The meaning of h@5_pos_x and h@5_slice_:x

[xcjames]

May I ask what the meanings of h@5_pos_x and h@5_slice_:x are?

I search the meaning (as follows) but don't know whether it is true or not, and they are not mentioned in the paper.

h@k_pos_x → HR@k measured on different prediction targets/tasks (e.g., next item, masked items, reconstruction).

Example: h@5_pos_0=0.06 means 6% of next-item predictions hit in top-5.

h@k_slice_:x → HR@k measured on specific data subsets (e.g., short sequences, long sequences, tail items).

Example: h@5_slice_:3=0.027 means only 2.7% hit rate in the 3rd subset of the dataset

[Russ-YANG]

same question

[EdoardoBotta]

these are debug metrics that looks at recall when matching only a subset of the semantic ID. For example, for the h@5_slice_:2 metric to count a prediction as "hit", it is enough that the prediction matches the first 2 tokens of the semantic id with the target example.