用機器學習分析並預測線上付費課程用戶的學習行為以及高價值用戶其特徵-以Hahow為例

Abstract

隨著科技日新月異，網路已深入家家戶戶，串流影音跟著興起，人們逐漸習慣於即時取得網路上的資訊，觀看各式的影音資源。資訊傳遞的即時性與方便性，帶動了教育形式的多元化，促使網路教育、網路學習直接將傳統課堂延伸到人們的手機和電腦，各種國內外線上課程與平臺如雨春筍般問世，2012 年被紐約時報稱作「MOOC 元年」，除了知名的 Coursera、Udacity、edX 之外，全球知名 企業如google、微軟等也相繼推出各類線上課程，說明遠距教育和數位學習的資源已經擁有爆炸性的成長且日漸普遍。 既然數位資源如今已相當豐富，若能藉由數據分析的作法，透過平臺上蒐集的資料深度挖掘學習者的學習行為及特性，便能進一步幫助線上課程平臺推廣客製化、豐富的內容給學習者，甚至更加精準掌握學生的觀點與吸收狀態，進而對課程做出改善與調整。 本研究與台灣知名線上課程平臺 Hahow 合作，取得用戶購買課程的紀錄與文字等從平臺上蒐集的資料，在文字上的語意挖掘、文字統計特徵、用戶學習行為等不同種類的資料上做個別的資料處理和特徵萃取，建立一個基於機器學習的混合式模型，用來預測用戶購買課程花費的總金額，藉以區分各階層價值的用戶，透過此預測模型，能幫助 Hahow 平臺瞭解和掌握高價值用戶的學習行為表現、學習特徵、觀點與態度，甚至由分析結果能進一步提供用戶更適合、高品質的課程，提升用戶整體的滿意度與黏著度。 綜合來說，此次分析 Hahow 平臺用戶學習行為的研究價值在於能提供台灣市面上的線上課程平臺，少數在學術領域用數據分析的方法，進行文字探勘與資料挖掘，瞭解用戶學習行為的一項實作。從中文評論、問答、繳交作業的情形和其他學習活動等特徵進行建模，預測高價值的用戶，進而掌握不同價值層級用戶的學習行為與習慣，長期來看，能促進 Hahow 網站往後上架的課程將更符合相關客 群與使用者需求。

With the rapid development of technology, people are gradually getting used to obtaining information on the Internet in real time and watching various video resources. The real-time and convenience of information transmission has led to the diversification of education forms, and promoted online education development. In 2012, in addition to the well-known MOOCs platform like Coursera and edX, many global companies such as Google and Microsoft have also launched various online courses, indicating that digital learning has been increasingly common nowadays. Now that digital resources are quite abundant, if data analysis can be used to deeply learn the learning behavior and characteristics of learners through the data collected on the platform, it can further help them to improve their online learning environment. In this study, we cooperated with Hahow, a well-known online course platform in Taiwan, to obtain the data collected from the platform such as the records and texts of users purchasing information, and use different methods such as semantic mining, text statistical characteristics to come up dozens of user learning behaviors features. Data processing and feature extraction create a hybrid model based on machine learning to predict the total amount of users spending on purchasing courses to distinguish users of different levels of value. This prediction model can help Hahow explore the performance of learning behaviors, learning characteristics, opinions and attitudes of high-value users. Moreover, the analysis results can further provide Hahow an insight to improve the overall satisfaction of users. In general, the research uses data analysis related methods in the academic field to conduct text exploration and data mining to understand user learning behavior in the online course platform. In the long run, it can bring extra value to companies that provide online courses in Taiwan and help them get to know more about their customers. With those information, they can build a more satisfying platform for both teacher and students.