線上隨選廣播節目提供聽眾一個由自己控制節目撥出時間的收聽選擇，然而如何在浩瀚的節目中找到自己想收聽的節目卻是一大難題。本研究之案例為一線上隨選廣播節目服務網站，其所收錄的節目本身並無節目的相關介紹，目前須依賴人工逐一由節目中分析主題並建立檔案。而節目主題的資訊分析(Analysis of information)，必需經由專業的訓練，才能將訊息中的特性、關鍵字製作成可供檢索的內容，以便後續重覆查詢。透過人工處理的方式，不僅成本高昂，且易造成分析結果品質不一。因此，本研究透過語音辨識、網路爬蟲、中文斷詞與機器學習方法，將廣播節目由聲音轉為文字文本，搭配網路爬蟲擷取時事中新創的詞彙，提升萃取文本特徵項的效果，優化文本特徵項的準確性。透過自動化的主題分析可以大幅節省人力並獲得一致品質。本研究同時以階層式分群等演算法，找出各個節目的關聯性，作為節目搜尋的基礎。此一階層性的節目關聯分析可方便使用者了解同一主題之不同觀點與完整的事件報導。 Online on-demand radio programs provide listeners with a choice of time to set aside their own programs, but how to find the program they want to listen to in the vast program is a big challenge. The case of this study is an online on-demand broadcast program service website, which contains the program itself is not related to the program, the current need to rely on the manual one by one in the program analysis and the establishment of the file. The Analysis of information on the subject matter of the program must be professionally trained to produce the characteristics of the message and the keywords to be retrieved for subsequent repetitive inquiries. Through the way of manual processing, not only costly, and easily lead to different quality of the results of the analysis. Therefore, this study transforms the broadcast program from the sound to the text through the speech recognition, the network reptile, the Chinese adverbial and the machine learning method, and uses the web crawler to extract the new words in the current affairs to enhance the effect of extracting the text feature , To optimize the accuracy of text features. Through the analysis of the theme of automation can save a lot of manpower and get consistent quality. In this study, the hierarchical clustering algorithm is used to find out the relevance of each program as the basis of program search. This hierarchical program association analysis facilitates the user to understand the different views of the same subject and the complete event coverage.
