Professor, Computer Science, UCSB
Crowdsourcing provides a fast and easy way for those with challenging problems to get help from real human "workers" across the Internet. The growing popularity of crowdsourced systems has a number of implications on the security and viability of today's online systems. In this talk, I'll present some of our recent results across multiple projects that study the various benefits and downsides of crowdsourced systems. First, I'll briefly describe some of our earlier work on using crowdsourced workers to detect fake user profiles in online social networks. We designed a novel crowdsourced Sybil detection system, and performed large user studies to quantify its benefits and performance properties. Second, I'll discuss the growing challenge of malicious crowdsourcing (or Crowdturfing), its current status, and our efforts to understand the efficacy and limitations of crowdturf defenses using machine learning-based. Finally, I'll present some early results of our upcoming paper on crowdsourced investments, where we analyze 8+ years of historical data on crowdsourced investment analysis sites. We show that in extremely challenging fields like stock market analysis, while wisdom of the crowd does not help directly to produce good results, but it can be extremely valuable in distinguishing real experts from the rest. Our system shows that by applying a very simple investment strategy based on mining crowdsourced investment sites, we can consistently and significantly outperform the stock market averages in each of the past 8+ years.