Search

Term weighting is one of the most commonly used approaches, which works by assigning weights to terms, that aims to improve the performance of information retrieval or text categorization tasks. In this paper, we present a novel term weighting technique, called discriminative feature spamming technique (DFST), which identifies distinctive terms, based on a term utility criteria (TUC), and then spams them to increase their discriminative power. The experimental results show that the DFST outperformed a set of time-tested term weighting schemes, from the information retrieval field. All the experiments were performed on the largest ever Roman Urdu (RU) dataset of 11000 reviews, which was collected and annotated for this work. In addition, a custom tokenizer was built, which further improved classification accuracy. A cross-scheme comparison was performed, which showed that the results obtained by using the newly proposed DFST, were statistically significant and better than previous approaches.

Autonomous systems are making their way to the market. The transition from tasks performed by humans to tasks performed by machines begs for an answer to one of the most challenging questions in this area of research: Will humans understand and trust what a machine does? Analyzing human and machine behaviors offers the foundational steps toward finding answers to this question. This paper contributes a novel methodology for transforming low-level actions by each agent into high-level categorization of strategies to contrast the behaviors of humans and machines using a computational red teaming environment with a red (evader) and a blue (pursuer) agent. Two orthogonal sources of uncertainty were examined: the uncertainty in the blue agent's situation awareness about the red, and the red agent's uncertainty resulting from deceptive actions by the blue. For each uncertainty source, two different experiments were conducted by varying the controller of the red agent. In one experiment, the red agent was controlled by one of the 34 human subjects; and in the second, by an evolved neural network. The blue agent was controlled by a scripted rule-based system. In this time-critical task, the results revealed that humans tend to follow systemic and consistent strategies, sometimes ignoring the information available to them. On the other hand, machines tend to evolve more complex and diverse strategies. This finding calls for new computational intelligence techniques to enable the fusion of these different strategies into forms that each party can understand and use effectively.

Games are becoming increasingly indispensable, not only for fun but also to support tasks that are more serious, such as education, strategic planning, and understanding of complex phenomena. Computational intelligence-based methods are contributing significantly to this development. Learning Classifier Systems (LCS) is a pioneering computational intelligence approach that combines machine learning methods with evolutionary computation, to learn problem solutions in the form of interpretable rules. These systems offer several advantages for game applications, including a powerful and flexible agent architecture built on a knowledge-based symbolic modeling engine; modeling flexibility that allows integrating domain knowledge and different machine learning mechanisms under a single computational framework; an ability to adapt to diverse game requirements; and an ability to learn and generate creative agent behaviors in real-time dynamic environments. We present a comprehensive and dedicated survey of LCS in computer games. The survey highlights the versatility and advantages of these systems by reviewing their application in a variety of games. The survey is organized according to a general game classification and provides an opportunity to bring this important research direction into the public eye. We discuss the strengths and weaknesses of the existing approaches and provide insights into important future research directions.