作　　者： ; (刘霞）;

机构地区： 广州科技贸易职业学院

出　　处： 《南京理工大学学报》 2023年第5期658-664,共7页

摘　　要： 由于直播数据量大、数据模态多样、实时变化性强,且直播数据具有热点效应,给直播资源的精准推荐提出了更高挑战。为了提高直播推荐的有效性,提出了一种基于多模态神经网络的直播推荐算法,为各种大型媒体平台提供技术保证。首先,通过用户观看时长和评分权重矩阵建立了直播推荐评分准则。然后,在文本数据样本的基础上,选取了视频、图片、音频等多模态特征作为输入数据样本。利用多模态神经网络建立用户和直播资源的推荐算法,并构建直播推荐正则损失函数。以损失函数最小值为优化对象,不断迭代获得最优算法。最后,输入多模态特征,获得直播TOP-K推荐序列,并对推荐序列进行性能评价。实验结果表明,通过合理设置神经网络的层数,所提直播推荐算法在5个不同媒体平台数据中,均获得了较高的推荐性能,为直播平台的精准资源推荐提供了有效的策略支持。 Due to the large amount of live streaming data,diverse data modes,strong real-time variability,and the hot spot effect of live streaming data,it poses a higher challenge to the accurate recommendation of live streaming resources.In order to improve the effectiveness of live streaming recommendation,a live streaming recommendation algorithm based on multi-modal neural network is proposed to provide technical guarantee for various large media platforms.Firstly,the scoring criterion of live streaming recommendation is established through the user’s viewing time and scoring weight matrix.Then,on the basis of text data samples,multi-modal features such as video,pictures and audio are selected as input data samples.The multi-modal neural network is used to establish the recommendation algorithm of users and live streaming resources,and the regular loss function of live streaming recommendation is constructed.Taking the minimum value of loss function as the optimization object,the optimal algorithm is obtained by iteration.Finally,the multi-modal features are input to obtain the live TOP-K recommendation sequence,and the performance of the recommendation sequence is evaluated.The experimental results show that the proposed live streaming recommendation algorithm achieves high recommendation performance in five different media platform data by reasonably setting the layers of neural network,which provides effective strategic support for accurate resource recommendation of live streaming platforms.

关 键 词： 直播推荐 神经网络 多模态 媒体平台

领　　域： [自动化与计算机技术—计算机应用技术] [自动化与计算机技术—计算机科学与技术]