Many online platforms employ networks of human workers to perform computational tasks that can be difficult for a machine to perform (e.g. recognising an object from an image). This approach can be referred to as social computation. However, systems that utilise social computation often suffer from a lack of transparency, which results in difficulties in the decision-making process (e.g. assessing reliability of outputs). This thesis investigates how the lack of transparency can be addressed by recording provenance, which includes descriptions of social computation workflows and their executions. In addition, it investigates the role of Semantic Web technologies in modelling and querying such provenance in order to support decision-making. Following analysis of several use-case scenarios, requirements for describing the provenance of a social computation are identified to provide the basis of the Social Computation Provenance model, SC-PROV. This model extends the W3C recommendation for modelling provenance on the Web (PROV) and the P-PLAN model for describing provenance of abstract workflows. To satisfy the identified provenance requirements, SC-PROV extends PROV and P-PLAN with a vocabulary for capturing social computation features such as social actors (e.g. workers and requesters), incentives (e.g. promises of monetary rewards received upon completion of a task), and conditions (e.g. constraints defining when an incentive should be awarded). The SC-PROV model is realised in an OWL ontology and used in a semantic annotation framework to capture the provenance of a simulated case study, which includes 46,665 diverse workflows. During the evaluation process, the SC-PROV vocabulary is used to construct provenance queries that support an example workflow selection metric based on trust assessments of various aspects of social computation workflows. The performance of the workflow selected by this metric is then evaluated against the performance of two control groups - one containing randomly selected workflows and the other containing workflows selected by a metric informed by provenance which lacks SCPROV descriptions. The examples described in this thesis establish the benefits of examining provenance as part of decision-making in the social computation domain, and illustrate the inability of current provenance models to fully support these processes. The evaluation of SC-PROV demonstrates its capabilities to produce provenance descriptions that extend to the social computation domain. The empirical evidence provided by the evaluation supports the conclusion that using SC-PROV enhances support for trust-based decision-making.
Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:678800 |
Date | January 2016 |
Creators | Markovic, Milan |
Publisher | University of Aberdeen |
Source Sets | Ethos UK |
Detected Language | English |
Type | Electronic Thesis or Dissertation |
Source | http://digitool.abdn.ac.uk:80/webclient/DeliveryManager?pid=228546 |
Page generated in 0.0016 seconds