Ask HN: Is there a realtime mechanical turk service? | Hacker News Reader

Ask HN: Is there a realtime mechanical turk service?

5 pointshackerews5 comments

I'm looking to get Yes/No categorization on images in realtime via a mechanical turk-like service. For instance - "is there a cat in this photo?". Unfortunately most services I've seen take minutes/hours/days for a response w/ a person, or don't get the categorization right for my data w/ an algorithm.

View on HN

Mturk sometimes has a bit of latency from issued HITs to cascade through the system, but you can issue anticipatory tasks to counteract this latency.

The more $$ you offer, the less time it will take someone to take it on.

Combining these two concepts, bundle dozens of tasks into a single HIT, and score based on completion and promptness. You're going to have to build your own rating software, but this is very possible to do. Just be up front that promptness counts; most of the time it doesn't.

I disagree, at least for certain tasks. For example, I had a data entry task that I wanted done in near real-time. It basically amounted to "Type in the 5 lines of text in this document." I ran some tests on MTurk and found zero relationship between amount of money offered and the speed with which people accepted the job.

(disclosure: I work at StintMint) Real-time is possible assuming low complexity of requirements and consistent request processing rate. StintMint offers a managed crowdsourcing application to businesses in Commerce and AI Tech. Do visit our website to get in touch. Would be happy to take over from there.

(Disclosure - I work for SquadRun) Near real time is doable if we have a sense of your volumes and your data flow. We work with Uber, Sephora, Teespring, etc. for their time-sensitive data operations. Happy to chat to understand more. Just drop us a line on our website.

(disclosure: I work for Alegion) Near realtime is doable. Standard approach via MTurk requires processing that takes additional time. At Alegion, we use techniques to improve upon the standard approach. Would love the opportunity to discuss your project with you

Reply on news.ycombinator.com