Understanding observation truth values

[TzahiJ]

I do not understand how to work with combined truth values of observation.

 Let me put the following example:

I want to match a dataset of people from a well curated database (D1) to information that was collected from the web, with much lower confidence(D2)

In D1 each person has a phone number(PSTN) and SSN, and the truth value is always 1.

In D2 people may have either SSN or PSTN, or both. Truth value for each observation is set to 0.25i:

The closed predicates are:

D1_PSTN(P,PSTN),D2_PSTN(P,PSTN)

D1_SSN(P,SSN), D2_SSN(P,SSN)

The open predicate is:

SamePerson(P1,P2)

Rules are:

5: !SamePerson(P1,P2)  -- prior

1.       10:  D1_PSTN(P1,PSTN) & D2_PSTN(P2,PSTN) -> SamePerson(P1,P2)

2.       10:  D1_SSN(P1,SSN) & D2_SSN(P2,SSN) -> SamePerson(P1,P2)

3.       1000: D1_PSTN(P1,PSTN) & D2_PSTN(P2,PSTN) & D1_SSN(P1,SSN) & D2_SSN(P2,SSN) -> SamePerson(P1,P2)

I think Rule number 3 is ineffective. It has a very high weight. Because it is unlikely that a person from D2 will have both PSNT and SSN like a person from D1 just by coincidence (even though the truth values are only 0.25).

But - Truth values are computed by Lukasiewicz t-norm. and between is computed as max(0,0.25+0.25-1), which is 0.

To my understanding – Giving high weight to a rule does not matter if the result has low truth value.

How can I make a rule that will assign high truth values when two low truth value observation appear in the same rule?

Thanks, Tzahi

[Eriq Augustine]

Hey Tzahi,

Just looking at this real quick, it seems to me like you need the "confidences" from D2 to be changable.

If the truth values for D2 can be modified, then PSL can use the evidence from D1 to raise the values in D2 and therefore make your third rule effective.

So I think you should make the D2_* predicates open and use a non-uniform prior to set the initial values at whatever initial confidence comes out of your D2 extractor.

(Also, a confidence of 0.25 is pretty low. Is their any way to increase the confidence that comes out of your D2 extractor?)

-eriq

--
You received this message because you are subscribed to the Google Groups "PSL Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to psl-users+...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.