SGER: Discovery of Research Trends and Classification in Domain-Specific Text: Application to Nanoscale Science and Engineering Field

摘要CTS-0417401A型。随着万维网的爆炸性增长和来自多个研究领域的海量数据，对帮助提取隐藏信息的知识发现技术的需求增加了。对这些数据的分析可能导致发现隐藏在数据中的新趋势和模式，而这些趋势和模式是传统的基于查询的系统不可能发现的。这些将提高预测、资源分配的效率和决策过程的质量。在我们对这个探索性研究项目的初步分析中，我们查询了美国国家科学基金会纳米科学与工程(NSE)摘要的250,000个在线奖项。检索到的数据被分析、清理并存储在基于网络的数据库中。该数据库可在http://128.164.158.138/bellaTest.上访问我们已经开始探索数据挖掘技术，以提取新的研究领域和新的研究课题。到目前为止，我们已经实现了流行的k-均值技术和奇异值分解。目前，我们正在分析初步实验的结果。在这个请求中，我们计划完成我们的初步挖掘技术的分析，我们还想探索另外两种聚类技术：自组织振子网络(Soon)聚类算法和E-CAST聚类技术。这两种技术都不需要任何关于集群数量的知识。它们也被成功地应用于其他领域。在这项探索性研究中，我们希望申请额外的资金来进行以下任务：-完成对我们最初的数据挖掘技术的分析。-使用预定义的现有七个概念集作为k-均值技术的种子进行其他实验。-由于k-Means需要指定聚类的数量，我们还计划探索另外两种不需要聚类数量的数据挖掘技术：自组织振荡器网络(Soon)和E-Cast聚类技术。-分析所有上述技术，并确定最能描述NSE领域新兴研究领域的关键术语。除了挖掘NSE奖项的新趋势和研究领域外，我们还计划为NSF网站添加新功能。其中包括以带有Hover特征的美国地图的形式显示奖项的详细信息，以及搜索最大的奖项、顶级资助的调查人员等。这一请求的智力优势和新贡献包括全面了解如何处理其他领域的小型、特殊用途的数据收集。学生、研究人员和教育工作者也可以使用我们的系统来发现新的研究领域。提出的系统具有更广泛的影响，可以应用于其他应用，如检索系统和数字图书馆系统。

AbstractCTS-0417401A. Bellaachia, George Washington UniversityWith the explosion of the World Wide Web and the massive amount of data available from several research areas, the need for knowledge discovery techniques that help extract hidden information has increased. Analysis of these data may lead to the discovery of new trends and patterns hidden within the data that would be impossible to discover by just traditional query-based systems. These would improve forecasting, efficiency of resource allocations, and quality of the decision making process. In our initial analysis of this exploratory research project, we have queried the 250,000 NSF online awards for nanoscale science and engineering (NSE) abstracts. The retrieved data were analyzed, cleaned, and stored in a web-based database. This database can be accessed at http://128.164.158.138/bellaTest. We have started the exploration of data mining techniques to extract new research areas and new research topics. So far we have implemented the popular k-mean technique along with the singular value decomposition. Currently, we are in the process of analyzing the results of our initial experiments. In this request, we are planning to finish the analysis of our preliminary mining techniques and we would also like to explore two other clustering techniques: the Self-Organizing Oscillators Network (SOON) clustering algorithm and the E-CAST clustering techniques. Both of these techniques do not require any knowledge about the number of clusters. They have also been successfully used in other domains. In this exploratory research, we would like to request additional fund to conduct the following tasks: - Finish the analysis of our initial data mining techniques. - Conduct other experiments using the predefined set of existing seven concepts as seeds to the k-mean technique. - Since k-mean requires the specification of the number of clusters, we are also planning to explore two other data mining techniques that do not require the number of clusters: the Self-Organizing Oscillators Network (SOON) and the E-CAST clustering techniques. - Analyze all the above techniques and determine key terms that best describe new emerging research areas in the field of NSE field. In addition to mining new trends and research areas in the NSE awards, we are also planning to add new features to the NSF website. These include the display of details of awards in the form a US map with Hover features, and searching largest awards, top funded investigators, etc. The intellectual merit and new contributions of this request include a comprehensive understanding of how to process small, special-purpose collection of data in other domains. Students, researchers, and educators can also use our system to discover new research areas. The proposed system has broader impacts that can be applied to other applications such retrieval systems and digital library systems.