本論文提出了一個可依照唐詩情境來做推薦的唐詩推薦系統。本系統先利用詩詞本體論加標唐詩並使用類神經網路來將唐詩分類，再使用論文內提出的唐詩相似度計算公式來推薦唐詩。詩詞本體論為經由詩詞專家所編寫而成，為一個記錄了數個不同的主題類別，並且建立出類別間關係的資料結構。不同於以往的唐詩作者、格式或是關鍵字，使用詩詞本體論對唐詩所加標過的資訊，可以將唐詩的意境給表達出來。本論文使用自我組織特徵映射圖網路做為唐詩的分類器，配合加標後的資訊，將唐詩作意境上的分類。分類之後，本論文再提出之新的唐詩間相似度的公式-基於本體論相似度量測法，計算唐詩的相似度，推薦唐詩。 本系統使用被廣泛認知的「三百首唐詩」來當成系統的測試資料。每筆輸入的唐詩皆經由詩詞本體論加標過，以顯示唐詩的主題。本論文除了提出了一個新的唐詩分類之外，並提供了許多視覺化的分析工具以供系統的管理員更容易的去了解每個分群的特性。 最後，本論文提出了一個新型態的唐詩分類，和一個新的計算唐詩間相似度的公式-基於本體論相似度量測法，來將唐詩做適當的推薦。經由實驗與比較，本論文將唐詩被分為11群，其平均純粹度為0.80。而經由基於本體論相似度量測法所推薦的結果經由系統的使用者評分，亦得到不錯的成績。 Tang poems refer to poems written during China Tang Dynasty. By this time, Tang poems became one important literature, providing a large amount of poets and poems. This paper presents a Tang poem recommendation system based on the poem ontology and artificial neural networks. First of all, the feature representation of a poem is based on the poem ontology information, rather than its author, form, and key words. Then, the self-organizing feature map algorithm is adopted to cluster Tang poems into several clusters. In our system, the database is consisted of 300 widely-known poems. Each poem is tagged with the poem ontology to reveal their information of themes. Our system not only provides a new way to cluster Tang poems but also several visual analysis tools for users to understand the characteristics of poem clusters. In this thesis, a new measure for calculating the similarity degrees among poems is proposed. Based on the measure, a new approach for clustering Tang poems into several different styles is developed. Experimental results demonstrated that poems could be clustered into 11 different styles. In addition, the performance of the proposed poem recommendation system was encouraging.
