Summary We introduce and test several micro‐ and macro‐optimizations to the Score‐at‐a‐Time approach to processing impact‐ordered postings lists in a search engine. Our micro‐optimizations are at the single‐assembly instruction level, but our macro‐optimizations are algorithmic. Overall, we see an improvement of 37% on our baseline (22% on state of the art). We experiment with parallel search and present evidence that we have hit the memory wall.

Large scale retrieval systems often employ cascaded ranking architectures, in which an initial set of candidate documents are iteratively refined and re-ranked by increasingly sophisticated and expensive ranking models. In this paper, we propose a unified framework for predicting a range of performance-sensitive parameters based on minimizing end-to-end effectiveness loss. The framework does not require relevance judgments for training, is amenable to predicting a wide range of parameters, allows for fine tuned efficiency-effectiveness trade-offs, and can be easily deployed in large scale search systems with minimal overhead. As a proof of concept, we show that the framework can accurately predict a number of performance parameters on a query-by-query basis, allowing efficient and effective retrieval, while simultaneously minimizing the tail latency of an early-stage candidate generation system. On the 50 million document ClueWeb09B collection, and across 25,000 queries, our hybrid system can achieve superior early-stage efficiency to fixed parameter systems without loss of effectiveness, and allows more finely-grained efficiency-effectiveness trade-offs across the multiple stages of the retrieval system.

“Based on theoretical reasoning it has been suggested that the reliability of findings published in the scientific literature decreases with the popularity of a research field” (Pfeiffer and Hoffmann, 2009). As we know, deep learning is very popular and the ability to reproduce results is an important part of science. There is growing concern within the deep learning community about the reproducibility of results that are presented. In this paper we present a number of controllable, yet unreported, effects that can substantially change the effectiveness of a sample model, and thusly the reproducibility of those results. Through these environmental effects we show that the commonly held belief that distribution of source code is all that is needed for reproducibility is not enough. Source code without a reproducible environment does not mean anything at all. In addition the range of results produced from these effects can be larger than the majority of incremental improvement reported.

As an empirical discipline, information access and retrieval research requires substantial software infrastructure to index and search large collections. This workshop is motivated by the desire to better align information retrieval research with the practice of building search applications from the perspective of open-source information retrieval systems. Our goal is to promote the use of Lucene for information access and retrieval research.

We present an empirical comparison between document-at-a-time (DaaT) and score-at-a-time (SaaT) document ranking strategies within a common framework. Although both strategies have been extensively explored, the literature lacks a fair, direct comparison: such a study has been difficult due to vastly different query evaluation mechanics and index organizations. Our work controls for score quantization, document processing, compression, implementation language, implementation effort, and a number of details, arriving at an empirical evaluation that fairly characterizes the performance of three specific techniques: WAND (DaaT), BMW (DaaT), and JASS (SaaT). Experiments reveal a number of interesting findings. The performance gap between WAND and BMW is not as clear as the literature suggests, and both methods are susceptible to tail queries that may take orders of magnitude longer than the median query to execute. Surprisingly, approximate query evaluation in WAND and BMW does not significantly reduce the risk of these tail queries. Overall, JASS is slightly slower than either WAND or BMW, but exhibits much lower variance in query latencies and is much less susceptible to tail query effects. Furthermore, JASS query latency is not particularly sensitive to the retrieval depth, making it an appealing solution for performance-sensitive applications where bounds on query latencies are desirable.

The time cost of searching with an inverted index is directly proportional to the number of postings processed and the cost of processing each posting. Dynamic pruning reduces the number of postings examined. Pre-calculation then quantization of term / document weights reduces the cost of evaluating each posting. The effect of quantization on precision, latency, and index size is examined herein. We show empirically that there is an ideal size (in bits) for storing the quantized scores. Increasing this adversely affects index size and search latency; decreasing it adversely affects precision. We observe a relationship between the collection size and ideal quantization size, and provide a way to determine the number of bits to use from the collection size.

The need for a search engine to deal with ambiguous queries has been known for a long time (diversification). However, it is only recently that this need has become a focus within information retrieval research. How to respond to indications that a result is relevant to a query (relevance feedback) has also been a long focus of research. When thinking about the results for a query as being clustered by topic, these two areas of information retrieval research appear to be opposed to each other. Interestingly though, they both appear to improve the performance of search engines, raising the question: they can be combined or made to work with each other?