面向移动环境的情感推荐隐式反馈偏好挖掘研究

Based on the preference information contained in user's implicity feedback behaviors, cognition burden can be unloaded. In addition, with low interference, and users would be satisfied with the interaction progress. Considering the affective recommendation system in mobile environment, researches are carried on from aspects based on portable intelligent terminal. Firstly, research on construction of implicity feedback behaviors database in different contexts. And based on the multidimensional scaling analysis theory, confirm behavior type, utility and redundancy of group users. Secondly, preference mining based on implicit feedback behaviors is researched. Propose preference computing method in different granularity. And combine them to improve the accuracy and rapidity. Fourthly, on the basis of XML technology and Emotional Markup Language, study on the storage and exchange of preference information to realize data sharing capability in multi-areas and multi-platforms. Moreover, EML-format preference information will be lightened in different services to reduce huge transmission volume, increase the access speed, service efficiency, and penetration rate.

基于用户隐式反馈行为蕴含的情感偏好信息，可以减少用户的认知负担，使交互过程具有低干扰性，提升用户体验。项目拟针对移动环境中情感推荐系统，基于便携式智能应用终端，从三个方面展开研究：1.研究多种不同情境诱导下的隐式反馈行为样本库的构建方法，并基于多元尺度分析等理论，探索群体用户隐式反馈行为的种类、效用和冗余性；2.研究基于隐式反馈行为的用户情感偏好挖掘，提出粗粒度和细粒度的情感偏好计算方法，并将两者结合以提高推荐的准确性和实时性；3.基于XML技术和情感标记语言，研究隐式反馈情感偏好特征的存储与交换方法，并探索EML格式的情感偏好信息的轻量化问题，降低大体积数据文件的传输量，提高访问速度和效率，从而提升服务速度与研究成果的普及率。

研究基于便携式移动终端的情感推荐隐式偏好挖掘的基本理论和相关算法，主要成果包括：对移动环境下不同场景中多种用户隐式反馈行为数据和情感偏好特征的获取，进行了详细的分析和研究，建立了多情境隐式反馈行为样本库；对不同多源异构隐式反馈数据进行组合，提取其特征向量，统一结构，便于后续构建推荐模型；根据不同推荐场景的需求，利用深度学习技术构建了推荐模型，挖掘不同粒度的用户情感偏好，为用户提供个性化推荐服务，提升用户满意度；对用户情感偏好特征描述和挖掘过程进行轻量化处理；并结合无人系统的应用，研究了基于推荐和偏好挖掘的多源异构设备管理问题、个性化服务资源推荐等；集成各个部分研究内容，构建了基于用户隐式反馈行为的用户情感偏好挖掘系统；为个性化推荐领域的发展提供了新思路和理论依据。