Crowdsourcing for Query Processing onWeb Data: A Case Study on the Skyline Operator

Abstract

In recent years, crowdsourcing has become a powerful tool to bring human intelligence into information processing. This is especially important for Web data which in contrast to well-maintained databases is almost always incomplete and may be distributed over a variety of sources. Crowdsourcing allows to tackle many problems which are not yet attainable using machine-based algorithms alone: in particular, it allows to perform database operators on incomplete data as human workers can be used to provide values during runtime. As this can become costly quickly, elaborate optimization is required. In this paper, we showcase how such optimizations can be performed for the popular skyline operator for preference queries. We present some heuristics-based approaches and compare them to crowdsourcing-based approaches using sophisticated optimization techniques while especially focusing on result correctness.

Skyline Queries Computation on Crowdsourced- Enabled Incomplete Database

“…The crowd-sourcing strategy is based on incorporating human workers to attain improved results, while the advanced heuristic offers an alternative offline solution for times when crowd-sourcing may not be a feasible option, for example when the missing data are not easily available for the crowd or the costs of crowd-sourcing are prohibitive. We conclude that the approach introduced in [4] has failed to generate an accurate value when the missing rate is very high. Besides, the approach incurred high time latency and monetary cost when estimating the missing values from the crowd.…”

“…However, the correctness of the skyline might be deteriorated when relying on heuristics rules to identify the skylines. It may be that certain dominated tuples are included in the skyline results (false positive) and/or certain tuples which should be included in the skyline results are omitted (false negative) [4]. Furthermore, it is most likely that the relative error between the actual and the estimated values become very high if heuristics rules cannot capture the semantic relationship between the attributes; also if the missing rate in the database is high as it impacts the quality of the estimated values.…”

“…These activities negatively influence the database contents and deteriorate their quality [1], [2], [3]. These factors impact on the completeness and the correctness of the query result [4], [5], [6], [7]. Some queries cannot be optimally answered through traditional database management techniques as the process of answering certain queries relies on information that is incomplete, imprecise, or uncertain.…”

“…In 2013 the work in [6] proposed a new approach aiming to overcome the limitation of the previous work by proposing a different heuristic approach than KNN. The new approach adopts a model named minimum value model and has been further extended in 2015 [4] by adding the two new strategies of crowdsourcing and advanced heuristics. The crowd-sourcing strategy is based on incorporating human workers to attain improved results, while the advanced heuristic offers an alternative offline solution for times when crowd-sourcing may not be a feasible option, for example when the missing data are not easily available for the crowd or the costs of crowd-sourcing are prohibitive.…”

Preference Inference Based on Pareto Models

“…For Pareto models, PDP and PCP are not mutually expressive. While Pareto orders are widely studied in fields like voting theory [10] (unanimity), allocation problems [1] (Pareto optimality), decision making, database queries [2,7] (skyline operator) and economics (Pareto efficiency), there exists no general study of PDP or PCP based on Pareto orders so far. Pareto orders give a natural way of comparing alternatives; one alternative is better than another if it is better on all relevant evaluation functions (different criteria by which the alternatives can be evaluated).…”