components of a printed document image, i.e. identification of heading, subheading, caption, article and photo. Our method consists of a preprocessing stage where fuzzy c-means clustering is used to segment the document image into printed (object) region and background. Then Hough transformation is used to find white-line dividers of object region and grid structure examination is used to extract the non-text portion. After that, we use horizontal histogram to find text lines and then we label different components. Our method gives promising results on printed document of different scripts.
Cited by 1
Jian Liang, David Doermann · 2003
The use of content features extracted from recognized text is valuable in labeling logical elements in documents without rigid layout structure, like business letters. This paper discusses a model-based approach to combining content features with other geometrical and presentation features for logical labeling. Models are automatically initialized and adaptively improved using training samples. Satisfactory experimental results are presented.
Cited by 0
Christin Seifert, Eva Ulbrich, Michael Granitzer · Lecture notes in computer science · 2011
Cited by 10
Jiadong Ren, Xiaopeng Yin, Xiaodan Guo · 2008
Abstract: Efficient evaluation of XML queries requires the determination of whether a relationship exists between two elements. A number of labeling schemes have been designed to meet the need. However, most of them have poor updating performance. In this paper, a new dynamic region-based labeling scheme is proposed which uses real numbers instead of integers to represent the region. Moreover, the scheme can adjust the codes of some nodes in some parts of the document tree according to the condition of updates. Our analysis shows this new labeling scheme provides efficient support for updates.
Cited by 0
Dang Van Thin, Cuong Xuan Chu, Christian Graf et al. · 2025
Repairing and maintaining car parts are crucial tasks in the automotive industry, requiring a mechanic to have all relevant technical documents available.However, retrieving the right documents from a huge database heavily depends on domain expertise and is time-consuming and error-prone.By labeling available documents according to the components they relate to, concise and accurate information can be retrieved efficiently.However, this is a challenging task as the relevance of a document to a particular component strongly depends on the context and the expertise of the domain specialist.Moreover, component terminology varies widely between different manufacturers.We address these challenges by utilizing Large Language Models (LLMs) to enrich and unify a component database via web mining, extracting relevant keywords, and leveraging hybrid search and LLM-based re-ranking to select the most relevant component for a document.We systematically evaluate our method using various LLMs on an expert-annotated dataset and demonstrate that it outperforms the baselines, which rely solely on LLM prompting.
Cited by 0Open access
2008
The National Library of Medicine (NLM) has developed an automated system, named Medical Article Records System (MARS), to process bibliographic data (title, authors, affiliation, abstract, etc.) in biomedical journal articles for its MEDLINE® database. This paper describes a labeling module in the MARS, which automatically extract the bibliographic data in biomedical journal articles. The labeling module is composed of two sub modules: General label type module (GLTM) and Arbitrary label type module (ALTM). Six label types, which are commonly used in the journals, are collected from several thousand journals. Journals are classified as general label types if label types of the journals belong to one of the six label types. Otherwise, journals are classified as arbitrary label types. The GLTM processes journals that belong to general label types and the ALTM processes journals that belong to arbitrary label types. Rule-based algorithms are used for both modules and the rules are derived from analysis of several journal articles and features extracted from the optical character recognition (OCR) results. There are 126 rules derived for the GLTM and 49 rules for the ALTM. Experiments conducted with several medical journal articles show relatively accurate labeling results.
Cited by 0
Seyednaser Nourashrafeddin, Evangelos Milios, Dirk V. Arnold · 2013
We propose an interactive text document method, which is based on term labeling. The algorithm asks the user to cluster the top keyterms associated with document clusters iteratively. The keyterm clusters are used to guide the clustering method. Rather than using standard clustering algorithms, we propose a new text clusterer using term clusters. Terms that exist in a document corpus are clustered. Using a greedy approach, the term clusters are distilled in order to remove non-discriminative general terms. We then present a heuristic approach to extract seed documents associated with each distilled term cluster. These seeds are finally used to cluster all documents. We compared our interactive term labeling to a baseline interactive term selection algorithm on some real standard text datasets. The experiments show that with a comparable amount of user effort, our term labeling is more effective than the baseline term selection method.
Cited by 15
Association for Computational Linguistics 2025, Chu, Cuong Xuan, Graf, Christian et al. · Underline Science Inc. · 2025
Repairing and maintaining car parts are crucial tasks in the automotive industry, requiring a mechanic to have all relevant technical documents available. However, retrieving the right documents from a huge database heavily depends on domain expertise and is time consuming and error-prone. By labeling available documents according to the components they relate to, concise and accurate information can be retrieved efficiently. However, this is a challenging task as the relevance of a document to a particular component strongly depends on the context and the expertise of the domain specialist. Moreover, component terminology varies widely between different manufacturers. We address these challenges by utilizing Large Language Models (LLMs) to enrich and unify a component database via web mining, extracting relevant keywords, and leveraging hybrid search and LLM-based re-ranking to select the most relevant component for a document. We systematically evaluate our method using various LLMs on an expert-annotated dataset and demonstrate that it outperforms the baselines, which rely solely on LLM prompting.
Cited by 0Open access
Yichao Zhou, James B. Wendt, Navneet Potti et al. · 2023
Building automatic extraction models for visually rich documents like invoices, receipts, bills, tax forms, etc. has received significant attention lately. A key bottleneck in developing extraction models for new document types is the cost of acquiring the several thousand high-quality labeled documents that are needed to train a model with acceptable accuracy. In this paper, we propose selective labeling as a solution to this problem. The key insight is to simplify the labeling task to provide “yes/no” labels for candidate extractions predicted by a model trained on partially labeled documents. We combine this with a custom active learning strategy to find the predictions that the model is most uncertain about. We show through experiments on document types drawn from 3 different domains that selective labeling can reduce the cost of acquiring labeled data by 10× with a negligible loss in accuracy.
Cited by 0Open access
Song Mao, Jong Woo Kim, George R. Thoma · Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE · 2003
The <i>Medical Article Records System</i> or MARS has been developed at the U.S. National Library of Medicine (NLM) for automated data entry of bibliographical information from medical journals into MEDLINE, the premier bibliographic citation database at NLM. Currently, a rule-based algorithm (called ZoneCzar) is used for labeling important bibliographical fields (title, author, affiliation, and abstract) on medical journal article page images. While rules have been created for medical journals with regular layout types, new rules have to be manually created for any input journals with arbitrary or new layout types. Therefore, it is of interest to label any journal articles independent of their layout styles. In this paper, we first describe a system (called ZoneMatch) for automated generation of crucial geometric and non-geometric features of important bibliographical fields based on string-matching and clustering techniques. The rule based algorithm is then modified to use these features to perform style-independent labeling. We then describe a performance evaluation method for quantitatively evaluating our algorithm and characterizing its error distributions. Experimental results show that the labeling performance of the rule-based algorithm is significantly improved when the generated features are used.
Cited by 7
José M. F. Moura, Solange Oliveira Rezende · 2010
Cited by 10
Emilee Rader · 2007
Cited by 0
Shahsavari Baboukani, Shiva · Data Archiving and Networked Services (DANS) · 2022
Cited by 0