標題: 基於文字熱門程度之銷售預測和點擊率預測資料探勘方法
Data Mining for Sales Forecasting and Click-Through-Rate Prediction Based on Word Popularity
作者: 歐 漢 尼
劉敦仁
Hani Omar
Dr. Duen-Ren Liu
資訊管理研究所
關鍵字: Forecasting;Prediction;Autoregressive Integrated Moving Average (ARIMA);Back Propagation Neural Network (BPNN);Google Search;Popularity;Latent Dirichlet Analysis (LDA);Forecasting;Prediction;Autoregressive Integrated Moving Average (ARIMA);Back Propagation Neural Network (BPNN);Google Search;Popularity;Latent Dirichlet Analysis (LDA)
公開日期: 2015
摘要: Internet technology has become a part of everyday life for retrieving data, contacting, entertainments, shopping, marketing, and some in the emerging business and developing world. Due to thousands of pages and services on the web, search engines are designed to search for information on the World Wide Web. The words of query are the main part in the retrieving results by search engines; and hence the word popularity is important to improve the correlated business for service providers. In this study, we first proposed a hybrid ARIMA and Back Propagation Neural Network for sales forecasting based on the popularity of article titles to enhance sales and operations planning. Publishing industries usually pick attractive titles and headlines for their stories to increase sales, since popular article titles and headlines can attract readers to buy or subscribe to magazines. The popularity of article titles are analyzed by using the search indexes obtained from Google search engine. We proposed a novel hybrid neural network model for sales forecasting based on the popularity of article titles, historical sales data, and the prediction result of Autoregressive Integrated Moving Average (ARIMA) forecasting method. Our proposed forecasting model is experimentally evaluated and the result shows that our proposed forecasting method outperforms conventional techniques which do not consider the popularity of title words. Second, we use the power of words of online advertisements, which impressed by search engines (where users add their queries for searching), to predict the users’ click-through rate (CTR) of advertisements. We use the important words in the queries which correlated to the advertisements and to boost the prediction performance. Also, we use the popularity of words to cope the cold-start problem when new users insert their query without having any knowledge about them using just their queries. Our proposed prediction model is evaluated and the result of the experiments shows that CTR prediction using word popularity outperform the prediction models without word popularity, and the same for cold start problem.
Internet technology has become a part of everyday life for retrieving data, contacting, entertainments, shopping, marketing, and some in the emerging business and developing world. Due to thousands of pages and services on the web, search engines are designed to search for information on the World Wide Web. The words of query are the main part in the retrieving results by search engines; and hence the word popularity is important to improve the correlated business for service providers. In this study, we first proposed a hybrid ARIMA and Back Propagation Neural Network for sales forecasting based on the popularity of article titles to enhance sales and operations planning. Publishing industries usually pick attractive titles and headlines for their stories to increase sales, since popular article titles and headlines can attract readers to buy or subscribe to magazines. The popularity of article titles are analyzed by using the search indexes obtained from Google search engine. We proposed a novel hybrid neural network model for sales forecasting based on the popularity of article titles, historical sales data, and the prediction result of Autoregressive Integrated Moving Average (ARIMA) forecasting method. Our proposed forecasting model is experimentally evaluated and the result shows that our proposed forecasting method outperforms conventional techniques which do not consider the popularity of title words. Second, we use the power of words of online advertisements, which impressed by search engines (where users add their queries for searching), to predict the users’ click-through rate (CTR) of advertisements. We use the important words in the queries which correlated to the advertisements and to boost the prediction performance. Also, we use the popularity of words to cope the cold-start problem when new users insert their query without having any knowledge about them using just their queries. Our proposed prediction model is evaluated and the result of the experiments shows that CTR prediction using word popularity outperform the prediction models without word popularity, and the same for cold start problem.
URI: http://etd.lib.nctu.edu.tw/cdrfb3/record/nctu/#GT079734806
http://hdl.handle.net/11536/141684
Appears in Collections:Thesis