發展整合文字探勘之商品需求預測模式

Abstract

現今產品的行銷手法改變且生命週期逐漸縮短，使得市場對汽車產品的需求快速浮動，導致上游半導體廠商的晶片商品需求被扭曲(長鞕效應)，商品需求預測變得更加的困難。本研究係將需求預測商品 – 汽車晶片的下游商品與產業關聯網路新聞資訊納入需求預測模型當中，發展一個結合文字探勘與資料探勘的商品需求預測模式。在所提出的模式中，利用網路服務及排程程式來蒐集新聞文字內容，並透過文字探勘的方法來處理其內容，並利用TF-IDF將非結構化文字內容轉化成向量，再結合前向式的類神經網路方法來建立需求預測的模型，最後利用該模型來做商品需求預測分析。 研究結果顯示，包含新聞文字內容的商品需求預測模式能夠有效的提昇預測的準確度，以幫助高階管理人員及分析人員能參考預測結果來調整其商品生產及庫存相關策略。此外，所蒐集的文字內容也可幫助用來找出市場需求方向及訴求，以降低風險與成本，來達到永續經營的目的。

The demand forecasting has become a challenge, which under the upstream semiconductor company IC product demand influence(bullwhip effect) of the marketing strategy changes and shorter product life cycle in the automotive products. The aim of this study is to take automotive IC downstream related market sentiment into consideration in developing a product demand prediction model which is integrated the text mining and data mining technology. In our propose model, the text content be collected by the network services and a custom console jobs, then use TF-IDF to transfer the unstructured news text content to vectors, which be integrated with Feed-Forward artificial neural network to build a demand forecasting model for the product demand analysis. According to the research result, it indicates the proposed model with additional market sentiment, can improve the forecasting accuracy and assist high management team and analyst to adjust the product production and inventory strategies. Moreover, the retrieved market sentiment can help to conceive the strategy of marketing demand.