NAIS

Reference:

Xiangnan He et al. “NAIS: Neural Attentive Item Similarity Model for Recommendation.” in TKDE 2018.

Reference code:

https://github.com/AaronHeee/Neural-Attentive-Item-Similarity-Model

class recbole.model.general_recommender.nais.NAIS(config, dataset)

Bases: GeneralRecommender

NAIS is an attention network, which is capable of distinguishing which historical items in a user profile are more important for a prediction. We just implement the model following the original author with a pointwise training mode.

Note

instead of forming a minibatch as all training instances of a randomly sampled user which is mentioned in the original paper, we still train the model by a randomly sampled interactions.

attention_mlp(inter, target)

layers of attention which support prod and concat

Parameters:

• inter (torch.Tensor) – the embedding of history items

• target (torch.Tensor) – the embedding of target items

Returns:

the result of attention

Return type:

torch.Tensor

calculate_loss(interaction)

Calculate the training loss for a batch data.

Parameters:

interaction (Interaction) – Interaction class of the batch.

Returns:

Training loss, shape: []

Return type:

torch.Tensor

forward(user, item)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

full_sort_predict(interaction)

full sort prediction function. Given users, calculate the scores between users and all candidate items.

Parameters:

interaction (Interaction) – Interaction class of the batch.

Returns:

Predicted scores for given users and all candidate items, shape: [n_batch_users * n_candidate_items]

Return type:

torch.Tensor

get_history_info(dataset)

get the user history interaction information

Parameters:

dataset (DataSet) – train dataset

Returns:

(history_item_matrix, history_lens, mask_mat)

Return type:

tuple

input_type = 1

inter_forward(user, item)

forward the model by interaction

mask_softmax(similarity, logits, bias, item_num, batch_mask_mat)

softmax the unmasked user history items and get the final output

Parameters:

• similarity (torch.Tensor) – the similarity between the history items and target items

• logits (torch.Tensor) – the initial weights of the history items

• item_num (torch.Tensor) – user history interaction lengths

• bias (torch.Tensor) – bias

• batch_mask_mat (torch.Tensor) – the mask of user history interactions

Returns:

final output

Return type:

torch.Tensor

predict(interaction)

Predict the scores between users and items.

Parameters:

interaction (Interaction) – Interaction class of the batch.

Returns:

Predicted scores for given users and items, shape: [batch_size]

Return type:

torch.Tensor

reg_loss()

calculate the reg loss for embedding layers and mlp layers

Returns:

reg loss

Return type:

torch.Tensor

softmax(similarity, logits, item_num, bias)

softmax the user history features and get the final output

Parameters:

• similarity (torch.Tensor) – the similarity between the history items and target items

• logits (torch.Tensor) – the initial weights of the history items

• item_num (torch.Tensor) – user history interaction lengths

• bias (torch.Tensor) – bias

Returns:

final output

Return type:

torch.Tensor

training: bool

user_forward(user_input, item_num, repeats=None, pred_slc=None)

forward the model by user

Parameters:

• user_input (torch.Tensor) – user input tensor

• item_num (torch.Tensor) – user history interaction lens

• repeats (int, optional) – the number of items to be evaluated

• pred_slc (torch.Tensor, optional) – continuous index which controls the current evaluation items, if pred_slc is None, it will evaluate all items

Returns:

result

Return type:

torch.Tensor