npy file containing embeddings, dictionary, and reverse dictionary:
train_skipgram.py - training using Skip-gram, using both purchase and play actions into account as user context.Įach script outputs an image with the game embeddings visualised using t-SNE, and a. train_cbow_weighted.py - same as above, but, only play actions are taken into consideration, and the label is selected based on time played (more time played the game - higher the probability of being selected). 
train_cbow.py - training using CBOW, using both purchase and play actions into account as user context. Explained: Deriving Mikolov et al.’s Negative-Sampling Word-Embedding Method. 
Distributed Representations of Words and Phrases and their Compositionality. Skip-gram: (Rocket League -> (Dota 2, CS: GO)), (CS: GO -> (Dota 2, Rocket League)), (Dota 2 -> (CS: GO, Rocket League))įor more reference, please have a look at this papers:. CBOW: ((Dota 2, CS: GO) -> Rocket League), ((Dota 2, Rocket League) -> CS: GO), ((CS: GO, Rocket League) -> Dota 2). For example if a user has three games: Dota 2, CS: GO, and Rocket League, this (input -> label) pairs can be generated: TensorFlow implementation of word2vec applied on dataset, using both CBOW and Skip-gram.Ĭontext for each game is extracted from the other games that the user owns.
0 kommentar(er)
