Efficient query evaluation using a two-level retrieval process

By Andrei Z. Broder, David Carmel, Michael Herscovici, Aya Soffer, Jason Y. Zien. 2003

468 citations recorded by the source corpus.

Subfields: Computer Networks and Communications, Information Systems, 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 Networks and Communications

In Information Systems

In Signal Processing

Citations in the corpus

Graph Convolutional Neural Networks for Web-Scale Recommender Systems (2018, methodology): “After the hash function is computed, retrieval of items can be implemented with a two-level retrieval process based on the Weak AND operator [5].”
A Survey on Trajectory Data Management, Analytics, and Learning (2020, background): “Spatial -textual KS [25, 52] keywords points TF-IDF Inverted list Top-: T:SK [105, 195] point, keywords points Aggregate IR-tree T:STT [140, 202] points, keywords trajectories Aggregate Grid-index, Inverted list”
A General SIMD-Based Approach to Accelerating Compression Algorithms (2015, methodology): “For OR queries, we considered WAND [Broder et al. 2003; Ding and Suel 2011] and MaxScore [Jonassen et al. 2011; Shan et al. 2012].”
Up next: retrieval methods for large scale related video suggestion (2014, methodology): “Since we cast the video suggestion problem as a retrieval over an inverted index, we can use query optimization algorithms [8, 14] to efficiently find the optimal video suggestions even in a very large corpus.”
Efficiency Implications of Term Weighting for Passage Retrieval (2020, methodology): “Along with a typical exhaustive ranked disjunction ( RankedOR ), we employ a number of efficient dynamic pruning algorithms, including MaxScore , WAND , and BMW with fixed-sized blocks of 40 elements.”
Effectiveness/efficiency tradeoffs for candidate generation in multi-stage retrieval architectures (2013, background): “By fixing the feature generation and reranking stages, we are able to isolate the end-to-end effectiveness and efficiency implications of different algorithms.”
Assessing efficiency–effectiveness tradeoffs in multi-stage retrieval systems without using relevance judgments (2015, methodology): “2000, 2001) which is easily amenable to a blockcompressed inverted index, and can be processed in a document-at-a-time manner similar to WAND (Broder et al. 2003).”
Query Driven Algorithm Selection in Early Stage Retrieval (2017, methodology): “An efficient method for disjunctive DaaT processing is theWeak-AND (Wand) algorithm [8].”
Condensed List Relevance Models (2015, methodology): “present a two-pass method for efficiently scoring documents for standard retrieval [3].”
Contextual Memory Trees (2018, background): “A basic approach for text documents is an inverted index (Knuth, 1997; Broder et al., 2003), which indexes a document by the words that appear therein.”
Data Matching (2012, background): “Real-time data matching has much in common with the functionality of text and Web search engines [21, 43, 303]: real-time and approximate matching and ranking of results.”
Learning joint query interpretation and response ranking (2012, background): “The index supports semistructured queries specified by: • an answer type t from among the 200,000 YAGO types, • a bag of words and phrases in a IDF-WAND (weakand) operator [11], and • a snippet window width.”
From MAXSCORE to Block-Max Wand: The Story of How Lucene Significantly Improved Query Evaluation Performance (2020, methodology): “This led to an effort to implement a general-purpose Wand [3], based on a previous implementation for BooleanQuery.”
Towards an Efficient and Effective Search Engine (2012, background): “[6] addressed this random scan problem by introducing a new document-at-a-time query processing algorithm called WAND which can smartly skip some unnecessary postings for fast scanning.”
From Puppy to Maturity: Experiences in Developing Terrier (2012, background): “In addition, in WAND, skipping is also supported by underlying posting list iterators in order to reduce disk IO and further increase efficiency.”
Knowledge Graph and Corpus Driven Segmentation and Answer Inference for Telegraphic Entity-seeking Queries (2014, background): “The index returned about 330,000 snippets on average for WAND threshold of 0.6.”