Update README.md
Browse files
README.md
CHANGED
@@ -1,3 +1,80 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
license: apache-2.0
|
3 |
---
|
|
|
1 |
+
|
2 |
+
# FINN.no Slate Dataset for Recommender Systems
|
3 |
+
> Data and helper functions for FINN.no slate dataset containing both viewed items and clicks from the FINN.no second hand marketplace.
|
4 |
+
|
5 |
+
Note: The dataset is originally hosted at https://github.com/finn-no/recsys_slates_dataset and this is a copy of the readme until this repo is properly created "huggingface-style".
|
6 |
+
|
7 |
+
We release the *FINN.no slate dataset* to improve recommender systems research.
|
8 |
+
The dataset includes both search and recommendation interactions between users and the platform over a 30 day period.
|
9 |
+
The dataset has logged both exposures and clicks, *including interactions where the user did not click on any of the items in the slate*.
|
10 |
+
To our knowledge there exists no such large-scale dataset, and we hope this contribution can help researchers constructing improved models and improve offline evaluation metrics.
|
11 |
+
|
12 |
+
![A visualization of a presented slate to the user on the frontpage of FINN.no](finn-frontpage.png)
|
13 |
+
|
14 |
+
For each user u and interaction step t we recorded all items in the visible slate ![equ](https://latex.codecogs.com/gif.latex?a_t^u(s_t^u) ) (up to the scroll length ![equ](https://latex.codecogs.com/gif.latex?s_t^u)), and the user's click response ![equ](https://latex.codecogs.com/gif.latex?c_t^u).
|
15 |
+
The dataset consists of 37.4 million interactions, |U| ≈ 2.3) million users and |I| ≈ 1.3 million items that belong to one of G = 290 item groups. For a detailed description of the data please see the [paper](https://arxiv.org/abs/2104.15046).
|
16 |
+
|
17 |
+
![A visualization of a presented slate to the user on the frontpage of FINN.no](interaction_illustration.png)
|
18 |
+
|
19 |
+
FINN.no is the leading marketplace in the Norwegian classifieds market and provides users with a platform to buy and sell general merchandise, cars, real estate, as well as house rentals and job offerings.
|
20 |
+
For questions, email [email protected] or file an issue.
|
21 |
+
|
22 |
+
## Install
|
23 |
+
|
24 |
+
`pip install recsys_slates_dataset`
|
25 |
+
|
26 |
+
## How to use
|
27 |
+
|
28 |
+
To download the generic numpy data files:
|
29 |
+
|
30 |
+
```
|
31 |
+
from recsys_slates_dataset import data_helper
|
32 |
+
data_helper.download_data_files(data_dir="data")
|
33 |
+
```
|
34 |
+
|
35 |
+
Download and prepare data into ready-to-use PyTorch dataloaders:
|
36 |
+
|
37 |
+
``` python
|
38 |
+
from recsys_slates_dataset import dataset_torch
|
39 |
+
ind2val, itemattr, dataloaders = dataset_torch.load_dataloaders(data_dir="data")
|
40 |
+
```
|
41 |
+
|
42 |
+
## Organization
|
43 |
+
The repository is organized as follows:
|
44 |
+
- The dataset is placed in `data/` and stored using git-lfs. We also provide an automatic download function in the pip package (preferred usage).
|
45 |
+
- The code open sourced from the article ["Dynamic Slate Recommendation with Gated Recurrent Units and Thompson Sampling"](https://arxiv.org/abs/2104.15046) is found in (`code_eide_et_al21/`). However, we are in the process of making the data more generally available which makes the code incompatible with the current (newer) version of the data. Please use [the v1.0 release of the repository](https://github.com/finn-no/recsys-slates-dataset/tree/v1.0) for a compatible version of the code and dataset.
|
46 |
+
|
47 |
+
## Quickstart dataset [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/finn-no/recsys-slates-dataset/blob/main/examples/quickstart-finn-recsys-slate-data.ipynb)
|
48 |
+
We provide a quickstart Jupyter notebook that runs on Google Colab (quickstart-finn-recsys-slate-data.ipynb) which includes all necessary steps above.
|
49 |
+
It gives a quick introduction to how to use the dataset.
|
50 |
+
|
51 |
+
## Example training scripts
|
52 |
+
We provide an example training jupyter notebook that implements a matrix factorization model with categorical loss that can be found in `examples/`.
|
53 |
+
It is also runnable using Google Colab: [![matrix_factorization.ipynb](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/finn-no/recsys-slates-dataset/blob/main/examples/matrix_factorization.ipynb)
|
54 |
+
There is ongoing work in progress to build additional examples and use them as benchmarks for the dataset.
|
55 |
+
|
56 |
+
### Dataset files
|
57 |
+
The dataset `data.npz` contains the following fields:
|
58 |
+
- userId: The unique identifier of the user.
|
59 |
+
- click: The items the user clicked on in each of the 20 presented slates.
|
60 |
+
- click_idx: The index the clicked item was on in each of the 20 presented slates.
|
61 |
+
- slate_lengths: The length of the 20 presented slates.
|
62 |
+
- slate: All the items in each of the 20 presented slates.
|
63 |
+
- interaction_type: The recommendation slate can be the result of a search query (1), a recommendation (2) or can be undefined (0).
|
64 |
+
|
65 |
+
The dataset `itemattr.npz` contains the categories ranging from 0 to 290. Corresponding with the 290 unique groups that the items belong to. These 290 unique groups are constructed using a combination of categorical information and the geographical location.
|
66 |
+
|
67 |
+
The dataset `ind2val.json` contains the mapping between the indices and the values of the categories (e.g. `"287": "JOB, Rogaland"`) and interaction types (e.g. `"1": "search"`).
|
68 |
+
## Citations
|
69 |
+
This repository accompanies the paper ["Dynamic Slate Recommendation with Gated Recurrent Units and Thompson Sampling"](https://arxiv.org/abs/2104.15046) by Simen Eide, David S. Leslie and Arnoldo Frigessi.
|
70 |
+
The article is under review, and the preprint can be obtained [here](https://arxiv.org/abs/2104.15046).
|
71 |
+
|
72 |
+
If you use either the code, data or paper, please consider citing the paper.
|
73 |
+
|
74 |
+
```
|
75 |
+
Eide, S., Leslie, D.S. & Frigessi, A. Dynamic slate recommendation with gated recurrent units and Thompson sampling. Data Min Knowl Disc (2022). https://doi.org/10.1007/s10618-022-00849-w
|
76 |
+
```
|
77 |
+
|
78 |
---
|
79 |
license: apache-2.0
|
80 |
---
|