在推荐系统领域，人们往往只关注“用户-项目”之间的关联关系，而较少考虑它们所处的上下文环境（如时间、位置、周围人员、情绪、活动状态、网络条件等等）。但是在许多应用场景下，仅仅依靠"用户-项目"二元关系并不能生成有效推荐。例如，有的用户更喜欢在"早上"而不是"中午"被推荐合适的新闻信息；有的用户在不同的心情可能会希望被推荐不同类型的音乐。
Several topics in the field of context-aware recommendation systems
- Context modeling techniques in recommender systems;
- User modeling based on context-aware in recommender system;
- Context recommended data set;
- Algorithm for detecting correlation of contextual data;
- The algorithm of incorporating contextual information into the recommendation process;
- An explicit association algorithm is established between contextual features and user ratings;
- Interacting with the context-aware referral system;
- The new application of the context-aware recommender system;
- Large-scale context-aware referral system;
- Evaluation of the context-aware recommendation system;
- Mobile context-aware referral system;
- Context-Aware group recommendations.
How the context gets
- Display Acquisition (explicitly): Obtain contextual information directly associated with a user or project by means of physical device awareness, user inquiries, user-initiated settings, etc.
- Implicit acquisition (implicitly): Get contextual information with an introduction to existing data or the surrounding environment: you can get time context information based on the user's interaction log with the system
- Inference acquisition: Through statistical methods or data mining techniques. The naïve Bayesian classifier or other predictive model can be used to infer whether the user is in "home" or "office"
Classification based on the traditional recommendation system partitioning method
- Context-aware recommendation generation based on collaborative filtering
The introduction of contextual information into user similarity, project similarity and model-based collaborative filtering is extended to contextual user preference similarity computation and model-based context-aware collaborative filtering , which is expected by increasing context constraints, Improve the accuracy of the similarity calculation or the model, and thus improve the recommended accuracy.
Document Chen A. Context-aware Collaborative filtering system:predicting the user's preferences in ubiquitous computing environm Ent. In:
Proc. of the LoCA 2005. LNCS 3479, Berlin:springer-verlag, 2005. 244?253. http://dl.acm.org/citation.cfm?id=1056836 [doi:10.1145/1056808.1056836]
It is not sufficient to think that "similar users have similar preferences", but should also focus on "the preference of other users in the context of the current context similar to the active user ", proposing the integration of contextual information, contextual correlation coefficients based on the project, user-context similarity, etc. into collaborative filtering technology.
- Content-based context-aware recommendation generation
Incorporating contextual information into content-based recommendations, with a focus on user preferences, contextual and project attributes, namely: Mining user preferences for different project attributes under different context conditions, and combining the attribute descriptions of each specific item to discover the degree of matching (or probability) between users, projects, and contexts This predicts potential contextual user preferences, and finally generates recommendations in conjunction with the user's current context. Therefore, after the context modeling and the context user preference extraction, the project attribute feature description and the matching degree calculation method become the key.
Context-aware Referral system
