Working with me?

If you are interested in any of these topics, contact me.

PhD Topic 1: Dynamic Team Formation

The concept of this PhD project is to create innovative co-operative learning algorithms that will be applied to learning processes. Traditionally, the education system gravitates around memorization, assimilation and theory. But the shifting of the global economy has introduced new market demands that forces to restructure education to encourage entrepreneurship, creativity and risk-taking. Co-operative learning, based on teamwork is the path to follow. Our investigation proposes to develop and transfer Artificial Intelligent methods for team formation, individual assessment and personalised learning that will become software modules to be integrated into emerging e-learning platforms.

The work starts from current state-of-the-art results on static team formation and composition developed in a previous industrial PhD by IIIA-CSIC and CMT. This project is aimed at extending the previous work by addressing the research of Artificial Intelligence methods for a new fundamental way of producing teams. We will study the development of balanced dynamic team formation techniques where the duration of tasks, the sets of tasks and possibly the set of students vary along time. Together with the dynamic team formation problem, the work will tackle the problem of automatically assessing individual competencies and the problem of adapting learning contents sequencing to each individual. A particularly interesting problem is ‘Alternating education’, that is, an instructive experience, co-planned by the school system and other institutions, to give students formative opportunities, with a high and qualified profile. Here the question is what team of students to send to a company to work with an already organised team to solve a particualr task.

These new methods will be implemented and validated at the industrial level through pilot tests in a network of schools. In order to facilitate the transfer of technology, we will create software components that can be integrated into the Human Resources, Operations and Learning Systems of different sectorial organizations to help them to fulfil the challenge to educate the new generations in a more dynamic and vibrant context.

Type of PhD. Industrial Doctorate in collaboration with Enzyme.

Incorporation: Summer 2019.

Advisors: Juan Antonio Rodriguez and Carles Sierra

PhD Topic 2: Tell me what you do and I will tell you who you are

The reputation of an entity, individual or company, is usually measured as an aggregate of opinions expressed by third parties about their behavior. In particular, opinions on their behavior in the fulfillment of commitments and contracts, or opinions on the satisfaction of behavioral expectations. The latter is very common when there is no explicit commitment (e.g. to give good food in a restaurant).

Normally, it is understood that opinions are expressed explicitly (e.g. Tripadvisor, eBay, Amazon) and in most cases opinions express personal perception of the quality of products and services. That is, explicit opinions about behavioral expectations. However, there is a practically unexplored field that is the extraction and aggregation of implicit opinions from the behavior of the entities. Also, the combination of explicit and implicit opinions has received little attention. To better understand what we are referring to with an implicit opinion, let’s think about the case of soccer. If Manchester United beats Chelsea by 5 to zero, we can consider that at the end of the game the opinion of Manchester is that Chelsea is a very bad team and Chelsea’s opinion is that Manchester is a very good team. The acts of the entities and the interactions between them (e.g. through social networks, financial transactions) are sources of implicit opinions that can be of great value to predict their future behavior.

The objective of this thesis is therefore (i) the development of methods for the extraction of opinions implicit in the interactions among entities, (ii) the development of methods of aggregation of implicit and explicit opinions, and (iii) the construction of a predictive model of behavior based on the reputation of the entities.

Type of PhD. Industrial Doctorate in collaboration with Strands.

Incorporation: Summer 2019.

Advisors: Nardine Osman and Carles Sierra

PhD Topic 3: Argumentation-based multiagent recommender system

Preferences of users and user profiles could be used as a good source of arguments supporting recommendations: You like books with political plots in them, I have recently read one of those that I loved, you only like red wines from Bordeaux and never pay much for a bottle, I strongly recommend Priorat wines, that also have full body and velvet taste. In the last decade a number of papers have proposed the use of argumentation in recommender systems [4, 5]. However, recommendations can be conflicting depending on the recommenders different experiences, likes and dislikes. This phd thesis will explore an interactive distributed model in which teams of recommenders become active sofπtware agents that accompany the recommendations with arguments. This approach is radically different with previous recommender system approaches (content-based or not) where the experiences of users are represented as passive data in a database. These recommender agents will be able then to engage in argumentative dialogues with other recommenders to agree (or not) on a recommendation to the user. The composition of the best team of recommenders is a key issue to avoid biases. A mediator may pair recommenders according to different criteria (e.g. similarity of profiles with the user, diversity of their experiences, diversity of personalities and/or competences) and then aggregate the outcomes of the different dialogues into a meaningful recommendation to the user that will be duly justified by arguments.

Initially, publicly available data sets will be used for validation of the approach, in particular, the Sushi data set [2], which offers a well-designed source of preference data that can be used to form arguments for supporting recommendations. The approach will address two main issues still remaining in the recommender systems – the sensitivity of the precision of preference estimation with respect to the measurement error in the profile data; and the effect of central tendency effect (the tendency to select neutral positions on rating scales [6]). A comparison with current implementations of content-based recommender systems [3] will be done. During the PhD at least two use cases wil be identified where the results of the PhD work will be applied and evaluated.

Type of PhD. Contract from a H2020 project

Incorporation: Immediate.

Advisors: Juan Antonio Rodriguez and Carles Sierra

Note: This PhD is aimed to become a double degree with Western Sydney University.