數位人文與傳統文獻的相互作用：《呂氏春秋‧十二紀》聚類分析與篇章結構新探

The Interplay of Digital Humanities and Traditional Documentology: A Clustering Analysis and Structural Exploration of the Twelve Chronicles in the Lüshi Chunqiu

聚類分析（clustering analysis）能夠通過非監督式機器學習算法（unsupervised machine learning algorithms）揭示文本的隱藏結構，已被廣泛地應用在不同類型的文本探勘（data mining）中。鑑於現時先秦兩漢文獻的研究主要停留在資料庫的單向檢索層次，對人工智能的各種工具還未有足夠的重視，本文嘗試運用聚類分析法重新考察《呂氏春秋．十二紀》的主題和結構這個聚訟紛紜的老問題，藉此展示數位人文與上古文獻之間的相互作用（interplay）和對話的可能性。文章首先回顧近數百年來有關〈十二紀〉內容主題和結構編排的不同意見，繼而介紹聚類分析的原理，以及本文選用的數位工具和參數設定，並把四紀的聚類結果與過去的研究加以比對，說明此一方法的效度。接著文章會聚焦在異質性文本的元素，分析其獨特的連接功能，揭示〈十二紀〉的結構模式。本文的成果顯示文本探勘確能為古代文獻研究開拓嶄新視野，不但有助發現新問題，亦能夠解決舊有問題，尚有不少空間值得繼續開發。

Clustering analysis, a method of revealing the hidden structure of texts through unsupervised machine learning algorithms, has been widely used in different types of data mining. In view of the fact that the study of pre-Qin and Han texts has largely remained at the level of one-way searches of databases and paid little attention to the various tools of artificial intelligence, this article attempts to use clustering analysis to reexamine the themes and structure of the ''Twelve Chronicles'' in the Lüshi Chunqiu, which are long-standing and controversial issues; it further seeks to demonstrate the possibility of interplay and dialogue between the digital humanities and ancient texts. We first revisit different views concerning the main themes and structural organizations of the ''Twelve Chronicles'' over the past few hundred years, and then introduce the concept of cluster analysis as well as the digital tools and parameters used in this article. To illustrate the validity of this approach, a comparison between the clustering results of four of the ''Chronicles'' and past studies is carried out. We next focus on the heterogeneous elements of the texts and analyze their unique connecting functions so as to demonstrate the structural patterns of the ''Twelve Chronicles.'' The results of this study show that text mining can indeed open up new horizons for the study of ancient Chinese texts. Text mining not only helps to identify new problems, but also makes it possible to solve old ones. Given this, we conclude that there is still much room for further development.