Abstract
This paper addresses the need to enhance transparency in ambient intelligent environments by developing more natural ways of interaction, which allow the users to communicate easily with the hidden networked devices rather than embedding obtrusive tablets and computing equipment throughout their surroundings. Ambient intelligence vision aims to realize digital environments that adapt to users in a responsive, transparent, and context-aware manner in order to enhance users' comfort. It is, therefore, appropriate to employ the paradigm of "computing with words" (CWWs), which aims to mimic the ability of humans to communicate transparently and manipulate perceptions via words. One of the daily activities that would increase the comfort levels of the users (especially people with disabilities) is cooking and performing tasks in the kitchen. Existing approaches on food preparation, cooking, and recipe recommendation stress on healthy eating and balanced meal choices while providing limited personalization features through the use of intrusive user interfaces. Herein, we present an application, which transparently interacts with users based on a novel CWWs approach in order to predict the recipe's difficulty level and to recommend an appropriate recipe depending on the user's mood, appetite, and spare time. The proposed CWWs framework is based on linear general type-2 (LGT2) fuzzy sets, which linearly quantify the linguistic modifiers in the third dimension in order to better represent the user perceptions while avoiding the drawbacks of type-1 and interval type-2 fuzzy sets. The LGT2-based CWWs framework can learn from user experiences and adapt to them in order to establish more natural human-machine interaction. We have carried numerous real-world experiments with various users in the University of Essex intelligent flat. The comparison analysis between interval type-2 fuzzy sets and LGT2 fuzzy sets demonstrates up to 55.43% improvement when general type-2 fuzzy sets are used than when interval type-2 fuzzy sets are used instead. The quantitative and qualitative analysis both show the success of the system in providing a natural interaction with the users for recommending food recipes where the quantitative analysis shows the high statistical correlation between the system output and the users' feedback; the qualitative analysis presents social science evaluation confirming the strong user acceptance of the system.