AutoNetica™ is the newest product in the Netica family. It is a system to deploy your Bayes nets over the internet as a question-answer system. Such a system asks the user questions, or provides a dashboard to enter relevant information, and presents the user with conclusions. At each stage of the process, the system chooses the best questions to ask based on the previous answers given, using Netica's sensitivity analysis. It may continuously provide conclusions, or wait until it reaches a certain level of confidence before presenting any information. Contact Norsys for full documentation and pricing information (although you may try it out now as described below). With AutoNetica, you can easily build an interactive website, also known as a HED system (Human- Electronic Dialog).
How to: Using Netica Application, version 5.0 or later, construct a Bayes net in the normal manner. Next, select a node of interest (called a Target node), then choose Netica → Build Website. Alternatively, you can right-click on the background and choose HED System → Add Net to Website. In a moment your browser will open with your working system. You can copy/paste the URL from your browser to publish your site to the world. Example site
Applications: AutoNetica™ can be used to build diagnostic systems, such as troubleshooting for some machinery or plant, or to provide medical, environmental, corporate, etc. diagnostics. It can be used to build systems that identify some particular item(s) from a large set based on indirect criterion, such as a product to be purchased from an inventory, a movie or song selection, restaurant choice, person, etc. It can build classification systems to determine which group an item falls into, based on its characteristics, such as the personality type of a person, probable voting patterns, credit worthiness, whether a plant or product is poisonous, etc. Although these are typical applications, any Bayes net can be run on AutoNetica.
AutoNetica vs. Decision Trees: AutoNetica™ is superior to old tree-based methods because it is easier to build and maintain a Bayes net than a decision tree. Bayes nets represent their knowledge in a form which is easy to understand, and which can combine human expertise with learning from data. Furthermore, their modularity allows combining sub-models into a complete model. Any changes in probability distributions can drastically alter the structure of the optimum decision tree, whilst they produce a small change in a Bayes net. Bayes nets allow users to skip questions, answer “unknown”, or provide likelihood answers, and still maintain optimal performance. And they can recover from incorrect answers, which often throw decision trees on the wrong track.
Constructing Nets for AutoNetica: When constructing the Bayes net, its nodes are divided into (possibly overlapping) groups: observable, target and intermediate. Each observable node becomes a potential question in the system, and each target node will result in an answer, or have its probabilities displayed. AutoNetica™ determines which questions are the most relevant to determine the values of the target results (using sensitivity measures). It then presents those questions first, either one at a time, or a few at a time in a scrolling list (depending on the configuration you set). The user may skip questions at any time, in which case the questions are moved to another part of the screen, to allow later answering if desired.
Since only the most relevant questions are asked, based on the previous answers, a HED session has the feel of an interactive dialog, rather than just filling out a form. Since the system is so judicious in choosing which question to ask from its pool of possible questions, it is okay to have a very large set of possible questions, further adding to its ability to reach good conclusions.
For more information on how Netica orders the results, see Sensitivity to Findings.
A working HED system can be built from a regular Bayes net with a single click, (Network → Build Website) although if you want to you can highly customize it. Its visual appearance, dimensions, colors, behavior, etc. can all be easily changed, and graphics can be added so that it is seamlessly integrated with a larger website.
Hosting: Your HED system, built as described above, is being hosted by Norsys, but if for security or other reasons you wish to host it yourself, that is also an option. Either way, if you wish, it can become part of your existing website. Alternately, the “web page” can just be local to the machine it is running on, allowing you to distribute your solution as a desktop application that does not require the internet.
To build your first HED system, be sure you have Netica 5.0 or later installed on your machine and have read the AutoNetica™ manual.
For more information, email Norsys.