Inferring Human Activity Preferences by Modeling Human Decision Segments
Author
Dosenovic, David
Term
4. term
Education
Publication year
2019
Submitted on
2019-06-12
Pages
68
Abstract
Denne afhandling undersøger, hvordan man kan modellere menneskers aktivitetspræferencer—hvad de vælger at gøre—ved hjælp af digitale spor fra en lokationsbaseret tjeneste. Studiet analyserer et historisk Foursquare-datasæt med check-ins fra Tokyo over ti og en halv måned. Afhandlingen foreslår, at aktivitetsvalg påvirkes af 'beslutningssegmenter', forstået som tilbagevendende sammenhænge eller grupperinger, der former beslutninger. For at indfange disse præferencer anvendes en multilags perceptron, en type neuralt netværk. To modelleringsmetoder sammenlignes, hver med et forskelligt sæt input-features til at repræsentere mulige beslutningssegmenter. Modellerne evalueres, og resultaterne præsenteres.
This thesis investigates how to model people’s activity preferences—what they choose to do—using digital traces from a location-based service. The study analyzes a historical Foursquare dataset of check-ins collected in Tokyo over ten and a half months. It proposes that activity choices are influenced by 'decision segments', understood as recurring contexts or groupings that shape decisions. To capture these preferences, the thesis applies a multilayer perceptron, a type of neural network. Two modeling approaches are compared, each using a different set of input features to represent possible decision segments. The models are evaluated, and the findings are reported.
[This abstract was generated with the help of AI]
Keywords
machine learning ; ai ; neural networks ; behavior ; data analysis ; data visualization ; modeling ; model ; it ; computer science
Documents