ciecus/music_lgb_recommend_kkbox

use lightgbm to recommend music,and the benchmark dataset is kkbox dataset wsdm2018

The music recommendation algorithm recommends the music algorithm for the user in view of the problem of information overload in the modern era. The music recommendation has features such as large item space, large user space, low product cost, high reusability of items, large demand for items, rapid change in item preferences, and high degree of socialization. Today's music recommendation algorithm mainly relies on the user's operation record of music to establish a similarity matrix between user and music, and performs user-based collaborative filtering recommendation or collaborative filtering based on items; this article adds operations to the music dimensions and user dimensions. Dimensional analysis, analysis of the user's operation source for the generation of music behavior, to predict the user's preference for music; for the screening of each dimension feature, the stochastic logistic regression in stability selection (Stability Selection) is used to obtain the score of each feature. The selected features were analyzed using the Light Gradient Boosting Tree (LightGBM) model. In addition, this paper innovatively uses numerical feature geography for forecasting. It is verified that the use of numerical features for modeling has a steady improvement in both effectiveness and robustness, and this analysis method is applicable to all classed learning with supervised learning, with a wide range of Application .# music_lgb_recommend_kkbox use lightgbm to recommend music,and the benchmark dataset is kkbox dataset wsdm2018