注意力机制与改进RNN的混合音乐推荐算法研究

Research on Music Recommendation Algorithm Based on Attention Mechanism and Improved RNN

该算法由独立循环神经网络算法与注意力机制共同组成的深度神经网络实现,并在数据预处理阶段对用户收听历史记录的音频使用散射变换进行预处理.通过散射变换提取用户收听历史记录音频特征,再将此特征与用户画像共同通过混合注意力机制的独立循环神经网络得出推荐列表.仿真结果表明,文中给出的算法与已有的IndRNN(循环神经网络)和基于LSTM(长短期记忆网络)的音乐推荐算法相比,分别提高了7.8%和20.9%的推荐准确度.

According to the problem that the existing music recommendation algorithm has low accuracy,this paper developed an attention mechanism and improved RNN hybrid music recommendation algorithm.The algorithm is implemented by a deep neural netw ork composed of an independent recurrent neural netw ork algorithm and an attention mechanism.In the data preprocessing stage,the audio of the user’s listening history is preprocessed using the scatter transform.The user’s history audio feature is extracted by scattering transform,this feature is combined w ith the user’s information to through a hybrid attentional and recurrent neural netw ork.The simulation experiment show s that the proposed algorithm improves the recommendation accuracy by 7.8%and 20.9%compared w ith the existing IndRNN(circular neural netw ork)and LSTM(long-short-term memory netw ork)-based music recommendation algorithms.