Estimating Intent Types for Search Result Diversification
Estimating Intent Types for Search Result Diversification
复制标题
DOI：
10.1007/978-3-642-45068-6_3
复制
发表时间：
2013-12
期刊：
影响因子：
--
通讯作者：
Kosetsu Tsukuda;T. Sakai;Zhicheng Dou;Katsumi Tanaka
中科院分区：
文献类型：
--
作者：
Kosetsu Tsukuda;T. Sakai;Zhicheng Dou;Katsumi Tanaka
给定歧义或未指定的查询，搜索结果多样化旨在在单个搜索引擎结果页面（SERP）中容纳不同的用户意图。虽然自动识别给定查询的不同意图是实现结果多样化的关键步骤，但对意图类型（信息与导航）的估计也很重要。如果可以区分信息意图和导航意图，那么搜索引擎可以为每个导航意图返回一个最佳URL，同时在SERP中为信息意图分配更多空间。在此基础上，我们提出了一种意图重要性感知和类型感知的搜索结果多样化框架。我们使用ntcirr -9日语意图子主题挖掘和文档排序测试集进行的实验表明：(a)我们的日语意图类型估计方法达到了64.4%的准确率；(b)我们提出的多样化方法在56个主题上的D-nDCG和DIN-nDCG分别达到了0.6373和0.5898，与ntir -9 INTENT日语文档排名的最高表现相比，这在统计上是显著的进步。此外，我们的相关性导向模型显著优于我们的多样性导向模型和Douet等人的原始模型。
Given an ambiguous or underspecified query, search result diversification aims at accommodating different user intents within a single Search Engine Result Page (SERP). While automatic identification of different intents for a given query is a crucial step for result diversification, also important is the estimation of intent types (informational vs. navigational). If it is possible to distinguish between informational and navigational intents, search engines can aim to return one best URL for each navigational intent, while allocating more space to the informational intents within the SERP. In light of the observations, we propose a new framework for search result diversification that is intent importance-aware and type-aware. Our experiments using the NTCIR-9 INTENT Japanese Subtopic Mining and Document Ranking test collections show that: (a) our intent type estimation method for Japanese achieves 64.4% accuracy; and (b) our proposed diversification method achieves 0.6373 in D-nDCG and 0.5898 in DIN-nDCG over 56 topics, which are statistically significant gains over the top performers of the NTCIR-9 INTENT Japanese Document Ranking runs. Moreover, our relevance oriented model significantly outperforms our diversity oriented model and the original model by Douet al..