User Behaviour Modelling Using Machine Learning and Deep Learning Techniques

User Behaviour Modelling Using Machine Learning and Deep Learning Techniques

Aitor Almeida https://morelab.deusto.es/people/members/aitor-almeida/

DeustoTech (www.deustotech.eu) is a private non-profit institution of the Faculty of Engineering at the University of Deusto for applied research in new technologies. Since 2005 DeustoTech mission is to support the ICT activity in business and society through research, the development of technologies, innovation and knowledge transfer. We focus our activity around TRLs 2-7 and articulate it into four applied fields: Industry, Mobility, Energy and Society, having a fifth, the Chair of Applied Mathematics, as a transversal activity and support for the previous four. We are characterized for working with data of heterogeneous nature, throughout its life cycle and in compliance with ethical principles and humanists who define the University of Deusto. The research group which will host the IF candidate is Deustek (http://www.morelab.deusto.es/), one of the largest and most successful research groups in the University. The group has a strong background in the application of Artificial Intelligence techniques to middleware for embedded and mobile system in order to foster context-aware reactivity and activity modelling and reaction. Deustek has worked in several projects dealing with Activity Recognition, Behavior Modelling, Big Data and Intelligent Environments like City4Age (H2020), Greensoul (H2020), SONOPA (AAL), THOFU (national), PIRAMIDE (national), SABESS (national) and Aztarnet (regional).

We are looking for PostDoc applications to complete our research skills in machine learning and deep learning applied to user behavior modelling and recognition. The group is currently embarked in several European H2020 projects and proposals which involve knowledge and expertise in this research area applied to several domains: intelligent environments, health applications, active ageing and social networks. The candidates are expected to have a strong base in machine learning and deep learning applied to sequential data, text, sensor data or images. A proven track record of expertise in the area by a good range of impact publications and open source contributions is expected. Candidates are expected to be able to code their proposed solutions and work on integrating them in the related projects. Experience in the execution of past projects and in the preparation of proposals, particularly European, will be valued. Additional knowledge in any of the following fields will be considered for accessing to the open position (please take into account that we do not expect experience in all of them): intelligent environments, Smart Cities, Ambient Assisted Living, Social Network Analysis, active ageing and health related applications.

• Information Sciences and Engineering (ENG)

Deustek is an officially recognized and financed research group within the Basque university system. The Deustek research unit is a multidisciplinary research group composed mainly by computer scientist and electric engineers but integrating also designers, HCI experts and lawyers. Our research interests are diverse and applied to multiple domains, but always with a strong focus on relevant societal challenges like citizen participation, active ageing, hate speech or sustainable development. The candidate will be integrated into the project team of one of our active research projects, working alongside the other members of the research unit. The candidate will actively participate in the preparation and execution of European and Industrial proposals and projects. His/her integration into the Deustek team will be an opportunity to develop as an applied research scientist in close contact with excellence research in the European domain and the Basque industrial sector. Our focus on competitive research projects will allow the candidate to develop a significant contact network while collaborating with European research groups. The candidate will receive research and financial support for publication in high impact journals and attendance to relevant conferences.

– Aitor Almeida, Rubén Mulero, Piercosimo Rametta, Vladimir Urosevic, Marina Andric, Luigi Patrono. (2019) \\\\\\\»A critical analysis of an IoT—aware AAL system for elderly monitoring\\\\\\\» In Future Generation Computer Systems. vol. 97. p. 598-619. DOI: 10.1016/j.future.2019.03.019. Impact factor: 4.639 (Q1) – Gorka Azkune, Aitor Almeida. (2018) \\\\\\\»A Scalable Hybrid Activity Recognition Approach for Intelligent Environments\\\\\\\» In IEEE Access. vol. 6. p. 41745-41759. DOI: 10.1109/ACCESS.2018.2861004. Impact factor: 3.557 (Q1) – Aritz Bilbao Jayo, Aitor Almeida, Diego López-de-Ipiña. (2016) \\\\\\\»Promotion of active ageing combining sensor and social network data\\\\\\\» In Journal of Biomedical Informatics. vol. 64. p. 108-115. DOI: 10.1016/j.jbi.2016.09.017. Impact factor: 2.447 (Q1) – Gorka Azkune, Aitor Almeida, Diego López-de-Ipiña, Liming Chen. (2015) \\\\\\\»Extending knowledge-driven activity models through data-driven learning techniques\\\\\\\» In Expert Systems with Applications. vol. 42. p. 3115-3128. DOI: 10.1016/j.eswa.2014.11.063. Impact factor: 2.981 (Q1) – Aitor Almeida, Gorka Azkune. (2018) \\\\\\\»Predicting Human Behaviour with Recurrent Neural Networks\\\\\\\» In Applied Sciences. vol. 8. p. 305. DOI: 10.3390/app8020305. Impact factor: 1.689 (Q3)

Behavior modelling and recognition is one the research areas within Deustek. During the recently closed City4Age H2020 project we created the base of our current behavior modelling research: a multi-level behavior model, a algorithm for sensor-based activity recognition and a behavior prediction algorithm. That research is currently being expanded in the newly approved FuturAAL project (Retos, national). In FuturAAL we will be working the next 3 years with the EHU in improving the activity recognition process both on sensor-based environments. We have also recently submitted 3 European and 1 regional proposals related with this area. In the Behavioromics (SC1-BHC-28-2019, H2020) proposal Deustek is leading the work package in charge of creating an expanded behavior model. In the BD4QoL proposal (SC1-DTH-01-2019, H2020) Deustek oversees creating the behavior recognition algorithms that will be used to ascertain the quality of life of post-treatment neck cancer patients. In the Toleration proposal (REC-RRAC-RACI-AG-2019, REC), which Deustek coordinates, we will create a behavior modeling algorithm to detect multi-media hate speech in social networks. Finally, in the PERUSE proposal (Elkartek, regional), which Deustek coordinates, we use behavior modelling algorithms to detect risk related to workplace stress. We are currently working on different proposal for the SC1-DTH-04-2020, SC1-DTH-02-2020 and SC1-BHC-29-2020 calls, with a strong focus on behavior modelling.

The research in this area is closely related to both behavioral and health sciences. In our previous projects we have worked closely with professionals from these areas (geriatricians, behavioral scientists, general practitioners…) in order to better understand the needs for behavior modelling in specific application domains (AAL, active ageing, prevention and detection of diseases…). We currently have a strong network of collaborators working on related areas with whom we work in our research. Ideally the prospective candidate should have previous experience working on related domains but is not essential. As part of our European research projects and proposals, the candidate can expect to collaborate with researchers and professionals in both of the aforementioned areas and having access to relevant pilot and user groups in order to gather data and test the developed models and algorithms. Members of the Deustek research group collaborate regularly in the interdisciplinary platforms (https://www.deusto.es/cs/Satellite/deustoresearch/en/home/interdisciplinary-research-platforms-1?cambioidioma=si) of the University of Deusto with researchers from other disciplines. Specifically, we take part in the Ageing and Wellbeing, B-creative, Strengthening Participation, and Social Justice and Inclusion platforms. The candidate will be expected to also take part in these platforms, helping him/her to interact with other research groups of the university.

Our interdisciplinarity will be mainly driven by our collaboration with other researchers and professionals in project consortia. Two of our current research projects offer interesting collaboration opportunities centered on behavior modelling and intelligent environments. In the FuturAAL project we are working alongside experts in computer vision from the EHU to improve the activity recognition in ego-centric vision systems. In the EU-SHAFE project we will contact stakeholders in seven different European regions to analyze the necessities and requirements for age-friendly cities. In the same way, our current H2020 proposals under evaluation include both medical researchers, behavioral scientists and relevant end-users, This will allow the candidate to interact with other European researchers working on different aspects of the behavior modelling problem. The advantages of this approach are twofold. On one hand the candidate will be able to have access to relevant collaborators in different areas and to end-users. On the other hand, the candidate will be able to expand her contact network, improving her capability to present competitive research proposals.

With more than 139 executed research projects the Deustek research group has a strong network of international collaborators, mainly centered in Europe. This network includes the academia, industry, public administrations, NGOs and end-user organizations with which we work continuously to further our research. The candidate will have access to this network to both work on the proposed research and to find new research opportunities in the form of new project proposals. The research proposed in this hosting offer is mainly aligned with the “Health, demographic change and wellbeing” societal challenge proposed in the H2020 programme, specifically in the “Better health and care, economic growth and sustainable health systems (BHC)” and “Digital transformation in health and care (DTH)” calls, which encompass most of the already presented proposals described before. The hosting offer is also aligned on one hand with the third SDG “Goal 3: Good health and well-being for people”, using the behavior modelling to provide personalized healthcare and improve the detection of early risks; and on the other hand with the eleventh goal “Goal 11: Sustainable cities and communities”, improving the inclusiveness and friendliness of the SmartCities by providing a better understanding of the citizens needs and behavior. It is planned for the candidate to apply for an ERC proposal at the of the hosting period, taking advantage of the CV improvement achieved during the hosting period.

The proposed offer is situated in the convergence of the biosciences and health binomial within the RIS3 strategy of the Basque Country. Is in this same convergence in which the Elkartek proposal that we have coordinated this year is located. In this proposal we have gathered industrial partners, research institutions and academic partners to work on the analysis of the workers behavior to improve health within the working environments. The candidate will have access to several stakeholders at both European and international level during the hosting: research groups working on behavior modelling in Canada, UK, Italy, Singapore, the Netherlands and Spain; research institutes in France, Italy and Spain; relevant industrial partners in Italy, the Netherlands, Serbia and Greece; and user associations and NGOs in Belgium, France, Italy and Spain.

The work developed during the hosting offer will have a significant social impact in three areas: 1- Promotion of active ageing: The improved algorithms for activity recognition will allow to monitor the daily life of elders, allowing to detect the harmful behaviors and assisting them in adopting more healthy behaviors 2- Improving of the elders independent living: advanced behavior models can improve the assistive technologies used to help the elders in their everyday life, allowing them to live independently for a longer period of time and delaying the internment in care facilities. 3- Minimize the impact of ageing in informal caretakers. By improving the assistive technologies we aim to reduce the burden of the caretakers to tackle the “caretaker syndrome”. To evaluate this impact our recent proposals in the H2020 BHC and DTH calls have significant user cohorts provided by end-user associations, which will ease the dissemination among the target users.

Several innovations are expected to be created as a direct result of the candidate’s work, as we will integrate him/her in our current research roadmap for behavior modelling in intelligent environments. The first result will be an extended multi-level user conduct model oriented to intelligent environments and health-related domains. Current models only contemplate three levels in user conduct (actions, activities and behaviors). During our work in the City4Age H2020 project, we identified the need to include more abstract conduct levels like habits and lifestyles. This will allow us to identify the repetitive abstract conducts related to early health risks in different domains. A second expected innovation will be work on a generalizable user conduct recognition ML model that will be able to integrate the knowledge extracted from multiple dataset in a single recognition mechanism. This will allow us to reuse multiple datasets created in different domains to extend and improve our algorithms. The result will be made available to both the research community and our industrial partners, allowing us to be more competitive in future project consortia at a European level.

One of the University of Deusto’s key duties is to be fully aware of problems within the institution itself and the society we live in. The university is taking the necessary steps to boost equality with the implementation of a Gender Equality Plan at institutional level with the firm objective of challenging and transforming gender roles and identities linked to professional careers, and work towards real institutional change. DeustoTech, as one of its institutions, is included into this service of social action and inclusion. The main aims consist of achieving full normalization, equal opportunities and gradually adopting the steps needed to ensure that the University of Deusto is an inclusive educational institution. Furthermore, the University of Deusto provides them with guidance on the transition to the labor market jointly with special job centers and companies at large. One of the main activity areas of the Deustek research group has always been working on relevant societal challenges. A significant number of our current proposals deal with working with vulnerable groups in order to promote their inclusion. In the Toleraction (REC) proposal we are working to fight hate-speech against religious minorities, specifically islamophobia and anti-semitism alongside the Human Rights Institute of the university. In the StandByMe (REC) proposal we are working on preventing sexual harassment and raising awareness in young adults, with a strong focus on gender issues.