Working paper, presented in CWEIST 2020
The reports of fake social media accounts have caused increasing concerns about the economic and social viability of social media. But the shadow economy around social media fake accounts is still poorly understood, due to the lack of data, transparency, and reliable way of detecting fake accounts. This research uses game-theoretical analysis to understand what makes social media influencers buy fake accounts, how the existence of fake accounts impact consumers, advertisers, social media platforms, and the overall social welfare. The central contribution of this paper is the characterization of equilibrium scenarios. We find that in a pooling equilibrium, only the influencer with low content quality (‘‘low type’’) buys fake accounts while the high type one does not. However, in the ‘‘costly-separating’’ equilibrium, the purchasing behavior flips, i.e., only the influencer with high content quality buys fake accounts while the low type does not. In addition, in the ‘‘costless separating’’ equilibrium, no influencer purchases fake accounts. In terms of the efficiency of the platform-initiated fake accounts detection, we find that the platform could under-detect, over-detect, or efficiently detect the fake accounts. Thus, we may not rely entirely on social media platforms to self-regulate their fake accounts.
Working paper, presented in CWEIST 2020
Working paper, under 2nd round review at Information Systems Research
Most crowdfunding sites rely on a single funding scheme (e.g., donation, reward, or equity) on their platform, while existing research on backer motivation suggests that backers can be motivated in multiple ways, through intrinsic and extrinsic rewards. Thus, a natural question pertains to whether the use of multiple funding schemes could enhance crowdfunding outcomes. This question critically depends on the interplay between extrinsic and intrinsic motivations. We provide empirical answers to this question by leveraging a natural experiment that took place on a leading reward-based crowdfunding platform that added a new donation option for all reward-based campaigns. Our results indicate that the addition of the donation option to reward-based campaigns has the impact of increasing their success relative to a control group of campaigns. This effect is observed even after accounting for inter-campaign differences via matched samples. To further account for inter-campaign differences, we track campaigns that experienced the site change during the fundraising process, and found that these campaigns gained more funds in the periods after the donation option was introduced. In addition, we find that the heightened crowdfunding success comes from an increase in the number of backers contributing to the campaigns. A finer analysis indicates that the increase in donations has a positive spillover effect on the reward-based contributions, leading to a better crowdfunding outcome. Theoretical and practical implications of our findings are discussed..
Working paper, presented in WITS 2019
Jump bidding is a prevent phenomenon in online auctions with important revenue implications. We propose a novel explanation for jump bidding based on budget constraints – budget-constrained bidders have incentives to jump their bid to increase their likelihood of winning in time-prioritized online auctions. We derive the conditions under which jump bidding is optimal for bidders at the margin, i.e. the current price is one increment away from their maximum bid. We find that the gap between a bidder’s budget and his true valuation and the bid increment jointly influence the likelihood of jump bidding. We propose a hybrid strategic-at-margin (SAM) bidding strategy for budget-constrained bidders. Our discrete-event simulations suggest that SAM outperforms alternative strategies of always bidding the minimum required bid or always jumping. We also propose a proportional SAM bidding strategy that requires less information about other bidders and still outperforms the strategy of bidding the minimum.
Artificial Intelligence (AI) Framework for Venture Capital Industry: How to Predict Startups’ Success in Primary Market
CEO @ DeepSearch.AI, funded by Angel round investment, ongoing
Asymmetry widely appears in venture capital industry including information, knowledge, relationship et al, which heavily affects investors’ decision making. Our startup focus on data-driven venture capital revolution, aiming to empowering investors to discover unicorns. Besides, our technical team supports various data scraping demands. If you need some data to do research or other things, feel free to contact me. We can help you!
Work at Zhongchou Inc. & Yuanshihui Inc., finished in December 2016
I spent one and a half years in crowdfunding industry, which are two platform experiences including reward-based and equity-based. I was involved as a software engineer at beginning, and then changed to a prodcut manager. My work mainly lied on product design, research and development. Besides, I also kept eyes on trends in industry. At the end of 2016, a paper about crowdfunding market which I coauthored was published in Financial Innovation.
Course Project, finished in December 2014, CS, University of Minnesota
This project is to help users better do city exploration with respect to what they are familiar with. In this project I go through the complete user experience design procedure: user research, prototype, cognitive walkthrough, heuristic evaluation, implementation.
Precision CrowdSourcing: Closing the Loop to Turn Information Consumers into Information Contributors
Funded by Google Research, finished in May 2015, GroupLens Research, CS, University of Minnesota
In this project, we propose a basic framework we call precision crowdsourcing – a systematic way of approaching the process of turning an information consumer into a long-term contributor through a series of requests, feedback, and interaction.
Empirical Finance & Financial Econometrics Workshop, finished in November 2011, CCER, Peking University
In this project, I proposed to use neural networks model to predict stock market trends, the performance of prediction is compared to other prediction models such as GARCH model
National Key Technology Project, finished in September 2012, EE, Tsinghua University
Due to high nonlinearity and memory effects of dual-band power amplifiers, I did behavioral modeling for concurrent dual-band power amplifier and digital pre-distortion based on 2D-Real Value Time Delayed Neural Networks, achieving an 10dB improvement in model accuracy than conventional approach.
Supported by Mitsubishi Heavy Industries, finished in May 2010, EE, Tsinghua University
Considering the complex process and high cost of radiation emission measurement in anechoic chamber, I developed a measurement system based on a current probe in common laboratory, reducing the cost of test dramatically while measurement accuracy is maintained.
National Student Research Training Program, finished in November 2008, EE, Southeast University
To eliminate the security risks in present wireless communication system, I designed a secure transmission approach from end to end based on SDIO encryption card, solving the security problem in wireless communication.