Eric C. Jensen

Hourly analysis of a very large topically categorized web query log

We review a query log of hundreds of millions of queries that constitute the total query traffic for an entire week of a general-purpose commercial web search service. Previously, query logs have been studied from a single, cumulative view. In contrast, our analysis shows changes in popularity and uniqueness of topically categorized queries across the hours of the day. We examine query traffic on an hourly basis by matching it against lists of queries that have been topically pre-categorized by human editors. This represents 13% of the query traffic. We show that query traffic from particular topical categories differs both from the query stream as a whole and from other categories. This analysis provides valuable insight for improving retrieval effectiveness and efficiency. It is also relevant to the development of enhanced query disambiguation, routing, and caching algorithms.

Repeatable evaluation of search services in dynamic environments

In dynamic environments, such as the World Wide Web, a changing document collection, query population, and set of search services demands frequent repetition of search effectiveness (relevance) evaluations. Reconstructing static test collections, such as in TREC, requires considerable human effort, as large collection sizes demand judgments deep into retrieved pools. In practice it is common to perform shallow evaluations over small numbers of live engines (often pairwise, engine A vs. engine B) without system pooling. Although these evaluations are not intended to construct reusable test collections, their utility depends on conclusions generalizing to the query population as a whole. We leverage the bootstrap estimate of the reproducibility probability of hypothesis tests in determining the query sample sizes required to ensure this, finding they are much larger than those required for static collections. We propose a semiautomatic evaluation framework to reduce this effort. We validate this framework against a manual evaluation of the top ten results of ten Web search engines across 896 queries in navigational and informational tasks. Augmenting manual judgments with pseudo-relevance judgments mined from Web taxonomies reduces both the chances of missing a correct pairwise conclusion, and those of finding an errant conclusion, by approximately 50%.

Automatic classification of Web queries using very large unlabeled query logs

Accurate topical classification of user queries allows for increased effectiveness and efficiency in general-purpose Web search systems. Such classification becomes critical if the system must route queries to a subset of topic-specific and resource-constrained back-end databases. Successful query classification poses a challenging problem, as Web queries are short, thus providing few features. This feature sparseness, coupled with the constantly changing distribution and vocabulary of queries, hinders traditional text classification. We attack this problem by combining multiple classifiers, including exact lookup and partial matching in databases of manually classified frequent queries, linear models trained by supervised learning, and a novel approach based on mining selectional preferences from a large unlabeled query log. Our approach classifies queries without using external sources of information, such as online Web directories or the contents of retrieved pages, making it viable for use in demanding operational environments, such as large-scale Web search services. We evaluate our approach using a large sample of queries from an operational Web search engine and show that our combined method increases recall by nearly 40% over the best single method while maintaining adequate precision. Additionally, we compare our results to those from the 2005 KDD Cup and find that we perform competitively despite our operational restrictions. This suggests it is possible to topically classify a significant portion of the query stream without requiring external sources of information, allowing for deployment in operationally restricted environments.

Improving automatic query classification via semi-supervised learning

Accurate topical classification of user queries allows for increased effectiveness and efficiency in general-purpose Web search systems. Such classification becomes critical if the system is to return results not just from a general Web collection but from topic-specific back-end databases as well. Maintaining sufficient classification recall is very difficult as Web queries are typically short, yielding few features per query. This feature sparseness coupled with the high query volumes typical for a large-scale search service makes manual and supervised learning approaches alone insufficient. We use an application of computational linguistics to develop an approach for mining the vast amount of unlabeled data in Web query logs to improve automatic topical Web query classification. We show that our approach in combination with manual matching and supervised learning allows us to classify a substantially larger proportion of queries than any single technique. We examine the performance of each approach on a real Web query stream and show that our combined method accurately classifies 46% of queries, outperforming the recall of best single approach by nearly 20%, with a 7% improvement in overall effectiveness.

Automatic web query classification using labeled and unlabeled training data

Accurate topical categorization of user queries allows for increased effectiveness, efficiency, and revenue potential in general-purpose web search systems. Such categorization becomes critical if the system is to return results not just from a general web collection but from topic-specific databases as well. Maintaining sufficient categorization recall is very difficult as web queries are typically short, yielding few features per query. We examine three approaches to topical categorization of general web queries: matching against a list of manually labeled queries, supervised learning of classifiers, and mining of selectional preference rules from large unlabeled query logs. Each approach has its advantages in tackling the web query classification recall problem, and combining the three techniques allows us to classify a substantially larger proportion of queries than any of the individual techniques. We examine the performance of each approach on a real web query stream and show that our combined method accurately classifies 46% of queries, outperforming the recall of the best single approach by nearly 20%, with a 7% improvement in overall effectiveness.

Fusion of effective retrieval strategies in the same information retrieval system

Prior efforts have shown that under certain situations retrieval effectiveness may be improved via the use of data fusion techniques. Although these improvements have been observed from the fusion of result sets from several distinct information retrieval systems, it has often been thought that fusing different document retrieval strategies in a single information retrieval system will lead to similar improvements. In this study, we show that this is not the case. We hold constant systemic differences such as parsing, stemming, phrase processing, and relevance feedback, and fuse result sets generated from highly effective retrieval strategies in the same information retrieval system. From this, we show that data fusion of highly effective retrieval strategies alone shows little or no improvement in retrieval effectiveness. Furthermore, we present a detailed analysis of the performance of modern data fusion approaches, and demonstrate the reasons why they do not perform well when applied to this problem. Detailed results and analyses are included to support our conclusions.

Varying approaches to topical web query classification

Topical classification of web queries has drawn recent interest because of the promise it offers in improving retrieval effectiveness and efficiency. However, much of this promise depends on whether classification is performed before or after the query is used to retrieve documents. We examine two previously unaddressed issues in query classification: pre versus post-retrieval classification effectiveness and the effect of training explicitly from classified queries versus bridging a classifier trained using a document taxonomy. Bridging classifiers map the categories of a document taxonomy onto those of a query classification problem to provide sufficient training data. We find that training classifiers explicitly from manually classified queries outperforms the bridged classifier by 48% in F1 score. Also, a pre-retrieval classifier using only the query terms performs merely 11% worse than the bridged classifier which requires snippets from retrieved documents.

Disproving the fusion hypothesis: an analysis of data fusion via effective information retrieval strategies

Many prior efforts have been devoted to the basic idea that data fusion techniques can improve retrieval effectiveness. Recent work in the area suggests that many approaches, particularly multiple-evidence combinations, can be a successful means of improving the effectiveness of a system. Unfortunately, the conditions favorable to effectiveness improvements have not been made clear. We examine popular data fusion techniques designed to achieve improvements in effectiveness and clarify the conditions required for data fusion to show improvement. We demonstrate that for fusion to improve effectiveness, the result sets being fused must contain a significant number of unique relevant documents. Furthermore, we show that for this improvement to be visible, these unique relevant documents must be highly ranked. In addition, we present a comprehensive discussion on why previous assumptions about the effectiveness of multiple-evidence techniques are misleading. Detailed empirical results and analysis are provided to support our conclusions.

Using titles and category names from editor-driven taxonomies for automatic evaluation

Evaluation of IR systems has always been difficult because of the need for manually assessed relevance judgments. The advent of large editor-driven taxonomies on the web opens the door to a new evaluation approach. We use the ODP (Open Directory Project) taxonomy to find sets of pseudo-relevant documents via one of two assumptions: 1) taxonomy entries are relevant to a given query if their editor-entered titles exactly match the query, or 2) all entries in a leaf-level taxonomy category are relevant to a given query if the category title exactly matches the query. We compare and contrast these two methodologies by evaluating six web search engines on a sample from an America Online log of ten million web queries, using MRR measures for the first method and precision-based measures for the second. We show that this technique is stable with respect to the query set selected and correlated with a reasonably large manual evaluation.

Predicting query difficulty on the web by learning visual clues

We describe a method for predicting query difficulty in a precision-oriented web search task. Our approach uses visual features from retrieved surrogate document representations (titles, snippets, etc.) to predict retrieval effectiveness for a query. By training a supervised machine learning algorithm with manually evaluated queries, visual clues indicative of relevance are discovered. We show that this approach has a moderate correlation of 0.57 with precision at 10 scores from manual relevance judgments of the top ten documents retrieved by ten web search engines over 896 queries. Our findings indicate that difficulty predictors which have been successful in recall-oriented ad-hoc search, such as clarity metrics, are not nearly as correlated with engine performance in precision-oriented tasks such as this, yielding a maximum correlation of 0.3. Additionally, relying only on visual clues avoids the need for collection statistics that are required by these prior approaches. This enables our approach to be employed in environments where these statistics are unavailable or costly to retrieve, such as metasearch.