Procedure:
We take up a BERT pre trained model. We use bert-as-a-service to get our inference graph. This serialised graph is then used for building the feature extractor. We further use this on Reuters Benchmark Corpus to generate embeddings and visualise them. Further a supervised model is trained using these features. A nearest neighbour accelerated search engine is built based on this model. We took up the IMDB movie summaries dataset. We vectorized the reviews and used our search engine to find nearest neighbour in the movie space! It thus recommends movies based on plot features.
Project Report: https://bit.ly/ProjectReportsOfTanishqWadhwani
If this side project of mine helped you in your journey, it would mean a lot to me if you could tap on the star button above.
It would help me boost my personal brand.
Thanks! :)
Wishing you all the very best!