Fragrantica perfume review clasifier (LSTM)

Train list

model name	padding length	num words	embedding	batch size	acc result
lstm_with_lemmatizated_01	200	5000	100	128	0.7450
lstm_with_lemmatizated_02	200	2500	100	256	0.7384
lstm_with_lemmatizated_03	200	1000	100	256	0.7029
lstm_with_lemmatizated_04	200	500	100	256	0.6727
lstm_with_lemmatizated_05	200	5000	200	256	0.7310
lstm_with_lemmatizated_06	200	2500	200	256	0.7358
lstm_with_lemmatizated_07	200	1000	200	256	0.7049
lstm_with_lemmatizated_08	200	500	200	256	0.6741
lstm_with_lemmatizated_09	200	5000	50	256	0.7452
lstm_with_lemmatizated_10	100	2500	50	256	0.7424
lstm_with_lemmatizated_11	100	1000	50	256	0.7085
lstm_with_lemmatizated_12	200	1000	200	512	0.7068
lstm_with_lemmatizated_13	100	5000	50	512	0.7429
lstm_with_lemmatizated_14	100	5000	50	512	0.7429
lstm_with_stopword_removed_01	200	5000	100	128	0.7390
lstm_with_stopword_removed_02	200	2500	100	256	0.7300
lstm_with_stopword_removed_03	200	1000	100	256	0.6960
lstm_with_stopword_removed_04	200	500	100	256	0.6693
lstm_with_stopword_removed_05	200	5000	200	256	0.7364
lstm_with_stopword_removed_06	200	2500	200	256	0.7365
lstm_with_stopword_removed_07	200	1000	200	256	0.6975
lstm_with_stopword_removed_08	200	500	200	256	0.6631
lstm_with_stopword_removed_09	200	5000	50	256	0.7391
lstm_with_stopword_removed_10	100	2500	50	256	0.7406
lstm_with_stopword_removed_11	100	1000	50	256	0.7009
lstm_with_stopword_removed_12	200	1000	200	512	0.7032
lstm_with_stopword_removed_13	100	5000	50	512	0.7032
lstm_with_stopword_removed_14	200	5000	100	512	0.7387
lstm_with_stopword_removed_15	200	1000	200	512	0.7441
bi_lstm_with_lemmatizated_01	200	5000	100	256

• lemmatizate 하면 미세하게 0.01 정도 accuracy 향상 효과가 있음
• num_words 는 대체로 큰것이 좋음. 근데 특이하게 embedding 이 클때는 오히려 num_words 가 적당한것이 좋았음
• batch_size 는 상관 없거나 512 가 나음
• padding_length 는 더 작은게 나아보임
• embedding 은 더 작은게 나음