基于用户行为与媒体情感分析的存储资源贝叶斯分配策略研究

Efficient allocation of video content storage resources is one of key problems for the transform of new media business of radio and television to the intelligent type. The existing resource allocation schemes only consider the single viewing behavior of users, it results the problems of heavy burden and repeat call in network resource storage, making it unable to meet the critical need of user experience promotion for intelligent radio and television. In order to solve the problem, this project will research the Bayes allocation method of storage resources based on the user behavior and media sentiment analysis. Through the merging of user behavior and real time experience, the optimal dynamic resource storage strategy will be formed. The research contents include: (1) construction of Bayes prior model based on forgetting law of user group, transforming the problem of storage resource allocation into that of solving the optimal strategy under the Bayes framework; (2) building the label metric model of user group behavior, by replacing the viewing time window with the label time window, the dynamic change rules are researched for the metric index function of user behavior and media sentiment; (3) through the optimal solution of adjustable two-peak model combined with user behavior and media sentiment, the storage resource allocation strategy is proposed under the Bayes framework, realizing the reasonable allocation and dynamic call of storage resources between edge-end and cloud-end. This subject aims to improve the user experience, realizing the theoretical breakthrough and technological innovation in the allocation of storage resources for intelligent radio and television.

视频内容存储资源的有效分配是影响广电新媒体业务向智慧型转移的关键问题之一，现有资源分配方案仅考虑用户单一收视行为，由此造成的网络资源存储负担重与重复调用问题使其无法满足智慧广电用户体验提升的核心需求。针对这一问题，本项目将研究基于用户行为与媒体情感分析的存储资源贝叶斯分配方法，通过融合分析用户行为与实时观后体验形成一种最优化的动态资源存储策略。具体研究内容包括：（1）基于用户群体遗忘规律的贝叶斯先验模型构建，将存储资源分配问题转化为贝叶斯框架下的最优策略求解问题；（2）构建用户群体行为标签度量模型，以标签时间窗代替收视时间窗，研究用户行为与媒体情感度量指标函数的动态变化规律；（3）通过最优求解结合用户行为与媒体情感的双峰可调模型，提出贝叶斯框架下的存储资源分配策略，实现边缘端与云端存储资源的合理分配与动态调用。本课题以提升用户体验为目标，以期实现智慧广电存储资源分配的理论突破与技术创新。

随着互联网和智能媒体的不断发展，媒体资源数据量激增，视频内容资源的合理存储与分配成为智慧广电满足用户个性化需求的关键。现有的广电云服务中所有资源存储在云端，给云端造成了巨大的存储负荷，且在用户与云端交互中存在网络资源重复调用、用户体验差等问题。针对上述问题，本项目提出基于用户行为与媒体情感分析的存储资源贝叶斯分配方法。主要研究内容包括：（1）构建基于用户群体遗忘规律的贝叶斯先验模型，以中国某地区居民的真实用户收视记录为支撑构建广电用户行为指标，利用艾宾浩斯记忆遗忘函数构建用户兴趣遗忘衰减模型，将存储资源分配问题转化为贝叶斯框架下的最优策略求解问题；（2）构建基于用户遗忘和重复增强的用户群体行为标签度量模型，在用户行为建模中以标签窗代替传统的收视时间窗，使用标签度量构建节目属性标签与节目情感标签，为资源分配策略提供标签度量指标依据；（3）构建基于双峰模型的最优资源分配策略，将用户行为分析与情感修正因素相结合，利用双峰模型实现边缘端与云端存储资源的合理分配。本项目基于贝叶斯框架达成了节约存储资源与用户群体行为研究精细化的平衡，最终得出了低负担云端存储、高资源利用率、良好用户体验的动态资源分配策略，实现了智慧广电存储资源动态分配策略的理论突破与技术创新。