Update README.md

README.md

@@ -25,6 +25,10 @@ configs:
   data_files: "merged_comments.parquet"
 - config_name: grammy_videos_lyrics
   data_files: "grammy_videos_lyrics.parquet"
+- config_name: item_factors
+  data_files: "item_factors.parquet"
+- config_name: user_factors
+  data_files: "user_factors.parquet"
 ---
 
 # Preprocessed Data for the ADA Project 2025
@@ -42,3 +46,16 @@ List of Files:
 * *timeseries_grammy.parquet*: This dataset describes the Timeseries evolution of channels belonging to Grammy Authors. We obtain this by filtering the Youniverse's `df_timeseries_en` split with the unique channels that can be found in the `metadata_grammy.parquet` dataset.
 * *merged_comments.parquet*: This dataset is given by filtering the `youtube_comments.tsv.gz` based on the feature `video_id`, that must be in the `music_video_ids` set.
 * *grammy_videos_lyrics.parquet*: This dataset is given by expanding the Grammy Dataset to also include the lyrics.
+
+### Collaborative Filtering Data
+
+The following items are obtained by performing Matrix Factorization on the `merged_comments.tsv.gz` split of the dataset. 
+As a first step the Sparse matrix of shape `(users, items) is created, with each entry corresponding to either a *0* or a *1* (where 1 means that the specific user commented in that specific video).
+Afterwards, we perform Matrix Factorization using the `implicit` library.
+The following files are the results of that:
+
+* *item_factors.parquet*: Latent Space representation of every `video_id`.
+* *user_factors.parquet*: Latent Space representation of every `author`.
+* *als_model.pkl*: Weights for the trained ALS model.
+* *user_id_map.pkl*: Mapping to go from the **original** `video_id` to the specific ID used for training in the ALS model.
+* *item_id_map.pkl*: Mapping to go from the **original** `author` to the specific ID used for training in the ALS model.