詩作風格知識庫之研究－以蘇軾近體詩為例

Abstract

現在人如果要創作詩詞，常受限於對詩作之韻目、格律與規範的不熟悉而怯步。更何況，要創作出具有古詩人風格韻味之詩作，更是一項難題。在以往，這常需要具有豐富詩作經驗與古詩人風格研究的專家才能達到此目的。因此，如何讓作詩更為簡單易學，並創作出兼具個人風格與古詩人風格之詩作，便成為一項熱門的研究課題。而隨著資訊技術的快速發展與成熟，運用資訊技術可以幫我們將原本棘手的問題變得更容易處理。

因此，在本篇論文中，我們以建立詩作知識庫（Knowledge Base）並結合資料探勘技術（Data Mining Technology）的方法來分析與建立古詩人之詩作風格知識庫，以使創作兼具個人風格與古詩人風格之詩作成為可能。知識庫技術的運用可使所分析之詩作規格與特徵等知識架構能易於新增、搜尋與管理，並可有效的搭配詩作系統來提供作詩時的詩作特徵參考資訊。因此，我們以北宋的大文學家蘇軾為例，提出詩作風格之多重萃取方法來萃取其生平創作風格特性以建立蘇軾風格知識庫。所提出之方法共分為4個步驟：1、詩作分類與詩人資料之建立：此為前置處理，依照專家之分析與意見將蘇軾詩作根據其既有風格先予以分類，以使後續之風格萃取能更加準確。並分析與整合蘇軾之生平資料（Profile），其中包含重要之參考屬性，例如：年代、居住地點、升官、貶職等等特性，以作為詩作特徵分類之參考。2、詩作規則之知識轉換：針對近體詩作之格律規範，再結合專家（Domain Expert）之專業分析來轉換並彙整成格律規則集（Rule Set）。3、詞彙知識庫之建立：使用蘇軾詩作進行斷詞、斷句，並利用詮釋資料技術（Metadata Technology）來描述蘇軾詞句以建立詞彙知識庫。4、潛在風格之萃取與分析：運用資料探勘技術分析蘇軾的平仄用韻方式，萃取蘇軾詩作之用韻風格。因此，運用所建立之格律規則集、詞彙知識庫、詩人資料庫與萃取之潛在風格資訊，再搭配詩作系統便可達成協助今人創作出融合個人風格與古詩人風格之詩作的目的。

Since the techniques such as meter and rhyme are usually required, it’s not easy to create poetry-ci; furthermore, if we want to create the poetry-ci like ancient poets', it’s obviously more difficult. Under the circumstances, how to compose the poetry-ci with one's own particularity as well as the style of ancient poets' will be a critical issue. As the information technology grows up rapidly, the issue should be able to be handled easily.

We hereby propose a method. First, we construct the Chinese Poetry-ci Knowledge Base; second, we analyze it by Data Mining technology, and then the Classical Poetry Knowledge Base could be made hereat. By the Technology of Knowledge Base, the structure of knowledge could be added, searched and managed; moreover, the referable information of writing poetry-ci could also be provided at the same time.

Now we propose a four step approach to construct Su-Shi's Classical Knowledge Base.

Step 1: Classify poetry-ci & construct poet's profile According to Su-Shi's style, we classify his poetry-ci by domain expert's analyses and opinions in order to extract his style as possible as we can. Besides, we also rearrange his profile including the following important attributions: his periods, inhabitiveness, preferment, degradation, etc.

Step 2: Transform the rules of poetry-ci We transfer the meter of jintishi to a rule set by domain expert's opinions.

Step 3: Construct Vocabulary Knowledge Base We segment Su-Shi's poetry-ci and use metadata technology to describe his terms so as to construct Vocabulary Knowledge Base.

Step 4: Analyze and extract Su-Shi's style We use data mining technology to analyze the way of PingZe which Su-Shi might prefer, and therefore the rules of rhyme could be extracted.

Combining the above-mentioned four step approach and using the automatic system of composing poetry-ci, it’s no more a dream to make poetry-ci which are similar to the ones of Su-Shis' in the coming future.