Word2Vec - Skipgram and CBOW

#Word2Vec #SkipGram #CBOW #DeepLearning
Word2Vec is a very popular algorithm for generating word embeddings. It preserves word relationships and is used with a lot of Deep Learning applications.
In this video we will learn about the working of word2vec and word embeddings. We will also learn about Skipgram and Continuous bag of words (CBOW ) which help in generating word2vec embeddings.
Word2Vec coupled with RNNs and CNNs are also used in building chatbots. They have lots of other use cases too.
Introduction: (0:00)
Why use word embeddings?: (0:14)
What is Word2vec?: (0:42)
Working of Word2vec?: (1:58)
CBOW and skipgram?: (2:48)
CBOW working ?: (3:36)
skip gram working ?: (5:32)

[@rma1563]

By far the best explanation of this topic. It's crazy you only took 7 minutes to explain what most people spend a lot more and still can't deliver. Thanks ❤

[@user-fy5go3rh8p]

This is the best explanation I've encountered so far. Thank you!

[@iindifferent]

Thank you. I was having a hard time understanding the concept from my uni and classes. After watching your video I went back and reread, and everything started to make more sense. Went back here watched this a second time and I think I have the hang of it now.

[@TheSemiColon]

Glad it helped!

[@sheshagirigh]

Thanks a ton. By far the best i could find after a lot of searching.. even better than few from stanford lectures!

[@tylerlozano152]

Thank you for the thorough, simple explanation.

[@subhamprasad6808]

Finally, I understood the concept of Word2Vec after watching this video. Thank you.

[@GunturBudiHerwanto]

Thank you sir! I always come back to this video when I forgot about the concept.

[@nax2kim2]

indexing for me 2:40 Word2Vec exam 3:06 CBOW 3:20 Skip Gram ----- 5:30 CBOW - working 5:50 Skip Gram - working 6:30 Getting word embeddings thx for this video :)

[@chihiroa1045]

Thank you so much! This is the most clear and organized tutorial I found on Word2Vec!

[@jiexiong8522]

Other word2vec videos are still intimidating even after a lot of graph and simplification. Your video is so friendly and helped me understand this key algorithm. Thanks!

[@fabricesimodefo8113]

Exactly what i was searching for ! so clear. Sometime you just need the neural network structure in details in graph or visually. Why don't many people do that ? Its the simplest way to understand what is happening in real in the code after

[@TheSemiColon]

This is what I needed when I was creating it, but did not find it anywhere :)

[@jusjosef]

Very simple, to the point explanation. Beautiful!

[@maqboolurrahimkhan]

Best and easy explanation of word2vec over the internet. Keep up the good work Thanks a ton

[@carlrobinson2926]

very nice explanation, not too long, straight to the point. thanks

[@bryancamilo5139]

Thank you, your explanation is great. Now I have understood the concept 😁

[@ajinkyajoshi2308]

Very well done!! Precise and to the point explanation!!

[@rainoorosmansaputratampubo2213]

Thank you so much. with this explanation I can understand it easier than read from books

[@skipintro9988]

Thanks, bro - this one is the easiest and simplest and quickest explanation on word2vec

[@Amf313]

Best explanation I saw through Internet to illustrate how Word2Vec works. Paper was a little bit hard to read; Andrew Ng's explanation was somewhat incomplete or at least ambigious to me, but your video made it clear. Thank you🙏

[@satyarajadasara9000]

Very nice video where everything was to the point! Keep posting such wonderful content!

[@pushkarmandot4426]

The best video. Explained the whole concept in a very short amount of time

[@absoluteanagha]

Love this! Such a great explanation!

[@bloodzitup]

Thanks, my lecturer had this video in his references for learning word2vec

[@varunjindal1520]

This is indeed very good video. To the point and covers what I needed to know. Thank you.

[@TheSemiColon]

Glad you found it useful, do share the word 🙂

[@MrStudent1978]

Absolutely beautiful explanation!! Very precise and very much informative....Thanks for your kindness. Sharing one's learning is the best thing that a person can do to contribute to the society. Lots of respects from Punjab India....

[@TheSemiColon]

Glad it was helpful!

[@ankursri21]

Thank you.. very well explained in shorter time.

[@OorakanaGleb]

Awesome explanation. Thanks!

[@MehdiMirzapour]

Thanks. It is really a brilliant explanation!

[@FTLC]

Thank you so much is was so confused before watching this video ,now its clear to me

[@coolbowties394]

Thanks so much for this thorough explanation!

[@TheSemiColon]

Glad it was helpful!

[@befesa1]

Thank you! Really good explanation:)

[@anujlahoty8022]

Simple and eloquent explanation.

[@mohajeramir]

this is the best explanation I have found. thank you

[@TheSemiColon]

Glad you found it useful, do share the word 🙂

[@MARTIN-101]

this was such an informative lecture, thank you.

[@nithin5238]

Very clear explanation man.. you deserve slow claps

[@aravindaraman8667]

Amazing explanation! Thanks a lot

[@HY-nt8nk]

Good work! Nicely explained.

[@theunknown2090]

Hey in cobw and skip gram Method there are 3 Weight metrics Which metric is selected as d embedding matrix ? And why

[@Zinghere]

Great explanation!

[@alialsaffar6090]

This was enlightening. Thank you!

[@mohajeramir]

this was excellent. Thank you

[@TheSemiColon]

Glad it was helpful!

[@AdityaPatilR]

If hope can set us free hope can set you free as well !! thank you for the explanation and following what you preach ;)

[@johncompassion9054]

4:50 "5X3 input matrix is shared by the context words". what do you mean by input matrix? Do you mean the weight matrix between the hidden layer (embedding) and the output layer? 5:18 "You take the weight matrix and it becomes the set of vectors". We have two weight matrices so which one? Also, I guess our vector embedding is the middle layer output values not weights. Correct me if I am wrong. Thank you.

[@prathimads2876]

Thank you so much Sir...

[@juanpablo87t]

Great Video, thank you! It is very clear how to extract the word embeddings in skip gram by multipliying the W matrix with the one hot vector of the corresponding word, however I can't figure how to extract them from the CBOW model as there are multiple W matrixes, could you give me a hint or a maybe a resource where this is explained?

[@ogsconnect1312]

I cannot say anything but excellent. Thank you

[@impracticaldev]

You earned a subsciption. Good luck!

[@sunjitrana374]

Nice explanation, Thanks for that!!! One question: How to decide optimal length of hidden layer? here in example its 3 and in general you said it's around 300.

[@renessadesouza5601]

Thank you so much

[@ashwinrameshbabu2418]

At time 5.28, cbow , hope gives 1x3 and set gives 1x3 dimension output. How are they combined into 1 (1x3) before sending to final layer?

[@md.prantohasan9630]

Excellent explanation in a very short time. Take

[@sadeenmahbubmobin7102]

reading material ta bujhay de amre akhn :3

[@haorao2464]

Thanks so much!

[@hardikajmani5088]

Very well explained

[@pranabsarkar]

Thanks a lot!

[@DangNguyen-xx3zi]

Appreciate the work put into this video, thank you!

[@TheSemiColon]

Glad it was helpful!

[@keno2055]

Why does the hidden layer at 4:59 have 3 nodes if we only care about the 2 adjacent nodes?

[@gouripeddivenkataasrithbha5148]

Truly the best resource on word2vec by far. I have only one doubt. What do you mean by size of a vector being three. Other than this, I was able to understand everything.

[@TheSemiColon]

the size of final vector for each word is the size of word vector.

[@aliqais4896]

thank you very much

[@parthpatel3900]

Wonderful video

[@Hellow_._]

how can we give all input vectors in one go to train the model?

[@tumul1474]

awesome !!

[@anindyavedant801]

I had a doubt, shouldn't the first weight matrix with which the input is multiplied be of dimensions 5x3 as all the connections need to be mapped to the hidden layer matrix and we have 5 inputs and 3 nodes in the hidden layer so the weights would be 5x3 and the second one would be vice versa i.e. 3x5

[@muhammedhassen4354]

easy way explanation gr8

[@himanshusrihsk4302]

Really very useful

[@iliasp4275]

thank you , The Semicolon.

[@057ahmadhilmand6]

i still dont get it, the word vector for each word is a matriks?

[@hashinitheldeniya1347]

can we cluster word phrases into groups using this word2vec technique?

[@hs_harsh]

Sir can you provide the link of slides used. That would be helpful. I'm a student at IIT Delhi and I have to deliver a similar lecture presentation. Thank you!

[@Simply-Charm]

Thank you

[@ms10596]

So helpful

[@hadrianarodriguez6666]

Thanks for the explanation! If I want to work with terms of two tokens, how can I do it?

[@TheSemiColon]

you may want to append them may be ?

[@gauharahmad2643]

Sir what do we mean by size of each vector in 4:37 ?

[@MultiAkshay009]

great work! 😍I am really thankful to you. But still I have a doubt with implementation part. 1) How to train the models for new datasets? 2) How to use both approaches differently CBOW and Skip-gram for training of the models? I badly need help with this. :(

[@TheSemiColon]

Thanks a lot. If you are implanting it from scratch then you have to encode each word of your database as a one hot vector train it using anyone of the algorithm skipgram or cbow and then pull out it's weights. Then multiply the weights with the one hot vector. The tensor flow official blog has a very nice example for it. You may use libraries like gensim to do it for you.

[@romanm7530]

Диктор просто огонь!

[@shikharkesarwani9051]

The weight matrix should be 5x3 (input to hidden) and 3x5 (hidden to output) @The Semicolon

[@Agrover112]

Wx+b hota hai

[@fahdciwan8709]

what is the purpose of multiplying the 3*5 Weight Matrix with the one-hot vector of the word? How does it improve the embeddings?

[@SameerKhan-ht4mx]

Basically the weight matrix is the word embedding

[@BrunoCPunto]

Awesome

[@naveenkinnal5413]

Just one question. So the final word vector size is the same as sliding window size?

[@TheSemiColon]

No, sliding window can be of any size.

[@nazrulhassan6310]

fabulous explanation but I need to do some more digging

[@arnav3674]

Good !

[@Mr.AIFella]

The matrices multiplication not correct. I think it should be 5x1 1x3 to be equal 5x3 to be multiplied by 3x1 to equal 5x1. Right?

[@TheEducationWorldUS]

nice explanation

[@randomforrest9251]

nice slides!

[@josephselwan1652]

it took me 10 times to understand it. but i finally did. lol what we do to get a job haha

[@qaisgafer3562]

Great

[@prajitvaghmaria3669]

Any idea how to create a deep learning chatbot with keras and tensorflow for WhatsApp platform using python from scratch ?

[@KARIVENKATARAMPHD]

nice

[@jatinsharma782]

Very Helpful 👍

[@qingyangluo7085]

how to get the word embedding vector using CBOW? what neighbour words do i plug in?

[@TheSemiColon]

You have to iterate over a corpus. Popular ones are Wikipedia, google news etc.

[@qingyangluo7085]

@@TheSemiColon Say I want to get the embedding vector of the word "love", this vector depends on what context/neighor words I plug in.

[@imanbio]

Plz fix the matrix sizes (3x5 should be 5x3 and vice versa..) - nice presentation

[@mohitagarwal437]

Best bhai aapne pura data science kar rakha hai kya ?

[@dhruvagarwal4477]

What is the meaning of vector size?

[@tobiascornille]

Which matrix is the embedding matrix in CBOW? W or W' ?

[@TheSemiColon]

it's W.

[@vionagetricahyo1268]

hey can you share this code ?

[@theacid1]

Thank you. My prof is unable to explain it.

[@AryanKhandal7399]

sir aswesome

[@_skeptik]

i didn't fully catch the difference between cbow and skipgram in this explanation

[@saikiran-mi3jc]

No much content in the channel to subscribe(i mean to say no playlist on nlp or cv ) .I came hear with lot of hopes. Content in the video is good.

[@fabricesimodefo8113]

typo 5:25 the input words should change to "set" and "free"

[@shazadkhanfu]

charan kaha hai aapke?