UZH-Logo

From active towards itive learning: using consideration information to improve labeling correctness


Bernstein, A; Li, J (2009). From active towards itive learning: using consideration information to improve labeling correctness. In: Human Computation Workshop, Paris, France, June 2009 - June 2009.

Abstract

Data mining techniques have become central to many applications. Most of those applications rely on so called supervised learning algorithms, which learn from given examples in the form of data with predefined labels (e.g., classes such as spam, not spam). Labeling, however, is oftentimes expensive, as it typically requires manual work by human experts. Active learning systems reduce the human effort by choosing the most informative instances for labeling. Unfortunately, research in psychology has shown conclusively that human decisions are inaccurate, easily biased by circumstances, and far from the oracle decision making assumed in active learning research. Based on these findings we show experimentally that (human) mistakes in labeling can significantly deteriorate the performance of active learning systems. To solve this problem, we introduce consideration information – a concept from marketing – into an active learning system to bias and improve the human’s labeling performance. Results (with simulated and human labelers) show that consideration information can indeed be used to exert a bias. Furthermore, we find that the choice of appropriate consideration information can be used to positively bias an expert and thereby improving the overall performance of the learning setting.

Data mining techniques have become central to many applications. Most of those applications rely on so called supervised learning algorithms, which learn from given examples in the form of data with predefined labels (e.g., classes such as spam, not spam). Labeling, however, is oftentimes expensive, as it typically requires manual work by human experts. Active learning systems reduce the human effort by choosing the most informative instances for labeling. Unfortunately, research in psychology has shown conclusively that human decisions are inaccurate, easily biased by circumstances, and far from the oracle decision making assumed in active learning research. Based on these findings we show experimentally that (human) mistakes in labeling can significantly deteriorate the performance of active learning systems. To solve this problem, we introduce consideration information – a concept from marketing – into an active learning system to bias and improve the human’s labeling performance. Results (with simulated and human labelers) show that consideration information can indeed be used to exert a bias. Furthermore, we find that the choice of appropriate consideration information can be used to positively bias an expert and thereby improving the overall performance of the learning setting.

Downloads

61 downloads since deposited on 04 Feb 2010
26 downloads since 12 months
Detailed statistics

Additional indexing

Item Type:Conference or Workshop Item (Paper), refereed, original work
Communities & Collections:03 Faculty of Economics > Department of Informatics
Dewey Decimal Classification:000 Computer science, knowledge & systems
Language:English
Event End Date:June 2009
Deposited On:04 Feb 2010 11:50
Last Modified:05 Apr 2016 13:39
Permanent URL: http://doi.org/10.5167/uzh-25869

Download

[img]
Preview
Filetype: PDF
Size: 2MB

TrendTerms

TrendTerms displays relevant terms of the abstract of this publication and related documents on a map. The terms and their relations were extracted from ZORA using word statistics. Their timelines are taken from ZORA as well. The bubble size of a term is proportional to the number of documents where the term occurs. Red, orange, yellow and green colors are used for terms that occur in the current document; red indicates high interlinkedness of a term with other terms, orange, yellow and green decreasing interlinkedness. Blue is used for terms that have a relation with the terms in this document, but occur in other documents.
You can navigate and zoom the map. Mouse-hovering a term displays its timeline, clicking it yields the associated documents.

Author Collaborations