Community Conundrum 25: Feature Visualization is now live! Read More

Conundrum 21 - Fashion reviews - Finale!

Community Manager
Community Manager
Conundrum 21 - Fashion reviews - Finale!

MichaelG_0-1598524061466.png

For this conundrum we will be bringing together results from the previous two fashion review conundrums and using them, along with a handy plugin, to build a model and predict some ratings!

 

Using the results of Conundrum #17 (processed dataset provided as well if you would like). Create a prediction task for the Rating of the review. There are a few options on how to do this:

  • Use the prepared dataset and dive straight into the Lab and using the Rating as target variable build a few models that predict it. A benchmark F1 score would be ~47%
  • Download and install the sentence embedding plugin. We would advise read the docs on how it works, but at a high level you can:
    • Download a pretrained model.
    • Use the downloaded model to create embeddings from your text.

After using the plugin you should be able to obtain higher performance in the lab.

 

  • In addition to the sentence embedding, you could use the results from Conundrum #20 and use the obtained sentiment score as an additional feature for the Machine Learning task.

You would likely need to combine the datasets containing the sentiment and the embeddings (tip: You can use the the review_id column to create a join).

Bonus: For some pretrained models, a maximum number of characters or tokens might exist. In order to estimate whether we can use the model, compute a custom Python metric in your datasets to probe the maximum number of tokens in the review_text_clean column.

 

Good luck!

I hope I helped! Do you Know that if I was Useful to you or Did something Outstanding you can Show your appreciation by giving me a KUDOS?
2 Replies
Level 6

OK, I've met the basic mark on this project.  F1 Score of .479. 

Cunondrum 21.jpg

 

Features Included.

Conundrum 21 Features.jpg

 

So this is doable.  Let's see how much further folks can go.

--Tom
Community Manager
Community Manager
Author

Nice work Tom! Thanks for sharing. 

I hope I helped! Do you Know that if I was Useful to you or Did something Outstanding you can Show your appreciation by giving me a KUDOS?
0 Kudos