知识找回场景下推荐系统模拟实现及评价研究

目的/意义]信息过载一直是知识工作者在搜集、处理和创造知识的过程中所面临的主要困境。这种困境导致的结果之一是很难回忆起曾经使用过的文档的内容细节及具体位置,而推荐系统则能减少这样的困难。通过研究对比不同推荐系统在这一任务下的优缺点,可以帮助知识工作者更好地完成回忆任务。方法/过程]基于相关理论,在同一场景（知识找回）模拟实现并测试了4种不同类型的推荐过程,包括基于内容的推荐CBR、基于协同过滤的推荐CFR、基于推理网络的推荐INR与融入了情境感知的推荐CAS,根据所确定的若干指标（精确性、情境相关性、预测性、多样性）对推荐效果进行比较。结果/结论]结果显示,以上推荐系统在帮助用户回忆并找回文档过程中都有各自的优势,而基于情境感知的推荐系统在情境相关性与预测用户行为方面具有较好的效果。

An Experimental Study on Simulation and Evaluation of Recommendation Systems for Knowledge Re-finding

Purpose/significance] Information overload has been always considered as the major barrier confronted by knowledge workers in the process of gathering, processing and producing information. One of its consequences is that it is hard to recall documents that ever used, while the recommendation system could reduce such difficulty. Comparing the recommendation efficiencies through representative recommendation mechanisms may assisst knowledge workers in accomplishing the task of knowledge re-finding.Method/process] Based on associated recommendation system theoies, this paper presents a simulation on 4 different recommendation procedures in an unified experimental scene(knowledge re-finding), the precedures includes CBR, CFR, INR and CAS. 4 evaluation criteria (precision, context-relevance, action-prediction, diversity) has been used to evaluate and compare the efficiency of corresponding recommendation systems.Result/conclusion] The results show that each recommendation procedure has its own advantages in knowledge re-finding from different perspectives, and CAS has advantages in both context-relevance and action-prediction.