基于用户查询日志的查询聚类

基于用户查询日志提出了新的查询聚类算法.用户查询日志数据量大,比通常用于查询聚类的查询展现日志和查询点击日志更加稠密,不易产生聚类小的问题,但噪声多,不容易处理.为发现相似查询并减少噪声影响,同一用户同一时段的多次查询(共现查询)之间认为具有较高相似概率.在这一假设基础上,利用查询共现关系建立查询的邻居查询向量空间.将查询用邻居查询向量表示,邻居查询向量的相似度作为聚类中的查询相似度.应用改进的基于密度聚类算法完成聚类.实验证明,95 262个查询组成数据集上,聚类算法实现查准率79.77%、查全率48.21%,平均聚类大小达到51.

Query clustering using user-query logs

A new query clustering method on user-query log was presented. Traditional clustering techniques focused on queries and click-through logs, which are often sparse. The average cluster size is often small. In contrast, the user-query log is much denser as well as noisier. To reduce the influence of the noises and discover similar queries, queries visited by the same user at the same session were assumed to be mostly similar. Based on the assumption, a new similarity measure using query co-occurrence relations was calculated to create query neighbor vector space. The queries were represented by vectors consisting of their neighbors. The similarity function for clustering was calculated based on the query neighbor vectors. An adjusted clustering method of density-based spatial clustering of applications with noise(DBSCAN) was applied to generate the clusters. Experiments on a real dataset of 95262 queries show that 79.77% precision and 48.21% recall is achieved and the average cluster size achieves 51.