**5. Conclusions**

Generating targeted, different, and personalized advertising to recommend a product to the customer in an unconventional way is not a trivial task. The main reason that customer tastes can become so complex is that they become unique and unrepeatable preferences; for this reason, achieving generalization of the entire process within the system is highly complex in terms of development and effectiveness.

The raw material of this work are the images in which the clients appear. The preprocessing of the images that will be entered into the deep learning models is one of the main aspects that will influence obtaining good results, since having an adequate processing in the data will allow the model to obtain a result with high precision using fewer resources.

The personalized retraining of the Single Shot Detection with the help of transfer of learning (in order to achieve good results without the need for extensive training) and its results, as well as the execution tests of the convolutional neural network to estimate age and gender and the neural network for Big Five confirm something that is very clear in the world of artificial intelligence: No model will be completely accurate in scenarios of the real world. There is no model that is perfect, that is why the field of deep learning has had periods of oblivion throughout history, and although recently, thanks to more powerful computers, as well as large data banks, it has been a relevant of study, since precisely obtaining an ideal model is one of the objectives that these fields seek to achieve.

With all this, face detection has an acceptable performance for the purposes of the system, although it could be improved, since it must be observed that the almost null existence of data sets available for commercial use or with free licenses is the main cause of not being able to refine or perfect the Single Shot Detection to detect faces in very difficult conditions, such as, people with glasses (whether dark or transparent), with hats, scarves, with tied hair in the case of women, and recently with a mouth mask. The recommendation generated by the system, in the end, is a suggestion based on certain parameters identified in a person, and clearly the ultimate decision to accept or reject it will be with the clients. The importance of this works lies in the "aggressiveness" in which it is suggested, and since it is simply a graphic that does not compromise the decision or intentions of the buyer, in addition to how attractive augmented reality can be for a public unfamiliar with this technology, it is considered to be more likely to arouse interest in Bubble Town®beverages rather than having some rejection or negative impact due to a breach of their personal data.

A complementary part of this system is to take into account the need to safeguard people's sensitive data (faces), stored in its database, to comply with business rules, and not incur any violations to Federal Law on the Protection of Personal Data. For this reason, a privacy notice is provided along with the system, and information that could be considered sensitive is encrypted to prevent its misuse.

**Author Contributions:** Conceptualization, methodology, M.A.M.-A. and C.A.D.; investigation and resources, M.A.M.-A., H.C. and C.A.D.; software, visualization and data curation, A.L.-C., E.R.- D. and V.L.M.-F.; validation H.C.; formal analysis, M.A.M.-A., C.A.D. and H.C.; writing–original draft preparation, M.A.M.-A. and A.L.-C.; writing–review and editing, H.C., E.R.-D. and V.L.M.-F.; supervision, M.A.M.-A., C.A.D. and H.C.; project administration and funding acquisition, M.A.M.A. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work has been possible thanks to the support of the Mexican governmen<sup>t</sup> through the FORDECYT-PRONACES program Consejo Nacional de Ciencia y Tecnología (CONACYT) under gran<sup>t</sup> APN2017–5241; the SIP-IPN research grants, SIP 2083, SIP 20210169, and SIP 20210189; IPN-COFAA and IPN-EDI.

**Informed Consent Statement:** Informed consent was obtained from all subjects involved in the study.

**Data Availability Statement:** The authors are committed to providing access to all the necessary information so that readers can fully reproduce the results presented in this work. For this, all the necessary information is available in the following repository at https://github.com/vicleo14/ PublicidadBT (accessed on 20 December 2021). A demo is also available at https://tinyurl.com/2p8 bf68s (accessed on 20 December 2021).

**Conflicts of Interest:** The authors declare no conflict of interest.
