Query chains: learning to rank from implicit feedback

By Filip Radlinski, Thorsten Joachims. 2005

584 citations recorded by the source corpus.

This paper presents a novel approach for using clickthrough data to learn ranked retrieval functions for web search results. We observe that users searching the web often perform a sequence, or chain, of queries with a similar information need. Using query chains, we generate new types of preference judgments from search engine logs, thus taking advantage of user intelligence in reformulating queries. To validate our method we perform a controlled user study comparing generated preference judgments to explicit relevance judgments. We also implemented a real-world search engine to test our approach, using a modified ranking SVM to learn an improved ranking function from preference data. Our results demonstrate significant improvements in the ranking given by the search engine. The learned rankings outperform both a static ranking function, as well as one trained without considering query chains.

Subfields: Computer Vision and Pattern Recognition, Information Systems, Information Retrieval & Search, Signal Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

In Information Systems

In Information Retrieval & Search

In Signal Processing

Citations in the corpus

Pretrained Transformers for Text Ranking: BERT and Beyond (2020, methodology): “n enabling this growth was a series of techniques for interpreting noisy user click data and converting them into training examples that could be fed into machine-learning algorithms [Joachims, 2002, Radlinski and Joachims, 2005]. As we lack the space for a detailed treatment of learning to rank, we”
Learning diverse rankings with multi-armed bandits (2008, background): “…of the Internet infrastructure, and has hence attracted signi.cant interest from the machine learning community (e.g. Herbrich et al., 2000; Burges et al., 2005; Radlinski & Joachims, 2005; Chu & Ghahramani, 2005; Metzler & Croft, 2005; Yue et al., 2007; Taylor et al., 2008).”
Click chain model in web search (2009, background): “CCM consistently outperforms two state-of-the-art competitors in a number of metrics, with over 9.7% better log-likelihood, over 6.2% better click perplexity and much more robust (up to 30”
Analyzing and evaluating query reformulation strategies in web search logs (2009, result): “Present Study 98.2% 61.3% 89.1% He [15] 60% 4 98% Jones [22] 87.3% Murray [27] 97.3% 4 76% Radlinski [31] 96.5% 4 92.3% Our focus on precision rather than recall resulted in 98.2% precision which is 38% higher than reported in He et al and slightly higher than Radlinski and Joachim’s 96.5%.”
Mining Query Logs: Turning Search Usage Data into Knowledge (2010, background): “Radlinski and Joachims [177] and Joachims et al.”
Efficient multiple-click models in web search (2009, methodology): “[15], and an algorithm was proposed to detect a sequence of reformulated queries from the same user to learn an improved function.”
Query suggestion using hitting time (2008, methodology): “Speci.cally, if vj = QT (the original query) and vi is a URL, we assign new weights for Wij by L p(QT |URL,User) wij. j.=j wij = .”
SemEval-2016 Task 3: Community Question Answering (2019, background): “ Our task goes in the direction of passage reranking, where automatic classiﬁers are normally applied to pairs of questions and answer passages to derive a relative order between passages, e.g., see (Radlinski and Joachims, 2005; Jeon et al., 2005; Shen and Lapata, 2007; Moschitti et al., 2007; Seve”
Identifying Web Search Query Reformulation using Concept based Matching (2013, methodology): “Deﬁnition : A Search Session is group of queries and clicks demarcated with a 30-minute inactivity timeout, such as that used in previous work (Downey et al., 2007; Radlinski and Joachims, 2005).”
Dueling Posterior Sampling for Preference-Based Reinforcement Learning (2019, background): “lsby et al., 2011; Eric et al., 2008), which is focused exclusively on learning an accurate model rather than maximizing utility of decision-making; learning with more structured preference feedback (Radlinski and Joachims, 2005; Shivaswamy and Joachims, 2012; Raman et al., 2013; Shivaswamy and Joac”
Neural information retrieval: at the end of the early years (2017, background): “Over the years, IR has made tremendous progress by learning from user behavior, either by introducing, e.g., click-based rankers (Radlinski and Joachims 2005) or, more abstractly, by using models that capture behavioral notions such as examination probability and attractiveness of search results…”
Generating query substitutions (2006, methodology): “Others using user query sessions as a source of information to improve retrieval include Furnas [10], who explicitly asked users whether their reformulations should be used for indexing, and Radlinski and Joachims [15] who use query chains followed by clicks as a source of relevance ordering infor-”