Watching inside the Screen: Digital Activity Monitoring for Task Recognition and Proactive Information Retrieval

Forskningsoutput: TidskriftsbidragArtikelVetenskapligPeer review

Sammanfattning

We investigate to what extent it is possible to infer a user’s work tasks by digital activity monitoring and use the task models for proactive information retrieval. Ten participants volunteered for the study, in which their computer screen was monitored and related logs were recorded for 14 days. Corresponding diary entries were collected to provide ground truth to the task detection method. We report two experiments using this data. The unsupervised task detection experiment was conducted to detect tasks using unsupervised topic modeling. The results show an average task detection accuracy of more than 70% by using rich screen monitoring data. The single-trial task detection and retrieval experiment utilized unseen user inputs in order to detect related work tasks and retrieve task-relevant information on-line. We report an average task detection accuracy of 95%, and the corresponding model-based document retrieval with Normalized Discounted Cumulative Gain of 98%. We discuss and provide insights regarding the types of digital tasks occurring in the data, the accuracy of task detection on different task types, and the role of using different data input such as application names, extracted keywords, and bag-of-words representations in the task detection process. We also discuss the implications of our results for ubiquitous user modeling and privacy.
Originalspråkengelska
Artikelnummer109
TidskriftProceedings of ACM on interactive, mobile, wearable and ubiquitous technologies
Volym1
Utgåva3
Antal sidor23
ISSN2474-9567
DOI
StatusPublicerad - 11 sep 2017
MoE-publikationstypA1 Tidskriftsartikel-refererad

Vetenskapsgrenar

  • 113 Data- och informationsvetenskap

Citera det här

@article{3027244dd4954e25b583651204816b29,
title = "Watching inside the Screen: Digital Activity Monitoring for Task Recognition and Proactive Information Retrieval",
abstract = "We investigate to what extent it is possible to infer a user’s work tasks by digital activity monitoring and use the task models for proactive information retrieval. Ten participants volunteered for the study, in which their computer screen was monitored and related logs were recorded for 14 days. Corresponding diary entries were collected to provide ground truth to the task detection method. We report two experiments using this data. The unsupervised task detection experiment was conducted to detect tasks using unsupervised topic modeling. The results show an average task detection accuracy of more than 70{\%} by using rich screen monitoring data. The single-trial task detection and retrieval experiment utilized unseen user inputs in order to detect related work tasks and retrieve task-relevant information on-line. We report an average task detection accuracy of 95{\%}, and the corresponding model-based document retrieval with Normalized Discounted Cumulative Gain of 98{\%}. We discuss and provide insights regarding the types of digital tasks occurring in the data, the accuracy of task detection on different task types, and the role of using different data input such as application names, extracted keywords, and bag-of-words representations in the task detection process. We also discuss the implications of our results for ubiquitous user modeling and privacy.",
keywords = "113 Computer and information sciences",
author = "Tung Vuong and Giulio Jacucci and Tuukka Ruotsalo",
year = "2017",
month = "9",
day = "11",
doi = "10.1145/3130974",
language = "English",
volume = "1",
journal = "Proceedings of ACM on interactive, mobile, wearable and ubiquitous technologies",
issn = "2474-9567",
publisher = "ACM",
number = "3",

}

TY - JOUR

T1 - Watching inside the Screen: Digital Activity Monitoring for Task Recognition and Proactive Information Retrieval

AU - Vuong, Tung

AU - Jacucci, Giulio

AU - Ruotsalo, Tuukka

PY - 2017/9/11

Y1 - 2017/9/11

N2 - We investigate to what extent it is possible to infer a user’s work tasks by digital activity monitoring and use the task models for proactive information retrieval. Ten participants volunteered for the study, in which their computer screen was monitored and related logs were recorded for 14 days. Corresponding diary entries were collected to provide ground truth to the task detection method. We report two experiments using this data. The unsupervised task detection experiment was conducted to detect tasks using unsupervised topic modeling. The results show an average task detection accuracy of more than 70% by using rich screen monitoring data. The single-trial task detection and retrieval experiment utilized unseen user inputs in order to detect related work tasks and retrieve task-relevant information on-line. We report an average task detection accuracy of 95%, and the corresponding model-based document retrieval with Normalized Discounted Cumulative Gain of 98%. We discuss and provide insights regarding the types of digital tasks occurring in the data, the accuracy of task detection on different task types, and the role of using different data input such as application names, extracted keywords, and bag-of-words representations in the task detection process. We also discuss the implications of our results for ubiquitous user modeling and privacy.

AB - We investigate to what extent it is possible to infer a user’s work tasks by digital activity monitoring and use the task models for proactive information retrieval. Ten participants volunteered for the study, in which their computer screen was monitored and related logs were recorded for 14 days. Corresponding diary entries were collected to provide ground truth to the task detection method. We report two experiments using this data. The unsupervised task detection experiment was conducted to detect tasks using unsupervised topic modeling. The results show an average task detection accuracy of more than 70% by using rich screen monitoring data. The single-trial task detection and retrieval experiment utilized unseen user inputs in order to detect related work tasks and retrieve task-relevant information on-line. We report an average task detection accuracy of 95%, and the corresponding model-based document retrieval with Normalized Discounted Cumulative Gain of 98%. We discuss and provide insights regarding the types of digital tasks occurring in the data, the accuracy of task detection on different task types, and the role of using different data input such as application names, extracted keywords, and bag-of-words representations in the task detection process. We also discuss the implications of our results for ubiquitous user modeling and privacy.

KW - 113 Computer and information sciences

U2 - 10.1145/3130974

DO - 10.1145/3130974

M3 - Article

VL - 1

JO - Proceedings of ACM on interactive, mobile, wearable and ubiquitous technologies

JF - Proceedings of ACM on interactive, mobile, wearable and ubiquitous technologies

SN - 2474-9567

IS - 3

M1 - 109

ER -