一個新的研究主題演變引用網路

Abstract

過去，人們自閱讀相關的期刊論文、專利或報導等資料中歸納得出研究領域的主題變化。然而隨著資訊爆炸的時代來臨，大量成長的文獻資料使得人工逐篇閱讀勞心勞力，主題演變的觀測變得更加地困難。因此，開發一套可協助使用者觀察研究主題演變的方法相當重要。 本研究提出一個新的研究主題引用網路，取代以往將單篇文獻作為節點並以文獻之間的引用關係作為連結依據的研究網路。我們抽取出文獻中作者所定義之關鍵字作為新引用網路的節點，並重新定義節點之間的引用關係以及權重。權重低於指定閥值的連結將會被刪除，而後針對網路內的節點進行聚類分析。每群節點集合是一個研究主題，藉由集合內的關鍵字內容可以得知研究主題的技術細節。最後，我們藉由檢測不同年度間的研究主題之引用關係來觀察其是否有演化關係。實驗結果顯示，新提出的引用網路可以從大量的文獻資料中抽取出可利用的研究主題，並協助領域內的專家觀察研究主題的演變。

In the past, people consumed related research thesis, journal papers or patent specifications to conclude the research trend of a topic area. However, with the coming of information explosion, consuming such a large number of data only by human seems to be inefficient. Therefore, it is important to develop a method to help users observe the evolution of research topics. In this study, we propose a novel citation network for detecting research trend to replace the past citation network that use a single document as a node and the citation relationships between documents as links. We extract the keywords which were specified by authors from the documents and redefine the citation relationships between nodes as well as the weight of edges. After deleting edges whose weight is lower than designated threshold, nodes in the final citation network would be clustered to find the research topics. Each cluster of nodes is considered as a research topic. The details of a research topic can be realized from the keywords in the cluster. Finally, we are able to observe the evolution of a topic area by checking the relationships among research topics in different years. The experiment shows that the proposed novel citation network can extract useful research topics from numerous literature and assist domain experts to observe the change of research topics.