情境感知推荐中的上下文宽松匹配方法研究

摘要

Context-aware recommendation is an effective recommending method,but it has the problem of hard to determine the appropriate context variables,and to use too many contextual variables will lead to the increase in dimensionality and the loss of accuracy in recommendation.We proposed a method,it decomposes the recommendation algorithm and matches the subject for the context with decomposed components,meanwhile,it adopts the context information matching strategy with loose match.Thus a hybrid recommendation algorithm is formed which integrates the context pre-filtering and modelling.It is proved by comparing its performance with other kinds of algorithms’that the method is effective in determining the effectual context variables and improving the accuracy of recommendation algo-rithms.%情境感知推荐是一种有效的推荐方法，但存在难以确定有效上下文变量的问题，使用过多的上下文变量会导致数据维度的上升及推荐准确性的下降。提出一种将推荐算法分解，以各分解部分为上下文匹配主体，且采取宽松匹配的上下文信息匹配策略。形成了一种融合上下文预过滤和上下文建模的混合推荐算法。通过同其他类型算法的性能对比实验证明了该方法在确定有效上下文变量及提高推荐算法准确性上的有效性。