Mcfromer—The end to end model to predict customer-purchase-intention

  • This is an end to end deep learning method Mcformer to utilize the customer clickstream data to predict the user purchase intention.
  • We aim to utilize the customer clickstream data to predict the customer purchase intention, the scenes as follows:
  • question

The framework of Mcformer

Framework of Mcformer

  • Introduction of Mcformer
  • In order to deal with multi-dimension clickstream sequence data, we proposed an end-to-end deep learning model, named Multi-channel for purchase transformer (Mcformer), to predict the customers’ purchasing intention. Figure 1 shows the model architecture of Mcformer. This model composed by four parts: embedding layer, multi-transformer layer, cross fusion layer and output layer. Embedding layer is used to embed the sparse one-hot vectors of the behavior data to dense vectors. After that, multi-channels transformer identify intra-information of each sequence. Then the cross fusion layer is applied to identify the inter-information of different sequences. Finally, Mcformer output the result by the multilayer perceptron.


  • sklearn
  • pandas
  • pytorch
  • cuda

Run the project

”’ bash python ”’


Our data is the real world data from, this dataset need to preprocessing, which need long time. If you need the data to verify our model, you could contact with us [email protected]

  • data

modify files to ensure the code work

And if you want to use your data, you have to provide

  • the number of items: ni
  • the number of category:nc
  • the number of types: nt
  • the number of hour/minutes: nh

you need to modify the files as follows:

  • cat_pad_unk = [[0,nc+1,nc+2], [0,nh+1,nh+2]]
  • item_pad_unk =[[0,ni+1,ni+2], [0,nh+1,nh+2]]
  • type_pad_unk =[[0,nt+1,nt+2], [0,nh+1,nh+2]]

the file location

if you want to visualize the training process ,you should change parameters.

The results

  • The result show that Mcformer get great performance in long sequence classification tasks.


View Github