JP4377390B2 - Agent device and program recording medium - Google Patents

Description

  The present invention relates to an agent device used in a multi-agent system and a program recording medium in which a program used to realize the agent device is recorded, and in particular, an agent device that can realize an efficient intelligent distributed environment, and The present invention also relates to a program recording medium on which a program used for realizing the agent device is recorded.

  An agent system is an autonomous software that uses a common language (for example, ACL (Agent Communication Language)) to link software to work on a computer system to check requests and assemble work. With the spread of the Internet, it is expected that various agent processes can be realized by handling information distributed on the Internet in an integrated manner.

  However, in order to link software in the agent world, it is necessary to know where an agent can provide a service that meets the request of the client. In addition, ACL only defines the grammar of the message and does not define terms for expressing the contents of the message. Therefore, it is necessary to have a translation function of this terminology.

  A multi-agent system is basically composed of an information requesting agent and an information providing agent. Conventionally, a mediating agent is prepared between these agents, and the requesting agent is sent to the mediating agent. In addition to having a function of selecting an agent that provides a service that meets the demands of the company, a translation function that absorbs differences in terminology between agents (called an ontology) was adopted. The intermediary agent is configured to have a function of integrating answers sent from a plurality of information providing agents.

  The present inventors have applied for a patent on a virtual catalog system that displays catalog information of product information and service information distributed in a network as one catalog information and announced it at an academic conference. The agent system of this virtual catalog system However, as shown in FIG. 27, a configuration is provided in which a mediating agent is prepared between a user agent (which connects the agent world and the browser) and a database agent (which connects the agent world and the database). The intermediary agent is configured to have a database agent selection function that manages the catalog information required by the user agent and a translation function that absorbs differences in terminology between agents. Yes. The intermediary agent is configured to have a function of integrating answers sent from a plurality of database agents.

  In other words, in the virtual catalog system shown in FIG. 27, the database agent can be handled in the virtual knowledge base by advertising (advertisement) the ability of the virtual knowledge base that abstracts the database to the mediating agent. It employs a configuration for notifying categories and ontologies used in the virtual knowledge base.

  In such a configuration, when the user issues a search request via the browser, the user agent converts the search request into an ACL message and sends it to the mediation agent. The database agent that can handle the ACL message is selected from the previously received advertisement information, and the terms of the ACL message are translated into the terms that can be handled by the selected database agent, and then the ACL message is sent to the database agent. .

  Upon receiving this search request ACL message, the database agent generates a SQL (database operation language) command from the ACL message, searches the database, assembles the search result as an ACL message, and sends it to the mediating agent. Upon receiving the search result ACL message, the mediation agent collects the search result responses sent from the plurality of database agents as one message and answers the user agent.

  In this way, the virtual catalog system disclosed by the present inventors has realized that catalog information of product information and service information distributed on the network is displayed as one catalog information. The configuration is such that a mediation agent is prepared between the user agent and the database agent, and this mediation agent has a database agent selection function for managing catalog information required by the user agent, and between the agents. It adopts a configuration that has a translation function that absorbs differences in terminology. The intermediary agent is configured to have a function of integrating answers sent from a plurality of database agents.

  Certainly, when constructing a multi-agent system, a mediation agent that performs mediation processing between agents is prepared, and this mediation agent has an agent selection function that provides a service that responds to the request of the requester. It is an extremely effective method to adopt a configuration in which a translation function that absorbs differences in terminology between agents and a function that integrates response results are provided.

  However, adopting such a configuration has the following problems.

  In other words, the functions of selecting an agent, the function of translating messages for exchanging with each selected agent, and the function of integrating multiple response messages are all concentrated on the mediating agent. When the load increases and a large amount of exchanges occur between agents, there is a problem that the responsiveness deteriorates.

  And even when there is no need to integrate reply messages, you must always go through a mediation agent even if you do not need a mediation agent, such as you must go through a mediation agent. There is a problem of waste.

  In order to solve this problem, it is conceivable to divide a plurality of functions of the mediation agent and make each agent in charge. For example, as shown in FIG. 28, it is possible to adopt a configuration in which the message translation function and the agent selection function are mediated by an agent other than the mediation agent.

  However, when this method is used, for example, to select the target agent and send a message to the agent, an agent having an agent selection function selects an access destination agent. Multiple agents, and so on, to the agent that has the message translation function, then asks the agent to translate the message and then passes the translated message to the selected agent. When a group of processes are executed over a number of times, the processes cannot be executed at once.

  From now on, the intermediary agent responsible for each function will perform exchanges with the intermediary agent acting as the organizer, and the number of exchanges will increase.

  The present invention has been made in view of such circumstances, and provides a new agent device capable of realizing an efficient intelligent distributed environment, and a new program in which a program used for realizing the agent device is recorded. It is intended to provide a program recording medium.

In order to achieve this object, the agent device of the present invention includes (1) receiving means for receiving a search request message sent from another agent device indicating search contents, and (2) search received by the receiving means. An extraction means for extracting a service list defined by a connection of paired data between the delivery destination information of the message and the processing information specified by the message, added to the request message; and (3) included in the search request message Selection means for extracting search contents and selecting database agent apparatuses corresponding to the search contents based on knowledge about capability information of the plurality of database agent apparatuses; and (4) whether a plurality of database agent apparatuses are selected by the selection means. If it is determined that multiple selections have been made, the service list extracted by the extraction means Create a detailed service list with a service list that combines delivery destination information that specifies an integrated agent device that integrates the search results of multiple database agent devices and processing information that instructs to integrate those search results. Refinement means, and (5) a transmission means for adding the refinement service list created by the refinement means to the search request message received by the reception means, and transmitting to the plurality of database agent devices selected by the selection means. Configure to include.
When this configuration is adopted, the service list of the search request message received by the receiving unit may include the term system type in the output of the search result. In this case, the refinement unit outputs the search result. And whether the terminology type used by each of the plurality of database agent devices selected by the selection means is different. , Before and after the service list designating the search processing added to the search request message received by the receiving means, processing information for instructing conversion of both terminology types, and a translation agent device for processing the conversion Create a detailed service list with a service list combined with the specified delivery destination information.
Next, the present invention will be described according to the principle configuration diagram of the present invention shown in FIG.

  In the figure, 1a is one or a plurality of first agent devices configured by technology related to the present invention, 1b is one or a plurality of second agent devices comprising the present invention, and 1c is related to the present invention. And a plurality of third agent devices configured by the technology.

  The first agent device 1a issues a processing request message that takes an abstract level description format, and the second agent device 1b embodies the processing of the message issued by the first agent device 1a. The agent device 1c is activated in order according to the message specified by the second agent device 1b, executes the processing requested by the first agent device 1a, and returns the processing result.

  Based on the knowledge of the system, the first agent device 1a generates a service list that is defined by the data pairing between the delivery destination information of the message transmitted to the other agent device and the processing information specified by the message. And a transmission unit 11 that adds the service list generated by the generation unit 10 to the message transmitted to the other agent device and transmits the message.

  This generation means 10 basically generates a service list that takes an abstract level description format, but if possible, generates a service list that takes a detail level description format, or a part of the detail level. A service list having an abstract level description format including the description format is generated.

  On the other hand, the second agent device 1b includes a receiving unit 20 that receives a message sent from another agent device, and an abstract level (may include a detail level) added to the message received by the receiving unit 20. Based on the knowledge of the system, the extracting means 21 for extracting the service list, the refining means 22 for refining the abstract level service list extracted by the extracting means 21, and the message received by the receiving means 20 are detailed. A transmission means 23 for adding a service list to be detailed by means 22 and transmitting it to another agent device is provided.

  The detailing unit 22 may perform detailing until the process is completed, or may perform detailing in a form in which a part of the detailing process is entrusted to a subsequent agent device. When there are a plurality of message delivery destinations, the message is duplicated, and the details are executed while recording the number of duplicates in the service list.

  On the other hand, the third agent device 1c has a receiving means 30 for receiving a message sent from another agent device, and a detail level (may include an abstract level) added to the message received by the receiving means 30. Extraction means 31 for extracting the service list, and execution means for executing processing specified by the processing information at the top of the service list when the delivery destination information at the top of the service list extracted by the extraction means 31 specifies the own device 32, a deletion unit 33 that deletes a service list part to be executed by the execution unit 32, and a service list excluded from the deletion process of the deletion unit 33 to the execution result message of the execution unit 32 to add another agent Transmitting means 34 for transmitting to the apparatus.

  When the number of replicas is recorded in the service list extracted by the extracting unit 31, the executing unit 32 executes a process specified by the processing information after messages corresponding to the number of replicas are output. In addition, when there is processing capability for one or more processing information following the processing information at the head of the service list, the processing specified by the processing information may also be executed.

  Further, the third agent device 1c, based on the storage unit 35 for storing the message received by the reception unit 30, the execution result of the execution unit 32 and the knowledge of the system, A correction means 36 for correcting one or both of the processing information, a determination means 37 for determining whether or not a message to be transmitted returns to the own apparatus by referring to the received service list, and a determination When the means 37 determines that the message returns to its own device, the exclusion means 38 removes the processing information to be executed by the own device from the service list of the message when the message returns. When the message to be removed by the exclusion means 38 returns, the processing information removed by the exclusion means 38 is displayed as the message. Provided in the the return means 39 for returning to the service list with the collection means 40 for collecting the processing state of the message of the agent device under its subject device.

  The exclusion means 38 may also remove processing information executed by other agent devices following processing information to be executed by the own device from the service list.

  Here, the functions of the first agent device 1a, the second agent device 1b, and the third agent device 1c are specifically realized by a program, and this program is recorded on a floppy disk or the like. Or recorded on a disk or the like connected via a line, and installed on the first agent device 1a, the second agent device 1b, or the third agent device 1c and operating on the memory from them. A multi-agent system constructed according to the present invention will be realized.

  In a multi-agent system including the second agent device 1b of the present invention configured as described above, when a processing request is issued from a user, the generation unit 10 of the first agent device 1a realizes the processing request. When the service means notifies the second agent device 1b of a message of a processing request, the service list is added to the message and transmitted.

  In response to the processing of the first agent device 1a, in the second agent device 1b, when the receiving means 20 receives the processing request message from the first agent device 1a, the extracting means 21 receives the processing request message. Extract the abstract level service list to be added to the message. In response, the refinement means 22 refines the extracted abstract level service list based on the knowledge of the system, and the transmission means 23 refines the service list in the message received by the reception means 20. To the third agent device 1c.

  In response to the processing of the second agent device 1b, in the third agent device 1c, when the receiving means 30 receives a processing request message from the second agent device 1b, the extracting means 31 receives the message. Extract a service list of detail level to be added to the message. In response to this, the execution means 32 executes the processing specified by the processing information at the top of the service list when the delivery destination information at the top of the extracted service list specifies its own device. The means 33 deletes the service list part to be executed by the execution means 32. When the execution result by the execution means 32 is obtained, the transmission means 34 adds the service list excluded from the deletion process of the deletion means 33 to the message of the execution result of the execution means 32, and another third agent device 1c. Send to.

  In this way, as shown in FIG. 2, the third agent device 1c is activated in accordance with the order of the delivery destination information included in the service list of the detailed level added to the message, and the processing information ( The processing specified by the service request service name, input status, and output status) is executed. Then, the third agent device 1c that is activated last responds the processing result to the first agent device 1a that is the processing request issuing source according to the delivery destination information that the service list has.

  At this time, the execution means 32 of the third agent device 1c performs the refinement while the detailing means 22 of the second agent device 1b duplicates the message and records the number of copies in the service list. The processing specified by the processing information is executed after the messages for the number of copies recorded in the received service list are prepared.

  Further, when the execution means 32 of the third agent device 1c has the processing capability for one or a plurality of processing information subsequent to the processing information at the head of the service list, the third third agent device 1c has another third destination as a delivery destination of the processing information. Instead of the agent device 1c, the processing specified by the processing information may be executed on behalf.

  Further, in the third agent device 1c, the storage unit 35 stores the message received by the reception unit 30. A service list indicating the behavior of the message is added to the stored message, so that when a failure occurs in the system, the failure analysis can be performed accurately and easily by referring to it. It becomes like this.

  The correction means 36 corrects (adds, deletes, or changes) one or both of the delivery destination information and the processing information of the received service list based on the execution result of the execution means 32 and the knowledge of the system. To do). This correction process makes it possible to cope with the case where the service list fluctuates depending on the execution result of the execution means 32.

  Further, when the determination unit 37 determines that the message returns to the device itself, the exclusion unit 38 executes the processing information (the other subsequent third information) to be executed when the message returns. In response to this message, the return means 39 receives the processing information removed by the excluding means 38 from the service list. Return to the service list of. By this process, it becomes possible to prevent the content of the self process from leaking to the other third agent device 1c.

  The collecting unit 40 collects the message processing status of the third agent device 1c under its own device. Since the service list indicating the behavior of the message is added to the message collected at this time, it is possible to accurately grasp the state of the third agent device 1c under its own device.

  As described above, according to the multi-agent system constructed in accordance with the present invention, message delivery destination information and a message are specified for a message exchanged between software such as agents distributed over a network. A service list defined by the connection of data to processing information is added, and the service list part that has been processed by the destination software is deleted, and the message with the service list is delivered from software to software. By adopting a configuration for executing processing, it becomes possible to construct an efficient distributed environment without placing a load on specific software.

  If it demonstrates concretely, according to the multi agent system comprised by this invention, the transmission of the message which is not one to one will be attained. Then, when selecting an agent that realizes a processing request, it is possible to extend the processing to select all processing performed across a plurality of agents.

  Then, when adding an agent, it is only necessary to register the addition of the additional agent with an agent such as a mediation agent. Also, for the complex processing performed using the additional agent, the agent such as a mediation agent is used. It is only necessary to add to the knowledge of the above, and since there is no need to change or add to other agents, it is possible to easily add / extend agents and add / extend complex processing contents based on them.

  Since the message processing state is assigned to the message itself, even when it is necessary to take a complicated processing procedure that spans multiple agents, one agent takes care of the processing procedure. Since it is not necessary to perform processing that distributes messages to multiple agents in order, the functions of the agents can be distributed, and the load can be avoided from being concentrated on one agent, and the load can be distributed efficiently. Is possible.

  Hereinafter, the present invention will be described in detail according to an embodiment applied to a virtual catalog system.

  FIG. 3 shows a system configuration of a virtual catalog system to which the present invention is applied.

  The virtual catalog system shown in this figure includes a network 900, a product / service database 100, a database agent 200, a browser 300, a user agent 400, a mediation agent 500, a first translation agent 600, a second The translation agent 700, the integration agent 800, and the network 900 are connected to each other.

  The database agent 200, the user agent 400, the mediation agent 500, the first translation agent 600, the second translation agent 700, and the integration agent 800 communicate with each other using ACL (Agent Communication Language). It is possible to communicate with each other.

  ACL is an inter-agent communication language that prescribes messages exchanged between agents. Knowledge representation language KIF (Knowledge Interchange Format) that expresses knowledge such as information contents and inter-agent communication transactions called performative. It consists of KQML (Knowledge Query and Manipulation Language), the language that defines the verb part, and Ontology, which indicates the type of terminology used in the agent.

  In order for an agent to communicate by ACL, it is necessary to abstract actual information as knowledge (virtual knowledge base: VKB) possessed by the agent. In the system of FIG. 3, the information handled is the product / service database 100, so that the information of the product / service database 100 is abstracted as the VKB possessed by the database agent 200.

  Access to the VKB consists of cutting out a part of the knowledge described in the VKB according to a condition and specifying an operation (extracting, deleting, rewriting, etc.) on the extracted knowledge. When accessing the VKB by ACL, the knowledge on the target VKB is extracted by the KIF relation, and the operation for the knowledge is designated by the KQML performative.

  This KIF relation includes "relation that associates a specific field of a VKB record with its value", "arithmetic relation that compares numerical values", and "relation that defines a logical combination of relations" ”And“ Relations that apply secondary conditions to obtain secondary results ”.

  The database agent 200 advertises its own ability to the mediation agent 500 in order to realize a virtual catalog system.

  The advertisement information includes a VKB name, a category that can be handled by the VKB, a field that appears in the VKB, an ontology used in the VKB, and a relation that can be used for accessing the VKB. An example of advertisement information is shown below.

((database Hyakumichihama market)
(=> (member? x Hyakumichihama market)
(isa? x produce)
(field-definition Hyakumichihama market product name 'is-text)
(field-definition Hyakumichihama market category name 'is-text)
(field-definition Hyakumichihama market category code 'is-number)
(field-definition Hyakumichihama market production name 'is-text)
...
(default-ontolgy standard.database.kif)
(allows-relational-db-query Momodohama market)))
In this advertisement information, the name of VKB is Hyakudohama market, the category handled is agricultural products, and there are product names / category names / category codes / producer names as fields, using standard ontologies, and between agents A negotiated relational database-like query method is available.

  As described with reference to FIG. 1, in the present invention, in a multi-agent system configured as shown in FIG. 3, a message is connected to a pair of data between message delivery destination information and processing information designated by the message. The service list defined in is added.

  FIG. 4 shows an example of a service list used when the present invention is applied to the virtual catalog system of FIG.

  In the service list shown in this figure, the service name to be executed by the agent at the delivery destination, the message status (such as the input ontology) before the process indicated by the service name is executed, and the process indicated by the service name are executed. Each service list unit is connected in a list form with paired data of the message status (output ontology, etc.) and the delivery destination agent name as one service list unit.

  This service list has the property of instructing the content and procedure of processing. For example, in the service list shown in FIG. 5A, first, the first translation agent 600 (t1) translates the search content specified by the message from the A ontology to the D ontology, and then the database agent. A search result is obtained by searching the search content at 200 (db1), and then the search result is translated from the D ontology to the A ontology by the second translation agent 700 (t2). The integrated agent 800 instructs that the search results searched by the three database agents 200 are integrated and the user agent 400 is answered.

  Here, when the service name, input state, output state, and delivery destination agent name recorded in the service list cannot be determined and are at an abstract level, as shown in FIG. "Will be attached.

  More specifically, the service name of the service list is declaratively describing the relationship between the input state of the service list and the output state of the service list. The contents of this relationship may be any one or more related to the processing in the agent, the processing to be performed on the message, the resetting and realization of the service list including the message delivery route and service name, and the transmission of the message to the next agent. Each agent performs processing depending on the specification and implementation of each agent while complying with the constraint.

  Anything is subject to processing indicated by the service name, but in the embodiment described below, for convenience of explanation, database search, translation to match the message to the delivery destination, delivery to the agent It is assumed that routes are set and multiple messages are merged.

  The main function of the user agent 400 is to convert a search request input through a browser into an ACL message, create an abstract level service list from the search request, and convert the abstract level service list into the ACL message. In addition, it is transmitted to the mediating agent 500.

  The main function of the mediation agent 500 is that when an ACL message having an abstract level service list is sent from the user agent 400, the ACL message is referred to by referring to the advertisement information sent from the database agent 200. The service list of the abstract level added to is concreteized by refining, and the service list of the sent ACL message is replaced with the service list of the detailed level, and then transmitted to the database agent 200.

  The main function of the database agent 200 is that when an ACL message having a detailed service list is sent from the intermediary agent 500, the ACL message is sent in response to the detailed service list added to the ACL message. The search content specified in the above is converted into an SQL (database operation language) command and a search process is executed. At this time, if necessary, the first translation agent 600 or the second translation agent 700 sends a translation request ACL. Send a message.

  The main functions of the first translation agent 600 and the second translation agent 700 are to combine the terminology used by the user agent 400 and the terminology used by the database agent 200, so that the user agent 400 Is to allow access to a plurality of database agents 200 having different terminology.

  Here, it is assumed that the first translation agent 600 executes the translation process from the A ontology to the D ontology, and the second translation agent 700 executes the translation process from the D ontology to the A ontology. .

  For example, as shown in FIG. 6, the first translation agent 600 is when the Hyakumichihama market database uses the D ontology, and the user issues a search request using the search screen of the A ontology system. Sometimes, using a translation function that performs translation from the A ontology to the D ontology, the item name “origin” used in the search screen of the A ontology system is converted to the “production area” used in the D ontology. At the same time, a translation process is executed such that the category code “01597218” input on the search screen of the A ontology system is converted to “30168945” used in the D ontology.

  7 shows an example of the processing flow executed by the user agent 400, FIGS. 8 and 9 show an example of the processing flow executed by the mediation agent 500, and FIG. 10 shows the database agent 200, the first translation agent 600, and the second FIG. 12 and FIG. 13 show an example of the processing flow executed by the integration agent 800.

  Next, according to these processing flows, the present invention will be described in detail.

  When the user agent 400 makes a search request for the product / service database 100 from the user, as shown in the processing flow of FIG. 7, first, in step 1, search is set as the service name and the ontology to be handled is input / An abstract service list is created by creating an output state and creating a service list unit in which the delivery agent name is undetermined and a service list unit in which the delivery agent name is its own agent name.

That is, if the ontology handled by the user agent 400 is an A ontology,
[ ^* Search, inA, outA, ^* ]
[ , , , Ua]
The abstract service list is created.

  Subsequently, in step 2, an ACL message having the search content as a message body is created. For example, when a request is made to search for a product / service database 100 with agricultural products as a category and search for producers handling apples of less than 1500 yen per 10 kg from Aomori Prefecture as the production area, An ACL message having such a search content as a message body is created.

  Subsequently, in step 3, the created abstract service list is added to the created ACL message and sent to the mediation agent 500, and the process is terminated.

  When receiving the ACL message of the search request issued by the user agent 400, the mediation agent 500 starts with the search contents of the received ACL message in step 1 as shown in the processing flow of FIGS. The category specified by the search request is extracted. In other words, when the search content includes agricultural products as a category, the fact that agricultural products are designated as a category is extracted.

  Subsequently, in step 2, by referring to the advertisement information from the stored database agent 200, the database agent 200 that handles the extracted category is selected. As described above, since the database agent 200 advertises capability information such as categories that can be handled to the mediating agent 500 in advance, the database agent 200 that handles the category specified by the search content is selected according to the advertisement information. It is.

  Subsequently, in step 3, it is determined whether there are a plurality of selected database agents 200. When determining that there are a plurality of database agents 200, the process proceeds to step 4, and the service list received from the user agent 400 has 2 A service name merge is added between two service list units, and the input state handled by the integrated agent 800 that executes this merge and the number of selected database agents 200 as denominators are “1”. Is registered as “1 / n”, and the output state is derived from the ontology of the output handled by the integrated agent 800 that executes this merge and “n × (1 / n) = 1”. “1” is registered, and the agent of the integration agent 800 that executes this merge is set as the destination agent name. To register the name.

That is, if the ontology handled by the integrated agent 800 is A ontology, the number of selected database agents 200 is “3”, and the agent name of the integrated agent 800 is m1, the service list received from the user agent 400 has 2 Between two service list units,
[Merge, outA: 1/3, out: 1, m1]
The service list unit is inserted.

  Subsequently, in step 5, the message is copied in accordance with the number of database agents 200 selected. That is, if the number of selected database agents 200 is n, the replication process is executed so that the number of messages is n. Messages duplicated by this duplication processing have the same ID.

  When the replication process in step 5 is completed and it is determined in step 3 that there is only one database agent 200 selected, the process proceeds to step 6 where the delivery destination agent possessed by the first service list unit in the service list In the name, the agent name of the selected database agent 200 is registered.

That is, when the agent names of the selected database agent 200 are db1, db2, and db3, the delivery destination agent names (users) of the first service list unit of the three messages created through step 4 / step 5 In the abstract service list sent from the agent 400, it was determined to be undecided)
[ ^* Search, inA, outA, db1]
[ ^* Search, inA, outA, db2]
[ ^* Search, inA, outA, db3]
In this way, the agent names db1, db2, and db3 are registered.

  Subsequently, in step 7, one of the selected database agents 200 is selected, and in step 8, it is determined whether or not all the database agents 200 have been extracted. 9), the service list detailed by the processes so far is added to the message and transmitted to the database agent 200, and the process is terminated.

  On the other hand, when it is determined in step 8 that the database agent 200 has been extracted, the process proceeds to step 10 (processing flow of FIG. 9), and the extracted information is referred to by referring to the advertisement information from the database agent 200 that is held. The ontology handled by the database agent 200 is checked, and in step 11, the ontology is registered in the input state and output state of the search service list.

That is, when the ontology handled by the selected database agent 200 is the D ontology, for example,
[ ^* Search, inA, outA, db1]
By registering the D ontology for the input state and output state of
[Search, inD, outD, db1]
In this way, the service list unit of search is refined.

  Subsequently, in step 12, whether or not the ontology handled by the extracted database agent 200 is different from the ontology used by the search request issuing user agent 400 (recorded in the abstract service list sent from the user agent 400). To check.

  When it is determined that the ontology handled by the database agent 200 is the same as the ontology used by the search request issuing user agent 400 according to this check processing, the process immediately returns to step 7.

  On the other hand, when it is determined that the ontology of the two is different, the process proceeds to step 13, where the service name translate is added before and after the search service list unit of the service list, and the input state of the translation source is added. After registering the ontology, register the ontology of the translation destination in its output state, register the agent name of the first translation agent 600 or the second translation agent 700 that executes this translate as the destination agent name, Return to Step 7.

In other words, the ontology used by the selected database agent 200 is the D ontology, the ontology used by the search request issuing user agent 400 is the A ontology, and the first translation agent that performs the translation process from the A ontology to the D ontology. If the agent name of 600 is t1, and the agent name of the second translation agent 700 that executes the translation process from the D ontology to the A ontology is t2, the search contents can be notified to the database agent 200. Before the search service list unit of the service list,
[Translate, inA, inD, t1]
In order to insert the service list unit and to notify the user agent 400 of the search result, after the search service list unit of the service list,
[Translate, outD, outA, t2]
The service list unit is inserted.

  In this way, when the mediation agent 500 receives a search request message having an abstract level service list from the user agent 400, as shown in FIG. 13 and FIG. 14, the abstract level service list is detailed. The message having the detailed service list is transmitted to the database agent 200.

  In FIG. 14, it is assumed that the translation processing by the first translation agent 600 and the second translation agent 700 is not performed because one database agent 200 uses the same ontology as the user agent 400.

  As shown in FIG. 15, the detailed level service list shown in FIG. 13 and FIG. 14 is first obtained from the A Ontology by the first translation agent 600 as necessary. The translation result is translated into the D ontology, and then the database agent 200 retrieves the retrieval content to obtain the retrieval result. Then, if necessary, the second translation agent 700 translates the retrieval result into the D ontology. The ontology is translated into the A ontology, and then the integrated agent 800 instructs to integrate the search results searched by the database agent 200 and to reply to the user agent 400.

  Next, processing executed by the database agent 200 / first translation agent 600 / second translation agent 700 / integrated agent 800 will be described.

  These agents basically execute the processing according to the processing flow of FIG. 10, but for the convenience of explanation, the integrated agent 800 is executed according to the detailed processing flow shown in FIGS. Will be explained.

  When the database agent 200 / first translation agent 600 / second translation agent 700 receives a message having a service list of a detailed level, first, as shown in the processing flow of FIG. The message body of the received message is taken out. In the case of this embodiment, the message body is a search content or a search result.

  Subsequently, in step 2, the service list possessed by the received message is taken out. As described above, this service list includes a service name to be executed by the delivery destination agent, an input state of the message, an output state of the message, and paired data of the delivery destination agent name in one service list unit. The service list unit is connected in a list form.

  Subsequently, in step 3, the head service list unit is extracted from the service list, and in step 4, it is determined whether or not the delivery agent name in the head service list unit indicates the own agent. When it is determined that the self agent is not pointed out, execution of the self process is not instructed, and the process is terminated as it is.

  On the other hand, when it is determined that it points to the own agent, the process proceeds to step 5 where the service name of the first service list unit is designated while referring to the designated input state and output state as necessary. Necessary changes are made to the message body by executing the process.

  That is, in the case of the first translation agent 600, the term system of the search content given in the message body is translated from the A ontology to the D ontology according to the designation of the service name “translate”. The product / service database 100 is searched according to the search content given in the message body according to the designation of the name, and if it is the second translation agent 700, the term of the search result given in the message body according to the designation of the service name translate The system is translated from D ontology to A ontology.

  Subsequently, in step 6, in response to the completion of the process, the output message of the processing result is assembled while deleting the first service list unit and adding the service list, and in step 7, the assembly output message is assembled. Is sent to the other agent and the process is terminated.

According to the processing flow of FIG. 10, the mediation agent 500 is shown in FIGS.
[Translate, inA, inD, t1]
[Search, inD, outD, db1]
[Translate, outD, outA, t2]
[Merge, outA: 1/3, out: 1, m1]
[ , , , Ua]
First, the first translation agent 600 is activated, and the search content given in the message body is translated from A ontology to D ontology as shown in FIG. 16 (a). Then, the database agent 200 is activated to obtain a search expressed in the D ontology, as shown in FIG. 16B, and then the second translation agent 700 is activated, and FIG. As shown, the search result searched by the database agent 200 is translated from the D ontology to the A ontology.

The service list at the end of this process is as shown in FIG.
[Merge, outA: 1/3, out: 1, m1]
[ , , , Ua]
It has become that.

  The integration agent 800 is activated in response to the merge of the service name of this service list.

  Next, processing executed by the integration agent 800 will be described according to the processing flow of FIGS. 11 and 12.

  When a message having a service list of a detailed level is sent, the integrated agent 800 first takes out the message body of the received message in step 1 as shown in the processing flow of FIG. 11 and FIG. . Since the message sent to the integrated agent 800 is a search result obtained by the database agent 200, in this step 1, when the message is addressed to its own agent, the search result is taken out.

  Subsequently, in step 2, the service list possessed by the received message is extracted, and in step 3, the first service list unit is extracted from the service list. Subsequently, in Step 4, when it is determined whether or not the delivery agent name in the head service list unit indicates the own agent, and when it is determined that the delivery agent name does not indicate the own agent, execution of the own process is instructed. Since it is not done, the processing is terminated as it is.

  On the other hand, when it is determined that it points to its own agent, the process proceeds to step 5 where the ID of the received message is extracted, and in step 6 it is determined whether or not a pool with the same message ID exists.

  If it is determined by the determination process in step 6 that there is no pool with the same message ID, the process proceeds to step 7 to create a pool corresponding to the message ID, and in step 8 the service list is output as the pool output value. The unit has an output value “1” (“1” derived from “n × (1 / n) = 1” registered in the output state in step 4 of the processing flow of FIG. 8). Subsequently, the process proceeds to step 10 where the message body (search result) extracted in step 1 is stored in the created pool.

  On the other hand, when it is determined by the determination process in step 6 that a pool with the same message ID exists, the process proceeds to step 9 to access the pool created for the message ID, and in step 10 the access is made. The message body (search result) extracted in step 1 is stored in the pool.

  When the storage process in step 10 is completed, the process proceeds to step 11 (process flow in FIG. 12), and the divided value of the service list unit extracted in step 3 (input state in step 4 of the process flow in FIG. 8). In step 12, it is determined whether or not the accumulated value has reached the output value “1” set in step 8.

  If it is determined by this determination processing that the cumulative value of the divided values of the first service list unit of the received message does not reach the output value “1”, the processing ends. On the other hand, when it is determined that the output value “1” is reached, that is, when it is determined that all of the copied messages are completed according to the processing flow of FIG. The message body (search result) is integrated into one, and in the subsequent step 14, the output message of the integrated integration result is transmitted to another agent, and the process is terminated.

  In this way, when there are a plurality of database agents 200 that handle search request categories, the integrated agent 800 performs processing to merge search results obtained from these database agents 200 as shown in FIG. .

In the output message of the integration result sent in this way,
[ , , , Ua]
In response to this service list, the user agent 200 that issued the search request obtains the search result integrated by the integrated agent 800.

  As a result, when a request to search for producers handling apples of less than 1500 yen per 10 kg is issued from Aomori Prefecture, the producers called Tsugaru Farm and Hirosaki Farm come from different database agents 200 When a search is made for a producer, the search results of the two producers are merged and returned to the user agent 400 that issued the search request, so that the user can search from one product / service database 100. Search results can be obtained.

  As can be seen from the description of this embodiment, the service list added to the message has the property that it represents a flow of a series of processes that undergo state transition while tracing the agent.

  From now on, if a configuration is adopted in which each agent stores messages received from other agents, the cause of the failure can be identified easily and accurately when a failure occurs. Each agent can easily and accurately grasp the progress of the process by referring to the message processing state of the agent under its control.

  FIG. 19 shows another example of the processing flow executed by the database agent 200 / first translation agent 600 / second translation agent 700 / integrated agent 800.

  The difference between this processing flow and the processing flow in FIG. 10 is that the processing flow in FIG. 10 employs a configuration in which only the processing specified in the first service list unit of the service list of the received message is executed. In the flow, as shown in step 6 and step 7, if the subsequent processing designated by the service list can be performed by the own agent, it is executed on behalf of the agent, and accordingly, in step 8 Therefore, the configuration is such that the executed service list unit portion is deleted from the service list.

For example, the database agent 200
[Search, inD, outD, db1]
[Translate, outD, outA, t2]
[Merge, outA: 1/3, out: 1, m1]
[ , , , Ua]
When the database agent 200 (db1) also has the translation function of the second translation agent 700 (t2), as shown in FIG. Instead, the translation process is also executed.

  Since the service list must be executed in order, even if the processing after the processing that cannot be executed by the local agent can be executed by the local agent, it cannot be executed on behalf of the local agent. When possible processing continues, it is possible to execute all of them on behalf of them.

  Such proxy processing is performed, for example, when the agent does not fully understand the capabilities of the agent requested to process the service list, or when multiple agents step through the same service list. However, optimization will be achieved by providing this proxy processing function.

  In this embodiment, the process that can be executed by oneself is executed in order. However, when there is a function that can execute a combination of processes that can be executed by one person at a time, the top of the service list is used. Multiple service list units can be taken out from a batch, and the matching between the service list unit and the process that can be executed by yourself is judged. If there is something that can be executed at once, it can be executed at once. preferable.

  FIGS. 21 and 22 show another example of the processing flow executed by the database agent 200 / first translation agent 600 / second translation agent 700 / integrated agent 800. FIG.

  The difference between this processing flow and the processing flow of FIG. 10 is that, in the processing flow of FIG. 10, after the processing of the head service list unit is executed, the output message of the processing result is transmitted as it is. In the flow, when it is determined whether there is one that specifies the own agent in the service list unit after the first service list unit, and when it is determined that there is one that specifies the own agent, that is, When it is determined that the message is returned to the local agent, a configuration in which the service name of the process is removed from the service list (the input state and output state may be removed) and the message is returned The point is that it is configured to execute it on its own agent.

  That is, when the process of designating the first service list unit of the service list of the received message addressed to itself is executed in step 7, the remaining service list includes the local agent name as the delivery destination in step 8. In step 9, the service name assigned to the service name is removed from the service list and saved.

  When receiving a message addressed to itself, in step 5, it is determined whether or not the service name in the first service list unit has been removed, and when determining that the service name has been removed, In step 6, the removed service name is processed.

  As can be seen from the above description, in the present invention, a message is not regarded as a single exchange, but as a series of exchanges that undergo state transitions while following various agents. Based on the technical idea of a message-oriented approach, a series of processing flow is defined with a focus on the message of what state transition the message gets to reach the target agent. Yes.

  In this approach of the present invention, various processing procedures are explicitly made externally difficult, so that processing conventionally performed by the mediating agent can be processed by another agent or determined by the mediating agent. The processing procedure that has been performed can be determined by another agent, and the number of message exchanges is also efficient.

  However, since all the processing performed on the message is written in the service list, the contents can be seen and misused. In addition, writing all the processing contents in the service list may increase the size of the message. Furthermore, there is a possibility that a security problem may occur when another agent performs the processing to be executed by the agent on behalf of the proxy processing described above.

  Therefore, in the processing flow of FIG. 21 and FIG. 22, processing information (service name, input state, and output state) that is executed when the message returns to its place is written back in the service list. Is hidden from other agents. The service list is restored when the user returns to his / her place.

  Here, the processing information executed by other agents following the processing information executed when returning to his / her place is also concealed by other agents and restored when returning to his / her place. You may take the structure of.

  Specifically, the restoration of the service list is realized by the following mechanism. That is, before sending a message, the agent saves processing information to be executed when it returns to its place in the service list storage pool in its own agent, and for the service list to be sent, the saved processing. Instead of information, a parameter for referring to the stored processing information is added. When the message is received, the reference parameter is recorded in the service list, and when it is determined that it is the reference parameter of the processing information to be stored, the processing information to be stored is restored to the service list. To do.

  According to the processing flow of FIG. 21 and FIG. 22, the contents of the self-processing can be hidden by other agents, the self-processing can be prevented from being delegated to other agents, and the size of the message can be reduced. In addition, when entering another system space (another knowledge space), it is possible to realize a function that guarantees that the user returns to the original system space via the agent used for entering.

  FIG. 22 shows another example of the processing flow executed by the database agent 200 / first translation agent 600 / second translation agent 700 / integrated agent 800.

  The difference between this processing flow and the processing flow of FIG. 10 is that the processing flow of FIG. 10 has a configuration in which the processing of the first service list unit is executed and then the service list unit is deleted. In the flow, when it is better to modify the remaining service list in accordance with the execution result of the process, a configuration is adopted in which it is modified.

  That is, in step 5, when the process for specifying the head service list unit of the service list of the received message addressed to itself is executed, it is determined in step 6 whether or not the service list should be corrected based on the execution result. If it is determined that it should be corrected, the service list is corrected in step 8.

  In the case of a specific example of an agent system that executes agent processing for travel, initially, when an unrestricted service list such as which airline's flights are used is created, hotel reservations are made. If the agent doing this determines that there is a discount incentive for the flight of a particular airline that is contracting with the hotel that has been booked, the name of the agent in the service list to be sent to the agent making the flight reservation will be The amendment is limited to airlines that can use discount incentives.

  According to this processing flow, the service list created initially can be dynamically changed to an appropriate one, so that a more practical agent system can be constructed.

  Next, details of the service list will be described.

  In the present invention, when an abstract level service list is created, it is refined according to the knowledge of the system obtained from advertisement information and the like.

  The details of the service list can be performed by a single agent, but a method may be employed in which the details are partially entrusted to one or more subsequent agents. It is efficient for subsequent agents who know the service better to refine the service list, and it is more efficient to refine the service list once the message processing has progressed to some extent. Because there is.

For example, when executing a travel agent process consisting of a hotel reservation and a flight reservation, as shown in FIG. 24, when the planning agent receives an abstract level service list of “ ^* travel search”, this abstract level By refining the service list of “C company hotel search / B company flight search” and passing it to the hotel search agent and flight search agent, C company hotel reservation and B company flight You may take the method of realizing the reservation.

Alternatively, as shown in FIG. 25, when the planning agent receives an abstract level service list of “ ^* travel search”, the abstract level service list is partially set as “ ^* hotel search / ^* flight search”. In response to this, the hotel search agent searches for a C company hotel in some detail, such as “Company C Hotel Search / ^* Flight Search”, and the flight search agent calls “Company B Flight Search”. By refining in such a manner, a method may be adopted in which a company B flight is searched and a company C hotel reservation and a company B flight reservation are realized.

  When adopting these methods, it is also possible to perform a plan using a plurality of planning agents having different algorithms, merge the results, and notify the user.

  Since it is not necessary to refine the service list all at once, according to the present invention, since one system can enter and use another system, distributed management of knowledge can be realized and distributed management can be performed. The feature is that the knowledge can be easily used.

  In other words, even if you do not know all the knowledge of other systems, if you can obtain a part of that knowledge, you can use that knowledge to refine the service list to enter that system, After that, you can refine the service list in the system and receive the processing results from the refined service list, so you can use the system without knowing all the knowledge of the system The distributed management of knowledge can be realized, and the knowledge that is distributed and managed can be easily used.

  For example, as shown in FIG. 26, when there are an organization A database and an organization B database, the mediation agent of organization B has knowledge about the database of organization A in addition to the knowledge of mediation agent of organization A. Even if it is not, the database of organization A can be accessed by requesting the mediation agent of organization A to refine the service list and receiving the processing result of the detailed service list.

  26 will be described in detail. The user agent of the organization B sends a search request message to the mediation agent of the organization B. At this point, it is specified that the service is a search request, and to which agent the search request is delivered is undefined.

Based on the service and knowledge, the mediation agent of organization B selects the agent in its own database that has a search request. At this time, the mediation agent of organization A is also selected. Next, each service is detailed. For the agent of the database of its own organization, the details such as “search → integration → user” are performed, and for the intermediary agent of the organization A, the abstraction level of “ ^* search” is performed. Thereafter, a message having a detailed service list is sent to another agent. At this time, the message to be sent to the mediation agent of organization A is sent after being converted into a format suitable for the knowledge space of organization A.

  Upon receiving a message from the mediation agent of organization B, the mediation agent of organization A selects an agent in its own database that has a search request based on the service and knowledge, and “search → conversion → integration → user” or the like. Send the message to the other agent after making the details.

  The database of organization A is searched by this detailed service list, the search result is returned to the integration agent of organization B via the conversion agent of organization A, and the user who operates the user agent of organization B The database A can also be a search target.

  Finally, knowledge distribution handled by the present invention will be described.

  The mediation environment with mediation agents is a powerful approach for linking disparate applications. The intermediary environment is summarized as follows.

  Create agents for each application and model the knowledge they handle. In other words, the definition of what is being done is given declaratively, and a fixed description is prepared for the data and knowledge format to be handled.

  Then, each agent notifies the mediating agent of what it does and how it can accept the exchange so that it can be accessed according to the created model.

  Based on the notified knowledge, the mediating agent converts the exchange according to the partner and selects the partner agent to which the message is transmitted. At the same time, the mediating agent assembles a larger processing flow based on the processing of individual agents. For example, a process flow is generated in which the message is converted and then sent to an appropriate agent, the returned message is converted and the questioning agent is answered. This assembly is performed by inference or using information given in advance.

  In order to be able to communicate with other agents in this intermediary environment, it is necessary to know at least the input / output format of the agent and the words used by the agent as knowledge.

  When a model is constructed in such an intermediary environment, a similar model space (knowledge space) can be created here and there. Then, it is necessary to connect the model spaces. In that case, it is impossible to redesign the whole. In addition, each model space is under separate management, and even after being connected, there is a possibility that each model space may be changed arbitrarily.

  From now on, when connecting model spaces, it becomes necessary as a function to be able to manage each knowledge space in a distributed manner. There are also various connections in this model space. There is a hierarchical connection method, and there is a connection method that seems to hang down in a specific detailed model space.

  The present invention provides a very effective means for the requirement of connecting such distributed model spaces.

  In other words, the service list used in the present invention can be left to the subsequent agent for details. From now on, it is not necessary to know all of another model space, and if there is knowledge about an agent (an agent such as an intermediary agent of organization A shown in FIG. 26) that serves as a window to that model space, a service list is sent to that agent. This is because it is possible to enter the model space by having the service list refined according to the knowledge of the model space after that.

  As described above, the present invention makes it possible to link the distributed model spaces without destroying the distributed environment, so that the present invention can be applied to any system that links software connected via a network. Become.

  Although the present invention has been described according to the illustrated embodiment, the present invention is not limited to this. For example, the present invention is not limited to an agent in a narrow sense and can be applied as it is to all software that exchanges messages in a distributed environment.

It is a principle block diagram of this invention. It is explanatory drawing of this invention. It is explanatory drawing of the system to which this invention is applied. It is explanatory drawing of a service list. It is explanatory drawing of a message. It is explanatory drawing of term conversion. It is the processing flow which a user agent performs. It is the processing flow which a mediation agent performs. It is the processing flow which a mediation agent performs. It is the processing flow which an agent performs. It is the processing flow which an integrated agent performs. It is the processing flow which an integrated agent performs. It is an example of a message. It is an example of a message. It is process explanatory drawing of an Example. It is an example of a message. It is an example of a message. It is an example of a message. It is the processing flow which an agent performs. It is an example of a message. It is the processing flow which an agent performs. It is the processing flow which an agent performs. It is the processing flow which an agent performs. It is explanatory drawing of detailing of a service list. It is explanatory drawing of detailing of a service list. It is explanatory drawing of distributed management of knowledge. It is explanatory drawing of a virtual catalog system. It is explanatory drawing of the system which distributed the function of the mediation agent.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1a 1st agent apparatus 1b 2nd agent apparatus 1c 3rd agent apparatus 10 Generation means 11 Transmission means 20 Reception means 21 Extraction means 22 Refinement means 23 Transmission means 30 Reception means 31 Extraction means 32 Execution means 33 Deletion means 34 Transmission means 35 Storage means 36 Correction means 37 Judgment means 38 Exclusion means 39 Restoration means 40 Collection means

Claims (4)

In an agent device used in a multi-agent system,
That will transmitted from another agent apparatus, a receiving means for receiving a search request message to indicate searches,
Is added to the search request message received in said receiving means, an extraction means for extracting destination information and the service list defined by the connection of the pair data of the specified processing information of the message of the message,
Selecting means for extracting the search content included in the search request message and selecting a database agent device corresponding to the search content based on knowledge about capability information of a plurality of database agent devices;
It is determined whether or not a plurality of database agent devices are selected by the selection means. If it is determined that a plurality of database agent devices are selected, the search results of the plurality of database agent devices are integrated into the service list extracted by the extraction means. and shipping information specifying an integrated agent apparatus, a detailed means to create a refinement service list obtained by adding a service list that combine the processing information for instructing to integrate those search results,
The received search request message the receiving means, by adding the refinement service list created in the refinement means, further comprising a transmitting means for transmitting to a plurality of databases Agent device selected in the selection means the agent device to feature. The agent device according to claim 1,
The service list of the received search request message includes the term system type in the output of the search result,
The detailed means determines whether the term system type in the output of the search result is different from the term system type used by each of the selected plurality of database agent devices. For the determined database agent device, before and after the service list designating the search processing added to the received search request message, processing information for instructing conversion of both terminology types, and agent device to feature a <br/> creating a refined list of services added to the service list of a combination of a specified delivery destination information translated agent device to handle the conversion. A program recording medium for recording a program used to realize an agent device used in a multi-agent system,
That will transmitted from another agent apparatus, a receiving process of receiving a search request message to indicate searches,
Is added to the search request message received by the receiving process, an extraction process of extracting the destination information and the service list defined by the connection of the pair data of the specified processing information of the message of the message,
A selection process for extracting the search content included in the search request message and selecting a database agent device corresponding to the search content based on knowledge about capability information of a plurality of database agent devices;
It is determined whether or not a plurality of database agent devices are selected in the selection process. If it is determined that a plurality of database agent devices are selected, the search results of the plurality of database agent devices are integrated into the service list extracted in the extraction process. and shipping information specifying an integrated agent apparatus, and their search results to create a refinement service list obtained by adding a service list that combine the processing information for instructing to integrate refinement process,
The search request message received by the receiving process, by adding the refinement service list created in the refinement process, perform a transmission processing to a computer to be sent to multiple databases Agent device selected in the selection process program recording medium according to feature that a program to is recorded. The program recording medium according to claim 3,
The service list of the received search request message includes the term system type in the output of the search result,
The detailed processing determines whether the term system type in the output of the search result is different from the term system type used by each of the selected plurality of database agent devices. For the determined database agent device, before and after the service list designating the search processing added to the received search request message, processing information for instructing conversion of both terminology types, and Create a detailed service list with a service list combined with delivery destination information that specifies the translation agent device that handles the conversion
A program recording medium characterized by the above.

Images