徐浩翔

Hsu, Hao-Hsiang

碩士論文(2022)

基於重要影響因素之銷售預測方法與技術開發

Development of Important influencing Factors-based Sales Forecasting Method and Technology

關鍵字 Keywords

sales forecasting、social media、big data、important influencing factors

摘要

銷售預測為零售業重要之商務活動，目的在透過銷售預測消除庫存過剩與商品缺貨之問題，以提升營業績效。銷售預測之相關因素眾多，包括如景氣指標之參考因素，以及商品銷售影響因素如商品特徵與售價、天氣、行銷活動等。這些因素對預測準確度影響不一，篩選重要之銷售預測相關因素以提高預測準確度有其必要性。隨著網際網路的發達與社群媒體的普及，社群媒體對於消費者購買商品行為的影響與日俱增，如何從網路社群媒體資料中，萃取出有助銷售預測之因素與特徵以提升銷售預測準確率，也是當前銷售預測的課題。

本研究提出一考量社群媒體影響力與因素重要性之銷售預測方法，並開發實現技術，再驗證技術之正確性與方法之有效性。本方法包括:相關因素之資料收集與特徵分析、社群媒體影響力分析、重要因素分析、資料前處理、銷售預測等步驟。

本研究使用Y企業所提供的商品銷售數據進行驗證。首先建構情感分析模型，用以分析句子的極性與強度，經過10折的交叉驗證後，情感模型的Macro-F1達到76.96%。接著針對兩間商店的兩個商品分別進行實驗，結果顯示先進行重要影響因素分析者之均方根誤差分別降低了6%、12%、7%與20%，而平均絕對誤差分別降低了6%、6%、3%與23%，確認本研究所提基於重要影響因素之銷售預測方法與技術開發可行且有效。針對社群媒體影響力分析模型進行實作，發現使用社群媒體影響力有助於提高預測準確度。

Abstract

Sales forecasting is an important business activity in the retail industry. The purpose of sales forecast is to eliminate the problem of over-stock and out-of-stock products in order to improve business performance. There are many factors related to sales forecasting, including reference factors such as economic indicators, as well as factors influencing product sales such as product characteristics and selling prices, weather, marketing activities, etc. These factors have varying influence on forecast accuracy, and it is necessary to select the important sales forecast-related factors to improve forecast accuracy. With the development of the Internet and the popularity of social media, the influence of social media on consumers' purchasing behavior is increasing day by day. How to extract the factors and characteristics that can help sales forecasting from online social media data to improve the accuracy of sales forecast is also a current issue in sales forecasting.

This research proposes a sales forecast method that takes into account the influence of social media and the importance of factors, and develops a realistic technique to test the validity of the technique and the method.

This research was conducted using the sales data provided by company Y. First of all, the sentiment model was constructed to analyze the polarity and intensity of the sentences, and the Macro-F1 of the sentiment model reached 76.96% after 10-fold cross-validation. Then, the experiment was conducted for two products in two stores. The results of the experiment proved that the RMSE was reduced by 6%, 12%, 7% and 20%, and the MAE was reduced by 6%, 6%, 3% and 23%, respectively, so that the sales forecasting method and technology development based on the significant influence factors is feasible and effective. Conducting a hands-on social media impact analysis model.It is found that social media influence can help to improve the accuracy of forecast, so the social media influence analysis model proposed in this research is feasible and effective.