Will Prodigy work for me?

[jay007 (Jason)]

I need to extract Product names, Colours, Sizes and Finish from a very large dataset. I have had some success using Pandas but there are so many patterns and variants in the Descriptions is very hard to get them all. This is were I thought I would build a model on the Products and I could tag the Colours, Sizes, Product names etc. then use Spacy. I don't mind paying the license for Prodigy but I also don't know if I I have Inception setup correctly. I used the Named entity with the Multi-Token Sequence Classifier (OpenNLP NER). Also with String matcher. While it does predict it feels like I will be at this for another week. Can some tell me if they have tried Inception and Prodigy and which is better. Does Prodigy use Named entity with the Multi-Token Sequence Classifier (OpenNLP NER) also?

[ryanwesslen]

Hi @jay007!

Thanks for your question and welcome to the Prodigy community

@jay007

I need to extract Product names, Colours, Sizes and Finish from a very large dataset. I have had some success using Pandas but there are so many patterns and variants in the Descriptions is very hard to get them all. This is were I thought I would build a model on the Products and I could tag the Colours, Sizes, Product names etc. then use Spacy. I don't mind paying the license for Prodigy but I also don't know if I I have Inception setup correctly.

Thanks for the background! While I haven't used INCEpTION, I have heard a few users who have used both (I'd be curious if other users have feedback).

@jay007

Does Prodigy use Named entity with the Multi-Token Sequence Classifier (OpenNLP NER) also?

But, from a quick look at the docs, it seems like while INCEpTION is built around OpenNLP (e.g., NER uses OpenNLP's Multi-Token Sequence Classifier), Prodigy is developed out-of-the-box around spaCy.

While you're not limited to spaCy in Prodigy -- in fact, you can export out your annotations (and then convert to OpenNLP's format) or write a custom recipe to extend Prodigy for an OpenNLP model-in-the-loop annotation workflow. But if your goal is speed, that may take a little time for you to make these modifications.

Are you open to using spaCy's NER component instead of OpenNLP? If so, then I definitely think then Prodigy may be worth a try. For more details on what the workflow may be, check out our Prodigy docs where we go through a quick example of how to install, run your first ner recipe, and train your first model.

If you want an even deeper dive on options (e.g., string/pattern matcher) and/or using a model-in-the-loop to correct or use active learning, we have detailed docs on ner on our docs page.

If you do buy Prodigy and realize it doesn't work, email us at contact@explosion.ai for a full refund. No questions asked.

Thanks again for considering Prodigy. Hope this helps!

[jay007 (Jason)]

Thanks for the reply. No I am just looking for something that works. The reason I asked about OpenNLP NER was if Prodigy used OpenNLP NER there would be no point. I have spent days testing Inception and not getting good results. The full refund is great, thanks. I will purchase and test. Ill report back here with results. Thanks again