Is there any way to improve Google NLP API result accuracy?

Question

I am using Google-Vision API to get text from images and using its result into NLP API. So far i intend to get Name,Location,address,email,contact number,job title,company name etc. when i scan a business card.Results so far are not that much accurate as sometimes the results are too vague,also NLP API return multiple entries for the same content text i.e Multiple value in names field,location field sometimes incorrect classifications too.Any suggestions on how i can improve its results?

Reference

1. Google vision API

2. Google language processing API

   Say for this business card

   VISION API results into

   NLP results into

Answer 1

Cloud Vision API returns the raw texts from the images, while the NL API focus on detecting known entities, thus two possible way to achieve what you are looking for :

1. Build an AutoML Natural Language Entity Extraction model - you want to label telephone and fax number differently as custom label

2. Write some custom logic to locate identified entities from NL API back to Vision output to see if it a "T" or "F".

3. Location is a vague term, it could be address, country, states, and etc, so if you are looking for address, use address. Also for fake address, the API will not return the entity