You are on page 1of 2

Different Approaches

Models Attempted Inferences


Bert 1) Do not get Contextual meaning for the given Attributes.
2) Do not have much data to train the model.
3) Do not make it to run in front end.
Hugging Face -gpt2 1) Do not distinguish Fact from Fiction.
2) Reflect Biases inherent to the systems they were trained on.
3) Do not use for the use case that required generated text to be true.
4) Do not have enough data to train the model.

RNN with LSTM 1) Do not have enough data to train the model.
2) Training RNN Completely tough task.
3) Gradient exploding and vanishing problems.
4) Do not get contextual meaning

Key to text -T5 model 1) Do not make it to run in front end


2) Do not have much data to train the model
3) Can get contextual meaning if we have proper data to train

Template Based 1) Can able to run in the front end with java script.
2) Can able to do in the browser.
3) Have to frame templates
4) Can make contextual meaning for the given attributes

Template Based Approach for NLG


1. We currently planned to do template based approach for Natural language
generation.From the given framed 160 attributes.
2. In template based approach, need to frame multiple default sentences for each
attributes .
3. By randomizing the sentences we can frame 150 to 300 words product
description.
For reference here attached some template sentences for some attributes.
Sneakers(attributes) - 1) For casual style a little extra and boost your streetwear look,
the classic and fashionable sneakers are your go-to option.
Monk Strap Formal(attribute) - 1) monk straps have evolved into one of the most
versatile pieces of footwear in a gentleman's closet.
2) The clean silhouette of a single monk strap fits well with traditional dress while the
laid back double buckle offers modern class with casual style.

ALBERTO TORRESI 3
*

ALBERTO TORRESI 3

You might also like