SymposiumPlanner-2011 is the newest in a series of Rule Responder instantiations for the Questions&Answers (Q&A) sections of the o cial websites of the RuleML Symposia. It supports committee members via personal agents based on member-encoded rule knowledge. The personal agents are invoked by an organizational agent which selects them using a responsibility assignment matrix. This paper describes SymposiumPlanner-2011, which goes beyond the previous instantiations by providing a more powerful user interface and reducing redundancy in the internal conference data repositories.
multi-agent system. This project successfully bridged EMERALD and Rule Responder, and added new functionalities such as the mapping of paper topics to Symposium tracks.
For the two installments of the 5th International RuleML Symposium8, ”RuleML 2011 - Europe” and ”RuleML 2011 - America”, our SymposiumPlanner-2011 employs two Sub-Organizational Agents (Sub-OAs) and an overarching SuperOrganizational Agent (Super-OA). Along with the personal agents, these agents handle other issues associated with Super-OA-to-Sub-OA delegation, intelligent decision-tree-like Q&A, e.g. about where to submit theory papers, etc. We upgraded the communication middleware and rule agents of SymposiumPlanner to their latest versions and complement the user client to issue queries via a controlled natural language.
The rest of the paper is structured as follows: Section 2 introduces the SymposiumPlanner-2011 system and highlights its novel features. Section 3 describes the implementation of SymposiumPlanner-2011 in detail. Based on our experience and analysis, we close with some suggestions about SymposiumPlanner’s future development. 2
The SymposiumPlanner-2011 agents support the organizing committee members in their organization tasks, such as: helping the program chair to monitor and possibly move important dates, finding contact information about selected chairs of the symposium, answering questions of participants about the conference, etc. The organization of the symposium committee is modeled by a coordinator, who is responsible for communicating with the executors which complete tasks on behalf of the symposium chairs. In SymposiumPlanner, the coordinator is implemented as an organizational agent (OA), which knows the responsibilities and the roles of each executor (Symposium chair). The OA manages the executors, which are implemented as personal agents (PAs).
The 5th International Symposium on Rules 2011 has two installments ”RuleML 2011 - Europe” (IJCAI 2011) and ”RuleML 2011 - America” (BRF 2011). To better manage disparate symposiums, SymposiumPlanner-2011 provides three Organizational Agents (OAs) to model the overall RuleML-2011 organization and its two sub-committees of the two Symposium installments: - The Super-OA delivers and filters queries and requested tasks to the Sub
OAs, single point of entry for that specific instalment of RuleML 2011. - The Sub-OAs manage the organization committee members of the two RuleML2011 installments. They filter, decide and delegate incoming queries to responsible personal agents in the appropriate sub-organization.
An ontological RAM represents the roles and responsibilities of the Sub-OAs and the personal agents in the virtual organization of RuleML Symposium 2011.
Table 1 gives a RAM fragment of the committee chairs of RuleML Symposium 2011. Negotiation and distributed coordination protocols are applied to manage and communicate with the organizations’ agents and external agents.
The personal agents are self-autonomous agents and have their own rulebased decision and behavioral logic on top of their personal information sources, Web services, vocabularies/ontologies and knowledge structures. The rules are executed by di↵ erent platform-specific rule engines which communicate via an Enterprise Service Bus (ESB) using standardized Reaction RuleML messages (event messages).
Besides consulting knowledge from the internal data repositories, which was very common in previous instantiations, the SymposiumPlanner-2011 rule agents access external data sources to reduce redundancy in the conference data via query languages such as SQL, SPARQL, etc. External data sources such as calendars, vocabulary definitions, databases, web pages, metadata sources, personal data are dynamically queried at runtime and used as facts in the internal knowledge base of an agent. For instance, in SymposiumPlanner-2011, the rule agents access data coming from the Semantic Web Dog Food RDF database, which contains information about the RuleML-2011 proceedings such as authors, papers, reviewers.
As a Web-based Q&A application, the previous instantiations of SymposiumPlanner allowed users to issue queries via selection of the queries in adaptable Web form menus. The potential queries needed to be defined before delivering the system to users. It becomes a real burden when there are many kinds of queries available. For the purpose of resolving this problem, SymposiumPlanner 2011 provides a translator service, which can automatically translate public interface descriptions of function rules to Reaction RuleML messages. Meanwhile, a controlled English has a restricted syntax and a restricted semantics described by a small set of construction and interpretation rules. It is not di cult for users to learn it. SymposiumPlanner 2011 also strives for allowing users to issue the queries in controlled English and then translate them into the standardized Reaction RuleML messages.
SymposiumPlanner-2011 provides three Organizational Agents (OAs) to model the overall RuleML-2011 organization and its two sub-committees of the two Symposium installments. The general architecture of SymposiumPlanner 2011 is shown in Figure 1. In SymposiumPlanner-2011, we use the latest Prova 3.1.3 rule engine for the OAs, which is now an OSGi bundle and can run in an OSGi container with just a few dependencies. It packs a lot of features, some of them are inspired by the latest developments in modern computational logic, functional programming, distributed systems, and event driven architectures, some of them completely new and original. Some of very unique features include: reaction groups for event processing, guards and guarded cut, dynamic branches in workflows, and etc. Meanwhile, we utilize latest Mule ESB 3.1, which includes major improvements to Cloud Connect, including custom schemas for each connector, much simpler invocation of connectors, a new polling mechanism, message enrichment capabilities, and a simple yet powerful logging facility. In what follows, we will detail the implementation of the SymposiumPlanner-2011 system.
Mule open-source ESB9 allows deploying the rule-based agents on the Mule
object broker and supports the communication in this rule-based agent processing
network via a multitude of transport protocols. Mule provides a distributable
object broker which follows the Staged Event Driven Architecture (SEDA)10
pattern to manage all sorts of service components. This design decouples event and
thread scheduling from application logic and avoids the high overhead associated
with thread-based concurrency models[
The distributed agent services which at their core run the rule engines are installed as Mule components which listen at configured endpoints, e.g., JMS message endpoints, HTTP ports, SOAP server/client addresses or JDBC database interfaces. Reaction RuleML11 is used as a common platform independent rule interchange format between the agents (and possibly other rule execution / inference services). The translator services are used to translate inbound and outbound messages from platform-independent Reaction RuleML and/or controlled natural language into the platform-specific rule engines execution syntaxes and vice versa. 3.2
Generally speaking, each agent service might run one or more arbitrary rule
engines. Here we only describe the rule engine Prova12 which is used for
implementing the complex logic of the organizational agents. Prova follows the spirit
and design of the recent W3C Semantic Web initiative and combines declarative
rules, ontologies and inference with dynamic object-oriented Java API calls and
access to external data sources via query languages such as SQL, SPARQL and
XQuery [
File Input / Output ..., fopen(File,Reader), ...
XML (DOM) document(DomTree,DocumentReader) :- XML(DocumenReader),...
SQL ... ,sql_select(DB,cla,[pdb_id,"1alx"],[px,Domain]).
RDF ...,rdf(http://...,"rdfs",Subject,"rdf_type","gene1_Gene"),...
XQuery ..., XQuery = ’for $name in StatisticsURL//Author[0]/@name/text() return $name’, xquery_select(XQuery,name(ExpertName)),...
SPARQL ...,sparql_select(SparqlQuery,...
Prova has its separation of logic, data access, and computation and its tight integration of Java and Semantic Web technologies. Due to the natural integration of Prova with Java, it o↵ ers an economic and compact way of specifying
10 http://www.eecs.harvard.edu/⇠ mdw/proj/seda/ 11 http://reaction.ruleml.org 12 http://prova.ws agents’ behavior while allowing for e cient Java-based extensions to improve performance of critical operations. The main language constructs of messaging reaction rules in Prova are: sendMsg, rcvMsg and rcvMult. For instance, the following query retrieves the tracks of the symposium RuleML 2011@IJCAI. getTracks(XID,Track):% look-up responsible agent (Program Chair) from RAM assigned(XID,Agent,ruleml2011ATijcai_ProgramChair,ruleml2011ATijcai_responsible), % send the query to personal agent sendMsg(XID,esb,Agent, "query", getTrack(Track)), % receive answers multiple times rcvMult(XID,esb,Agent, "answer", substitutions(Track)). 3.3
Reaction RuleML acts as an interchange language between distributed rule
agents in SymposiumPlanner. It is a general, practical, compact and user-friendly
XML-serialized sub-language of RuleML for the family of reaction rules and
incorporates various kinds of production, action, reaction, and knowledge
representation temporal/event/action logic rules as well as (complex) event/action
messages into the native RuleML syntax using a system of step-wise extensions
[
Distributed rule agents can be programmed by the proprietary languages and engines. Reaction RuleML provides a translator service framework which translates the rule messages from and to Reaction RuleML, controlled natural language and platform-specific rule languages, such as Prova, Drools, POSL, EMERALD, etc. For example, the query in Attempto Controlled English (ACE) ”What is the contact-information of the general-chair-of-RuleML-2011-IJCAI?” can be translated into a Reaction RuleML query message, in order to get the contact information of the RuleML 2011@IJCAI general chair. <RuleML> <Message mode="outbound" directive="query-sync"> <oid><Ind>SymposiumPlannerSystem</Ind></oid> <protocol><Ind>esb</Ind></protocol> <sender><Ind>User</Ind></sender> <content> <Atom> <Rel>getContact</Rel> <Ind>ruleml2011ATijcai_GeneralChair</Ind> <Var>Contact</Var> </Atom> </content> </Message> </RuleML>
This Reaction RuleML message is then translated into a Prova message query which is executed in the engine:
The SymposiumPlanner user client supports two solutions to construct queries. The straightforward solution uses an XML based file, which describes publicly accessibly rule functions together with their mode and type declarations, to create HTML forms. After user initializes the parameters, the translator service combines the structure of function rules and the parameters values to create the standardized Reaction RuleML message.
SymposiumPlanner user client also allows issuing queries in a controlled
natural language (Attempto Controlled English [
By parsing the DRS and applying the domain-specific rules which map named objects to constants and predicate relations to atomic predicates. The core element of the result is shown as follows: <Atom> <Rel>getAuthorsOfPaper</Rel> <Ind>’Rule-based Distributed and Agent Systems’</Ind> <Var>author</Var> </Atom> 4
The communication middleware Mule uses SEDA, which decomposes the
processes of Q&A in SymposiumPlanner with event-driven stages connected by
explicit queues. Theses queues enable Mule to decouple the receiver of a message
from the other steps in processing the message. That means the operations of
SymposiumPlanner lie mostly on the SEDA processing mechanism and its
performance. As a well developed framework, the performance of SEDA has been
evaluated by many e↵ orts. In [
In future, while there is always a potential need to add more human support to the system in order to enhance flexibility, we will also aim at achieving more e ciency in terms of responses from agents as they automate redundant tasks that human users can avoid and be able to respond to increasingly complex queries. This may also lead to the need for peer to peer communication between PAs in order to help each other in answering queries posed by external agents. In such a scenario we can see PAs using FOAF-like profiles to advertise their capability of solving any query and cooperating with other PAs and the OA to achieve a collaborative environment of query response. In the case of a complex query, the query can be decomposed into parts, the OA can then delegate parts of the decomposed query to relevant PAs to solve individually. The OA can then eventually assemble the responses from PAs into a complete solution and send it back to the External Agent. In view of the aforementioned and proposed developments in our e↵ ort to build a framework to assist human organizations, we can see the SymposiumPlanner truly provides the basis for our gradual transformation of workplaces into an e cient and productive environment.