摘要
传统推荐方法主要以现有数据所在范围为基础进行歌曲推荐,导致用户的满意度较低,为此,提出基于数据驱动遗传算法的自动化隐式评分歌曲推荐方法。首先将用户的历史数据作为基础矩阵,通过进行基础矩阵分解,得到用户对歌曲的隐式评分结果,采用互信息特征选择方法构建了特征选择模型,将无关特征或相似的冗余特征进行过滤,以此确保遗传算法输出的推荐结果具有更高的准确性。最后利用遗传算法中的深度神经网络模型实现对最终推荐结果的计算,将选择的特征值作为深度神经网络模型的学习驱动,对挖掘到的用户原始数据的隐式评分结果进行训练,并根据Sigmoid函数的输出结果判断歌曲是否满足推荐标准。测试结果表明,设计方法推荐结果中,用户对歌曲歌词、旋律、节奏以及情感的满意度分别达到了90.19%、90.94%、91.61%和90.89%,均明显高于对比方法。
Traditional recommendation methods recommend songs mainly based on the range of existing data,resulting in low user satisfaction.An automatic implicit scoring song recommendation method based on data-driven genetic algorithm is proposed.Firstly,the historical data of users were taken as the basic matrix,and the implicit scoring results of songs by users were obtained through basic matrix factorization.A feature selection model was constructed by using mutual information feature selection method to filter irrelevant features or similar redundant features,so as to ensure that the recommendation results output by genetic algorithm have higher accuracy.Finally,the deep neural network model in genetic algorithm was used to calculate the final recommendation results.The selected eigenvalues were used as the learning driver of the deep neural network model to train the implicit scoring results of the user's original data,and judge whether the song met the recommendation criteria according to the output results of the Sigmoid function.The test results showed that users'satisfaction with song lyrics,melody,rhythm and emotion reached to 90.19%,90.94%,91.61%and 90.89%,respectively,which were significantly higher than those in the comparison method.
作者
张研
ZHANG Yan(Anhui Vocational College of Art,Hefei,Anhui 230001,China)
出处
《河北北方学院学报(自然科学版)》
2023年第1期10-14,共5页
Journal of Hebei North University:Natural Science Edition
基金
安徽省高等学校省级质量工程项目“幼儿舞蹈创编中音乐元素的重要性研究”(2021jyxm1356)。