Recommending healthy bundles in a food context
Authors
Dyhrberg, Kristian Bjerglund ; Nielsen, Mark Vinsløv ; Eliasen, Bogi Hansen
Term
4. term
Education
Publication year
2018
Submitted on
2018-06-14
Pages
78
Abstract
Dette speciale undersøger, hvordan man kan anbefale sunde bundter af aftensmadopskrifter – altså grupper af opskrifter, der foreslås sammen. Specialet udvikler et sundhedsmål, der vurderer opskrifter ud fra fordelingen af deres ingredienser. En eksisterende algoritme til at samle opskrifter i bundter tilpasses, så den kan kombinere sundhedsmålet med en trænet SVD-model (en maskinlæringsmetode, der finder mønstre i præferencer). Forskellige versioner af algoritmen testes i et offline-setup for at se, hvilken der klarer sig bedst. Derudover sammenlignes SVD-modellens nøjagtighed med en KNN-algoritme (k-nærmeste naboer) for at vurdere, om KNN er et brugbart alternativ. Fordi bundter kan være udfordrende at præsentere, gennemføres også et brugerstudie med 16 deltagere, der afprøver tre forskellige layouts for at finde det, der opfattes bedst.
This thesis explores how to recommend healthy bundles of dinner recipes—groups of recipes suggested together. It introduces a health measure that scores recipes based on the balance of their ingredients. A bundling algorithm from prior work is adapted to incorporate this health score together with a trained SVD model (a machine-learning method that learns patterns in preferences). Several versions of the algorithm are evaluated in an offline setting to see which performs best. The accuracy of the SVD model is also compared with a KNN approach (k-nearest neighbors) to assess whether KNN is a viable alternative. Because bundles can be challenging to present, a user study with 16 participants tests three different layouts to find which is perceived best.
[This abstract was generated with the help of AI]
Keywords
Documents