Apply classic ml algorithm for text embeddings and classifier, apply recommendation model for building recommendation system
http://jmcauley.ucsd.edu/data/amazon/
In order to save time, I use the spacy module, which has built-in functionality to calculate the average vector of a sequence of words.
While in another project named Stocksight, NLTK module has been applied. The link below shows the difference between these two:
What's more, there are other famous NLP libraries worth mentioning. It can be checked out as follows:
https://elitedatascience.com/python-nlp-libraries
As for the evaluation metrics, it's important to understand the difference between all metrics. Here is a good reference (in Chinese though):
https://www.cnblogs.com/futurehau/p/6109772.html
A really good slide for recommendation system - Tensorrec: https://www.slideshare.net/JamesKirk58/boston-ml-architecting-recommender-systems
Ans here is the sourde code of Tensorrec:
https://github.com/jfkirk/tensorrec
Hyperparameter tuning for size of embeddings plays a significant role in filtering recommendation model. This article introduces the function of embedding layer (in Chinese):
https://blog.csdn.net/u010412858/article/details/77848878
A clear introduction about the difference between these two: