| Summary: | 碩士 === 國立臺北科技大學 === 資訊與財金管理系 === 107 === Sales forecasting is one of critical management tools for companies to make informed business decisions. The traditional sales forecasting methods can be divided into two categories: quantitative forecasting method and qualitative forecasting method. The quantitative forecasting method uses historical data mathematical method to predict, and the qualitative analysis method uses wisdom and opinion forecast by experts and department heads. Both methods have not performed well on commodity sales forecasts. In the past, many studies have pointed out that consumer reviews affect consumer purchases, and many studies use consumer digital reviews to predict product sales.
Therefore, this study proposed a sales forecasting model that combined commentary sentiment analysis with Long Short-Term Memory Network (LSTM), in addition to exploring the impact of combining commodity reviews and historical sales data on e-commerce sales forecasts, as well as predicting future sales of commodities with the short-term demand characteristics.
The comments and the sales figures were collected from “taobao.com”, the comments were converted to the ratings of confidence, “positive” and “negative” through sentiment analysis, in the training stage. The sales forecasting model was trained by historical data for predicting sales volume in the next period, but also using time-series sequence. The study designed multiple conditions in order to validate the accuracy of the model and compare the difference between the various length of training time-series, windows size, using and without using sentiment analysis of comments. This study assumed the LSTM model combined with sentiment analysis will have better performance in sales forecasting, and the results of experiments were consistent with the expectation.
In addition to providing corporate decision support about the future, the contribution of this research also attempts to improve the accuracy of sales forecasts by combining qualitative and quantitative data, and provide future research directions for relevant researchers in the future.
|
|---|
