社交网络环境下基于协同过滤的上下文感知推荐系统研究

With the explosive growth of information and rapid development of social network, collaborative filtering recommender system is facing the key problems in the recommendation process, for instance, the problem of scalability, context information processing, as well as a user preference model. In this situation, how to achieve an efficient collaborative filtering recommendation model in the social network environment has become a focal problem. The results of existing research on various problems in the recommender system are mainly carried out from these aspects: Hadoop data-storage system, MapReduce programming, data mining, user modeling, collaborative filtering model and optimization model, etc. However, the lack of context information processing in the design of the recommender system seriously restricts the performance of the system, the reason for which is that collaborative filtering model still confronts with the problem of data sparsity and the user preference model faces the cold-start problem, even though the context information is so rich. In order to probe into the influence of context data on the performance of collaborative filtering model, this study aims to the context-aware recommender systems with collaborative filtering in social network, and focuses on the key issues such as the accurate extraction of user preference from the context information, overcoming the data sparsity problem and cold-start problem, the evaluation of system performance, the system development and realization, and so on. As the same time, we also make full use of the context data to improve the adaptability of the collaborative filtering recommender system in the social network environment, which is significant to enhance the recommendation quality and efficiency of the recommender system.

信息的爆炸式增长及社交网络的快速发展，使得协同过滤推荐系统面临着可扩展性、上下文信息处理、用户偏好建模等问题，实现社交网络环境下高效的协同过滤推荐已成为热点问题。现有研究主要从Hadoop储存、MapReduce编程、数据挖掘、用户建模、协同过滤模型建立与优化等方面来解决推荐系统中的各种问题，但在系统设计中却缺少对上下文信息的处理，以至于使得系统即使在上下文数据很丰富时协同过滤模型仍面临数据稀疏性问题，用户偏好模型面临冷启动问题等，这相当程度上制约了系统的性能。为解决社交环境下上下文数据对协同过滤模型的性能影响问题，本项目面向社交网络环境下基于协同过滤的上下文感知推荐系统，对上下文用户偏好精确提取、缓解高维数据稀疏性、冷启动问题、系统性能评价、系统开发与实现等问题进行重点研究，充分利用上下文数据改进协同过滤推荐系统对社交网络环境的适应能力，对提高推荐系统的推荐质量和推荐效率具有重要的意义。

本项目针对社交网络环境下基于协同过滤的上下文感知推荐系统的精确提取上下文用户偏好、缓解高维数据稀疏问题、社交网络信息融入基于协同过滤的上下文感知推荐系统、利用上下文提高社交网络推荐系统精确度、有效解决推荐系统的评价等问题进行研究。提出基于分析上下文和相似用户社交网络关系的上下文用户偏好提取方法、基于协同过滤和高阶奇异值分解技术的上下文推荐方法以及基于社交网络分析的上下文感知推荐系统的定义，同时考虑社交网络信息和上下文信息对推荐系统的影响。然后面向个性化网络服务领域，提出社交网络环境下的上下文感知推荐方法，利用上下文相似用户偏好和社交网络关系，发现用户的最近邻居，并基于协同过滤预测潜在上下文用户偏好以生成推荐。最后，利用数据集展开实验对比并对推荐效果进行评估。社交网络环境下基于协同过滤的上下文感知推荐系统研究具有重要的理论意义与实用价值。