I am trying to solve a NER (Named-Entity-Recognition) problem. I have input sequence as my input which contains approx 20-400 words. Sequence contains mostly English words. Please have a look at this question, as this one is related. So, entire sequence/words are not co-related with each other (so I am not sure if LSTM or similar could be useful)
I am aware about IOB format being used for NER. Data-set I have is separate entities. e.g. If you have a look at the question link above, target classes are price, product_id etc. I have list of prices, product_ids, usage_instructions etc. For NER as far as I know, I will have to annotate whole sequence (e.g. IOB format).
1. Is there any way, where I can use data-set I already have for training, without annotating whole sequences manually.
2. What kind of model should I use? Because as you can see each & every words are not co-related with each other. Also since individual entities are not co-related, I am expecting to get correct result even if their order is different.
P.S. Let me know if some details are missing. Thanks for your help!