Concept Similarity Measures the Understanding Between two Agents Jesús-M. Olivares-Ceja Adolfo...

18
Concept Similarity Measures the Understanding Between two Agents Jesús-M. Olivares-Ceja Adolfo Guzmán-Arenas www.jesusolivares.com [email protected] CIC - IPN MEXICO Research Center for Computing Centro de Investigación en Computación
  • date post

    21-Dec-2015
  • Category

    Documents

  • view

    213
  • download

    0

Transcript of Concept Similarity Measures the Understanding Between two Agents Jesús-M. Olivares-Ceja Adolfo...

Concept Similarity Measures the Understanding Between two Agents

Jesús-M. Olivares-Ceja

Adolfo Guzmán-Arenaswww.jesusolivares.com [email protected]

CIC - IPN

MEXICO

Research Center for ComputingCentro de Investigación en Computación

Agenda

Finding similar concepts: SIM algorithm

Degree of Understanding (du)

Finding similar concepts

thing (thing)

hardware(tool,hardware)

key (key)

data (data, item)

data_key (key)hammer (hammer)

thing (thing)

tool (tool)

tool_key (key)

vegetable (vegetable)

screwdriver (screwdriver)

citric(citric)

apple (apple)

lemon (lemon)

orange (orange)

wrench (key, wrench)

Finding similar concepts

thing (thing)

hardware(tool,hardware)

key (key)

data (data, item)

data_key (key)hammer (hammer)

thing (thing)

tool (tool)

tool_key (key)

vegetable (vegetable)

screwdriver (screwdriver)

citric(citric)

apple (apple)

lemon (lemon)

orange (orange)

wrench (key, wrench)

concept words

Finding similar concepts

thing (thing)

hardware(tool,hardware)

key (key)

data (data, item)

data_key (key)hammer (hammer)

thing (thing)

tool (tool)

tool_key (key)

vegetable (vegetable)

screwdriver (screwdriver)

citric(citric)

apple (apple)

lemon (lemon)

orange (orange)

wrench (key, wrench)

concept words

concept words

SIM algorithm

SIM has four cases

[Olivares 2002][Guzmán 2003]

A: Both parent and concept maps

B: Parent map, concept does not

C: Concept map, parent does not

D: None maps

SIM algorithm: case A

sv=1

fruit_vegetable(fruit, vegetable)

garlic_onion(garlic, onion)

vegetable(vegetable)

onion (onion)

white_onion (onion)

yellow_onion (onion)

purple_onion (onion)

SIM algorithm: case B

sv=0.75

citrus(citrus, citric)

lime(lime)

[color=green, texture=smooth. ...]

citric(citric)

celementine (clementine)

[color=green, texture=smooth,...]

SIM algorithm: case C

season_fruit(season_fruit)

apple(apple)

[shape=round, color=yellow. ...]

solid_food(solid_food)

apple (apple)

[shape=round, color=red,...]

sv=0.75

SIM algorithm: case Dthing

(thing, something)

perishable(perishable)

data_model(data_model, database)

relational_model (relational_model)

frozen_food (frozen_food)

meat(meat)

fruit_vegetable (...)

sv=0

Degree of Understanding (du)

du(B, A) = 1/ |A| * sim(cA, B)

du=0.53

Degree of Understanding (du)

du(A, B) = 1/ |B| * sim(cB, A)

du=0.17

Degree of Understanding (du)

du(B, C) = 1/ |C| * sim(cC, B)

du=0.07

Conclusion

No common ontology is needed for inter-agent communication

It is possible to discover concept cB which is most similar to CA

(sim discovers CB with sv [0,1])

It is possible for an agent to know how well it understands another agent

(du measures that, du [0,1])

Future Work

Improve knowledge representation expresiveness

Automatic knowledge extraction using NL techniques to build ontologies

Ontology merging (learning between two agents)

SIM should be tested with real knowledge to be tunned

Thank you

Gracias a ti

Tlazokamati

[email protected]

[email protected]

SIM algorithm: sv case B

Call sim recursively to confirm that pB is the ancestor of cA

If the pB' found is thing sv=0 otherwise

(a special son of pB is searched in OB that)

most properties matches (using sim)

sv = fraction of similar properties

if not found

(check among the sons or grandsons of the father of pB)

sv = (fraction of similar properties) * (0.8)

or

sv = (fraction of similar properties) * (0.8)2

if still not found

return “son_of” pB, sv = 0.5

SIM algorithm: sv case C

Properties of cB and sons are compared (using sim)

if more or equal than 0.8 is obtained cB is returned

otherwise

return “probably” cB and sv calculated

but if sv == 0 then return “not found” and sv = 0