Research on Computing Word Similarity in Pre-Qin Classics Language Network Oriented to Digital Humanities

Abstract

At present, there is relatively little research on ancient Chinese texts in the field of digital humanities, and ancient Chinese information processing urgently needs new algorithms. To realize the word similarity calculation of pre-Qin classics, a total of 25 pre-Qin classics were first mapped into a language network. Based on local relative entropy, we proposed an improved weighted network node similarity calculation method (LREW). This method judges the similarity based on the local network characteristics of the nodes, and the degree of the nodes and the weight information of the edges between the nodes are considered. We used the relative entropy to calculate the difference in the amount of information between different nodes. After experimental comparison, compared with the existing LRE and RE algorithms based on relative entropy, the proposed LREW method can achieve the best results in calculating the similarity between words in the pre-Qin classics. Compared with CN, Jaccard, Salton, and CDSim algorithms based on common neighbor nodes, although the accuracy of LREW is low, the comprehensiveness of the similar word recognition is high, which can ensure that potential similar nodes in the network will not be missed.