3
$\begingroup$

I am looking in to building a kind of troubleshooting web application. It would be a form that starts with a first question. Depending on the answer, you get a follow up question and so on until the app has qualified your problem in to a small group of problems. To me it sounds a bit like a decision tree but what I have read about them is that it is the internal structure of a model and not what I am looking for. My guess is that a model needs all the input variables at once and not like I am looking for that feeds it one parameter at a time.

At this time I do not know of any data available. With the client we could create the desired resulting problem groups and the questions as well.

Would it be possible to solve this with the help of AI instead of hand coding a lot of case switch statements? If so could you point me to what to read up on?

$\endgroup$
  • $\begingroup$ I could absolutely call it a troubleshooting app. The domain is in vetrenary medicine. Instead of calling the vet you would get the client to fill in this web form. It would determine if it is a serious state of your pet and then let the vet call the client. In other cases it would tell the client to book an appointment later that week. So the information in the app is static but constantly evolving to make better decisions. But that would be a manual task. $\endgroup$ – Martin Larsson Jun 3 '18 at 12:57
1
$\begingroup$

In addition to Manuel' answer,if you're planning to build a trouble shooting system and that's web based application,then I regard it as a rule based expert system,which can be developed to improve web based problem solving ability. Besides that,several experiments have been conducted to evaluate the performance of such systems. Therefore I would like to get a copy of this article and read through for your own benefit. https://pdfs.semanticscholar.org/644f/486b639fb5fd9cda71ac719d636acb4d9142.pdf

In conclusion,it's possible to implement such application by AI based on the above.

$\endgroup$
1
$\begingroup$

are you sure you don't want to use decision-trees? because it is the most relevant one i can think of for your project. you can check "akinator" , it is almost like what you have described.

$\endgroup$
  • 1
    $\begingroup$ O that sounds interesting. Do you have any links to the technical details on how they built Akinator? Or can you explain in more detail what one might need to do to build a solution like that? $\endgroup$ – Martin Larsson Jun 3 '18 at 9:14
  • 1
    $\begingroup$ go check en.akinator.com , if you are able to implement decision trees on your dataset then it is almost done. $\endgroup$ – riemann77 Jun 3 '18 at 9:58
0
$\begingroup$

The idea with the decision tree was a good guess, but the better approach is “model-based-reasoning for fault diagnosis”. That's sounds a bit complicated so let us explain it from bottom up. The first modul what we need is a prediction engine. That is a piece of software which contains domain specific knowledge, for example about car repairing. A ready to run example can be seen in one of the youtube-clips about “car repairing games”. In such games, the player has at the beginning a malfunction car with the wrong color, broken wheels and non-working engine and he must initiate some actions to fix the problem. The game-engine (model based reasoning) is calculating the future state if the player had done an action.

According to AI literature such a model is usually stored in a PDDL file, but it is also possible to use more advanced technology like the Loom knowledge representation language. The second modul is a planner, which determines automatically which actions are needed to bring the system into a given state. Both submoduls combined (symbolic game engine plus planner) are able to work as a troubleshooting expert system.

The most difficult part is to transform the domain knowledge into a prediction engine. In real games this is done manually in software projects. The idea is to write a simulator which behaves like the original. For example, if the car has an empty tank and the user is sending the command “refuel”, then the simulator reacts to the input visually.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.