學生對 明尼苏达大学 提供的 Recommender Systems: Evaluation and Metrics 的評價和反饋

My issues about the previous courses in this specialization seem to have been addressed in this one. The assignment in the end is a real good one. The creators of this course have done well to evolve a really thought-provoking and relevant assignment. The course itself helps one develop the appropriate thought process, which comes in handy while deciding upon a metric for a problem at hand.

This is a good course to understand the main approaches to recsys evaluation, their pros and cons. Guest interviews were useful. It would be cool to have reading lists of the articles suggested by the lecturers and guests somewhere. No partial credit for answers in the final task was a bummer, because it took away a reflection opportunity, intended by the authors. Maybe it makes sense to split the last exercise question by question in order to give an opportunity to think and reflect.

If you are new to Recommender Systems evaluation, and would like to first know why we do what we do in evaluating a recommender system, go for this course! Each and every approach is explained in vivid details, stripped to the bare essentials so you can see the skeleton of that approach! The only shortcoming, in my opinion was that i felt the codes in honours content in Lenskit could've been further explained. But, all in all, a wonderful place to start!

It is not perfect but best of specialisation so far. It is a little bit philosophical rather than technical and formal, but it was exactly meet my current personal needs. Can not be recommended as a first and only introduction to a topic of an evaluation and metrics of recommender systems.

P.S. Exercises and quizzes, both main and honour, are somewhat eccentric.

I think this is the most relevant course in the specialization. I find myselt in a situation where I need to evaluate a recommender system I developed, and the topics and material discussed throughout the course gave me many insights. Not everything in RSs is about ML!

The course presents the different metrics used for evaluating recommender systems. Moreover, they show many real-life applications where these metrics could be applied and the trade-offs of them. It also includes interviews with experts on the field.

I was working on a cross-domain recommendation system where i would recommend books to a user whose movie ratings have been given. I made the algorithm but didn't have any idea as to how to evaluate it but this course helped me through. Thanks

What an excellent course. A recommender system with a low error (RMSE, for example) does not mean to have a good recommender system because of its lag of novelty, serendipity and diversity.

Wonderful course provide realtime examples of the pros and cons of each approach and metric, very useful and enjoyable

That last assignment is great for a better understanding of the metrics.

wonderful!!! They teach a lot what I did not expect!

nice to learn excel statistic

execellent

pYTHON

Good lecturers. Pretty well designed course too. Major problem is that I couldn't get the project in the honours to build or compile - the build instructions were poor and outdated (and they should really have a git repo). Feel that would have been a good learning exercise. Would have liked more practical assignments in general.

The course is interesting because it makes you ask the right questions about recommender systems design. Overfall, there's no great theory behind recommend systems, it's mostly about understanding users' and business' needs, and lecturers do a great job to explain that!

not an easy course, specifically the honors track. the information is good, but not presented as well as in the previous two courses. Also there are errors in the honors assignment that make it unnecessarily difficult and you spend a lot of time on irrelevant things.

the part of offline evaluation is really good and practical as well. However, although knowing online evaluation is a more complex subject, I felt it lacked a little bit how to put all this knowledge in practice.

Very Informative. Still, there isn't too much complete evaluation example invovled. It would be a great help if this course can provide some breakdown/design of a recommendaer evaluation system.

Very good. But left out 1 star because one honors assignment did not have the material(base code) to download. Repeated questions were not answered in forum.

A lot of very in detail theories and metrics. I wish it could have more hands on experience.

This course was very helpful for giving me a breadth of exposure to various ways to look at evaluating recommender systems. Having faced a very similar problem evaluating a recommender system for a legal document search/suggestion engine (like Google News for lawyers), this gave me a proper "birds eye" perspective on that problem that I wish I had before. We faced exactly the same problem you describe of finding the proper tradeoff between precision and recall, or search vs. discovery.

BUT what is lacking here is teaching us how to go implement these different evaluation metrics in practice. Sadly I don't feel any more equipped to go back to that legal search engine client and guide them toward a very concrete decision about the right metrics to use. I would just come with a mix of new opinions of metrics they should consider -- but how should they choose? what offline evaluation should we do? what online experiment could we run to decide? etc. If you had run us through problem set/assignments involving real-world situations like this, where we had to calculate these different metrics (given sample data) and come up with compelling cases for different metrics to use for evaluation, I would feel otherwise.

That said thank you for your hard work putting the course/specialization together. I hope my feedback helps constructively, but don't see it as criticism. It's because I am very enthusiastic about what you've been teaching me -- and I plan to go implement it for new clients of mine in my Data Science consulting practice (www.waterwaydata.com) -- that I only want the course to be the best it can be for others too.

Loved the first part of the course where they introduced many relevant evaluation metrics (root mean square, Spearman, ROC, Precision/Recall, .etc). However, offline/online evaluations were vaguely explained and lacked depth. I really wish there were more concrete, written examples. The final quiz was abstract, weird, and difficult to understand.

The first two weeks are fantastic, up until evaluation metrics stop being covered. After that nothing concrete is said and very little is to be learned. Skip after that.

The computer assignment is lack of explanation.