電子商務平台銷售件數影響因子分析模型: Yahoo購物商城電子商務平台之實證研究

The Determinants of Sales Quantity on E-Commerce Platform ：An Empirical Study of Yahoo Mall

江明吉(Ming-Chi Chiang)

指導教授 ： 黃俊堯

國立臺灣大學/管理學院/商學研究所/碩士(2016年)

https://doi.org/10.6342/NTU201600912

摘要

本研究以電商平台Yahoo超級商城為研究對象， 從2015年7月19日蒐集至2015年9月24日共68天，190項商品，52間網路店家觀察值為研究對象，探討影響電商平台銷售量的因素。本研究利用所蒐集之前半月資料(7/19-8/18)，利用多元線性迴歸進行分析並建立模型，以分析各項因素對銷售件數影響之程度以外將迴歸模型進行殘差分析，並確認誤差項呈隨機分佈以確保模型預測能力。

確認模型預測能力後，利用後半月資料(8/18-9/24)將變數帶入模型中得出模型銷售預測值，並與實際值比較，利用T檢定測驗其差異程度來檢驗模型預測的準確程度。本研究的主要發現如下:

1.本研究找出了不論依照何商品別或商店別歸納變數，皆影響銷售件數的最顯著因子「售價」、「問與答人數」、「當日排序」。消費者在購物時會進行詳細的比價選擇售價較低的產品，並且依照先前購買或產生興趣的人對於產品的詢問以及討論程度來判斷此產品之熱門程度選擇產品。「當日排序」則與銷售件數呈現負相關，代表著排序越小越靠前之產品越容易被消費者看見，而消費者有偏好選擇滿足其需求的前提下，資訊搜尋成本較低之產品。此結果與文獻探討中消費者行為模式會對產品產生知覺(Awareness)、興趣(Interest)後產生搜尋(Search)與分享(Share)並產生行動(Action)相符。

2.本研究所建立之銷售預測模型，經T檢定之後預測值與實際值無顯著差異， 且依照商店別分類的變數所產生的模型預測能力高於依照商品別分類的模型，電商平台營運的店家可依照此模型來預測未來可能的銷售績效。

3.根據本研究之結果顯示，商店別的銷售件數差異比商品別分類更為明顯，店家可將資源投入在提升店家曝光與排序優先度、整體店家商品數量等，依照此模型產生之建議，來修改其營運方向及資源配置比例，以提高銷售量。

關鍵字

電商平台 ； 銷售預測 ； 多元迴歸 ； 資訊搜尋成本 ； AISAS

並列摘要

This study provides an insight of the factors that influences the sales quantity of the product， which sold on e-commerce platform. The research sample include 190 items and 52 online stores from the e-commerce platform website， Yahoo Mall for the period since July 19 to September 24， 2015. The research conducts a sales predict model by using data from July 19 to August 18 with multiple linear regression method. And put the last data (From August 19 to September 24) into the model to get the prospective sales quantity. Analysis the predict ability by using T test to check the difference between the prospective sales quantity and the real quantity.

The following are the main findings from the result of analysis:

1.The statistical significant factors that influences the sales quantity under two different way to conclude the data by product or by store are the “Price” ， “Q&A quantity” and the “Daily rating.” The customers will choose the lower price products and rating by the quantity of people who discussing this product to evaluate the quantity of the product. Daily rating is negative relative to sales quantity， which represents the smaller the rating number ; the easier this product can be seen by customers. And the customer will prefer to buy the product which can fit their demand with lower searching cost.

2.The sales prospective quantity by the model this study made is not statistical significant different with the real sales quantity with the T test. The predict ability of the model with the data conclude by stores is higher than conclude by products. The stores on the e-commerce platform can follow the result of the model to realize and predict the future sales quantity.
3. Base on the result of the study， the sales quantity concluded by stores is more divergence than concluded by product. The stores can put resource on espousing the brand ，rating and the product quantity that stores have. And follow the result of the model to reallocate resource and operation way to raise the sales quantity.

並列關鍵字

e-commerce platform ； sales predict ； multiple linear regression ； searching cost ； AISAS

參考文獻

葉昱呈(2011), 探討網路口碑傳播與社會關係連結對於線上購物意願之影響

Chiang Wei-yu Kevin， Zhang Dongsong，Zhou Lina (2004)Predicting and explaining patronage behavior toward web and traditional stores using neural networks: a comparative analysis with logistic regression .

Dai Wei， Peng Geng， Liu Ying， Li Shuaipeng (2014) A Prediction Study on E-commerce Sales Based on Structure Time Series Model and Web Search Data ， Chinese Control and Decision Conference (CCDC)

Jiang, Hongbo(2015),Platform Based E-Commerce Credit Evaluation System Based on Fuzzy Comprehensive Evaluation

Li Zhou1 and Qing-yi Chen(2014),Evaluation of E-commerce Performance in SMEs based on Vector Auto Regression Model