Crowds and Clouds
- 1. Our Data Ourselves: MobileMiner •20 Young coders from “Young Rewired State” were issued with Android smartphones. •Developed MobileMiner together, an app that records the behaviour of other apps. •Return their data at hack-days. •Discuss their attitudes to privacy before and after confronting them with their data.
- 2. OpenCellID and OSM are croud-sourced • Full GPS is too invasive, and consumes power. • Avoid use of Google location API. • OpenCellId provides locations of (many) cell towers. • http://opencellid.org
- 3. NervousNet from ETHZ: •http://www.nervousnet.ethz.ch •NervousNet hub mobile app polls various physical sensors at a user- defined rate. •Data is pushed to one or more remote “proxies”. •Outputs of sensors combined into “virtual sensors”. •Small custom deployment at CCC Congres.
- 4. Crowd Sourcing: Brand Disambiguation ● Search YouTube for “apple and blackberry”. ● Some videos will compare the firms and their prooducts. ● Many will be about pies, jam, crumbles, etc...
- 5. Crowd Sourcing: Brand Disambiguation ● Can discriminating between the two be easily automated? ● Does the word “cook” in video descriptions imply they're about pies? ● Not really! ● What about “slice of pie”? ● Better, but even that gets used as an “amusing” metaphor for market-share sometimes:
- 6. Crowd Sourcing: Brand Disambiguation ● A machine-learning approach? Natural language processing? ● http://ianozsvald.com/category/socialmediabranddisambiguator/ ● Much of machine learning is “supervised”: ● Train a model with labeled “gold-standard” examples. ● Test the model by using it to classify more labled examples. ● Use the model to classify unlabeled examples in production. ● Use crowd sourcing to get labelled training/test examples. ● http://crowdcrafting.org/project/piesorphones/
- 7. Crowd Sourcing: Alternative Platforms ● See the GATE NLP crowd sourcing tutorial: ● https://gate.ac.uk/wiki/TrainingCourseJune2016/ ● Amazon Mechanical Turk is now US only. ● CrowdFlower is available, sign-up is free. (https://www.crowdflower.com/) ● Cannot launch tasks with no payment. ● SurveyMonkey has a free trial, but surveys are not true crowd sourcing. ( https://www.surveymonkey.co.uk/) ● Proflific is geared towards research, no fee trial. (https://www.prolific.ac/) ● CrowdCrafting is entirely free, but tasks must be unpaid. (https://crowdcrafting.org)
- 8. CrowdCrafting ● Sign up and create a new project.
- 9. CrowdCrafting ● Tasks can be populated by media from cloud storage.
- 10. CrowdCrafting ● Various task templates are available.
- 11. CrowdCrafting ● Can import from YouTube playlists and Twitter searches.
- 12. CrowdCrafting ● Pastebin can be used to host.csv files easily.
- 13. CrowdCrafting ● Editing tasks can be straight-forward...
- 14. CrowdCrafting ● ...sometimes more advanced editing is needed.
- 15. A Crowd Sourcing Task... ● In groups, think of a crowd-sourcing task that would be of genuine use or interest. ● How (in)tractable would your task be to automation? ● How would you get data to populate the task? ● How would you find workers to complete the task? ● Could the task be framed such that there are non-financial incentives?
- 16. NervousNet: http://www.nervousnet.ethz.ch Our Data, Ourselves at PyData 2015: https://youtu.be/hjjniizB794 Follow us on Twitter: @KingsBSD Read our blog: http://big-social-data.net/ Slideshare: http://www.slideshare.net/kingsBSD/