GMF implementation inspired by https://github.com/hexiangnan/neural_collaborative_filtering/blob/master/GMF.py

gmf.py

regs = regs or [0, 0]
user_input = Input(shape=(1,), dtype="int32", name="user_input")
item_input = Input(shape=(1,), dtype="int32", name="item_input")

user_embeds = Embedding(
    input_dim=num_users,
    output_dim=embed_size,
    embeddings_initializer="normal",
    embeddings_regularizer=l2(regs[0]),
    input_length=1,
)
item_embeds = Embedding(
    input_dim=num_items,
    output_dim=embed_size,
    embeddings_initializer="normal",
    embeddings_regularizer=l2(regs[1]),
    input_length=1,
)
user_flattened = Flatten()(user_embeds(user_input))
item_flattened = Flatten()(item_embeds(item_input))
predict_vector = multiply([user_flattened, item_flattened])
prediction = Dense(
    1,
    activation="sigmoid",
    kernel_initializer="lecun_uniform",
    name="prediction",
)(predict_vector)
model = Model(inputs=[user_input, item_input], outputs=prediction)