Sunday, March 23, 2008

PhD Findings and Conclusion

Findings

The thesis covered the following areas :-


  • Enabling people to create software visually.
  • Creating design abstractions familiar to domain experts e.g. diagrams for engineers.
  • Ensuring interoperability using open standards.
  • Automating user to computer translation process.

This post explains how the alternative approach of User Driven Modelling/Programming used for this thesis to develop models and modelling capabilities compared to that of spreadsheet development, used within other projects. The alternative approach was outlined in this thesis, of using open standards ontologies/taxonomies and a web interface for developing decision support models for design and costing. The stepped translation approach designed and implemented in this thesis enabled structured modelling, and visualisation using interactive technologies.


  • Step 1 - Ontology
  • Step 2 - Modelling Tool
  • Step 3 - Interactive Visualisation

This stepped translation solved problems of the spreadsheet approach as indicated in the table below, and then in more detail in following sections -


Improvement - Achieved By
Maintenance - Structuring and Translation
Extensibility - Structuring and Visualisation
Ease of Use - Visualisation, Interaction, and Translation
Sharing of Information - Shared Ontology and Interoperability


The 3 step translation process created ensures translation of domain level modelling into open standard representation and software and vice versa.

Maintenance

The use of a centralised information source makes these models more reliable than the standalone spreadsheet. This centralised structure was easier to manage than updating multiple instances of the spreadsheets used by different people and ensuring they all contain the same information. So the first task was to build a system for collaborative model building. As a piece of information can then only belong to a unique location, the problems arising from duplicate pieces of information are eliminated. The models have only the functionality that is added by the model builder so there are not other side effects to keep track of, as there are with generic functionality within spreadsheets. Enabling people to create software visually makes it easier for model builders and model users to keep track of any information they are responsible for. Translation from the ontology to models and visualisations ensures one change will affect all stages, so this makes maintenance easier.


Extensibility

Creating the infrastructure for the collaborative model building system took much more time than it did for the spreadsheet system, but having done so it is quicker and easier to create further models. This is because of the facilities provided for model builders and end-users to customise the software in any one of the three step translation process. This means progress has been made in making it possible for non-programmers to build models. It also indicates that the extra research and development time taken was worth it in the long term, most of this time involved productive research and this can be used in future projects. The use of open standards in this thesis for information and models ensures there should be a development path, whatever changes there may be in the software market. This use of open standards also ensures that the system can link with most environments used by others. The translation and visualisation approach ensures that new models can be added using the existing ontology, and that design changes in the ontology and translation can enable modelling of different problems. So if there is a new problem to be modelled there are two ways to achieve this.

Ease of Use

Many people now are familiar with web pages and at least the basics of how to navigate them, and by creating such an environment, and standardising the navigation to those ways commonly used over the web, it is possible to ease usability. The models contain only the functionality that is added by the model builder unlike the spreadsheets which had generic functionality that was not required and led to user’s confusion. New Web 2.0 interaction technologies have allowed production of a rich user interface for web programs in a similar way to single computer applications. This means information held in an ontology and translated through modelling tools can be made available as interactive applications for many users. Translation allowed the same user interface to be provided in multiple tools and computer languages. Also this research showed that it was possible to provide user interfaces and visualisation differently as appropriate according to the type of user, the situation, or the kind of information to be shown.

Sharing of Information

The use of open standards languages for representing information makes it much easier to represent information in a way that makes it accessible both to people and software. Ontology based modelling tools use these open standards and so ensure dependable translation, interoperability, and sharing of information. Web browsers make it possible to share information with many users at once, and so this enables collaboration. Structuring of the information using standardised languages makes it easier to search and visualise the information. This ensuring of interoperability is important for long term use of the overall modelling system.

Conclusion

Within this thesis it is argued that there is a need for software developers to create programs that enable users to solve problems themselves. In effect this involves production of a system to create systems. This approach can widen programming participation by including computer literate non-programmers. This is a reaction to the increased complexity of real world problems and software systems, which makes development of software solutions impractical without greater involvement from end-users. It is difficult for developers to foresee every need of users and use of the software produced, so it makes sense to enable more end-user customisation. It is also argued that the research for this thesis has been a step towards making end-user programming possible. The research ideas look complex at first glance but this research is all about simplifying software development.

The approach of developing decision support models for design and costing using a spreadsheet was compared to the alternative approach of using open standards taxonomies and a web interface for this purpose. The conclusion is that although use of spreadsheets allows for the creation of models relatively quickly they are beset by problems. These relate to Maintenance, Extensibility, Ease of Use, and Sharing of Information. The spreadsheet example and the explanation in the thesis represent problems currently experienced throughout software and computer use.

The alternative approach for this thesis of User Driven Modelling/Programming involves the development of a system, where a model builder, via visual editing of library taxonomies can undertake maintenance and extension of information. Dealing with this proof of concept has indicated that it is easier to maintain, search and share information using this approach than it was using spreadsheets. This also enables much more of the maintenance task to be left to users, who can also customise the system. Creating the infrastructure has taken much more time than it did for the spreadsheet system, but having done so it is much quicker and easier to create further models. This indicates that the extra research and development time taken though far exceeding what would have been required for a spreadsheet modelling project is well worth it in the long term. Also the use of a centralised information source makes these models more reliable than the standalone spreadsheet, standalone decision support models created individually may contain out of date information. In addition, since a well constructed ontology implies that a piece of information can only belong to a unique location, the problems arising from duplicate pieces of information are eliminated. It is also much easier to create models once the infrastructure is in place; this can enable users to develop models. The ability to visualise, search and share information using structured languages and web pages is a huge advantage for creation of dynamic structured views and decision support models over the web.

This research was a test case for a whole new approach that could be possible, of collaborative end-user programming by domain experts. The end-user programmers can use a visual interface where the visualisation of the software exactly matches the structure of the software itself, making translation between user and computer, and vice versa, much more practical. For this reason highly structured visualisations were preferred over web spreadsheets. Semantic Web languages are ideal for representing graphs and trees in an open standard way. The spatial, and tree/graph visualisations used both have the same underlying semantics, and therefore can both be translated to computer languages. In fact it would be much better in the long run to use the Semantic Web languages as standardised programming languages for such problems as this would avoid the need to further translate into other programming languages, and systems. The advantage to this is that of using Semantic Web languages for representation of information, meta-programming, and translation to a visual display for users. The use of Semantic Web languages as a connectivity environment for connecting information, and for connecting users to the information held in Semantic Web data sources enables an environment that could be made easier to use, install and maintain.

More generally a new approach is required to software creation. This approach should involve developers creating software systems that enable users to perform high level programming, and model the problem for which they are the experts. This is an alternative to the provision by developers of modelling solutions that try to provide an out of the box solution that just needs ’tweaking’. Such an out of the box system is not practical considering both increases in complexity of manufactured products, and of software systems themselves. Feedback from publishing the research examples behind this thesis and working with industrial partners indicates that people like to work on their own solutions, providing they are computer literate and confident they have domain knowledge that the developers do not possess. This is true for software development in general, not just in the domain of engineering. Research cited in this thesis from others involved in end-user programming confirms this.

For proving the hypothesis that it is possible to create an end-user programming environment, usable by non programmers, it has been found that structuring and relating of information is all important in this solution. To achieve this, it was only necessary to link the information visually via equations, and store these results for reuse and collaboration. If users can understand and navigate relationships, and add new relationships they can model most problems. It was important to design a visual interface that is intuitive to use, and allows for proper interpretation of the results. Feedback has indicated that users can navigate this structure and manipulate it. This is preferable to ’black box’ solutions that hide information. There are no dead ends or blocks to expanding and improving this approach. To make the system easier to use it is only necessary to trial continually better interfaces, and to assist by providing guidance to the user. There was not sufficient time and resources to expand this research much to areas outside engineering modelling, but there is scope for researchers to improve end-user programming for engineering modelling systems, and to expand the research into other areas.

No comments: