- 學位論文
以社群網路之情緒分析預測汽車銷售市場
Vehicle Sales Forecasting by Sentiment Analysis Data
摘要
汽車產業是國家經濟發展的重要指標之一，美國汽車市場更是全球的關注焦點，牽連製造業與零組件供應，產能利用率的增加可以提振就業機會，對景氣環境的發展有深遠的影響。而今社交網站的興起，讓人們更易於討論產品、品牌與服務等，實現了一種意見交流的平台，有許多研究或公司企圖從中挖掘寶貴的意見、探取潛在的商機。 本研究以美國地區的新車銷售為研究標的，自社群網路獲取大眾對於買車的情緒觀點，且考量到市場經濟亦會影響民眾消費力，還選用了道瓊工業平均指數和標準普爾500指數之兩大美股指標，來實現預測輕型汽車和整體汽車兩種銷售數量。本研究使用最小平方支持向量迴歸（LSSVR）法來建構汽車銷售預測模型，採用平均絕對百分比誤差（MAPE）與均方根誤差（RMSE）為衡量指標，比較預測效果是否能優於天真預測（Naïve）、指數平滑（ES）、Holt-Winters、整合移動平均自迴歸模型（ARIMA）和季節性整合移動平均自迴歸模型（SARIMA）等時間序列法，並進一步實驗原始數據與經季節調整的數據。預測結果顯示，使用情緒分析混合股市資料集作為LSSVR模型的輸入因子時，模型有最佳的準確率，預測效果優於時間序列模型；而且對股市資料集與汽車銷售量做季節性調整，模型更能有效提高預測準確度。
並列摘要
The automobile industry plays an important role when it comes to economic development of a country. America’s automobile industry has been a bright spot in the global economy. It is related to a wide range of neighboring industries, such as steel, transportation, and component supply. The increase of capacity utilization can create more job opportunities that will have a far-reaching impact on economic structure. For years, social media websites have been spreading widely, and make it easier for people to share their comments about products and services of brands. People are now able to seek information and exchange their opinions in social media, that eventually may effect one’s purchase intention and behavior. For these reasons, many researchers and practitioners increasingly resort to social media to obtain valuable information and potential business opportunities. In this study, we focus on monthly new vehicle sales in the US. The dataset consists of three kinds of input variables. One is sentiment data set, another is stock market index data set and the other is the combination of the first two data sets. The sentiment data set use tweets data to get the public opinion about buying cars by sentiment analysis. Dow Jones Industrial Average (DJI) and Standard & Poor's 500 (S&P 500) are selected as the performance of purchasing power and market economy. The aim of this study is to predict light vehicle and total vehicle sales in the US, and the Least Squares Support Vector Regression (LSSVR) method is used to construct a vehicle sales forecasting model. For forecasting accuracy comparison, the mean absolute percentage error (MAPE) and root mean square error (RMSE) are computed. Then, compare with five time series models: Naïve、Exponential smoothing、Holt-Winters、ARIMA and SARIMA. Furthermore, this paper explores the usefulness of raw data and seasonally adjusted data for vehicle sales forecasting. Our empirical analysis indicates that using the combination of sentiment and stock market index data as the input variables of LSSVR model has the best predictive results. Comparisons of prediction accuracy demonstrate that our model outperforms other time series models. Moreover, LSSVR models with seasonally adjusted data perform better than unadjusted raw data, and the prediction accuracy of the proposed method is improved by approximately 35%.