针对个性化推荐系统中用户偏好的学习与高维稀疏数据处理问题.受到隐马尔可夫模型(HMM)结构特征启发,采用一种考虑上下文的两阶段用户偏好收集推理策略的个性化推荐算法.选择MD算法对系统历史评分信息进行挖掘处理,提取用户偏好分布频繁三项集作为隐含状态,将用户评分项目序列看作观测状态,从而抽象为一个HMM模型,结合BP神经网络进行第一阶段的HMM模型的用户偏好学习与推理.然后根据第一阶段的学习训练生成最优推荐集合.实验结果表明基于HMM的推荐算法比传统推荐算法具有更好的适应性和推荐质量.%According to the personalized recommendation system user preference learning and high dimensional sparse data processing .Inspired by hidden Markov model ' s ( Hidden Markov Model, HMM) structure,this paper introduced a two stage user preference context collection reasoning strategy of personalized recommendation algorithm .Chose the MD algorithm to extract the information of the system history score , extracts the frequent three items of the user prefer-ence distribution as the implicit state , and considers the user′s scoring sequence as the observa-tion state , which is abstracted into a HMM model , user preference learning and reasoning for the first stage of the HMM model combined with BP neural network .Then according to the training of generating optimal first stage recommendation set .The experimental results showed that the recommended algorithm HMM based on the traditional recommendation algorithm had better adaptability and recommendation quality .
展开▼