International Journal on Artificial Intelligence Tools

This paper reviews an experiment in human-computer interaction, where interaction takes place when humans attempt to teach a computer to play a strategy board game. We show that while individually learned models can be shown to improve the playing performance of the computer, their straightforward composition results in diluting what was earlier learned. This observation suggests that interaction cannot be easily distributed when one hopes to harness multiple human experts to develop a quality computer player. This is related to similar approaches in robot task learning and to classic approaches to human learning and reinforces the need to develop tools that facilitate the mix of human-based tuition and computer self-learning.

Human-Computer Interaction modelling can benefit from machine learning. This paper presents a case study of the use of machine learning for the development of two interrelated Bayesian Networks for the purposes of modelling student interactions within Intelligent Learning Environments. The models predict (a) whether a given student's interaction is effective in terms of learning and (b) whether a student can answer correctly questions in an intelligent learning environment without requesting help. After discussing the requirements for these models, the paper presents the particular techniques used to pre-process and learn from the data. The case study discusses the models learned based on data collected from student interactions on their own time and location. The paper concludes by discussing the application of the models and directions for future work.

This paper presents and discusses a survey of existing tools to support design and evaluation of websites, with special emphasis on improving the information navigation process. The amount of information of today's websites, the continuous evolution of the medium and the heterogeneity of typical users' profiles, make the website design task particularly hard. The presented tools are mainly based on recent models of Web usage behavior, and involve various natural language and semantic similarity modeling methods. Validation studies of the presented tools have shown that they can support effectively various phases of the website design lifecycle including information structuring, hyperlink evaluation and assessment of alternative designs. In this paper, existing techniques are discussed, the aspects of Web design that (should and) could be better supported are identified and suggestions are made on extensions of existing approaches to better support the usability evaluation process.

This paper proposes a novel computational model that allows the dynamic adaptation of the information to be presented to the user in the frame of a web application. In adaptive web-based hypermedia systems, content and presentation are usually dealt with independently and sequentially, generating content first and then adapting its presentation. By contrast, our approach is based on the idea that it is impossible to select the best content and the best presentation independently, but effective user-system interaction can only be achieved through a joint assessment of content characteristics and presentation alternatives. Consequently, the goal of the approach is to determine the best match between a set of contents eligible to be presented to the user and a collection of possible presentations. To this end, the quality of a match is assessed on the basis of a set of user interface design principles focused on the relations existing between content and presentation. These principles express different and often contrasting exigencies and are generally articulated in vague and merely qualitative terms. In our approach they are formalized as soft constraints to be satisfied to the highest possible degree. A fuzzy approach is adopted to compute the degree to which a match satisfies each principle and to aggregate the results of the application of a set of interface design principles to a match. The proposed approach is substantiated through a running example in the area of e-commerce.

The evaluation of subjective aspects of HCI, such as human-likeness, likeability or users' emotions towards computers is still quite a neglected issue, especially in the field of non-task oriented conversational systems (chatterbots). In this paper we try to bridge this gap by proposing a new methodology of evaluation. The methods presented were tested in our research on humor-equipped chatterbots. We describe them in details, discuss their drawbacks and usability. In one of the presented methods we used an emotiveness analysis system, which itself can be considered an AI tool, as it was used to detect users' emotions towards conversational systems, and to perform their automatic evaluation. We also propose some methods that we have not used yet, which, however, seem applicable in this field, such as brain scanning techniques. Finally, we give some ideas that should be addressed in the future.

In previous work, we introduced a novel concept of a generalised event, an abstract event, which we define as a change of state of abstract predicates that represent knowledge about the surrounding world. Abstract predicates are defined by formulae in temporal first-order logic (Abstract Event Specification Language (AESL)) whose leaf predicates represent low-level sensor-derived knowledge. Abstract events are detected by Rete Networks structured as a deductive knowledge-base. Current Abstract Event detectors cannot express sufficiently well certain high-level situations, such activity derived from user trajectories.

In this work we introduce a novel type of abstract event detector, a hidden Markov Model detector (hMM-detector). hMM-detectors are implemented as pattern recognition engines that use several stochastic models, hidden Markov Models (hMMs), in order to classify observed activities to the most likely activity class. We link hMM-detectors with AESL by specifying a new AESL operator for defining hMM-based Abstract Events, thus increasing AESL's expressive power.

We describe the experimental evaluation of the above work that was carried out at the University of Cambridge. hMM-Detectors were trained and tested with real data from the Active BAT location system. We evaluate the expressiveness of the enhanced AESL by discussing three case studies in healthcare that relate to continuous monitoring of elderly or injured patients. We demonstrate that AESL can be used in order to improve the dependability of continuous monitoring of patients and the provision of high-quality healthcare.