糖心原创

Ph.D. Committee:听 Drs. Amit Sheth, Advisor, Keke Chen, Krishnaprasad Thirunarayan, Valerie Shalin (Department of Psychology) and Brandon Minnery (Kairos Research)

ABSTRACT

The wisdom of the crowd is a well-known example of collective intelligence wherein听an aggregated judgment of a group of individuals is superior to that of an individual.听The aggregated judgment is surprisingly accurate to predict the outcome of a range of听events from geopolitical forecasting to stock price index. Recent research studies have听shown that participants鈥� previous performance data contributes to the identification of a听subset of participants that can collectively predict an accurate outcome. In the absence of听such performance data, researchers have explored the role of human-perceived diversity听to assemble an intelligent crowd.听The online social networks are becoming increasingly popular in sharing and seeking听domain-specific knowledge. Tapping the crowd wisdom on online social networks can听help prediction for several real-world tasks. Independent and contextually diverse crowd听selection using social media data imposes unique challenges such as,

• Complementing short and potentially noisy social media data with domain specific knowledge.
• Lack of labeled data in evaluating closely connected social media participants.
• Combining social media text, indicating a diverse perspective, and network features, indicating potential influence, in diverse crowd selection.
• Interpretable diversity measure to understand the type of diversity that can enhance the performance of a crowd.

This dissertation first provides several data-driven measures from social-media data and听shows that participant diversity can be inferred from social media data and that it can听benefit performance in the real world prediction tasks. Domain-specific knowledge graphs听provide the foundational basis to evaluate and drive contextually diverse crowd selection听and complementing short social media text. Novel group detection and crowd selection听algorithms incorporating text, network, and knowledge-graph can automatically select听diverse crowd and also provide recognizable diversity interpretation. It is shown that such听a diverse crowd can accurately predict an outcome of real-world events. These results have听implications for numerous domains that utilize aggregated judgments - from consumer听reviews to econometrics, to geopolitical forecasting and intelligence analysis.