Home>

Environment: MacOS 10.12.6, Python 3.7.4, Tensorflow 1.13.2

I want to implement a recommendation function often found on EC sites

After learning the deep learning RNN (including LSTM), I've come to the stage of trying it out, but the data used in the learning program was just an enumeration of numbers or a language model.

I don't know what kind of data I can prepare for deep learning, even if I try to learn it with my own data.

The ultimate goal is to make a trial function of the recommend feature on the EC site: "People who have purchased this product (and later) have also purchased this product".

Is the original data like this?
If not, add it.

user_id datetime p_id post_num
100321 2019-01-01 19:02:41 673482349878 110-0003
101957 2019-03-11 10:21:07 687468758448 778-0012
107277 2019-05-30 12:33:91 645058389846 106-0023
・ ・ ・ ・

For example, in a learning program

(x_train, y_train), (x_test, y_test) = mnist.load_data ()
print (x_train [: 1], y_train [: 1])
# result
[[3.6908443e + 00 7.1141434e-01 -2.1140914e + 00 -4.1410069e + 00
  -4.5744715e + 00 -3.4319086e + 00 -1.9507914e + 00 -1.1070668e + 00
    ・
    ・
    ・
   6.6552722e-01 1.0359968e + 00 1.4922866e + 00 -1.9050734e + 00]] [0]

is a simple number.

However, if the result I want is a product number or more, you can't find a pattern that is easy to understand even if you google it.

Can you give me some hints?

Append

I understand that this article is somehow close to what I want to do.
I have organized the steps to learn LSTM with Keras

Isn't there any problem even if the numbers 1-14 in this example are just strings?

  • Answer # 1

    I ’m not going to answer, so I ’ll close it.

Related articles