医学文献主题新颖性探测方法对比分析

目的：探讨应用新颖性探测模型评估医学文献主题新颖性的可行性，对比分析2种新颖性探测方法（词重叠法和基于共词的逆文档频率量化法）的优劣。方法：选取生物医学领域8个研究主题，从PubMed数据库收集文献，构建2种新颖性探测模型，结合文献主题新颖性的专家分析结果，利用ROC曲线及AUC值对2种新颖性探测模型的可行性进行评估。结果：词重叠法的新颖度计算结果波动幅度较大，能够更好地将文献内容间差异表现在数据上。基于ROC曲线及AUC值分析，词重叠法对于判断新颖文献具有一定准确性，基于共词特性的逆文档频率量化法对于判断新颖文献准确性较低。结论：两种新颖性探测方法得出的新颖度计算结果呈中度相关，二者的均值差异有统计学意义，前者的表现优于后者。

Comparative analysis of subject novelty detection methods in medical literature

Objective To study the feasibility of novelty detection model in assessing the subject novelty of medical literature and comparatively analyze the advantages and disadvantages of words-overlap algorithm and co-words-based inverse file frequency quantitative algorithm. Methods Two novelty detection models were established for the 8 research subjects in PubMed-covered literature. The feasibility of two novelty detection models in assessing the subject novelty of medical literature was assessed according to the subject novelty of literature analyzed by experts, ROC curves and AUC values. Results Words-overlap algorithm showed that the fluctuating amplitude of subject novelty was rather high, which can thus reflect the difference between the contents in literature on the data. ROC curves and AUC values-based analysis revealed a high accuracy of words-overlap algorithm for judging the novelty of literature while co-words-based inverse file frequency quantitative algorithm displayed a low accuracy for judging the novelty of literature. Conclusion The novelty of literature detected with the two novelty detection methods is moderately related. The mean novelty value detected with the two novelty detection methods is of statistical significance. However, the novelty of literature detected with words-overlap algorithm is higher than that detected with co-words-based inverse file frequency quantitative algorithm.