That is a very good question that I had in mind myself when I started designing the product support system.
 
The advantage of having problem-solution pairs predetermined or explicitly created lies in the fact that there is a high success factor in choosing a correct solution. However, this strategy implies that a new solution is needed for each new answer, not only in terms of a new case but also of new documents. The continuous generation of new solutions degrades response times and increases document databases though, which results in document management and indexing difficulties.
 
In the case that a proposed solution is related to any existing one, not only the success factor can be undermined but also search algorithms are needed for optimizing the search efficiency and indexing capabilities (especially once the knowledge base, including the data and case bases, becomes significantly large).
 
Therefore, I decided to choose a “hybrid” strategy. The solutions have two basic parts, the case chosen and the document related to that case. I have abstracted both the cases and the documents into two major fields, “static” or “dynamic”. If the static fields are different then a new abstracted case and document are needed otherwise the problem refers to the same abstracted case and document. Different problems therefore can be paired with the same abstracted solution. This is similar to having ranges or thresholds in which several problems can be included. The abstracted solution is specialised according the dynamic parts of the problem, giving an accurate response to the user’s query.
 
As a safeguard (ensuring the correct solution is presented to the user), the user can give feedback to the system depending on whether he/she is satisfied with the presented solution, so that the solution can be either reformed either within the limits of the same abstracted solution or by generating a new solution.   
 
I hope this point is clarified now.

Your query actually is only partially answered till today and still remains an open research issue.
 
Traditional measures for IT systems should include response times etc. However, in the case of a product support system msec differences are not as important as usability and solution (documentation) quality. Usability studies and methodologies are in place already for performance support systems and interactive electronic technical manuals (product support systems have not been introduced till now). However, a comparison study has not been carried till now merely because the details from other studies are not publicly available.
 
The documentation quality is still an open research theme. General guidelines from organizations such as IEEE are in place but quantifiable measures have not yet been defined. This is one of the current issues that we also investigate.

This sounds like a good compromise and should balance the size of the database with its flexibility of response. The only drawback could be getting user feedback to evaluate the quality of the response, if the first response is not perfect, we would have to rely on people being patient enough to try again instead of giving up, still unsatisfied.

Is there such an existing software? Its sounds as though it may be a project of its own!

Rich.

We hope that after some relatively small usage of the system the number of inappropriate answers will be minimised since the problem solution pairs will become more. However,you are absolutely right that this could be a project on its own since there are a lot of different things that should be considered, such as how to present the feedback (form?) to the user, ask questions for clarification (?), how to deal with inaccurate comments and suggestions, how to solve conflicts after re-evaluation etc. The advantage is that a plethora of different techniques could be used as enablers.

Currently we are investigating Bayesian networks because of their good learning algorithms and causal characteristics. However, neural networks and genetic algorithms could be also researched as alternatives.

To the best of my knowledge there is not a dedicated software that can satisfy the aforementioned requirements, although built-in "satisfaction questionnaires" are included in several systems. However, these questionnaires have been evaluated manually till now.