Great!

great!

thanks

thanks

Thanks

Great!

Thanks

Cool!!

bagus

fghrt

.....

grate

Great

Nice!

GOOD

good

nice

Good

good

good

5*

no

bk

Having previously completed the Deep Learning Specialization, I came to this course with the intention of completing the whole NLP specialization, rather than because I was especially interested in the content of this first course from that specialization.

The Deep Learning Specialization sets a high standard of teaching quality and I have to say I found this course is not quite to the same standard. It's pretty good but not as good. The instructors are very knowledgeable, they make the effort to explain each topic clearly and they do a pretty good job of that.

What I felt could be improved is providing context of where each topic fits into the broader picture of both the theory and current practice of NLP. I was often left feeling, why are we spending time on this particular topic? Is this technique used in current practice or is it just of didactic or historical interest? Great teachers always have the broader context in mind and make sure that students see how everything fits into the bigger picture and why it is worth studying.

Although techniques were clearly explained, I felt that the underlying concepts were sometimes less well explained. An example is vector representations of words: we were shown the use of vector arithmetic to find analogies, but without much in the way of explanation of how this is possible. To me, this was the wrong way around: it makes more sense to me to first build an understanding of the representations, then introduce the remarkable result that these representations allow finding analogies.

In this course, sentences are represented as a "bag of words". This is processing natural language in the way a food processor processes food: chopping it up into a word soup. Since one of the most fundamental aspects of language is its structure, this might seem a hopeless approach. However it gives surprisingly good results for some simple tasks such as classifying tweets as having positive or negative sentiment. If you've done course 5 of the Deep Learning Specialization (Sequence Models), this will feel like a step backwards. There's no deep learning in this course. But I signed up for the course knowing that, so I can't criticise it on that basis. I'm taking the view that this course lays the foundations for more advanced and current topics in the subsequent courses in the specialization and I look forward to getting onto those.

The labs and assignments generally work smoothly. There are a few inconsistencies and a couple of the hints were a bit misleading but generally OK. It's a bit paint-by-numbers though, filling in bits of code within functions rather than working out for yourself how to structure the code.