以經驗數據分析架構教材分享模型與設計獎勵誘因實驗

Abstract

在九年一貫教育體系與一綱多本政策下，近年來國內教材分享網站蓬勃發展。個人電腦普及讓老師能夠方便地在教材分享系統交流教材與教學經驗，教師可以增進教學能力並提供高品質教育以增進下一代孩童的競爭力。但是，目前缺少鼓勵分享行為的有效誘因。 本論文探討國小教師在台灣學術網路(TANet)上教師面對不同誘因和不同教材品質與數量所展現的行為模型，目的為提出架構處方性模型之方式以進行誘因設計，提高社群內資源數量與品質。模型中探討的行為，包含四項行為：網路社群的加入與離開及社群資源的上傳。過去文獻探討中發現使用者行為受社群教材質與量及獎勵誘因制度的影響，因此研究議題分為兩項：(1) 無誘因制度下，以教育網站思摩特之實證資料為基礎，使用者集體行為(collective behavior)模型 (2)固定社群資源下，架構使用者對獎酬制度反應模型之實驗設計。 為架構使用者集體行為模型，我們延伸過去以S型曲線探討上傳機率與時間關係的Bass Model[Lai05]。利用S型曲線具有在前段緩慢成長，累積一定數量活時間後中段迅速成長，後段漸趨飽和成長停滯的特性，設計S型使用者集體行為模型探討上傳、下載、加入和離開的機率與社群中教材質與量的關係。並在無誘因制度的前提下，利用專業教師網路社群思摩特(SCTNET)的經驗數據，進行迴歸分析，以曲線配適法導出其模型參數。經由分析架構出教師在不同教材數量與質量下的集體下載機率與加入機率為S型函數模型，此一模型可以用來提供研究者與網路管理者未來在預測在面對不同教材數量與質量下，集體教師可能下載機率與加入機率估計方式。 為架構使用者對獎酬制度反應模型之實驗設計，我們設計以相較分等的獎酬制度，在固定社群資源下，在實驗室同儕對同儕式教育分享系統上進行小型實驗，探討在不同的獎酬比例與獎酬金額兩因子下，使用者加入、離開、上傳與下載的反應模型。我們替換Bass模型中S型曲線的橫軸以預測在固定獎酬比例下，使用者面對不同獎勵金額的變化。並依據期望理論與預期效用理論捕捉使用者面對固定獎酬金額下，不同獎酬比例的影響。未來研究者可藉由此實驗設計於實驗室中架設的小型環境中，獲取使用者在不同獎酬比例與獎酬金額下的使用者加入、離開、上傳與下載行為的資料。藉由獲得的資料幫助架構對網站管理中獎酬制度設計的處方性模型。

Teaching material sharing (TMS) websites grow rapidly in Taiwan due to one-outline-multiple-texts policy under nine-year compulsory curriculum. The prevalence of personal computers provides convenient way to share teaching resources and experiences on TMS system. Teachers can enhance their teaching abilities and provide high-quality education to increase the competitiveness of our next generation. However, no effective incentives to encourage sharing activities are proposed. This thesis focuses on modeling TMS behavior over different reward incentive and different quality and quantity of teaching materials on TANet among elementary teachers. We aim to propose a prescriptive model for incentive design to establish a high quality and sustainably increasing quantity for teaching material sharing system. User behaviors discussed in this thesis including joining in to the community, leaving from the community, uploading and downloading teaching materials. Recent researcher investigated teacher behavior on knowledge sharing mainly influenced by website resources and reward incentive. Hence we consider two issues that are critical to the TMS: (1) Collective behavior modeling with SCTNet empirical data without reward incentive. (2)Experiment design for modeling response to reward incentive under constant website resources. We extend the Bass Model [Lai05] to S-shaped curve model for constructing collective behavior model. Bass Model uses the S-curve for describing the relation between uploading probability over time. The S-curve has a characteristic: slow start in the initial period, fast growth when the amount accumulated to a level and saturation in final. Based on this characteristic, we propose an S-shaped model which describes the relation between website resources and collective behavior which includes the probabilities of joining, download and upload. With empirical data from SCTNet without reward incentive, we analyze empirical data obtained from SCTNet and use regression to derive the parameter of the S-shaped model. This prescriptive model could serve as an estimation method for future network manager to predict the collective behavior under different teaching material quantity and quality. To design experiment for modeling response to monetary reward, the reward policy we consider is relatively ranking policy. Under constant community resources, we design small experiment on P2P TMS system in lab to model the user response including the probability of joining, upload and download over different reward proportion and reward budget. We replace the horizontal axis of the S-curve in Bass Model [Bas69] to conjecture the variation of user response including upload, download and upload under different reward budget in fixed reward proportion. We also base on the Expectancy theory [Ger04] and Expected Utility Theory [Nic02] to sketch the user response under different proportion in reward fixed reward. Through the Two-Factor factorial lab experiment, future researcher can obtain the user response data, which includes the probability of joining, download and upload, under different reward budget and different proportion of reward winner. Such results may help construct the modules of reward policy management in the prescriptive modeling framework for incentive design.