|
Descripción
|
|
|---|---|
| The customization of applications and services today is an example of the importance of being aware of the user, her behavior and her context by the service providers. Particularize for each user is a complex task so that machine learning algorithms are used to classify or cluster users in profiles based on parameters that users of the same group share or whose difference is minimal. In this way, there are user groups or clusters that roughly define the behavior of every user in the same cluster from a series of parameters that describe a model. In this work, since the main line of research carried out previously was focused on indoor localization, modeling the users behavior from indoor users trajectories has been proposed. This modeling is based on both Markov and non-Markov models. Markov models only take into account the current state of the user position to estimate the next position but non-Markov models are memory-aware and take into account previous states as well as the current one. Proposed models, and more particularly the Markov model and the model with memory 1, which only takes into account the immediately previous and the current states to estimate the following user position, are evaluated in this work to check how models could represent, in an indirect way, the behavior of users by groups. The evaluation of the models is done using a classification algorithm, which is not evaluated in this work, and classifying trajectories that are generated from the proposed models and evaluating the percentage of correctly classified trajectories according to the following two parameters: number of trajectories needed to generate the proposed models and number of trajectories used to train the classifier. In addition, it is intended to estimate the number of trajectories necessary to ensure, with a certain confidence degree, the profile a user belongs to if she is identified and classified from her trajectories. As trajectories are defined as a succession of positions, the symbolic location paradigm has been addressed, which allows us to estimate the user position using wireless technology when indoors. Thus, three algorithms are proposed, which are easy and quick to implement, calibrate and configure, and enable the deployment of a localization system in real environments in a practical way. Within the possible wireless technologies, we decided to use a Bluetooth Low Energy beacon-based system that have a great autonomy, what enables to have a stable infrastructure and with little maintenance, non invasive and easy to deploy; while the user carries a mobile device that runs a localization application that detects the beacons and estimate its own position. The proposed algorithms have been designed, developed and evaluated to estimate their usefulness in controlled scenarios and their possible installation in real scenarios. In addition, it has been proposed the design and development of a mobile application of personalization from the user position, where the localization algorithms proposed throughout this work have been implemented. On one hand, the actual implementation in a controlled environment has been used for its subsequent deployment in a demo environment, where banking or retail businesses could be installed and use the localization system for, for example, the personalization of services as a function of the position of the user from their trajectories; or in a real environment like a real supermarket, in Nice, where an infrastructure has been deployed in part of the enclosure for the evaluation of the complete solution in an uncontrolled environment. On the other hand, this application offers tools for user tracking and construction of paths or sending personalized notifications depending on the position of the user, as well as an informative dashboard that allows to know the position data in each zone of the different users. | |
|
Internacional
|
No |
|
ISBN
|
|
|
Tipo de Tesis
|
Doctoral |
|
Calificación
|
Sobresaliente cum laude |
|
Fecha
|
12/09/2017 |