Re: It thinks that R1 P2 R2

>> > See the comment inline.
>> >
>> >> Consider the following statements...
>> >>
>> >> R1 P1 R2
>> >> R1 rdf:type C1
>> >> R2 rdf:type C2
>> >>
>> >> P2 rdfs:subPropertyOf P1
>> >> P2 rdfs:domain C1
>> >> P2 rdfs:range C2
>> >>
>> >> My engine thinks that
>> >>
>> >> R1 P2 R2
>> >
>> > The imaginary engine is not good enough.
>>
>> Anything I can do to improve the imaginary engine? In fact, I may use 
>> this
>> "inference" as an application in my project. Let me describe here...
>>
>> A part of the schema...
>>
>> <rdf:Property rdf:ID="connectSignalTo" xml:base="&e;">
>>   <rdfs:domain rdf:resource="#SignalOutput"/>
>>   <rdfs:range rdf:resource="#SignalInput"/>
>> </rdf:Property>
>>
>> before...
>>
>> <gom:Point rdf:nodeID="A0">
>>   <rdf:type rdf:resource="&e;SignalOutput"/>
>> </gom:Point>
>> <gom:Point rdf:nodeID="A1">
>>   <rdf:type rdf:resource="&e;SignalInput"/>
>> </gom:Point>
>>
>> creating a connection...
>>
>> <rdf:Description rdf:nodeID="A0">
>>   <gom:connectTo rdf:nodeID="A1"/>
>> </rdf:Description>
>>
>> "think"...
>>
>> <rdf:Description rdf:nodeID="A0">
>>   <e:connectSignalTo rdf:nodeID="A1"/>
>> </rdf:Description>
>>
>> "Think" is the only solution I devise for the above problem. The 
>> imaginary
>> proposes to the end user that the drawing connection, gom:connectTo, may 
>> be
>> a signal connection, e:connectSignalTo.
>>
>
> I can see the situation.
>
> General speaking,  there doesn't exist any neat solution to this fuzzy 
> issue.
>
> Thinking:
>
> What is the sufficient condition to make the assertion P2(R1, R2) true? 
> even if under some precondition such as P1(R1, R2), subPropertyOf(P2, P1), 
> R1 and R2 satifying the domain and range constraints of P2.
>

Yes, I agree with you that there exists a problem of the determination of 
the sufficient condition. Whether the statement P2(R1,R2) is a true 
assertion, should it be determined by the end-user. In the situation I 
described above, the imagined statement P2(R1,R2) will be prompted to the 
end-user. The end-user may configure the engine to accept the imaged 
statement manually or automatically. In fact, it's only a proposed feature 
of my engine, implementation of the imaginary feature is not really done 
yet.

>
> Probability-based reasoning can be used to partially solve this problem, 
> Or
>
> Machine learning, Bayesian network, Neural Network, Or
>
> Soft Computing Technique
>
> [snip]
>

I also agree with you that there are many technologies for learning 
thoughts. My experience of using some of them was not really satisfactory 
:(. For example, one time I helped a friend to "attack" the XOR problem 
using Neural Network in MS Excel. I failed to find out the correct 
parameters to solve the problem. It seems that I need to spend more time to 
get familar with any neural network. The same adverse situation may be 
encountered by some other users.

I think out the following formulae when I sit inside a bus today...

Score = (a*x + b*y + c*z) * w

where
a, b, c are parameters to be configured.
x is the percentage of rdfs:domain matched.
y is the percentage of rdfs:range matched.
z is the percentage of rdfs:subPropertyOf matched.
w is either 1 or 0, it means that whether the imagined statement should be 
rejected by some other reasons.

Can neural network apply to this kind of formulae?

>> >> Jeremy
>> >>
>> >
>> > Yuzhong Qu
>> >
>>
>> Jeremy
>>
>
> Yuzhong Qu
>

Jeremy 

Received on Sunday, 13 March 2005 12:44:22 UTC