rnn-playlist-prediction

Playlist prediction with recurrent neural networks on Cornell Playlist Data Set

Experiments with Softmax outputs

On Small Data Set

Model Name	func	func_1	func_2	func_3	func_4	func_5
SimpleGRU	29.64%	42.83%	59.58%	70.79%	77.60%	82.07%
BiLSTM	30.16%	43.52%	60.52%	71.56%	78.38%	82.85%
BiLSTM_with_Attention	31.42%	44.32%	60.75%	71.85%	78.36%	82.65%

On Big Data Set

Model Name	func	func_1	func_2	func_3	func_4	func_5
SimpleGRU	17.55%	26.16%	37.59%	45.82%	52.10%	57.11%
BiLSTM	18.26%	27.16%	39.05%	47.47%	53.92%	59.24%
BiLSTM_with_Attention	20.74%	29.83%	41.46%	49.85%	56.38%	61.69%

On Complete Data Set

Model Name	func	func_1	func_2	func_3	func_4	func_5
BiLSTM_with_Attention	28.50%	38.85%	52.79%	62.27%	68.63%	73.63%

Details

Model Name	Notebook
BiLSTM	Notebook
SimpleGRU	Notebook
BiLSTM_with_Attention	Notebook

Experiments with Word2Vec embeddings as inputs and outputs

Model Name	mae	mse	acc	mape	cosine_proximity	categorical_crossentropy	Details
SimpleGRU	0.59	0.74	0.25	313.03	-0.65	-40.60	Notebook
BiLSTM	0.59	0.74	0.26	499.65	-0.66	-37.26	Notebook