LFC-BERT Sementic-relative-distance

[yangheng95]

LFC-BERT Sementic-relative-distance

[yangheng95]

However, the difference could be ignored to some extent.

[yangheng95]

I am sorry for this puzzle, but I am somehow didnt catch your meaning. please refer to the latest implementation of CDW vec cacculation at

PyABSA/pyabsa/module/apc/dataset_utils/apc_utils.py

Line 291 in f16de40

local_context_begin = max(0, aspect_begin - SRD)

, it is clear to understand and accurate.

[Pager07]

The new implementation of CDW vec on the apc_utils file is very clear thanks.
The link here references your code, (presumably your old code).

I notice the CDW vec generated using this implementation here is different from CDW vec from this implementation here.

Could you tell me which one is the correct?

[yangheng95]

Both codes of the old version are not very accurate, because sometimes the aspect location calculated is wrong (e.g., some word/token is the same as the beginning of an aspect will be regarded as the aspect and calculate position, which has been solved in PyABSA). To answer your question, |position_of_token - average_position_of_aspect| +(average_length_of_aspect /2) seems to be more reasonable but not completely right.

[ningmiaokai]

The new implementation of CDW vec on the apc_utils file is very clear thanks. The link here references your code, (presumably your old code).

I notice the CDW vec generated using this implementation here is different from CDW vec from this implementation here.

Could you tell me which one is the correct?

Hello, could you tell me what's the meaning of opt in here.
Thank you!