基于LSTM-Att方法的音乐流行趋势预测

Music Trend Forecast Based on LSTM-Att Method

利用循环神经网络的分支长短时记忆网络与注意力机制结合的模型进行音乐流行趋势的预测。首先,分析了传统的支持向量机以及循环神经网络等方法在预测时间序列数据上不能捕捉长时间序列信息等不足之处;其次,基于以上分析建立了长短时记忆网络加注意力机制结合的预测模型,针对所要预测的未来两个月歌手歌曲播放量,对数据集进行分析及相关属性选取、归一化等预处理,选取组合相应的歌曲日播放量、连续3天播放均值作为相应时间点的样本构建神经网络训练集;最后,设计实现了基于长短时记忆网络加注意力机制相结合的预测模型实验。实验结果表明,所使用的预测模型较传统的机器学习方法支持向量机以及长短时记忆网络等在均方根误差和平均绝对误差两个指标上取得了较为明显的提升。

Prediction of music popularity trend is carried out by using a model that combines the long short-time memory network of the cyclic neural network and the attention mechanism.Firstly,the shortcomings of traditional support vector machines and cyclic neural networks in predicting time series data like not capturing long time series information are analyzed.Secondly,based on the above analysis,the prediction model of the combination of long short-time memory network and attention mechanism is established.According to the predicted amount of songs played by artists in the next two months,the data set is analyzed and pre-processed such as relevant attribute selection and normalization,and the daily amount of songs played in combination and the average amount of songs played in 3 consecutive days are selected as samples at corresponding time points to construct the neural network training set.Finally,the prediction model experiment based on the combination of long short-time memory network and attention mechanism is designed,which shows that the proposed prediction model has achieved significant improvement over the traditional machine learning method support vector machine and long short-term memory network in terms of root mean square error and average absolute error.