shenweichen / shenweichen/DeepCTR-Torch
Making DIEN dataset
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 3.5k
- Forks
- 733
- PR merge metrics
- No merged PRs in 30d
Description
Describe the question(问题描述)
As I am processing data to use DIEN model, I reckon that data should have some different format compared to DeepFM due to the user behavior sequence list if I am correct.
Because we will have sequence behavior (item user click history), I guess the dataset should be like one user one row?
For example, it would be like below.
user_id item_sequence target_ad
0 [20, 30, 22] 3
1 [11, 45, 2] 10
2 [77, 35, 64] 4
3 [20, 30, 22] 7
4 [20, 30, 22] 16
5 [20, 30, 22] 1
But in DeepFM case, we do not use user behavior sequence, so many rows can have same user ID I guess? The example of what I am saying is as below: (user 1 and 5 have multiple rows)
user_id clicked_item_id target_ad
1 5 3
1 6 10
1 5 4
5 8 7
5 11 16
9 2 1
So in general, DIEN dataset would have number of row = number of user in this case whereas DeepFM can have arbitrary number of row as long as data exists?
And as we have to put target ad according to DIEN paper, can I take out the last sequence of original item_sequence and put it as target ad? Because with sequence history, the last item sequence should be predicted if it was classification problem.
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
The issue names no files, tests, or entry points; begin by reviewing the DIEN paper and the repository's DIEN implementation to determine the required sequence and target formats. Done means reaching an agreed dataset structure and documenting or validating it against the model's expected inputs.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python, pytorch
- Domain
- machine-learning
- Issue type
- Feature
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 20/100