摘要 台灣圖書業普遍存在對於新書銷售量不明確的問題。對於圖書經銷商而言,在新書未來銷售量無法掌握下,容易因為依賴經驗法則訂貨而造成存貨過多或是存貨不足的問題,其帶來的負面影響將是成本的增加與收益的減少。對於資本額不高的圖書經銷商產業而言,無疑是一項嚴重的考驗。因此,有效的新書銷售預測模式更顯的重要。 對於新書而言,因為該產品不具有實際銷售紀錄所以圖書採購人員在資訊不足的情況下將無法充分掌控最適的訂購數量。本研究透過分析影響圖書銷售量的相關因素,利用案例式推理(CBR)的案例比對方式將新書和舊書進行因子比對,擷取和該新書屬性最相近的舊書,並以該舊書的銷售軌跡來作為新書的銷售預測資料。為提升案例式推理的預測效果,本研究嘗試加入分群的概念,將龐大的資料庫利用類神經網路中的SOM分群法加以分群(簡稱S-CBR),以增加傳統案例式推理法在案例比對與銷售預測上效率和效果。 最後,將S-CBR法之預測結果嘗試和K-CBR(以K-mean分群)、傳統CBR法、批量訂購法進行比較,透過準確度與成本面等兩大層面來分析本研究所建構的S-CBR預測效果。結果發現,S-CBR法皆相較於K-CBR法、傳統CBR法與批量訂購法在本個案資料的預測效果上有較佳的表現。
ABSTRACT The generic problem in book industry in Taiwan is uncertain sales of new released books. Making an order by relying on past sale records can easily result in over stock or shortage for book retailers. It also increases cost and depresses profit. No doubt it’s a severe challenge for low capital book-retail industry. Therefore, an effective forecast model is critical. Because there are no past sales records of new released books, purchasers in book industry won’t be able to know the most likely amount of an order, which can satisfy future demands. This research analyzes the factors, which affect book sales, tests the factors similarity of past and new released books by using case retrieval of Case-Based Reasoning (CBR), and estimates the sales of new released books by using the sales records from past released books that have similar topics. In order to increase the effectiveness of the forecast from CBR, this research introduces the concept of clustering, which divides the gigantic sales database by utilizing SOM of neural network, S-CBR for short. This concept increases the effectiveness and efficiency of case retrieval in traditional CBR. Finally, the result of the forecast of S-CBR is compared with the results of K-CBR, which is divided by K-mean, and traditional CBR. The conclusion is that S-CBR is more accurate in the forecast of the data than K-CBR or traditional CBR is.