Occasion Aware Recommendation

This is the reading summary of paper Time to Shop for Valentine’s Day: Shopping Occasions and Sequential Recommendation in E-commerce.

You can find the corresponding code on Github Occasion-Aware-Recommenation

Overall model

$$ \mathbf{U} = { u_1, u_2, …, u_n} \ \mathbf{P} = { p_1, p_2, …, p_C}
\mathbf{T} = { t_1, t_2, …, t_M} \

\mathbf{H^{u_1}} = {(p_1^{u_1}, t_1^{u_1}), (p_2^{u_1}, t_2^{u_1}),…(p_{|H^{u_1}|}^{u_1}, t_{|H^{u_1}|}^{u_1})} $$

\[\text{query } \mathbf{q} \text{ and key-value pairs } \mathbf{P} = \{(K_l, v_l)|l \in [1,L]\} \\ \text{output:} o = \sum_{l=1}^{l}{\alpha_{ql} v_l}, \text{where } \alpha_{ql} = \frac{exp(s(q,k_l))}{\sum_{l=1}^{L}{exp(s(q,k_l))}}\]

Intrinsic preference modeling

• For user u, stack his/her items into a single array in time order \(\mathbf{P^u} = (p_1^u, p_2^u, ..., p_{|P^u|}^u)\\\)

• each item has itself’s embedding and its position embedding, item bought at the same time should have sam position embedding