Partisan Reputation Framework

[edit] Campaign Application Model

[edit] Campaign Phases

In order to enable both bottom up grass roots campaigns and also top-down projects and formal CBPR-type collaborations, we create a framework that lets entities:

• Create
  • The campaign is created with a specific objective. Attributes associated with a campaign include its sampling modality requirement, space and time resolutions, the life-time of the campaign, and budget associated with running the campaign.
• Recruitment
  • Gather the necessary individuals to run the campaign in terms of people to collect samples [based on space, time, reputation], and classifiers to analyze data.
• Sampling
  • Trigger samples opportunistically using two-legged autonomous mobile nodes [humans carrying cell phones].
    • System listens to published locations of citizen-sensors.
***Trigger sampling based on spatial and temporal coverage needs. 
***Adjust windows, triggers via messages to achieve coverage. 
***Pass samples to distributed analysts who verify/classify
• Verification
  • Human and automatic classifiers get rid of data that is not pertinent to the campaign [noise] and contribute to updating the sampling strategy.
• Auditing
  • Users that sample data have the ability to audit data or act as auditors for others. Exploratory tools, such as ImageScape, can be used in this process.
*Analysis and Presentation
  • Pass audited samples to distributed analysts who further verify/classify data. Accept and post data to shared repositories [data can be used for multiple campaigns] and provide visualization of data [maps, galleries].

campaigns in a semi-automatic distributed manner using cell phones as platforms for data gathering with focus on ‘human in the loop’ data gathering and analysis.

[edit] Threat Models

• Creation / Presentation
  • Action
    • Initiator creates a campaign that should run using the Partisan framework and then shares the output.
  • Threats
    • Initiator creates a campaign and does not use the data for the purpose stated.
    • Initiator does not define campaign requirements properly.
    • Initiator creates campaign with poor quality output.
    • New user creates campaign campaign [no prior behavior known].
• Recruitment/Execution
  • Action 
***Users are needed for sampling based on campaign .
  • Threats
    • Participant is a new user and does not perform sampling properly.
    • Participant provides data but quality is not good.
    • Participant does not provide data.
    • Participant provides data at variables rates.
    • Participants do not respond to feedback for certain contexts.
    • Participants data does not agree with others in same context.
    • Groups of participants try to coerce the system intentionally.

[edit] Reputation System

[edit] Metrics

• Quantity - How many campaigns were completed by campaign initiators and also the number of samples that participants actually took.
• Quality - What are the quality of sensor readings from individuals. Quality of output from a campaign initiation.
• Subjective - User feedback to indicate campaign effectiveness / process.
• Coverage - Sensor readings in rare contexts are more valuable.
• Confidence - Values that agree with others in a particular context vs being different are considered more trustworthy.

[edit] User Specific Metrics

• Creation, Recruitment, and Presentation
  • Number of campaigns created
  • Number of campaigns initiated
  • Requirements vs resources obtained/used in execution
  • Requirements vs resources used in output of campaign
  • User feedback based on quality of campaign output
• Recruitment and Execution
  • Number of campaigns volunteered for
  • Number of campaigns executed
  • Quality/confidence of data provided
  • Quantity of data provided
  • Coverage provided

[edit] Ideas for Reputation Calculation

• Training earns you more reputation.
• Audits based on other users helps in determining validity of data.
• Variance of differences important to consider.
• Challenge/Response to proactively checkup on users.
• Out of campaign measurements could work as well.
• Space and time presence could be a positive indicator.
• Associations [social network] can be used to determine or give 
 weight to a person’s reputation.
• Aging and forgetting can be used for reputation where new events 
 count for more and old events age.
• Reputation can be based on phases [can you analyze to see if 
 there are local or global patterns associated].
• Reputation is dynamic - changes based on context [space, time] 
 and also resolution [street, zip, city, state, country].
• Privacy protection [aggregation, perturbation] can reduce utility.
• Balance exists between using a person too much / too little.

[edit] Incentives

• Level of Service
  • Campaign Creation - The number of campaigns and the resources a campaign can use [space, time, length, user count].
  • Access to Output - Which campaign outputs can a user access and to what extent. Varies in terms of number, time, and usage.
• Rewards
  • Point system where there is a cost for creating campaigns and benefit for contributing to campaigns [sampling, auditing]. These vary based on the utility of the data.
• Out of Band
  • Real money, free service on cell phone, etc...
  • Banned from usage.
  • Civic Duty