Duplicate Article/Plagiarism Discovery

描述（由申请人提供）：由于发现的机会很少，预算减少，但持续的出版压力，重复出版物和剽窃的不道德行为是显著的。由于编辑和审稿人没有可靠的方法来识别现有的和潜在的重复科学文章，这意味着这种情况可以不受限制，直到现在。eTBLAST是一个网络上所有人都可以使用的文本相似度搜索工具，它被用来证明我们可以通过系统地将每个Medline摘要（或综述中的摘要）与所有其他Medline记录进行比较，以高灵敏度和特异性检测假定的重复/剽窃文章。我们假设，严格识别这种行为的提供者，对重复文章进行详尽的标记，以及为编辑、审稿人、授权官员等定制的搜索工具的可用性，以便在接受发表之前发现潜在的问题手稿，这将是一个实质性的威慑，最终提高所有人报告的科学质量。我们将通过以下具体目标来解决这个问题：1)改进统计预测、阈值、签名和算法，以最大限度地提高我们在Medline中检测假定的重复和剽窃文章的效率。2)系统地检查Medline的每条记录，并与其他记录进行比较，建立一个公共数据库，其中包含已被人工审查/验证的可疑文章，以分配重复的概率。3)分析趋势，比率和任何统计上相关的分布，以了解和解决这种行为的根本原因。4)创建一个安全的资源，对所有期刊/审稿人开放，从而使他们能够在接受之前评估新颖性和与先前出版物的可能重叠。

DESCRIPTION (provided by applicant): With little chance for discovery and decreasing budgets, yet sustained pressure to publish, the unethical practices of duplicate publications and plagiarism are significant. With no robust method to identify existing and potential duplicate scientific articles by editors and reviewers means that this can go unchecked, until now. eTBLAST, a text similarity search tool available to all on the web, has been used to demonstrate that we can detect with high sensitivity and specificity putative duplicate/plagiarized articles by systematically comparing each Medline abstract (or abstract in review) to all other Medline records. We hypothesize that rigorous identification of purveyors of this behavior, the exhaustive tagging of duplicate articles and the availability of a search tool customized for use by editors, reviewers, granting officials, etc. to detect potential problem manuscripts before they are accepted for publication will be a substantial deterrent, ultimately improving the quality of reported science for all. We will address this through the following specific aims: 1) Refine statistical predictors, thresholds, signatures and algorithms to maximize the efficiency by which we can detect putative duplicate and plagiarized articles within Medline. 2) Systematically check every Medline record against every other to develop a public database of questionable articles that have been reviewed/verified manually to assign a probability of duplication. 3) Perform an analysis of trends, rates and any statistically relevant distributions to understand and address root causes for this behavior. 4) Create a secure resource that is available and open to all journals/reviewers, thus enabling them to estimate novelty and probable overlap with previous publications prior to acceptance.