基于词嵌入的国家自然科学基金学科交叉知识发现方法--以“人工智能”与“信息管理”为例

Method of Discovering Interdisciplinary Knowledge of the National Natural Science Foundation of China Based on Word Embedding:A Case Study on Artificial Intelligence and Information Management

学科交叉的研究是促进各种复杂科学问题解决的重要途径。本文利用国家自然科学基金所资助项目中人工智能学科与信息管理学科关键词之间的共现关系,通过word2vec相关模型,将相应关键词映射到低维向量空间。利用关键词向量计算关键词之间的关系大小,获取量化的关键词共现关系。通过PageRank算法,计算关键词在共现网络中的重要性。利用DBSCAN聚类算法等对项目中未出现过的具有跨学科性质的关键词共现进行分析,并结合关键词重要度等文字信息和可视化信息分析潜在学科交叉知识。研究结果发现,本文提出的模型能较好地将潜在的学科交叉知识提取出来,并可以利用跨学科关键词共现关系大小,进行学科交叉知识的过滤及排序,其结果具有一定的可解释性和合理性,为探索学科交叉知识发现方法以及发现潜在交叉学科成长提供了新的研究思路。

Interdisciplinary research is an important way to promote the resolution of various complex scientific problems.In this paper,the keyword co-occurrence relationship in the disciplines of artificial intelligence and information management in the projects funded by the National Natural Science Foundation of China is used to map the corresponding keywords to the low-dimensional vector space through word2vec correlation model.The keyword vector is used to calculate the relationship between keywords and obtain the quantized keyword co-occurrence relationship.PageRank algorithm is utilized to calculate the importance of keywords in the co-occurrence network.DBSCAN clustering algorithm is used to analyze keyword co-occurrence with an interdisciplinary nature that did not appear in the project,and textual information such as keyword importance is combined with visual information to analyze potential interdisciplinary knowledge.The experiment shows that the model proposed in this study can extract the potential interdisciplinary knowledge well and can filter and sort the interdisciplinary knowledge by using the interdisciplinary keyword co-occurrence relation.The results are interpretable and reasonable and provide new research ideas for exploring methods in discovering interdisciplinary knowledge and identifying its potential growth.