Hacker News Reader

Cool work! I think peer review has a ton of potential and is certainly something grows better with scale.

My favorite course in college used a similar approach to scaling grading, described (very) briefly here: http://commons.library.upenn.edu/sites/default/files/picture...

Thanks for the comment! I looked at WHOOPPEE a while ago and it is great work. We have thought a lot about doing rankings instead of grading/assessing each piece of work individually with evaluation criteria.

On one hand, I believe that it is easier with rankings to get more numerically correct grades. Humans are bad at rating things, but binary rankings are doable.

The challenge with rankings for peer review, is two-fold. The first problem is that a ranking does not directly tell you where you are lacking or how to improve. The second problem is that quality is generally not 1-dimensional (as the ranking would be).

These problems can be solved by letting students rank papers according to multiple criteria (consequently generating multiple rankings) and then combining these rankings. The challenge with this solution is that doing multiple rankings often ends up becoming a quite time-consuming and overwhelming task. Also, from the research I have looked at about this, students generally find it hard to pick one or the other, often requesting the "I don't really know" option making the problem somewhat less trivial.

Congrats David! I was in the course last year when we were ~150 students. I really enjoyed both giving and reading the peer feedback from others. I was also surprised how there would be this dynamic of exchanging tips and tricks on what could have been done better between anonymized peers. All the best with the future of peergrade! /Jakob

Hi Jakob – thank you – we're thrilled to hear that!

We, at Peergrade, strongly believe that there's a lot to be gained if students share more of their experience (knowledge gained, questions, struggles) when engaging with the class material. Through sharing those learning experiences students will learn faster and our aspiration is that Peergrade will be the facilitator of exactly that sharing and improved learning.

Great work, I like the idea of not requiring students to have an account to login, but instead use a code. I also feel as if this is a great way to introduce the youth to the concept of "Code Review" which I believe is critical to secure coding and their future.

Thanks! We originally built Peergrade just with email/password login because we started with a focus only on higher education. When we got the first users from schools that didn't have emails, we had to rethink the login flow. Using a code has a lot of advantages since it makes it a lot easier to get started - but often students realize later (for example a week later) that they would love to have saved their feedback, which is now impossible.

I totally agree about code-review! I actually wrote a post on Medium about teaching code review with peer feedback: https://medium.com/@davidkofoedwind/teaching-code-review-in-...

This is great, congrats! How did you get your first few customers? How do you plan on monetizing? I realize Peergrade is free, but, as an educator, do you have any advice with respect to selling into K-12 vs higher-ed?

Thank you!

First customers were through personal network, we then quickly transitioned into cold emails to universities in our region.

We are already making money, primarily through sales to institutions in higher-ed. The longer term plan is continuing down that path, with institution sales making the majority of revenue, but with individual teacher plans available as well. The freemium model is to get teachers hooked, and then have their school, department, or university purchase as they see increased use and a need for integration with their existing school system (LMS). This is a path we have seen for a bunch of our current customers. Especially integrations seems to drive institutions to buy.

K-12 vs Higher-ed is a complex beast, but an analogy can be drawn to Consumer vs B2B. K-12 is high-volume, low price, low touch deals and have some similar characteristics to the consumer market (virality + simple products). Higher-ed is lower-volume, larger deals sizes, but more hands-on sales. Currently our sales and marketing efforts are focused on higher-ed because it does seem to have the best ROI.

What type of Quality Control features do you have in place to ensure that peers take the time to provide quality feedback? I have often found that incentives alone do not suffice.

Great question! There is a few different things we do. First of all, students can flag parts of the feedback for moderation by a teacher. Secondly we calculate the inter-rater agreement between reviewers on a submission - that way teachers can look at submissions with a high disagreement.

Of the slightly more complex parts, we have some algorithms that can detect biased students (similar to this https://arxiv.org/abs/0711.3964). Another simple trick is that if one reviewer says your work is good, another that it is bad and you don't flag the feedback saying your work is bad, then the review stating your work is good is probably too nice.

The final thing we are working with is making sure that students are honest when they assess feedback quality. To do this, we use some natural language processing to estimate the quality of feedback so we can see if our model is in agreement with the student that reviewed the feedback.

Basically, we do a lot of automated outlier detection and highlight outliers to teachers for manual moderation. Not all of this is in production yet (in the end, students are actually quite honest), but all of it is working in non-production environments on actual data.

I used it for one of my courses a little over a year ago; I don't recall seeing any flag, but maybe that's a newish feature.

My primary problem was that the professor often created a question or two that 9/10 groups hadn't addressed, and failed to add questions addressing the value of their work. Meaning a group may have created a product worthy of a 95+/100 score, but failed to address some poorly made question by the professor. Although you could argue that flaw lies upon the professor rather than your product :)

Another problem was that it was very easy to figure out the owner of the assignment, which basically meant that many of the students would give top score to their friends regardless of the quality of their work.

Finally, I hated the psychology that played into effect when grading the peers. Especially because I couldn't help but think that the professor would be able to associate the comments/scores to me (resulting in me giving overly positive scores to terrible assignments)..

That being said, I think what you're building is really great, and if I were a professor then I'd likely want to use it myself.

Thanks for your comment! We love feedback (part of our DNA) ;).

One of our biggest challenges is making a product that is flexible for teachers but which is hard to misuse. Bad assessment rubrics give bad student experiences, and unfortunately we can't make them for the teachers. We are in the process of writing a small booklet for teachers on how to make rubrics better.

The challenge with anonymity is especially a problem in smaller classes. We make sure to strip metadata from the submissions, so if people don't put their names in the documents, the only way to determine the author is to know the content or the writing style. It sounds like the challenge in your course was that people were working on different projects and that you knew who was working on what?

The psychology part is the most tricky. We want to make a setup where the incentives are correct, but where people don't feel scared of giving feedback to each other. The way we primarily do this is through feedback-on-feedback where receivers of feedback are asked to review the quality of the feedback they received, and then we let the teachers moderate this and students flag problems. I don't think we are completely there yet, but I can definitely see that we are making progress on solving this :).

Yeah, it was a group project with a lot of presentations, so could instantly recognize who the project belonged to just by looking at group number/easily recognizable pictures. Not sure if there were other identifiable information.. but you say peergrade doesn't display this, so I guess here it's again the fault of the professor to let students upload entire documents rather than letting them copy parts of the text..

That being said, I do seem to recall that the uploading process was somewhat simplified, meaning that we had no choice but to upload everything together, rather than splitting the uploading part into different sections. I mean, if the assignment was a research paper, then instead of uploading it all together (usually in the form of a pdf), it could be divided into different sections, e.g. introduction, related works, etc. (with only text it would be much harder to recognize the author, and if person A reviewed introduction of PersonB and related works of PersonC then it would be even more difficult to identify).. this could also go a long way towards solving the poorly made questions by the professor since it would be guaranteed that all the questions were relevant and completed (not sure if that makes sense, and I wouldn't be surprised if I remember wrong and all of this is already possible ^_^)

About feedback on feedback, it sounds good in theory, but I recall many wouldn't bother looking at it (especially because most would just leave "ok", "good", "I agree" etc as their grading comment), and those who did would just skim it without taking any actions to correct poor feedback where it was clear the person grading it hadn't bother to read the content in detail. But I of course have no idea if our behavior was the norm rather than the exception :)

In fact, one of my biggest issues with grading in general (second only to being graded based on a stupid test/presentation at the end of the semester rather than the assignments completed throughout the course) is that teachers will often be biased, and I think you have a real shot at fixing this by ensuring that even the teachers don't know the author of the assignment they grade.

You are right, at this point we only allow students to submit one piece of work. The reason is to limit confusion on the reviewers side and to ensure that proper context is kept - it is not always possible to review the conclusion without the introduction. That being said, we are looking at alternative solutions.

In relation to the feedback on feedback, it is again a question of how the teacher implements it. In my own course, the quality of feedback a student provides counts for 30% of their final grade, that makes incentives very clear.

I totally agree with teacher bias being a huge problem. We have a lot of teachers telling us that they are surprised when Peergrade highlights their own personal biases! We are building a way for teachers to also grade people anonymously to further this :).

Congrats! I remember watching your YC experience through Medium Series. The journey seems super exciting :).

Thank you! Happy to hear someone was actually reading that. The experience with using Medium Series as an author was not very pleasing to say it the least.

I've heard of a few other guys doing that... or maybe it was you already :-p.

Why are you better than the others ?

Hope it was us! I think we are better for a few reasons.

1) We are feature-wise the most complete peer feedback solution on the market. This matters because very few classes and courses are the same and teachers like flexibility where product fit to their teaching.

2) Secondly we spend a lot more time thinking about the user experience for teachers and students in our product. Getting accurate grades and having a lot of smart features is of course important, but making it easy to use and setup is more important.

3) It seems that most other peer feedback solutions are built with a narrow mindset about just scaling grading. We are building Peergrade with a focus on feedback over grades and always looking at improving the way students interact with and learn from each other.

Hope that was an answer to your question :)

Launch HN: Peergrade (YC S17) – Student to Student Feedback