item-Graph2vec

a Fast and Efficient Method for Item Embedding from the Item Co-occurrence Graph, with Applications to Collaborative Filtering

Usage

###Anime dataset

1. Compile fasttext under root of the project

2. Download Anime dataset: https://www.kaggle.com/datasets/hernan4444/anime-recommendation-database-2020

   Data initialization

   $ python item2vec-input.py Anime && python Graph-item2vec-input.py Anime
   

3. Train model by running fasttext and pecanpy.

   $ ./fastText/fasttext skipgram -input ./item2vec-Anime-input.txt -output ./models/item2vec-Anime-model -minCount 5 -epoch 50 -neg 100
   

   $ pecanpy --input Graph-item2vec-Anime-input.txt --output ./models/Graph-item2vec-Anime-model.emb --mode SparseOTF --verbose --weighted --directed --p 1 --q 0.001 --walk-length 200 --num-walks 40 --dimensions 100
   

4. Anime accuracy testing

   $ cd ./Anime
   $ python genre.py 
   $ python result.py --top 1 -- num 10
   

###Movielens dataset

1. Download movielens dataset: https://grouplens.org/datasets/movielens/25m/ Data initialization

   $ python item2vec-input.py Movielens && python Graph-item2vec-input.py Movielens
   

2. Train model by running fasttext and pecanpy.

   $ ./fastText/fasttext skipgram -input ./item2vec-Movielens-input.txt -output ./models/item2vec-Movielens-model -minCount 5 -epoch 50 -neg 100
   

   $ pecanpy --input Graph-item2vec-Movielens-input.txt --output ./models/Graph-item2vec-Movielens-model.emb --mode SparseOTF --verbose --weighted --directed --p 1 --q 0.001 --walk-length 200 --num-walks 40 --dimensions 100
   

3. Movielens accuracy testing

   $ cd ./movielens
   $ python result.py --top 1 -- num 10
   

###Douban dataset

The operation is similar to the previous two data sets，specific can refer to: https://github.com/lujiaying/MovieTaster-Open

###t-SNE

After performing Douban data accuracy test

```
$ cd ./douban
$ python t-SNE.py 
```

###artificial dataset

```
$ cd ./artificial dataset
$ python deal.py --m 10000 --n 10000 --den 0.001
$ python change.py 
```

Then, the model is still generated using fasttext and pecanpy，refer to above.

##Related environment configuration

This code base is implemented in Python 3.6 and up