Modeling Life Cycle Dynamics of Content and Membership of User-Generated Content Sharing Networks with Two Case Studies

• Main Authors: Rong-Huei Chen, 陳榮輝
• Other Authors: Shi-Chung Chang
• Format: Others
• Language: en_US
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/8b228d

博士 === 國立臺灣大學 === 電機工程學研究所 === 106 === User-generated content sharing networks (UGCSNets), in which members are content contributors as well as users, have had a significant impact on the sharing economy and on society via the sharing and reuse of contents. In current practice, many new UGCSNets failed to grow. Some successful UGCSNets have also been declining. There are many forms of UGCSNets. In this work, our focus is a content sharing aspect in UGCSNets such as knowledge sharing networks (Wikipedia) in which users share their contents. Social aspects and characteristics such as connecting to others in Facebook are not addressed. Managing a UGCSNet for sustainability requires a quantitative grasp of how individual members’ participation and sharing affect and are affected by each other so that the growth and decline time and speed of UGCSNets can be predicted and proactive actions can be taken as early as possible. Qualitative descriptions from literature presented the positive and negative feedback loops. In the positive loop, members contribute contents, enabling a UGCSNet to provide positive benefits which are the basis to attract and retain members. However, the network size may have a negative impact because of free-riding and social loafing. The overall impact on the dynamics of UGCSNet is a combination of positive and negative effects. Motivated by findings of qualitative descriptions from literature and available empirical data, this dissertation aims at quantitatively modeling the life cycle dynamics of UGCSNets from growth to decline by exploiting available data and takes a teaching material sharing network (TMSN) and Wikipedia as conveyer problems. The main challenges are understanding and modeling the positive and negative loops based on qualitative descriptions from the literature and limited collective data. Specific challenges are as follows: C1) Modeling individual participation and sharing and their effects on membership and total content volume using only collective statistics; C2) Modeling the feedback effect from membership and shared contents to individual sharing and participation; C3) Designing quantitative model of how new contents affect individual behaviors and life cycle evolution; C4) Designing quantitative model of how free riding and social loafing negatively affect individual behaviors and life cycle evolution; and C5) Extracting and updating model parameters with the evolution of a UGCSNet given limited data availability. To overcome the challenges, this dissertation proposed several methodologies: M1) A structure of generalized Bass diffusion model-embedded stochastic difference equations (GBDSDEs) quantitatively models the qualitative descriptions linking members, content sharing, and ability to attract and retain members. M2) The GBDSDEs capture the positive feedback effect from membership and contents and negative feedback effect from free riding and social loafing to the membership and content evolutions in UGCSNet life cycle. i) Two difference equations describe the membership and content volume evolutions separately using individual participation and content uploading as stochastic drives. ii) The drive models are an innovative generalization of the Bass diffusion model (BDM) and innovatively model how the individual willingness to participate and share is affected by the total membership and contents. iii) Within the structure, an auto-regression model is used to capture the usefulness of contents (PI), where the PI of new uploaded contents is discounted as time elapses. iv) A monotonically decreasing quadratic model is used to capture the decrease of individual sharing willingness with the free riding effect. M3) A quasi-bootstrap-based nonlinear least square (QBNLS) method is adopted to extract from the limited early stage empirical data and periodically update the model parameters as the UGCSNet evolves. The dissertation demonstrates values of our quantitative modeling approaches: R1) Capability of robust prediction of life cycle stage transition time and speed based on limited empirical data. Analyses of the coefficients of determination show good fits between model predictions and actual outcomes for both SCTNet and Wikipedia. The simulation validations demonstrate robustness of GBDSDEs in predicting time and speed of UGCSNet growth and decline. R2) Application of what-if analysis for impact estimation of management strategies on life cycle dynamics. Applications of the modeling approach to what-if analyses demonstrate the ability to predict and assess the effects of specific managerial strategies such as the initial content volume and the number of founding altruistic members on the growth and the new shared content volume at situation stage on the decline of a UGCSNet. The approaches have the potential to assess at what stage of life-cycle a UGCSNet is at and evaluate the influences of managerial strategies on network life cycle.