1 / 24

Semantic Resolution in a Simple E-Commerce Application

This paper explores the concept of semantic resolution in e-commerce applications, including a simple scenario, semantic mapping algorithms, and agent system architecture. The goal is to understand the meaning of messages exchanged between software agents.

nbrister
Download Presentation

Semantic Resolution in a Simple E-Commerce Application

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Semantic Resolution in a Simple E-Commerce Application Zhongli Ding 12/03/2001

  2. Outline • Motivation • A Simple Scenario • Operations for Semantic Resolution • Agent System Architecture • Semantic Resolution Protocol • Semantic Mapping Algorithms • Future Direction

  3. Motivation • Understanding the meaning of messages exchanged between software agents is very important to interoperable multi-agents systems. • Forcing all agents to use some common vocabulary defined in shared ontologies will be an oversimplified solution since all agents are designed and used somewhat independently. • Semantic differences should be allowed and be resolved when they arise during agent run-time interaction.

  4. A Simple Scenario

  5. A Simple Scenario • Shared Ontology: ONT-0 • Buyer Agent (A1): ONT-1 • Seller Agent (A2): ONT-2 • ONT-1 and ONT-2 are based on ONT-0 • Semantic Resolution: The process of identifying the meaning of terms defined in different ontologies and matching these terms semantically. • Ontology Representation Language: DAML+OIL

  6. Operations for Semantic Resolution • Semantic Querying: The process of obtaining the description of a term/concept from a different ontology via some agent communication messages. • Semantic Mapping: The hypothesize-and-test process of map or re-classify the description got about source concept into one or more concepts defined in the target ontology. Measures of partial match are needed.

  7. Agent System Architecture

  8. Agent System Architecture • Reasoning Engine Module • Performs semantic matching (including partial matching, ranking, etc.). • Enforces the communication protocol (determining the performative, content, and values of other parameters for the next message). • Will be implemented by XSB, a prolog engine.

  9. Agent System Architecture • Message Decoding Module • Converts DAMLized message to RDF triples as Java objects (using RDFAPI, a DAML parser developed by DAML group). • Inserts RDF triples into XSB (using YAJXB, a tool developed by DAML group that bridges Java and XSB). • These triples will trigger appropriate prolog rules in XSB and start the reasoning for the agent’s next action. • Message Encoding Module • Constructs FIPA ACL messages based on reasoning results from XSB, using predefined classes of message types. Note: FIPA ACL messages are also DAML+OIL encoded.

  10. Agent System Architecture • Communication Module:This module will be implemented in Java. We choose MQ Series, a IBM Message Oriented Midware, as the message transport mechanism. MQ Series provides guaranteed message delivery and a set of APIs for Java. The main task of this module is: • Accesses MQ Series (open channel, wait, write/read messages). • Passes messages between MQ Series and Message Decoding/Encoding modules. • Agent Control Module • Coordinates operations and interactions among these components. Will be implemented in Java.

  11. Semantic Resolution Protocol We need: • An ACL to encode messages: FIPA ACL • A content language to encode the content of messages: DAML+OIL • A communication protocol that specifies how these messages can be used for meaningful conversations: “Semantic Resolution Protocol (SRP)”

  12. Performatives call-for-proposal (CFP) propose accept-proposal reject-proposal terminate inform not-understood query-if confirm disconfirm The first 5 performatives are for RFQ. The rests are for semantic querying and mapping. Semantic Resolution Protocol

  13. Semantic Resolution Protocol

  14. Semantic Resolution Protocol

  15. Semantic Resolution Protocol

  16. Semantic Mapping Algorithms • Use heuristic methods for approximating partial matches. • We have 3 algorithms: • Subsumption • Search for Plausible Subsumees • Hypothesize-and-Test

  17. Subsumption • subsumption(A, B, theta) is an extension of the structural comparison for subsumption operation in DL, based on some heuristic measures. It returns a numeric score, theta, in [0, 1] that quantifies the degree that concept A subsumes concept B, or -1 if A and B are disjoint. • A concept has a list of all primitive super-classes P and a list of all properties R. A subsumes B only if: • Pa is a subset of Pb . • Constraints on Rb is compatible with (i.e., is at least as strict as) that of Ra.

  18. Subsumption • Denote Sa and Sb as the lists of primitive super-classes of A and B, respectively; and Pa and Pb be the lists of properties of A and B, including those inherited from their superclasses, respectively.

  19. Subsumption • Comparing the super-class lists Sa and Sb • Check if any member Sa_i in Sa is logically inconsistent with any member Sb_j in Sb. If inconsistency is detected, then A cannot subsume B. Otherwise, we proceed to compute a heuristic measure of the degree that Sa is a subset of Sb: • If this measure is –1, then the entire matching process stops (no comparison of properties will be performed), and returns –1, meaning that A cannot subsume B.

  20. Subsumption • Comparing the property lists Pa and Pb • Identify all matching pairs between Pa_i in Pa and Pb_j in Pb. Pa_i matches Pb_j if they have the identical property name, including the name space; or Pb_j is a sub-property of Pa_i (or vice versa). • Compute compatibility measure for each matching pair Pa_i in Pa and Pb_j in Pb. If their constraints are incompatible, a measure of –1 is given to that Pa_i. Otherwise, use some heuristic rule to compute a positive measure for that pair.

  21. Subsumption • Combining comparison results • Combine all the positive measures: • Combine all the negative measures: • Combine C1 and C2: • Normalize C (where i is over all |Pa| + 1 weights): • CN is returned as the final score of A subsuming B.

  22. Search for Plausible Subsumees • search(A) searches the concept space of the target ontology and returns a ranking of all candidate target concepts based on the degree they subsume the source concept A. • We choose to use the backtracking mechanism of prolog to conduct a depth-first search for all partial subsumes of a given source concept and rank them by their respective theta values.

  23. Hypothesize-and-Test • HT(A) conducts semantic mapping as hypothesize-and-test cycles and iteratively improves the solution quality.

  24. Future Direction • Develop a more formal treatment for approximating semantic mapping with partially matched concepts. One approach is to incorporate probability theory, in particular the Bayesian belief network, into the ontology class hierarchies. • Develop a theoretical model of semantic resolution for differentiated ontologies. This model views the semantic resolution as evidential reasoning, in which the evidences are incrementally accumulated via semantic querying and the solution gradually emerges through semantic mapping with each new evidence.

More Related