Weighted Slotted RuleML for Similarity Matching in AgentMatcher Information Agents Harold Boley, NRC...

10
Weighted Slotted RuleML for Similarity Matching in AgentMatcher Information Agents Harold Boley, NRC IIT e- Business Virendra Bhavsar, UNB, Faculty of Computer Science 2 November 2002 Revised: 26 February 2005

Transcript of Weighted Slotted RuleML for Similarity Matching in AgentMatcher Information Agents Harold Boley, NRC...

Page 1: Weighted Slotted RuleML for Similarity Matching in AgentMatcher Information Agents Harold Boley, NRC IIT e-Business Virendra Bhavsar, UNB, Faculty of Computer.

Weighted Slotted RuleML forSimilarity Matching in AgentMatcher

Information Agents

Harold Boley, NRC IIT e-BusinessVirendra Bhavsar, UNB, Faculty of Computer Science

2 November 2002Revised: 26 February 2005

Page 2: Weighted Slotted RuleML for Similarity Matching in AgentMatcher Information Agents Harold Boley, NRC IIT e-Business Virendra Bhavsar, UNB, Faculty of Computer.

26-Feb-05RuleML 0.882

Slotted RuleML Via the Name-Giving Metarole slot (I)

Here is a self-explaining 'metarole' (slot) representationof the slotted, RDF-like Jess fact

(automobile (make Ford) (model Explorer) (year 1999))

in Object-Oriented (OO) RuleML – part of RuleML 0.88 –with user slots named via the first subelement:

<Atom> <opr><Rel>automobile</Rel></opr> <slot><Ind>make</Ind><Ind>Ford</Ind></slot> <slot><Ind>model</Ind><Ind>Explorer</Ind></slot> <slot><Ind>year</Ind><Ind>1999</Ind></slot></Atom>

Page 3: Weighted Slotted RuleML for Similarity Matching in AgentMatcher Information Agents Harold Boley, NRC IIT e-Business Virendra Bhavsar, UNB, Faculty of Computer.

26-Feb-05RuleML 0.883

Slotted RuleML Via the Name-Giving Metarole slot (II)

This 'Slotted RuleML' notation corresponds to the'positionalized' ruleml-datalog notation

<Atom> <opr><Rel>automobile</Rel></opr> <Ind>Ford</Ind> <Ind>Explorer</Ind> <Ind>1999</Ind></Atom>

if the 'roles' of the make, model, and year positionsare remembered somewhere else (signature declaration)

Page 4: Weighted Slotted RuleML for Similarity Matching in AgentMatcher Information Agents Harold Boley, NRC IIT e-Business Virendra Bhavsar, UNB, Faculty of Computer.

26-Feb-05RuleML 0.884

Slotted RuleML Via the Name-Giving Metarole slot (III)

Note that a mix of an ordered sequence of args androle-unordered args will thus be possible in facts as well:

<Atom> <opr><Rel>automobile</Rel></opr> <Ind>Ford</Ind> <Ind>Explorer</Ind> <Ind>1999</Ind> <slot><Ind>mileage</Ind><Ind>65000</Ind></slot> <slot><Ind>color</Ind><Ind>white</Ind></slot></Atom>

Also, Ind elements can be replaced by (typed) Var elements in all of the positional, slotted and mixed notations

Page 5: Weighted Slotted RuleML for Similarity Matching in AgentMatcher Information Agents Harold Boley, NRC IIT e-Business Virendra Bhavsar, UNB, Faculty of Computer.

26-Feb-05RuleML 0.885

Extending Slotted RuleML for Weighted AgentMatcher Keyphrases

AgentMatcher’s weighted keyphrases as Jess-likerole-weighted fact:

(automobile (make 0.7 Ford) (model 0.1 Explorer) (year 0.2 1999))

Weighted Object-Oriented (WOO) RuleML – part of RuleML 0.88 – extends user slots by an XML attribute for such weights:

<Atom> <opr><Rel>automobile</Rel></opr> <slot weight="0.7"><Ind>make</Ind><Ind>Ford</Ind></slot> <slot weight="0.1"><Ind>model</Ind><Ind>Explorer</Ind></slot> <slot weight="0.2"><Ind>year</Ind><Ind>1999</Ind></slot></Atom>

Page 6: Weighted Slotted RuleML for Similarity Matching in AgentMatcher Information Agents Harold Boley, NRC IIT e-Business Virendra Bhavsar, UNB, Faculty of Computer.

26-Feb-05RuleML 0.886

Role-Weighted Slotted RuleML Queries for AgentMatcher Similarity Matching

AgentMatcher agent can then use a Jess-likerole-weighted query:

(automobile (make 0.5 Ford) (model 0.1 Explorer) (year 0.4 Y))

RuleML 0.88 can use a corresponding query, and similaritymatching with the above fact succeeds, binding Y = 1999:

<Atom> <opr><Rel>automobile</Rel></opr> <slot weight="0.5"><Ind>make</Ind><Ind>Ford</Ind></slot> <slot weight="0.1"><Ind>model</Ind><Ind>Explorer</Ind></slot> <slot weight="0.4"><Ind>year</Ind><Var>Y</Var></slot></Atom>

Page 7: Weighted Slotted RuleML for Similarity Matching in AgentMatcher Information Agents Harold Boley, NRC IIT e-Business Virendra Bhavsar, UNB, Faculty of Computer.

26-Feb-05RuleML 0.887

Flat Feature Terms in Slotted RuleML

The minimal 'metarole' (slot) representation of the slotted,RDF-like (untyped) feature term, similar to F-logic term,

auto[make -> Ford; model -> Explorer; year -> 1999]

in RuleML 0.88 with user slots named via the first subelement:

<Cterm> <opc><Ctor>auto</Ctor></opc> <slot><Ind>make</Ind><Ind>Ford</Ind></slot> <slot><Ind>model</Ind><Ind>Explorer</Ind></slot> <slot><Ind>year</Ind><Ind>1999</Ind></slot></Cterm>

Can also be used as an argument in facts

Page 8: Weighted Slotted RuleML for Similarity Matching in AgentMatcher Information Agents Harold Boley, NRC IIT e-Business Virendra Bhavsar, UNB, Faculty of Computer.

26-Feb-05RuleML 0.888

Nested Feature Terms in Slotted RuleML

The representation of the nested (untyped) feature term

vehicle[winter -> auto[make -> Ford; model -> Explorer; year -> 1999] summer -> cycle[make -> Honda; model -> Magna; year -> 2002]]

in RuleML 0.88 with user slots named via the first subelement:<Cterm> <opc><Ctor>vehicle</Ctor></opc> <slot><Ind>winter</Ind> <Cterm> <opc><Ctor>auto</Ctor></opc> <slot><Ind>make</Ind><Ind>Ford</Ind></slot> <slot><Ind>model</Ind><Ind>Explorer</Ind></slot> <slot><Ind>year</Ind><Ind>1999</Ind></slot> </Cterm> </slot> <slot><Ind>summer</Ind> <Cterm> <opc><Ctor>cycle</Ctor></opc> <slot><Ind>make</Ind><Ind>Honda</Ind></slot> <slot><Ind>model</Ind><Ind>Magna</Ind></slot> <slot><Ind>year</Ind><Ind>2002</Ind></slot> </Cterm> </slot></Cterm>

Page 9: Weighted Slotted RuleML for Similarity Matching in AgentMatcher Information Agents Harold Boley, NRC IIT e-Business Virendra Bhavsar, UNB, Faculty of Computer.

26-Feb-05RuleML 0.889

Flat Weighted Feature Terms in Slotted RuleML for AgentMatcher Keyphrases

Representation of AgentMatcher’s weights in an extended feature term:

auto[make -0.7-> Ford; model -0.1-> Explorer; year -0.2-> 1999]

RuleML 0.88 user slots can again be extended by an XML attribute for capturing weights:<Cterm> <opc><Ctor>auto</Ctor></opc> <slot weight="0.7"><Ind>make</Ind><Ind>Ford</Ind></slot> <slot weight="0.1"><Ind>model</Ind><Ind>Explorer</Ind></slot> <slot weight="0.2"><Ind>year</Ind><Ind>1999</Ind></slot></Cterm>

Can also be used as an argument in facts

Page 10: Weighted Slotted RuleML for Similarity Matching in AgentMatcher Information Agents Harold Boley, NRC IIT e-Business Virendra Bhavsar, UNB, Faculty of Computer.

26-Feb-05RuleML 0.8810

Nested Weighted Feature Terms inSlotted RuleML for AgentMatcher Trees

Representation of AgentMatcher’s weights in a nested feature term:

vehicle[winter -0.6-> auto[make -0.7-> Ford; model -0.1-> Explorer; year -0.2-> 1999] summer -0.4-> cycle[make -0.5-> Honda; model -0.2-> Magna; year -0.3-> 2002]]

RuleML 0.88 with the XML attribute for capturing weights:<Cterm> <opc><Ctor>vehicle</Ctor></opc> <slot weight="0.6"><Ind>winter</Ind> <Cterm> <opc><Ctor>auto</Ctor></opc> <slot weight="0.7"><Ind>make</Ind><Ind>Ford</Ind></slot> <slot weight="0.1"><Ind>model</Ind><Ind>Explorer</Ind></slot> <slot weight="0.2 "><Ind>year</Ind><Ind>1999</Ind></slot> </Cterm> </slot> <slot weight="0.4"><Ind>summer</Ind> <Cterm> <opc><Ctor>cycle</Ctor></opc> <slot weight="0.5"><Ind>make</Ind><Ind>Honda</Ind></slot> <slot weight="0.2"><Ind>model</Ind><Ind>Magna</Ind></slot> <slot weight="0.3"><Ind>year</Ind><Ind>2002</Ind></slot> </Cterm> </slot></Cterm>