Intent feature discovery using Q&A corpus and web data
Intent feature discovery using Q&A corpus and web data
复制标题
使用 Q 进行意图特征发现
DOI：
10.1145/2108616.2108665
复制
发表时间：
2010
期刊：
影响因子：
--
通讯作者：
Katsumi Tanaka
中科院分区：
文献类型：
--
作者：
Soungwoong Yoon;A. Jatowt;Katsumi Tanaka
网络搜索环境中的用户意图被定义为用户的信息需求，并被认为是通过分析过去的数据（例如查询、点击历史和用户配置文件）来发现的。然而，即使在相同的查询中，用户也可能有不同的意图。在本文中，我们试图通过寻找意图的特征来发现意图的特征。我们的假设是，如果用户表达的查询清楚地指出了某种意图，他/她可以使用该查询到达预期的网页。我们使用意图演化过程（称为多意图模型）来概念化意图特征的这种功能。我们使用网络问答语料库分析收集候选意图特征，并提出使用由点击链模型支持的搜索引擎索引的自动判断方法来证明候选意图特征的适应性。实验结果表明，可以有效地提取意图特征，并为意图发现提供证据，而无需人工监督。
User intent in Web search environment is defined as user's information need, and believed to be found by analyzing past data such as queries, click histories and user profiles. However, users may have different intents even in the same queries. In this paper, we attempt to discover the characteristics of intent through finding its features. Our assumption is that if a user expresses the query which clearly points out to certain intent, s/he can reach an intended Web page using that query. We conceptualize this functionality of intent features using intent evolution procedure, called multiple intent model. We collect candidate intent features using Web Q&A corpus analysis, and suggest the automated judgment method using search engine indexes powered by Click Chain Model to demonstrate the adaptability of candidate intent features. Experimental results show that intent features can be extracted efficiently and provide evidences toward intent discovery without human supervision.