Nick Craswell

Effective site finding using link anchor information

Link-based ranking methods have been described in the literature and applied in commercial Web search engines. However, according to recent TREC experiments, they are no better than traditional content-based methods. We conduct a different type of experiment, in which the task is to find the main entry point of a specific Web site. In our experiments, ranking based on link anchor text is twice as effective as ranking based on document content, even though both methods used the same BM25 formula. We obtained these results using two sets of 100 queries on a 18.5 million document set and another set of 100 on a 0.4 million document set. This site finding effectiveness begins to explain why many search engines have adopted link methods. It also opens a rich new area for effectiveness improvement, where traditional methods fail.

Measuring Search Engine Quality

The effectiveness of twenty public search engines is evaluated using TREC-inspired methods and a set of 54 queries taken from real Web search logs. The World Wide Web is taken as the test collection and a combination of crawler and text retrieval system is evaluated. The engines are compared on a range of measures derivable from binary relevance judgments of the first seven live results returned. Statistical testing reveals a significant difference between engines and high intercorrelations between measures. Surprisingly, given the dynamic nature of the Web and the time elapsed, there is also a high correlation between results of this study and a previous study by Gordon and Pathak. For nearly all engines, there is a gradual decline in precision at increasing cutoff after some initial fluctuation. Performance of the engines as a group is found to be inferior to the group of participants in the TREC-8 Large Web task, although the best engines approach the median of those systems. Shortcomings of current Web search evaluation methodology are identified and recommendations are made for future improvements. In particular, the present study and its predecessors deal with queries which are assumed to derive from a need to find a selection of documents relevant to a topic. By contrast, real Web search reflects a range of other information need types which require different judging and different measures.

Engineering a multi-purpose test collection for web retrieval experiments

Past research into text retrieval methods for the Web has been restricted by the lack of a test collection capable of supporting experiments which are both realistic and reproducible. The 1.69 million document WT10g collection is proposed as a multi-purpose testbed for experiments with these attributes, in distributed IR, hyperlink algorithms and conventional ad hoc retrieval.WT10g was constructed by selecting from a superset of documents in such a way that desirable corpus properties were preserved or optimised. These properties include: a high degree of inter-server connectivity, integrity of server holdings, inclusion of documents related to a very wide spread of likely queries, and a realistic distribution of server holding sizes. We confirm that WT10g contains exploitable link information using a site (homepage) finding experiment. Our results show that, on this task, Okapi BM25 works better on propagated link anchor text than on full text.WT10g was used in TREC-9 and TREC-2000 and both topic relevance and homepage finding queries and judgments are available.

Relevance weighting for query independent evidence

A query independent feature, relating perhaps to document content, linkage or usage, can be transformed into a static, per-document relevance weight for use in ranking. The challenge is to find a good function to transform feature values into relevance scores. This paper presents FLOE, a simple density analysis method for modelling the shape of the transformation required, based on training data and without assuming independence between feature and baseline. For a new query independent feature, it addresses the questions: is it required for ranking, what sort of transformation is appropriate and, after adding it, how successful was the chosen transformation? Based on this we apply sigmoid transformations to PageRank, indegree, URL Length and ClickDistance, tested in combination with a BM25 baseline.

Results and challenges in Web search evaluation

Abstract A frozen 18.5 million page snapshot of part of the Web has been created to enable and encourage meaningful and reproducible evaluation of Web search systems and techniques. This collection is being used in an evaluation framework within the Text Retrieval Conference (TREC) and will hopefully provide convincing answers to questions such as, “Can link information result in better rankings?”, “Do longer queries result in better answers?”, and, “Do TREC systems work well on Web data?” The snapshot and associated evaluation methods are described and an invitation is extended to participate. Preliminary results are presented for an effectivess comparison of six TREC systems working on the snapshot collection against five well-known Web search systems working over the current Web. These suggest that the standard of document rankings produced by public Web search engines is by no means state-of-the-art.

A comparative analysis of cascade measures for novelty and diversity

Traditional editorial effectiveness measures, such as nDCG, remain standard for Web search evaluation. Unfortunately, these traditional measures can inappropriately reward redundant information and can fail to reflect the broad range of user needs that can underlie a Web query. To address these deficiencies, several researchers have recently proposed effectiveness measures for novelty and diversity. Many of these measures are based on simple cascade models of user behavior, which operate by considering the relationship between successive elements of a result list. The properties of these measures are still poorly understood, and it is not clear from prior research that they work as intended. In this paper we examine the properties and performance of cascade measures with the goal of validating them as tools for measuring effectiveness. We explore their commonalities and differences, placing them in a unified framework; we discuss their theoretical difficulties and limitations, and compare the measures experimentally, contrasting them against traditional measures and against other approaches to measuring novelty. Data collected by the TREC 2009 Web Track is used as the basis for our experimental comparison. Our results indicate that these measures reward systems that achieve an balance between novelty and overall precision in their result lists, as intended. Nonetheless, other measures provide insights not captured by the cascade measures, and we suggest that future evaluation efforts continue to report a variety of measures.

Optimisation methods for ranking functions with multiple parameters

Optimising the parameters of ranking functions with respect to standard IR rank-dependent cost functions has eluded satisfactory analytical treatment. We build on recent advances in alternative differentiable pairwise cost functions, and show that these techniques can be successfully applied to tuning the parameters of an existing family of IR scoring functions (BM25), in the sense that we cannot do better using sensible search heuristics that directly optimize the rank-based cost function NDCG. We also demonstrate how the size of training set affects the number of parameters we can hope to tune this way.

Inferring query intent from reformulations and clicks

Many researchers have noted that web search queries are often ambiguous or unclear. We present an approach for identifying the popular meanings of queries using web search logs and user click behavior. We show our approach to produce more complete and user-centric intents than expert judges by evaluating on TREC queries. This approach was also used by the TREC 2009 Web Track judges to obtain more representative topic descriptions from real queries.

Comparing the sensitivity of information retrieval metrics

Information retrieval effectiveness is usually evaluated using measures such as Normalized Discounted Cumulative Gain (NDCG), Mean Average Precision (MAP) and Precision at some cutoff (Precision@k) on a set of judged queries. Recent research has suggested an alternative, evaluating information retrieval systems based on user behavior. Particularly promising are experiments that interleave two rankings and track user clicks. According to a recent study, interleaving experiments can identify large differences in retrieval effectiveness with much better reliability than other click-based methods. We study interleaving in more detail, comparing it with traditional measures in terms of reliability, sensitivity and agreement. To detect very small differences in retrieval effectiveness, a reliable outcome with standard metrics requires about 5,000 judged queries, and this is about as reliable as interleaving with 50,000 user impressions. Amongst the traditional measures, NDCG has the strongest correlation with interleaving. Finally, we present some new forms of analysis, including an approach to enhance interleaving sensitivity.

Query-independent evidence in home page finding

Hyperlink recommendation evidence, that is, evidence based on the structure of a webs link graph, is widely exploited by commercial Web search systems. However there is little published work to support its popularity. Another form of query-independent evidence, URL-type, has been shown to be beneficial on a home page finding task. We compared the usefulness of these types of evidence on the home page finding task, combined with both content and anchor text baselines. Our experiments made use of five query sets spanning three corpora---one enterprise crawl, and the WT10g and VLC2 Web test collections.We found that, in optimal conditions, all of the query-independent methods studied (in-degree, URL-type, and two variants of PageRank) offered a better than random improvement on a content-only baseline. However, only URL-type offered a better than random improvement on an anchor text baseline. In realistic settings, for either baseline, only URL-type offered consistent gains. In combination with URL-type the anchor text baseline was more useful for finding popular home pages, but URL-type with content was more useful for finding randomly selected home pages. We conclude that a general home page finding system should combine evidence from document content, anchor text, and URL-type classification.