數位口碑影響力建構之流行性商品銷售預測模型

Abstract

銷售預測在企業流程中扮演關鍵的角色。傳統以歷史資料為基礎的銷售預測方法只限用於需求穩定的商品，而對於銷售起伏不定的流行性商品而言，現存的預測方法表現不盡理想。然而，流行性商品因其需求起伏不定的特性，比一般長銷型商品更需要精確的預測才能確保企業的利益。 許多研究結果顯示數位口碑影響消費者的購買行為，而不同類型的數位口碑其影響力也不盡相同。本研究藉由分析網路評論的類別、評論人特質和評論影響力等面向以釐清數位口碑和消費者行為及產品銷售之間的關係。 本研究對於銷售預測主要有兩項貢獻：首先我們提出了一個結合語意傾向分析（Polarity Mining）、語意強度分析（Intensity Mining）及影響力分析的網路評論分類方法，並建立一套分析數位口碑差異的架構。另一方面，本研究從數位口碑影響力的觀點出發，分析網路評價對於消費者購買決策的影響。 本研究以台灣知名藥妝連鎖店之實際銷售資料為例，實行的結果顯示，本研究所提出的考量網路評論影響力的模型適用於具話題性、且能引發網路熱烈討論的流行性商品之銷售預測，其預測的結果遠比傳統的方法（例如：移動平均法等）優異。此結果說明數位口碑可被視為企業的一種無形資產，其與商品銷售之間有高度的關聯性，並可以運用於增進銷售預測的準確度。

Sales forecasting is one of the most critical parts of business procedure since it is the foundation of other operations. Traditional forecasting techniques are only suitable for products with stable demand. For those products with unpredictable sales trends, i.e. fashion products, the forecasting accuracy of traditional techniques are not acceptable. However, for these products, it is more necessary to construct a forecasting method in order to ensure enterprise profit. Prior research shows that there is a strong relationship between product sales and online word-of-mouth. Besides, some studies are concerned the extent of word-of-mouth impact to be different among different review categories. In this study, we try to figure out how word-of-mouth affects products sales by means of analyzing review properties, reviewer characteristics and reviews influences. This study contributes to the sales forecasting research in two folds. A novel classification model which involves polarity mining, intensity mining and influence analysis is proposed. We provide a theoretical framework to understand the difference between review categories. In addition, we introduced review influence on sales forecasting for fashion products and verified that the significant relationship between online word-of-mouth and consumer behavior. The proposed model is evaluated by using real data from a well-known cosmetic retailer in Taiwan. The experimental results demonstrate that the model is especially suitable for fashion products with abundant online reviews. It also shows in this study that the forecasting models adopting the refined review influence model outperforms the traditional time series forecasting models. Overall, this study contributes to the literature by proposing a new aspect of review classification, and introducing review influence on sales forecasting for fashion products. The result is favorable and shows that online word-of-mouth is a type of virtual currency that affects the product sales and can be applied on sales forecasting.