Sentiment Analysis and Music Recommendations Systems Description: a review-based recommendation system with natural language processing (NLP) and Collaborative Filtering on 1.5M review data and 75k music meta data, utilizing Python via keras framework on an AWS EC2 instance.
-
Trained a convolutional neural network (CNN) for Sentiment Analysis on the review texts classifying the reviews as positive and negative, summarized keyword using NLTK, visualized word embeddings, and gained 91.4% test accuracy.
The most significant words in review texts:
The most significant words in summary texts:
-
Trained a stacked Autoencoder with SGD on tokenized and embedded text matrix for dimensionality reduction.
-
Trained a Deep Neural Network with regularization for the recommender with MAE below 0.4 and MSE below 40.
