Dieter Vogts

Marking student programs using graph similarity

We present a novel approach to the automated marking of student programming assignments. Our technique quantifies the structural similarity between unmarked student submissions and marked solutions, and is the basis by which we assign marks. This is accomplished through an efficient novel graph similarity measure (AssignSim). Our experiments show good correlation of assigned marks with that of a human marker.

The impact of an adaptive user interface on reducing driver distraction

This paper discusses the impact of an adaptive prototype in-car communication system (ICCS), called MIMI (Multimodal Interface for Mobile Info-communication), on driver distraction. Existing ICCSs attempt to minimise the visual and manual distraction, but more research needs to be done to reduce cognitive distraction. MIMI was designed to address usability and safety issues with existing ICCSs. Few ICCSs available today consider the drivers context in the design of the user interface. An adaptive user interface (AUI) was designed and integrated into a conventional dialogue system in order to prevent the driver from receiving calls and sending text messages under high distraction conditions. The current distraction level is detected by a neural network using the driving speed and steering wheel angle of the car as inputs. An adaptive version of MIMI was compared to a non-adaptive version in a user study conducted using a simple driving simulator. The results obtained showed that the adaptive version provided several usability and safety benefits, including reducing the cognitive load, and that the users preferred the adaptive version.

Comparison of the effects of professional and pedagogical program development environments on novice programmers

The program development environment a novice programmer uses has been identified as having a significant impact on the outcome of the process of learning to program. Professional program development environments are developed specifically for professional programmers and are often utilised by educational institutions to teach novice programmers. Pedagogical program development environments have been created to assist novice programmers in the process of learning to program. An empirical experiment was conducted to compare a representative professional program development environment with a representative pedagogical program development environment to determine how novice programmers are affected by the program development environment utilised. It was found that low risk novice programmers benefited significantly in terms of academic performance, while high risk novice programmers benefited significantly in terms of programming behaviour by using a pedagogical program development environment. The results indicate that a significant improvement in programming behaviour does not imply a significant increase in academic performance as well.

Design and evaluation of a multimodal interface for in-car communication systems

In-car communication systems (ICCS) are becoming a common feature in todays cars, however they can cause distraction due to usability issues and competition for driver attention. Several multimodal interfaces have been designed and implemented using speech in order to allocate the drivers hands and eyes solely to the driving task. A lack of feedback and flexibility in some ICCS still contributes to increasing the cognitive load of the driver and negatively affecting the user experience. This paper discusses the design and evaluation of MIMI (Multimodal Interface for Mobile Info-communication), a multimodal prototype that was designed to address the usability issues with existing ICCS. Several communication tasks were evaluated using the following metrics: time on task, task completion, workload, flexibility, error prevention and feedback. Results obtained show high rates of perceived usability in terms of time on task and task completion rates due to the flexibility, feedback and error prevention provided by MIMI. The results show that multimodal input can contribute to preventing errors; however commands where speech input cannot be combined with manual input (e.g. dictating a phone number), are still negatively affected by speech recognition errors.

Using machine learning to predict the driving context whilst driving

This paper discusses how the driving context (driving events and distraction level) can be determined using a mobile phone equipped with several sensors. The majority of existing in-car communication systems (ICCS) available today are built-in and do not use the driving context. This creates two issues: firstly, the use of an ICCS is limited to specific cars and secondly, the drivers safety remains an issue, as the driving context is not taken into account. This paper discusses two experiments in which data are collected, trained and tested in order to create a model that predicts driving events and distraction level. A mobile, context-aware application was built using the MIMIC (Multimodal Interface for Mobile Info-communication with Context) Framework. The Inference Engine uses information from several sources, namely mobile sensors, GPS and weather information, to infer both the driving event and the distraction level. The results obtained showed that the driving events and the distraction level can be accurately predicted. The driving events were predicted using the IB1 technique with an accuracy of 92.25%. In the second experiment, the distraction level was predicted with 95.16% accuracy, using the KStar (decision tree) technique. An analysis of the decision tree showed that some variables were more important than others in predicting the driving context. These variables included the speed and direction, as well as acceleration, magnetic field and orientation.

Are mobile in-car communication systems feasible?: a usability study

The issue of driver distraction remains critical despite efforts that aim to reduce its effects. In-Car Communication Systems (ICCS) were introduced to address visual and manual distraction occurring when using a mobile phone whilst driving. ICCS running on mobile phone have increased the number of people using ICCS as they can be installed at no cost and the quality of speech recognition on mobile devices is improving. Little research, however, has been conducted to investigate usability problems with mobile ICCS. This paper proposes a new model to address some of the issues found with current mobile ICCS, called the multimodal interface for mobile info-communication with context (MIMIC). This paper discusses the design and usability evaluation of a prototype mobile ICCS, designed using MIMIC. Several tasks were evaluated using different metrics including time on task, task completion, task success, number of errors, flexibility, user satisfaction and workload. Results obtained show that the users gave a good overall rating to the mobile ICCS, which indicates that users will easily accept such technology. Future work will include redesigning the speech user interface in order to address the usability issues found with the current prototype.

A sketch-based articulated figure animation tool

Animating models using traditional 3D modelling and animation packages is complex and difficult for novice users to use. A sketch-based articulated figure animation tool is introduced that provides users with an intuitive interface to pose and animate skeletons of any topology easily. An isomorphic mapping between the sketch and the skeleton is determined and a recursive algorithm poses the skeleton so that it matches the pose sketched. It is shown how sketch- and depth ambiguity can be resolved using disambiguation information and pressure sensitivity. The tool developed allowed novice users to successfully and easily sketch stick figures to pose articulated figures for sketch-based animation.

When Similarity Measures Lie

Do similarity or distance measures ever go wrong? The inherent subjectivity in similarity discernment has long supported the view that all judgements of similarity are equally valid, and that any selected similarity measure may only be considered more effective in some chosen domain. This paper presents evidence that such a view is incorrect for structural similarity comparisons. Similarity and distance measures occasionally do go wrong, and produce judgements that can be considered as errors in judgement. This claim is supported by a novel method for assessing the quality of similarity and distance functions, which is based on relative scale of similarity with respect to chosen reference objects. The method may be applied in any domain, and is demonstrated for common measures of structural similarity in graphs. Finally, the paper identifies three distinct kinds of relative similarity judgement errors, and shows how the distribution of these errors is related to graph properties under common similarity measures.

Sketch-based interfaces: drawings to data

Portable devices are ubiquitous and users are expecting more sophisticated and natural interactions, where the interface mimics and extends real world objects, such as paper. Visual communication languages are used on these media to convey a wealth of information, which can be analysed and automatically extracted. This research investigates how a digital alternative to a static medium can seamlessly and easily provide electronic support. The case study of the research, the pre-production phase of computer generated film making, was chosen as it has a comprehensive visual language on which to draw. A framework for sketch-based interaction is proposed, and a prototype system is evaluated on a number of different criteria. The main findings of the evaluation were that users found the sketch-based interaction intuitive, easy to learn and preferred it to standard interaction techniques. The main limiting factors of a sketch-based interaction technique were identified as adequately dealing with recognition errors and managing algorithm execution time so that the system remained interactive. A set of guidelines is provided to help researchers and developers in creating tools using similar techniques.

Models towards a hybrid conversational agent for contact centres

The usability of touch-tone Interactive Voice Response (IVR) systems is dismal. Clients would rather speak to a contact centre agent than navigate through the menu structure found in these systems. Contact centres, due to a variety of reasons, most notably high personnel costs, tend to utilise IVR as their solution for automation. IVR is an example of a simple forward chaining rule based expert system. An evaluation was conducted to determine whether a natural language interface would provide a more effective automation technique in comparison to current techniques utilised by contact centres. This evaluation compared the advantages and disadvantages of a natural language interface and a rule based expert system interface (modelled to resemble an IVR) and concluded that a hybrid system utilising a combination of these techniques would provide a better solution. This paper discusses two models that could be employed in the combination of a rule based expert system with a natural language interface.