Initialization of Type Words

[anushkasw]

I had a question regarding the initialization of type words. According to the code: if self.args.init_type_words: so_word = [a[0] for a in self.tokenizer(["[obj]","[sub]"], add_special_tokens=False)['input_ids']] meaning_word = [a[0] for a in self.tokenizer(["person","organization", "location", "date", "country"], add_special_tokens=False)['input_ids']]

The meaning words are initialized with certain entity types. While these are the probable entity types for the TACRED dataset, the same is not true for the SemEval dataset.

I wanted to know how this initialization affects the working of the algorithm on other datasets like SemEval. Should we change this initialization based on the dataset?

[njcx-ai]

Thanks for your attention. The choice of initialization has some influence on the final model performance, but the impact is not significant. To establish a stronger correspondence between the type words and the specific task at hand, it is necessary to adapt the initialization based on the characteristics of the dataset.

[anushkasw]

Got it. So, did you use a different set of initialization while training the semeval dataset or the same ones?

[njcx-ai]

The same one.

[zxlzr]

hi buddy, do you have any further questions?

[yccckid]

if I change a dataset, Should I change the entity type, i.e., ["person","organization", "location", "date", "country"] into my own entiy type? Is this better than not change?