Music recommender systems are an integral part of our daily life. Recent research has seen a significant effort around black-box recommender based approaches such as Deep Reinforcement Learning (DRL). These advances have led, together with the increasing concerns around users’ data collection and privacy, to a strong interest in building responsible recommender systems. A key element of a successful music recommender system is modelling how users interact with streamed content. By first understanding these interactions, insights can be drawn to enable the construction of more transparent and responsible systems. An example of these interactions is skipping behaviour, a signal that can measure users’ satisfaction, dissatisfaction, or lack of interest. In this paper, we study the utility of users’ historical data for the task of sequentially predicting users’ skipping behaviour. To this end, we adapt DRL for this classification task, followed by a post-hoc explainability (SHAP) and ablation analysis of the input state representation. Experimental results from a real-world music streaming dataset (Spotify) demonstrate the effectiveness of our approach in this task by outperforming state-of-the-art models. A comprehensive analysis of our approach and of users’ historical data reveals a temporal data leakage problem in the dataset. Our findings indicate that, overall, users’ behaviour features are the most discriminative in how our proposed DRL model predicts music skips. Content and contextual features have a lesser effect. This suggests that a limited amount of user data should be collected and leveraged to predict skipping behaviour.

The ability to skip songs is a core feature in modern online streaming services. Its introduction has led to a new music listening paradigm and has changed the way users interact with the underlying services. Thus, understanding their skipping activity during listening sessions has acquired considerable importance. This is because such implicit feedback signal can be considered a measure of users' satisfaction (dissatisfaction or lack of interest), affecting their engagement with the platforms. Prior work has mainly focused on analysing the skipping activity at an individual song level. In this work, we investigate different behaviours during entire listening sessions with regards to the users' session-based skipping activity. To this end, we propose a data transformation and clustering-based approach to identify and categorise skipping types. Experimental results on the real-world music streaming dataset (Spotify) indicate four main types of session skipping behaviour. A subsequent analysis of short, medium, and long listening sessions demonstrate that these session skipping types are consistent across sessions of varying length. Furthermore, we discuss their distributional differences under various listening context information, i.e. day types (i.e. weekday and weekend), times of the day, and playlist types.

This paper introduces an information-theoretic method for selecting a subset of problems which gives the most information about a group of problem-solving algorithms. This method was tested on the games in the General Video Game AI (GVGAI) framework, allowing us to identify a smaller set of games that still gives a large amount of information about the abilities of different game-playing agents. This approach can be used to make agent testing more efficient. We can achieve almost as good discriminatory accuracy when testing on only a handful of games as when testing on more than a hundred games, something which is often computationally infeasible. Furthermore, this method can be extended to study the dimensions of the effective variance in game design between these games, allowing us to identify which games differentiate between agents in the most complementary ways.

Ensemble Decision Systems offer a unique form of decision making that allows a collection of algorithms to reason together about a problem. Each individual algorithm has its own inherent strengths and weaknesses, and often it is difficult to overcome the weaknesses, while retaining the strengths. Instead of altering the properties of the algorithm, the Ensemble Decision System augments the performance with other algorithms that have complementing strengths. This work outlines different options for building an Ensemble Decision System as well as providing analysis on its performance compared to the individual components of the system with interesting results, showing an increase in the generality of the algorithms without significantly impeding performance.

This paper presents a new optimization technique for keyboard layouts based on Pareto front optimization. We used this multifactorial technique to create two new touchscreen phone keyboard layouts based on three design metrics: minimizing finger travel distance in order to maximize text entry speed, a new metric to maximize the quality of spell correction by reducing tap ambiguity, and maximizing familiarity through a similarity function with the standard Qwerty layout. The paper describes the optimization process and resulting layouts for a standard trapezoid shaped keyboard and a more rectangular layout. Fitts' law modelling shows a predicted 11% improvement in entry speed without taking into account the significantly improved error correction potential and the subsequent effect on speed. In initial user tests typing speed dropped from approx. 21 wpm with Qwerty to 13 wpm (64%) on first use of our layout but recovered to 18 wpm (85%) within four short trial sessions, and was still improving. NASA TLX forms showed no significant difference on load between Qwerty and our new layout use in the fourth session. Together we believe this shows the new layouts are faster and can be quickly adopted by users.

This work investigates the application of Evolutionary Computation (EC) to the induction of generalised policies used to solve AI planning problems. A policy is defined as an ordered list of rules that specifies which action to perform under which conditions; a solution (plan) to a planning problem is a sequence of actions suggested by the policy. We compare an evolved policy with one produced by a state-of-the art approximate policy iteration approach. We discuss the relative merits of the two approaches with a focus on the impact of the knowledge representation and the learning strategy. In particular we note that a strategy commonly and successfully used for the induction of classification rules, that of Iterative Rule Learning, is not necessarily an optimal strategy for the induction of generalised policies aimed at minimising the number of actions in a plan.

This paper compares supervised and unsupervised learning mechanisms for the emergence of cooperative multiagent spatial coordination using a top-down approach. By observing the global performance of a group of homogeneous agents-supported by a nonglobal knowledge of their environment-we attempt to extract information about the minimum size of the agent neurocontroller and the type of learning mechanism that collectively generate high-performing and robust behaviors with minimal computational effort. Consequently, a methodology for obtaining controllers of minimal size is introduced and a comparative study between supervised and unsupervised learning mechanisms for the generation of successful collective behaviors is presented. We have developed a prototype simulated world for our studies. This case study is primarily a computer games inspired world but its main features are also biologically plausible. The two specific tasks that the agents are tested in are the competing strategies of obstacle-avoidance and target-achievement. We demonstrate that cooperative behavior among agents, which is supported only by limited communication, appears to be necessary for the problem's efficient solution and that learning by rewarding the behavior of agent groups constitutes a more efficient and computationally preferred generic approach than supervised learning approaches in such complex multiagent worlds

This paper presents a comparative study between genetic and probabilistic search approaches of evolutionary computation. They are both applied for optimizing the behavior of multiple neural-controlled homogeneous agents whose spatial coordination tasks can only be successfully achieved through emergent cooperation. Both approaches demonstrate effective solutions of high performance; however, the genetic search approach appears to be both more robust and computationally preferred for this multi-agent case study.

In this paper, we introduce the first stage of experiments on neuro-evolution mechanisms applied to predator/prey multicharacter computer games. Our test-bed is a computer game where the prey (i.e. player) has to avoid its predators by escaping through an exit without getting killed. By viewing the game from the predators' (i.e. opponents') perspective, we attempt offline to evolve neural-controlled opponents capable of playing effectively against computer-guided fixed strategy players. Their efficiency is based on cooperation which emerges from an abstract type of partial interaction with their environment. In addition, investigation of behavior generalization demonstrated the crucial contribution of playing strategies in the development of successful predator behaviors. However, emergent well-behaved opponents trained offline with fixed strategies do not make the game interesting to play. We therefore present an evolutionary mechanism for opponents that keep learning from a player while playing against it (i.e. online) and we demonstrate its efficiency and robustness in increasing the predators' performance while altering their behavior as long as the game is played. Computer game opponents following this online learning approach show high adaptability to changing player strategies, which provides evidence for the approach's effectiveness and interest against human players.

LIA