Design and implementation of an evolutional data collecting system for the atomic and molecular databases

原子和分子数据库在基础科学和工业应用的各个领域都很有用。根据目前的数据库开发程序，原子和分子数据主要是通过数据中心的科学家工作人员的辛勤工作从科学论文中收集的。在本研究中，我们探索了信息技术在原子分子数据库演进发展中的一种新的应用，即通过互联网自动收集科学论文。我们以电子激发和电离截面的书目数据库为训练样本，开发了一个可以识别未知文献中原子和分子数据存在的机器学习软件。为了刻画文章的特点，我们首先使用了训练和测试用例中的术语频率。其次，我们使用原子和分子状态的信息以及与原子和分子物理相对应的技术术语，这被发现有助于提高识别的准确性。我们研究了识别科学论文电子形式中原子和离子物种、电荷状态、电子构型、光谱项的表达的方法。我们调查了在《物理》杂志上发表的不同物理领域的论文中每一种表达的出现频率。从A到E版本的日记。核物种和分子的表达也被分开识别，这被发现有助于区分原子物理论文与核物理论文以及凝聚态物理论文。结合互联网搜索引擎，目前的结果将使人们不仅可以从文章中收集原子和分子数据，而且可以在广泛的研究领域收集更广泛的科学信息。

Atomic and molecular database is useful for wide variety of fields of basic science and industrial applications. According to the present procedure of database development, atomic and molecular data are mainly collected from scientific papers through laborious work of staff scientists in the data centers. In this study, we have investigated a new application of information technology for evolutional development of atomic and molecular database based on automatic collection of scientific papers through the internet. We have developed a machine learning software which can recognize the existence of atomic and molecular data in unknown articles, using the bibliographic database of electron excitation and ionization cross sections as training examples. In order to characterize the articles, we firstly use the frequency of terms in the training and test examples. Secondly, we use information on the atomic and molecular states and technical terms corresponding to the atomic and molecular physics, which is found to be useful for improving the accuracy of the recognition. We study methods to recognize expressions for atomic and ionic species, charge states, electron configuration, spectral terms in the electronic form of scientific papers. We investigate frequency of appearance of each expression in the papers from different fields of physic published in Phys. Rev. A to E journals. Expression for nuclear species and molecules are also recognized separately, which are found to be useful to distinguish papers for atomic physics from those for the nuclear physics as well as physics of the condensed matter. Combining with the internet search engines, present result will make one possible to collect not only atomic and molecular data from articles but broader scientific information over a wide range of research fields.