The sum of its parts: reducing sparsity in click estimation with query segments

The critical task of predicting clicks on search advertisements is typically addressed by learning from historical click data. When enough history is observed for a given query-ad pair, future clicks can be accurately modeled. However, based on the empirical distribution of queries, sufficient historical information is unavailable for many query-ad pairs. The sparsity of data for new and rare queries makes it difficult to accurately estimate clicks for a significant portion of typical search engine traffic. In this paper we provide analysis to motivate modeling approaches that can reduce the sparsity of the large space of user search queries. We then propose methods to improve click and relevance models for sponsored search by mining click behavior for partial user queries. We aggregate click history for individual query words, as well as for phrases extracted with a CRF model. The new models show significant improvement in clicks and revenue compared to state-of-the-art baselines trained on several months of query logs. Results are reported on live traffic of a commercial search engine, in addition to results from offline evaluation.     

Variational bayes for modeling score distributions

Empirical modeling of the score distributions associated with retrieved documents is an essential task for many retrieval applications. In this work, we propose modeling the relevant documents’ scores by a mixture of Gaussians and the non-relevant scores by a Gamma distribution. Applying Variational Bayes we automatically trade-off the goodness-of-fit with the complexity of the model. We test our model on traditional retrieval functions and actual search engines submitted to TREC. We demonstrate the utility of our model in inferring precision-recall curves. In all experiments our model outperforms the dominant exponential-Gaussian model.     

Retrieval constraints and word frequency distributions a log-logistic model for IR

We first present in this paper an analytical view of heuristic retrieval constraints which yields simple tests to determine whether a retrieval function satisfies the constraints or not. We then review empirical findings on word frequency distributions and the central role played by burstiness in this context. This leads us to propose a formal definition of burstiness which can be used to characterize probability distributions with respect to this phenomenon. We then introduce the family of information-based IR models which naturally captures heuristic retrieval constraints when the underlying probability distribution is bursty and propose a new IR model within this family, based on the log-logistic distribution. The experiments we conduct on several collections illustrate the good behavior of the log-logistic IR model: It significantly outperforms the Jelinek-Mercer and Dirichlet prior language models on most collections we have used, with both short and long queries and for both the MAP and the precision at 10 documents. It also compares favorably to BM25 and has similar performance to classical DFR models such as InL2 and PL2.     

Managing the Progressive Publishing Company During Market and Technology Transitions

As new media technologies and platforms emerge and take hold in our society, traditional publishers are wondering: What’s in this new content climate for me? The simple answer is: a lot. The digital world, mobile content delivery mechanisms and the public’s increasing comfort—even preference for—a media menu from which they can pick and choose what they want and how they want to receive it, brings exciting and potentially lucrative opportunities. For publishers who understand how to leverage their brand and create authentic, identifiable value in the eyes of the customer, risk can be reduced and new revenue streams built. Here are four best practices to position your publishing company for growth.     

Popular E-Content at The New York Public Library: Successes and Challenges

Updated from a presentation given at Biblionext.it in Rome in April 2011, this article will highlight The New York Public Library’s success with e-books and other forms of popular e-content and our efforts to stay one step ahead of the consumer shift from print reading to e-reading. Consumer e-reading is dominated by Amazon.com in the US, followed by one of the largest chain bookstores, BarnesandNoble.com. The availability of digital versions of very popular titles, coupled with the explosion of e-readers, tablets, and smartphones that are priced competitively and fairly easy to use, are helping move a lot of Americans into the e-book world. Last month, Amazon.com announced they sold more e-books than physical books for the first time ever. Print books are not going away, but our experience is that it is clear e-books are no longer just an extra format to offer, they are integral to our future.     

Is there an ethics of algorithms?

We argue that some algorithms are value-laden, and that two or more persons who accept different value-judgments may have a rational reason to design such algorithms differently. We exemplify our claim by discussing a set of algorithms used in medical image analysis: In these algorithms it is often necessary to set certain thresholds for whether e.g. a cell should count as diseased or not, and the chosen threshold will partly depend on the software designer’s preference between avoiding false positives and false negatives. This preference ultimately depends on a number of value-judgments. In the last section of the paper we discuss some general principles for dealing with ethical issues in algorithm-design.     

Taking a Different Route: Contribution of Articulation and Metacognition to Intervention With At-Risk Third-Grade Readers

This study compared the impact of 2 types of small-group interventions targeting below-level 3rd-grade students. The study compared WordWork decoding and spelling instruction (Calfee, Miller, Norman, Wilson, &; Trainin, 2006 Calfee, R. C., Miller, R. G., Norman, K., Wilson, K., &; Trainin, G. (2006). Learning to do educational research. In M. A. Constas &; R. J. Sterenberg (Eds.), Translating theory and research into educational practice: Developments in content domains, large scale reform, and intellectual capacity (pp. 77–103). New York, NY: Routledge. [Google Scholar]; Calfee &; Patrick, 1995 Calfee, R. C., &; Patrick, C. L. (1995). Teach our children well: Bringing K–12 education into the 21st Century. Stanford, CA: Stanford Alumni Assoc. [Google Scholar]), which includes attention to articulation and metacognition, with a more traditional phonological awareness program (Torgesen &; Bryant, 1993 Torgesen, J. K., &; Bryant, B. R. (1993). Phonological awareness training for reading. Austin, TX: Pro-Ed. [Google Scholar]). University education students delivered the interventions with equal instructional time and fidelity, to both groups. Students in both intervention groups engaged in repeated readings of connected text to promote transfer. The impact of training was assessed through decoding, spelling, and oral reading fluency measures. Results indicated that WordWork produced more positive results in decoding, spelling, and fluency, and had significantly fewer treatment resisters.