基于海量搜索历史数据的用户兴趣模型

User interest model based on mass historical search data

针对互联网搜索引擎环境中,基于海量搜索历史数据分析用户兴趣的问题,提出一种改进的用户兴趣模型。该模型根据用户搜索的历史数据,结合向量空间模型(VSM)和TF-IDF算法,递归地回溯出用户兴趣权重列表。为解决用户兴趣变化和时间性能的问题,该模型引入时间遗忘机制进行动态更新,并在Hadoop分布式系统架构下利用Map Reduce分布式编程模型进行实现。实验结果表明,改进的用户兴趣模型的查准率和召回率都能达到50%,具有较好的可行性和可用性。

To solve the problem of analyzing users＇ interests with mass historical search data in Internet search engine environment, an improved user interest model was proposed. The user interest weight list was backtracked recursively by the proposed model which combined Vector Space Model（ VSM） with TF-IDF algorithm according to the users＇ historical search data. A forgotten mechanism was introduced to update the proposed model dynamically when users＇ interests changed. The Map Reduce programming model under the Hadoop distributed system framework was used to update the time performance.Experimental results show that the precision and recall of the improved user interest model are close to 50%, which is both practical and useful.