In this paper, a new property data-driven based weight calculation method is proposed, which is applicable to vectors consisting both categorical and continuous attributes. According to the query vector, a corresponding search space is specified. In this space, data distribution is analyzed dynamically, thus to gain the discrimination degree and then the weight of each attribute. To evaluating our method, we experiment with an actual database containing several attributes of clothes. Experiments show our method improve the accuracy and the results can better fit users' expectations.%本文针对混合类型元素组成的向量,即包含值域离散型、值域连续型元素的向量,提出了一种基于数据驱动的属性权重计算方法。根据查询向量的取值确定搜索空间范围,并统计搜索空间内属性取值分布情况,动态的统计出各个属性在搜索中的区分度大小,进而计算出各属性在相似度计算时所占权重值,并将权重值引入到基于向量距离的检索中。本文利用服装数据库对检索方法进行评估,实验结果表明基于数据驱动的权重计算方法很好地分析出属性区分度,使检索结果更加符合用户预期,取得较好的结果。
展开▼