植基於詞彙數量關係探討軍事新聞主題－以青年日報為例

Exploring the Topics of Military News Based on the Quantitative Relationship of Terms Using the Online Youth Daily News Dataset

陳良駒(Liang-Chu Chen) ； 陳日鑫(Jih-Hsin Chen)

《資訊管理展望》 12卷1期 (2010/06) Pp. 21-42

https://doi.org/10.29715/JIMCSA.201006.0002

摘要

文件聚類係指透過特徵擷取與關鍵詞彙的連結，計算文件主題內容相似度，以進行文件自動歸類的技術。一般來說，中文不像其他拼音語言一樣詞性分明，以致於在特徵的擷取與識別上較為困難，對於文件聚類效果影響較大。目前已有相關文獻針對中文文件、新聞、網頁等資料進行研討，但尚未有文獻從國軍的角度建構新聞主題之關連。
本論文目的如下：（一）整合數量關聯規則與聚合演算法進行中文文件聚類，並與傳統向量空間模型及關聯規則方法進行比較；（二）探討軍事新聞主題之相關性，並建構軍事新聞檢索的雛型介面。本研究以青年日報為探勘對象，發現使用數量關聯規則方法於中文新聞文件之聚類分析的整體準確率(F-measure)優於關聯規則與向量空間模型，亦即加入「數量」的概念於關聯規則的方法中，對關鍵詞進行規則探勘，能找出較高關連度的詞組，有效提昇文件聚類的精準度。最後依據關鍵詞與文件之間的關連程度描繪出新聞主題關係圖與文件地圖，藉以瞭解新聞主題發展的相關性，並提供相關新聞的推薦。

關鍵字

關聯規則 ； 數量關聯規則 ； 向量空間模型 ； 軍事新聞 ； 青年日報

並列摘要

Document clustering is an automatic classification technique which can retrieve the features of documents, construct the linkages of terms, and calculate the similarity of contents. For Chinese documents, the difficulty in capturing and identifying the document features has an adverse effect on the classification/clustering results.
The purpose of this study is as follows: (1) to propose an approach which can integrate the quantitative association rule (QAR) with the hierarchical agglomerative clustering (HAC) algorithm for clustering Chinese news. The result is compared with the traditional vector space model (VSM) and the association rule (AR) methods; (2) to explore the topic relationship of military news, and to build a retrieval interface for showing the similarity among military news. In this study we adopt the online Youth Daily News as the source of text. The finding of this study indicates that the F-measure value of using QAR to cluster Chinese news is superior to that of the approaches of AR and VSM. That is, using the ”quantitative” concept in text mining is helpful for acquiring rules efficiently, constructing term-based relationship effectively, and clustering documents accurately. This study depicts the topic structure and the document map which are constructed by the relationship between keywords and documents for understanding the development of military news. Besides, this study develops a prototype system for searching military news and recommending related information.

並列關鍵字

Association Rule ； Quantitative Association Rule ； Vector Space Model ； Military News ； Online Youth Daily News