论文部分内容阅读
【目的】降低文献–作者二分网络在投影为合著网络过程中的信息丢失影响,形成适应特定二分网络的合著关系预测指标和方法,提高预测准确率和结果可解释性。【方法】首先构建文献–作者二分网络及其投影合著网络;接着抽取二分网络中的二阶路径和三阶路径表示作者间的关联关系;最后利用逻辑回归方法学习不同路径对于合著关系预测的贡献,由此形成文献–作者二分网络中基于路径组合的合著关系预测指标。【结果】在图书情报领域的实验证实,文献–作者二分网络在投影为合著网络过程中存在较大的信息丢失,并以合著关系预测准确率变化进行定量计算;逻辑回归方法适合学习不同路径对于合著关系预测的贡献,由此形成的路径组合指标准确率远远高出其他指标,并且预测结果更易解释。【局限】其他的多阶路径尚未引入到该模型中,方法通用性还需在其他领域进行验证。【结论】合著关系预测应直接在文献–作者二分网络上进行,以降低投影过程中的信息丢失影响;文献–作者二分网络上的路径组合指标是合著关系预测的最优指标;该方法可扩展应用到其他类型的二分网络中,如专利–发明人二分网络。
【Objective】 The objective of this study is to reduce the influence of information loss caused by the literature-author dichotomy network in the process of co-authoring the projection network, and to form the forecasting index and method of co-author relations to adapt to a specific dichotomous network so as to improve the prediction accuracy and interpretability of results. 【Method】 Firstly, document-author dichotomy network and its projection co-author network were constructed; then the second-order and third-order paths in dichotomous networks were extracted to represent the association between authors; finally, logistic regression was used to study the prediction of co-ownership relationship Thus forming the co-relation prediction index based on path combination in the literature-author dichotomy network. [Results] Experiments in the field of library and information have confirmed that there is a large information loss in the literature-author dichotomy network in the process of co-authoring the projection and the quantitative calculation is made based on the change of the prediction accuracy of the co-author relations. The logical regression method is suitable for learning different The contribution of the path to the prediction of the co-ownership relationship results in a much higher accuracy of the path combination indicator than other indicators, and the prediction result is easier to explain. [Limitations] Other multi-stage paths have not yet been introduced into the model, and the versatility of the method needs to be verified in other areas. 【Conclusion】 The co-author relationship prediction should be conducted directly on the literature-author dichotomy network to reduce the impact of information loss during the projection process. The path-combination index on the literature-author dichotomy network is the optimal index for co-author relationship prediction. Scalable to other types of dichotomous networks, such as the patented-inventor dichotomy.