The cost of search and evaluation in problem-solving social networks : an experimental study

• Ano de defesa: 2016
• Autor(a) principal: Farenzena, Daniel Scain
• Orientador(a): Lamb, Luis da Cunha
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Palavras-chave em Português: Interface homem-maquina; Teoria : Jogos; Redes sociais
• Palavras-chave em Inglês: Social computing; Human-computer interface; Game theory; Social experiments
• Link de acesso: http://hdl.handle.net/10183/148280

Online networks of individuals have been used to solve a number of problems in a scale that would not be possible if not within a connected, virtual and social environment such as the internet. However, the quality of solutions provided by individuals of an online network can vary significantly thus making work quality unreliable. This dissertation investigates factors that can influence the quality of the work output of individuals in online social networks. Specifically, we show that when solving tasks with small duration (under 5 minutes), also known as microtasks, individuals decision making will be strongly biased by costs of searching (and evaluating) options rather than financial or non-financial incentives. Indeed, we are able to show that we can influence individuals decisions, when solving problems, by rearranging elements visually to modify an the search sequence of an individual, be it by designing the virtual work environment or manipulating which options are first shown in non-controlled environments such as the Amazon Mechanical Turk labor market. We performed several experiments in online networks where individuals are invited to work on tasks with varying degrees of difficulty within three settings: mathematical games with objective truth (Sudoku and SAT instances), surveys with subjective evaluation (public policy polling) and labor markets (Amazon Mechanical Turk). We show that the time spent solving problems and the user interface are more relevant to the quality of work output than previous research have assumed and that individuals do not change this behavior while solving the sets of problems. Finally, to complement our study of online problem-solving, we present additional experiments in an online labor market (Amazon Mechanical Turk) that agrees with our networked experiments, shedding new light on how and why people solve problems.