Table 11 Frequently used notations
From: Improving session-based recommendation with contrastive learning
Notation | Description |
---|---|
\({\Theta }\) | Parameters in PIE-CL |
\({\mathcal {F}}\) | Activation function |
B | The batch size |
T | The sessionâ€™s length |
K | K value in objective evaluations |
L | Clip value for session truncation |
N | The size of the memory bank |
\({\mathcal {S}}/ x_i \) | Session set / i-th item |
\({\mathcal {V}}/ v_i\) | Item embedding set / i-th item |
t | The last time step of the session |
n | The total number of items in the item set \({\mathcal {V}}\) |
\(\mathbf {e}_i\) | The embedding of i-th item |
\(\mathbf {p}_i\) | The embedding of position for i-th item |
\({\mathcal {A}}\) | The affinity matrix for importance extraction |
\(\mathbf {Q}\) | The query matrix for importance extraction |
\(\mathbf {K}\) | The key matrix for importance extraction |
d | The dimensionality of hidden state |
\(\alpha _i\) | The attention weight of i-th item |
\(\mathbf {z}_l\) | The long-term preference |
\(\mathbf {z}_s\) | The short-term preference |
\(\mathbf {z}\) | The individual preference |
\({\hat{z}}_i\) | The matching score with a certain item |
\({\hat{y}}_i\) | The matching probability with a certain item |
\({\mathcal {L}}\) | The cross-entropy loss |
\({\mathcal {L}}_{CL}^i\) | The proposed contrastive learning loss for i-th record |
\(\mathbf {z}_i^+\) | The mean vector of positive session representation |
\(\mathbf {z}_i^-\) | The mean vector of negative session representation |
\(\mathbf {x}_{it}\) | The target representation for i-th record |
M | The sum of contrastive samples |
x | The simplified representation for target item |
\(x^+\) | The simplified representation for matching score |
\(x^-\) | The simplified representation for non-matching score |
\(\tau \) | The temperature parameter |
\(\lambda \) | The hyper-parameter for losses compromise |