科研机构名称归一化实现

Realization of Research Institution Name Normalization

[目的／意义]机构名称的数目多且较为繁杂，机构名称归一化可将同一机构的规范名称以及不同时段、不同表达形式的非规范名称汇集在一起，提高查询检索的查全率和查准率；有利于建立与其他系统之间的互操作，实现资源的共享。[方法／过程]在分析机构名称字符串的特点和基于K-means算法的基础上，利用编辑距离算法实现一级机构名称的初步聚类，然后利用初步聚类结果并基于TF-IDF算法计算机构名称各词项的权值，从而基于K-means算法将机构名称围绕聚类中心抱团聚簇，并对每一个簇的机构名称赋予唯一标识符。[结果／结论]该方法可实现同一机构实体不同形式的规范名称的归一，提高机构名称聚类的准确率，但对K取值、距离测度方法的选取仍有待优化。

[ Purpose/significance] Institution names are numerous and complicated. The normalization of institution names brings the authoritative name and the informal ones （ both at different times and in different ways of expression ） of the same institution together, enhancing comprehensiveness and accuracy of searches, promoting interoperability with other systems, and thus realizing resource sharing. [ Method/process] Based on the analysis of institution names＇ characteristic and K-means algorithm, this paper utilizes the edit distance similarity algorithm to achieve name normalization of institution names. Then uses TF-IDF to calculate the weight of each item, around the cluster center to normalize institution name based of K-means algorithm and gives the unique identifier to every cluster. [ Result/conclusion] It achieves name normalization of the same institution name in different forms. And it improves the precision of institution name cluster, but the choice of K value and distance measurement method still needs to be optimized.