vanzytay / kdd2018_mpcn Goto Github PK

For example, a sample (user reviews, item reviews, rating) represented the input for model MPCN obviously exists a review wrote by such user for such item. As TransNets mentioned, that review should not be present in development and test. I also saw that "We would like to emphasize that, when building user and item representations using their respective reviews, all reviews belonging to interactions from the test and development sets were not included." in section 4.1 of your paper. But I didn't find out how to deal with the above problem in your code. Could you give me some enlightenment?

I apologize for disturbing you if it's a problem too easy.

Hi,
I've read a lot of your papers and some of them are excellent ideas.
I would like to ask whether the results reported in your paper are the results under seed 1337 or the average of the results of other multiple seeds? I noticed that a lot of your projects use seed 1337.

I don't know what this preprocess for file, please give a hand.

Hi,
I am working on the Amazon data set. Can you share the data prep of the Amazon? It will be very helpful to me.
Thank you very much.

Hi! Thanks for share your work. when I run prep_amazon2.py, it don't work.
NameError: name 'build_word_index' is not defined
Can you share this file.
Thank you!

I read your paper Reasoning_ with Sarcasm by Reading In-between in ACL2018. I deal with the dataset Reddit ,and I can't get the similar results.I hope that you can share me the source code inlcuding the dataset.Thank you very much!

Hi!

Thanks for sharing your great work!

Can you share your running instructions and code for data prep? Would be extremely helpful!

Thanks so much :)

Hi,
Very brilliant job! I tried to understand your code, but I have some problems.

    embed = tf.reshape(embed, [-1, smax, _dims])
    lengths = tf.reshape(lengths, [-1])
    return embed, lengths

I think the code here is different from the comments.
In the code ,the embed should be [bsz x (num_docs * seq_len) x dim] => [(bsz * num_docs) x seq_len x dim] and lengths tensor also seems to have a problem. You mentioned in the article that it is not a combination of comments into documents, but a hierarchical structure, but I found that your lengths tensor does not seem to be used.

Thanks !

Hi, thanks for sharing your implementation.
I tried to reuse your code on Amazon datasets, but I didn't get the performance as expected.
I tried to use Glove pretrained embedding at first and then randomly initialize the embeddings.
For the max length of each review, I firstly used the number covering 90% lengths but later set the maximum length as 50.
I doubt whether I made some mistakes in parameter settings.

Hi!
Thx for you sharing your code.
Can you share running instruction of yelp?
Thx.

Hi! Thanks for share your work In Github. when I run prep_amazon2.py and train.py, it don't work.
NameError: name 'build_word_index' and 'tylib_tokenize' are not defined
NameError: name name 'get_rec_config'is not defined' are not defined
Can you share theses files.
Thank you!