Τι είναι το SASRec και ποια είναι η αρχιτεκτονική του;

SASRec (Self-Attentive Sequential Recommendation) is a model for recommendation systems that uses the self-attention mechanism from transformers to predict the next element in a sequence of user actions (e.g., purchases or views).

The architecture of SASRec is based on:

• Input: a sequence of user interactions with items, encoded into a vector representation.
• Positional embeddings to account for the order of items in the sequence.
• Multiple layers of self-attention that allow the model to consider dependencies between items at different distances.
• Feed-forward layers after self-attention for nonlinear transformations.
• Output: probabilities for the next item the user might select.

The architecture roughly looks like this:

Input sequence -> Embedding + Positional Encoding ->
Multi-head Self-Attention -> Feed Forward ->
Output logits for next item prediction

SASRec works well with long sequences and dynamically highlights important items in the user's history without using recurrent networks.