Winter Milestone 5 Team Enigma

From crowdresearch
Revision as of 06:34, 14 February 2016 by Dineshdhakal (Talk | contribs) (System details)

Jump to: navigation, search

System (for task feed and open gov write up)

We're going to borrow systems section from this paper as an example: Vaish R, Wyngarden K, Chen J, et al. Twitch crowdsourcing: crowd contributions in short bursts of time. Proceedings of the 32nd annual ACM conference on Human factors in computing systems. ACM, 2014: 3645-3654. Please note how this section was divided into different parts. Please follow the same template.

Brief introduction of the system

There is a massive amount of information necessary for a healthy crowdsourcing marketplace — for example accurate reputation ratings, skill tags on tasks, and hourly wage estimates for tasks — that is privately held by individuals, but rarely shared. We introduce Boomerang, an interactive task feed for a crowdsourcing marketplace, that incentivizes accurate sharing of this information by making the information directly impact their future tasks or workers. Requesters' ratings of workers, and their skill classifications of tasks, are used to give early access to workers who that requester rates highly and who are experts in that skill, so giving a high rating to a mediocre worker dooms the requester to more mediocre work from that worker. Workers' ratings of requesters are used to rank their high-rated requesters at the top of the task feed, and their estimates of active work time are used to estimate their hourly wage on other tasks on the platform.

How is the system solving critical problems

Finding relevant tasks that pay out well and in time is a fundamental need for the workers. Meanwhile, finding skilled workers that deliver good quality results is requesters' aim. Boomerang aims to solve this problem by creating a feedback system that affect the users directly on the system in future. While requester reputation is the major criteria for workers trying to find tasks, there are other preferences and signals that workers look at while trying to find relevant tasks. One such indicator is the hourly wage equivalent that the task would pay. We extend the boomerang idea to include hourly rate as one of the signals for ranking tasks on the task feed. The system works by taking from workers as input, the time it takes them to complete a task. In addition to the workers' and requesters' ratings, we use the hourly wages as a way to bump up the tasks on to the task feed. Another signal for bumping up the task on the feed would be the worker's categorical rating as perceived by the requester. The bumping up of tasks will be valid only for requesters' that are in the same ratings zone, eg:- all requesters that have received tick + ratings only. This system increases interactions between workers and requesters that not only have good reputation but also keep posting tasks that are more aligned with requesters's skills and interests and pay fair wages.

The hourly wage model for the tasks on the feed will be built over the existing task completion times reported by the workers for submitted tasks. The incentive for workers to report accurate time is the accuracy of hourly wage calculated for other tasks, and effective ranking of the task on their feed.

Introducing modules of the system

Below, we introduce the various modules of the suggested extension to Boomerang .

Module 1: Estimated Duration of Tasks (Hourly Wage)

Problem/Limitations

The Boomerang model ensures that the problem of reputation inflation is kept in check by incentivizing users to give accurate ratings that directly affect their future activity on the crowdsourcing platform. While the overall experience of the workers with requesters might have been positive, the workers will still be benefited by a more granular preferences for the tasks showing up on their task feed. Workers find it helpful to know how much time a particular task would take before attempting that task. Moreover workers are known to track their hourly wage rates through third party extensions or scripts. The question remains how reliable are these durations reported by other workers on third party forums and platforms. There should be an incentive for the workers to report time taken to complete a task accurately.

Module preview

Currently workers depend on third party sites to find out estimated time it would take them to finish a task. These reported durations are not always accurate as there is practically no personal incentive for workers to accurately measure and report the same. We let the workers report a completion time for each task they complete. Based on these reported times, we build a model to predict the hourly wage of the individual worker for the tasks that occur on their task feed. This model takes into account the time it took for other workers to complete the same task and the time it took the same worker to complete other tasks.

System details

  • Every worker on the system will be asked to provide a input for the amount of time it took them to complete a task while submitting the same. The worker will be provided a stopwatch timer on the task screen to track the time. The timer starts once the task is accepted, and the worker can pause and resume the timer. On clicking on submit, a confirmation with recorded time is displayed to the worker. At this point the worker can either submit the recorded value or can modify the time as he/she sees fit. This also allows the workers to enter time for tasks of longer duration where they might not be on the system the entire time, or took breaks in between etc.
  • The system calculates a mean completion time X for every active task on the platform. This is the mean time taken by all the workers who have attempted this task. For new tasks, the mean time is as reported on the prototype tasks phase. If the requester did not opt for a prototype task, we would simply have to wait for the first worker to attempt this task.
  • The system also calculates, for every task submitted by the worker, a signed deviation from the mean time X. All such deviations in the same task category are then averaged out to get a single negative/positive value of a mean deviation per category. Let's denote this by D(i) for category i.
  • For every task on the worker's feed belonging to a given category i, we use D(i) in a linear regression model to predict the estimated time for the task. We use the simplest model by adding the above mean deviation D(i) to the mean completion time X for a task t. Thus, the Estimated Time for task t is given by ET(t) = X + D(i)
    • Hourly wage can simply be calculated by dividing the reward value by the Estimated Time.

Module 2: Photo Ranking

Problem/Limitations

Beyond harnessing local observations via Census, we wanted to demonstrate that twitch crowdsourcing could support traditional crowdsourcing tasks such as image ranking (e.g., Matchin [17]). Needfinding interviews and prototyping sessions with ten product design students at Stanford University indicated that product designers not only need photographs for their design mockups, but they also enjoy looking at the photographs. Twitch harnesses this interest to help rank photos and encourage contribution of new photos.

Module details

Photo Ranking crowdsources a ranking of stock photos for themes from a Creative Commons-licensed image library. The Twitch task displays two images related to a theme (e.g., Nature Panorama) per unlock and asks the user to slide to select the one they prefer (Figure 1). Pairwise ranking is considered faster and more accurate than rating [17]. The application regularly updates with new photos. Users can optionally contribute new photos to the database by taking a photo instead of rating one. Contributed photos must be relevant to the day’s photo theme, such as Nature Panorama, Soccer, or Beautiful Trash. Contributing a photo takes longer than the average Twitch task, but provides an opportunity for motivated individuals to enter the competition and get their photos rated. Like with Census, users receive instant feedback through a popup message to display how many other users agreed with their selection. We envision a web interface where all uploaded images can be browsed, downloaded and ranked. This data can also connect to computer vision research by providing high-quality images of object categories and scenes to create better classifiers.

Module 3: Structuring the Web

Problem/Limitations

Search engines no longer only return documents — they now aim to return direct answers [6,9]. However, despite massive undertakings such as the Google Knowledge Graph [36], Bing Satori [37] and Freebase [7], much of the knowledge on the web remains unstructured and unavailable for interactive applications. For example, searching for ‘Weird Al Yankovic born’ in a search engine such as Google returns a direct result ‘1959’ drawn from the knowledge base; however, searching for the equally relevant ‘Weird Al Yankovic first song’, ‘Weird Al Yankovic band members’, or ‘Weird Al Yankovic bestselling album’ returns a long string of documents but no direct answer, even though the answers are readily available on the performer’s Wikipedia page.

Module preview

To enable direct answers, we need structured data that is computer-readable. While crowdsourced undertakings such as Freebase and dbPedia have captured much structured data, they tend to only acquire high-level information and do not have enough contributors to achieve significant depth on any single entity. Likewise, while information extraction systems such as ReVerb [14] automatically draw such information from the text of the Wikipedia page, their error rates are currently too high to trust. Crowdsourcing can help such systems identify errors to improve future accuracy [18]. Therefore, we apply twitch crowdsourcing to produce both structured data for interactive applications and training data for information extraction systems.

Module details

Contributors to online efforts are drawn to goals that allow them to exhibit their unique expertise [2]. Thus, we allow users to help create structured data for topics of interest. The user can specify any topic on Wikipedia that they are interested in or want to learn about, for example HCI, the Godfather films, or their local city. To do so within a oneto-two second time limit, we draw on mixed-initiative information extraction systems (e.g., [18]) and ask users to help vet automatic extractions. When a user unlocks his or her phone, Structuring the Web displays a high-confidence extraction generated using ReVerb, and its source statement from the selected Wikipedia page (Figure 1). The user indicates with one swipe whether the extraction is correct with respect to the statement. ReVerb produces an extraction in SubjectRelationship-Object format: for example, if the source statement is “Stanford University was founded in 1885 by Leland Stanford as a memorial to their son”, ReVerb returns {Stanford University}, {was founded in}, {1885} and Twitch displays this structure. To minimize cognitive load and time requirements, the application filters only include short source sentences and uses color coding to match extractions with the source text. In Structuring the Web, the instant feedback upon accepting an extraction shows the user their progress growing a knowledge tree of verified facts (Figure 5). Rejecting an extraction instead scrolls the user down the article as far as their most recent extraction source, demonstrating the user’s progress in processing the article. In the future, we envision that search engines can utilize this data to answer a wider range of factual queries.