查询背后的信息需求被定义为查询意图,搜索引擎可以根据不同的查询意图,提供多样性的服务,优化检索效果.查询意图的识别多被看成是一种分类问题,现有的大多数方法都基于查询串文本本身的特征和查询的用户点击数据特征.这两种方法存在如下的困难:对于查询串的文本特征,查询比较短,特征比较稀疏,要进行比较准确的理解会比较困难;对于用户点击特征,由于用户提交查询的长尾性分布,大多数查询的提交次数都是较少的,对于这些查询,要判别它们的意图是比较困难的.为了克服长尾查询上查询意图判断的不可靠问题,本文提出利用查询结果的相关性分数的分布作为特征来判断查询意图的方法.这种方法依赖查询结果的特征,比查询串本身的特征更加丰富;同时不依赖于用户的点击数据,因此可以克服长尾查询上的困难.结果表明,使用结果分数分布,可以提高意图判别的准确程度.%Query intent defined as the information need behind the query, search engines can provide diversity servicefor queries based on their intent to improve effectiveness. Query intent recognition always treated as a classificationproblem; State - of - art solutions most rely on the text of queries and the click information of users. Those two methods have following disadvantages: for the properties of the text of queries, the queries are usually short and lack of properties, so it is hard to understand the intent correctly. For the click through information, most queries have been submitted a few times because of the distribution of queries submitted. Therefore, it is difficult to distinguish their query intent. In order to solve the unreliability of the judgments about the query intent of long tail queries; we propose that using the distribution of correlation scores of result of queries to judge the query intent. This method relies on the result of queries and we can get richer properties than queries themselves. Meanwhile, we do not need the information data of users so that we can overcome the difficulty of long tail queries. The result shows that we could increase the accuracy of classification by using the distribution of scores of query results.
展开▼
