Please use this identifier to cite or link to this item: http://hdl.handle.net/10397/85174
DC FieldValueLanguage
dc.contributorDepartment of Computing-
dc.creatorShen, Jiaxing-
dc.identifier.urihttps://theses.lib.polyu.edu.hk/handle/200/10083-
dc.language.isoEnglish-
dc.titleData-driven analytics of human dynamics using privacy-sensitive data-
dc.typeThesis-
dcterms.abstractHuman dynamics is interdisciplinary research which has been extensively investigated in various disciplines from different dimensions. As a result, it leads to somewhat different research focuses like human mobility, international and domestic migration, and population change. In this dissertation, we focus on human dynamics in computer science which refers to human activities and human interactions. The rapid development of digital information technologies, like communication technology, sensing technology, and mobile technology, has enabled a mobile and big data era for human dynamics research. These technologies keep track of our lives with digital records of places we go, products we buy, and people we meet. Human dynamics research with data from limited observations or confined experiments has transformed into tons of data records on human communications, interactions, and activities in the naturalistic environment. In Chapter 2, we study the possibility of user profile inference using privacy-sensitive audio. The contributions are three folds. First, we propose a privacy-sensitive modality for gender identification. The effectiveness and robustness are improved by ensemble feature selection and a two-stage classification. Second, an adaptive correlation-based multichannel VAD algorithm for privacy-sensitive audio is proposed. Last, we bring new insights of gender difference in interruption through analysis of group conversation in natural settings. In Chapter 3, we utilize the WiFi data to infer relational contextual information. One of our contributions is an effective heuristic that could significantly improve the detection performance of shopping groups. The heuristic indicates APs under which groups appear more frequently and barely separate should have larger weights in measuring customer similarity. The second contribution is to apply matrix factorization to detect groups without extra clustering processes. Matrix factorization could properly handle data issues in the measured similarity including noise filtering and data completion. Besides, imposing a sparsity constraint to the factorization process could derive the clustering results directly. In Chapter 4, we explore the relative contextual information based on the WiFi data and study the impact of human presence on wireless coverage. We identified the correlation between wireless coverage and the number of on-site people. Another contribution is the two observations of heuristics which could improve room-level localization. On the one hand, the duration of visit in different shops is different. On the other, different shops have different popularity in attracting customers at different time slots. These two features can be exploited to distinguish locations with similar wireless fingerprints.-
dcterms.accessRightsopen access-
dcterms.educationLevelPh.D.-
dcterms.extentxvii, 120 pages : color illustrations-
dcterms.issued2019-
dcterms.LCSHHong Kong Polytechnic University -- Dissertations-
dcterms.LCSHData mining-
dcterms.LCSHBig data-
dcterms.LCSHInformation science -- Social aspects-
dcterms.LCSHComputer networks -- Social aspects-
dcterms.LCSHInformation science -- Statistical methods-
Appears in Collections:Thesis
Show simple item record

Page views

46
Last Week
1
Last month
Citations as of Mar 24, 2024

Google ScholarTM

Check


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.