I am interested in developing an Expert system for HR which will utilize Case-based reasoning to solve HR issues eg. compensation of employees. Data mining will also be a key element.

Ok, so you want to learn both rule-base design, and fuzzy logic design. What you do is build a fussy logic or neural network front-end, that you train with case based reasoning, and a Rule base depending on the typical HR decisions that need to be made.

However, you could get a much better system if you looked into the HR cognitive engine that was createded by Professor Sun.

Well, it's a matter of how rigid you want the expert system to be, isn't it. Rule base engines are notorious for expanding to exceed the capacity of their storage whenever something ambiguous comes along. (If it is self-learning) If you don't need it to learn, then it is more rigid and can't deal with new options. This is part of the problem with rule bases.

No one for instance has ever figured out how to implement Asimov's three laws of robotics in a rulebase. The terms are WAY too vague.

There is not best software. There is best theory, the so called axiomatic approach (utility theory, decision making in the sense of Professor H. Raiffa). The best solution is the prescriptive approach in regards to the normative axiomatic approach of von Neuman, Fishburn, Keeney, Kahneman, Tversky... In my opinion you could use the stochastic programming in realizing the prescriptive approach in decision making and decision support. In these conditions you will resolve the qualitative human preferences based problems with mathematical exactness.

Yes, well that is the classical A.I. approach, calculate it to death. For those of you who don't know all the big words, stochastic programming is using things like the monte carlo random walk approach to finding the right decision. By saying that it will deal with qualitative human preference based problems with mathematical exactness, he is really saying, That human qualitative preferences are beyond the brute force approach, and need to be dealt with using a heuristic such as random walks.

Because he doesn't recognize that fuzzy logic is a heuristic, the issue must be mathematical exactness, even though random walks are hardly mathematically exact, being to some level random.

A better strategy, is to use something like a Parallel Meta-Heuristic for the search. This does not guarantee the "Best" strategy, but it gets very close to it, and takes a lot less time than random walks.

The problem is, just how much new stuff you are going to have to learn to tool up to use it. Each technique has its own learning curve, and an HR system by its nature has too many soft parameters for the classic rulebase to deal with. Spend a little time looking at the heuristic tools and trying to figure out which ones will do the job the best.

Dear colleague, we could use Potential Function Method, Robins Monro procedure and/or other stochastic procedure following the ideas of Aizerman and Vapnik with a pseudo random sequence(not so big words, real working methods). Starting from here, we can use the gambling approach and we exactly know the learning points in the machine learning stochastic approach developed by Aizerman and Braverman. In this case with 64 or 128 (45 minutes) or 256 expert answers (human preferences in the gambling approach-yes, not or equivalent) we determine analytically the von Neumann utility function. The pseudo random sequence permits repetition and even correction of the expert answers.

Well, you certainly can quote chapter and verse, But not being a large fan of Von Neumans work, I am not terribly impressed when I compare the results against Sun's Clarion Cognitive Architecture.

45 minutes? on WHAT platform prey tell?

Dear colleague, thank you for your opinion!

I continue to stand in my position.

Perhaps you could find the following book and paper of Professor R. Keeney, about the conditions of utility function evaluation and applications in DSS:

1. Value-driven expert systems for decision support,

Ralph L. Keeney, Systems Science Department, University of Southern California, Los Angeles, CA 90089, USA;

2. R.L. Keeney, H. Raiffa: Decisions with multiple objectives–preferences and value tradeoffs, Cambridge University Press, Cambridge & New York.

Dear colleague, thank you for your opinion!

I continue to stand in my position.

Perhaps you could find the following book and paper of Professor R. Keeney, about the conditions of utility function evaluation and applications in DSS:

1. Value-driven expert systems for decision support,

Ralph L. Keeney, Systems Science Department, University of Southern California, Los Angeles, CA 90089, USA;

2. R.L. Keeney, H. Raiffa: Decisions with multiple objectives–preferences and value tradeoffs, Cambridge University Press, Cambridge & New York.

Well thankyou for the citation, I am sure some here will be pleased to follow up on it....

You didn't say what platform it took 45 minutes on.

You do realize that 45 minutes on a cray multiprocessor is not the same as 45 minutes on a pentium i7, or on a pentium 4? I wasn't actually trying to convince you to change your mind, but to note that there is more than one viewpoint as to best theory, and that you represented a more classical A.I. approach, than I represent. But since the idea here is to answer the question, I think that the dialoge is more important than the selection.

Without knowing that there are a range of answers, the individual might grab the first one.

(Rulebases) without looking to see if it worked best.

You could find the answer in the published paper “Value based decisions and correction of ambiguous expert preferences: an expected utility approach” (4 pages), http://www.researchgate.net/profile/Yuri_Pavlov/publications/.

I think the same that the dialoge is more important than the selection.

Could someone provide a paper, book or link to the HR cognitive engine createded by Professor Sun? Thank you.

You might try http://cnd.memphis.edu/ijcn2009/tutorials/helie.pdf Professor Sun worked out of Rensselear Polytechnique, but most of the documentation was done by Sebastien Helie. There are a number of tutorials based on the architecture. A Google search on "Clarion Cognitive Architecture"

seems quite productive:

Google claims about 267,000 hits, and Google Scholar about 14,000 but you can expect many of them to be reprints, or citations.

Let me know if you can't access Google and I can offer more

Value-driven design is a systems engineering strategy based on microeconomics which enables multidisciplinary design optimization. Value-driven design is being developed by the American Institute of Aeronautics and Astronautics, through a program committee of government, industry and academic representatives. In parallel, the US Defense Advanced Research Projects Agency has promulgated an identical strategy, calling it Value centric design. At this point, the terms value-driven design and value centric design are interchangeable. Intelligent DSS value-driven systems could be based on measurements of human notions, preferences and could be integrated with model-driven DSS which emphasizes access to and manipulation of a statistical, financial, optimization, or simulation model..

Value-driven expert systems for decision support,

Ralph L. Keeney, Systems Science Department, University of Southern California, Los Angeles, CA 90089, USA;

Thankyou Yuri