Does lookup data need training data for each data?

Rasa Open Source
1

Hi, I am working on a bot and there is a alot of products in lookup table, so I just want to ask do I have to add training data for each and every product mentioned in lookup because there are many lookups for every item. So the probability of count if (n_products times n_colors) times training data. If there is any optimize solution please feel free to suggest.

2

Just so I understand correctly, you’ve got a lookup table that you’re using with a regex entity extractor and you’re worried that you’re only capturing part of the entity?

3

Lookup table has products_category, products_sub_category, color, gender, brand names. This bot is being used for e-comm product search from my own database. So, in training data. Here is training data that I added:

 intent: product_inquiry
  examples: |
    - I want a product
    - I want to buy something
    - show me [Laptop](sub_category)
    - show me [Mobiles](sub_category)
    - show me [Shoes](sub_category)
    - show me [Monitors](sub_category)
    - I want a 17 inch [laptop](sub_category)
    - [laptops](sub_category) starting from [20000](min)
    - [smartphones](sub_category) starting from [20000](min)
    - can you show me [rose gold](color) [smartphones](sub_category) starting from [5000](min)
    - [pink](color) [shoes](sub_category)
    - [blue](color) [shoes](sub_category) for [men](gender)
    - [green](color) [shoes](sub_category) for sports
    - [shoes](sub_category) for running
    - Show me [Men's Clothing](sub_category)
    - i would like to see [Men's clothing](sub_category)
    - Show me [Women's Clothing](sub_category)
    - can you show me [Women's clothing](sub_category)
    - 19 inch [monitor](sub_category) from [hp](brand)
    - show me some [monitors](sub_category)
    - gaming [monitors](sub_category) under [7000](max)
    - [HP](brand) [Laptops](sub_category) with [Black](color)
    - [women](gender) sports [shoes](sub_category) starting from [1000](min)
    - [sports shoes](sub_category) under [700](max)
    - [casual shoes](sub_category) under [800](max) and starting from [200](min)

so lookup table such as sub category has more than 10 examples and brand names are also having more than 10 names. So, how do I write my training data that it predict even the name which is not added in training data but it is inside lookup table. Here, I am not using any regex right now.

4

It sounds like you may want to use the RegexEntityextractor. You can put all strings that match your entities into a lookup table which in turn can be detected by this component. As long as it appears in the lookup it doesn’t need to also be in your training data.

There’s also more information on lookups and extractors here.

5

Thanks, I forgot to add use_lookup_table = True. Now, I added it is working perfectly and I don’t need more training data too.