GliNER marking pronouns as PERSON

[manojkumarvohra]

GliNER is tagging pronouns like I, you, we, they as PERSON entity type.

These are not named entities and should not be picked by the model.

Only named entities (person names/nouns) should be picked.

[urchade]

Unfortunately, this was the case in the training dataset . Try to change the label name

[manojkumarvohra]

when you say changing label name ... can you please elaborate on it?

[urchade]

In instead of using "person" try another alternative like "person name"

[manojkumarvohra]

Unfortunately its not working with a changed label.

However increasing the threshold does seems to ignore pronouns but that would mean trading off the detection rate?

[urchade]

there are infinite possible label name, try to find the one that works

or change the model you use. Which model version is this ?

[manojkumarvohra]

I am using gliner_multi_pii-v1 through a spacy pipe integrated with Presidio.

I see presidio follows its own entity labels.

[urchade]

if you are interested in pii, you can check this blogpost: https://ai.gopubby.com/zero-shot-named-entity-recognition-ner-c81f82cb0f70

[manojkumarvohra]

Thanks for sharing. Let me try other models in our integration .... especially the gliner_medium_news-v2.1

[manojkumarvohra]

Hi,

While I have closed this issue, can you please help me understand if it can detect entities from chinese simplified text?

Officially on the page its mentioned to support 6 languages : English, French, German, Spanish, Italian and Portugese.