KR20020027707A - Agent system for artficial intelligence type and method for driving the same and computer readable medium the same - Google Patents

Abstract

PURPOSE: An artificial intelligence agent system, a driving method thereof, and a recorder thereof are provided to efficiently use resource of user computers based on the natural language processing of users by using a P-to-P(Peer-to-Peer) network technology. CONSTITUTION: The system comprises an agent interface unit(100) for outputting an instruction via a second path and information to a user by receiving an instruction from the user, a natural language processing unit(200) for converting the instruction offered via a first path into a specific format and feeding back the converted instruction to the agent interface unit, a work scheduler unit(300) for performing search schedule work by the instruction inputted via the second path and offering information to the agent interface unit, a knowledge base unit(500) for storing the information offered by users, storing the profiles of the users based on the stored information, and adapting to user environments, and an agent cooperation unit(400) for providing information to the agent interface unit, checking the knowledge base, transmitting instructions to an outside system, and maintaining agent profiles.

Description

Agent system for artficial intelligence type and method for driving the same and computer readable medium the same}

The present invention relates to an artificial intelligent agent system and a driving method thereof, and more particularly, to an artificial intelligent agent system, a driving method thereof, and a computer storing the same for learning a user's personality and simplifying a repetitive task based on natural language processing. Relates to a readable recording medium.

In general, the Internet is an on-line computer network spread across the globe, connected to governments, businesses, schools, and general networks and individuals, and is currently a network with more than 3 million computers.

The Internet began with funding from the US government in 1984, and was initially a public network limited to a limited number of people, including government and school researchers, but is now a universal network open to anyone. In Korea, computer communication companies use one-on-one information service network, Unitel, Hitel, Chollian, etc. to connect one-on-one message (e-mail) and one-to-many message delivery (Board) It offers a wide range of services, including Bulletin Board, One-Line Chat, File Transfer, and Internet Search.

In particular, general users access the Internet search sites such as Yahoo, Lycos, and Empass to obtain the desired information in order to search the vast amount of information on the Internet backbone network. It also uses a meta-based search tool that does not own its own database and has the principle of showing the user the results of the search terms entered by other search tools. There are Savvy Search, MetaCrawler, Missed Search, and Metaglider.

However, in the case of a general search engine, a search engine that performs a natural language search such as Empas, or a meta search engine, the user is provided with the corresponding information according to the ranking closest to the search at the request of the search, and the user who receives the information is provided. On the side, even if the information is not close to the information requested by the search is provided a problem that takes unnecessary time.

SUMMARY OF THE INVENTION The present invention has been made in an effort to solve such a conventional problem, and an object of the present invention is to be based on a user's natural language processing, to provide search information closest to a user's requested data, and to interwork with a user's characteristics. By using P-to-P network technology, we will provide artificial intelligence agent system for efficient utilization of user computer resources through agent-to-agent collaboration.

Another object of the present invention is to provide a method of driving the artificial intelligent agent system.

It is still another object of the present invention to provide a computer readable recording medium storing the above-described method for driving an artificial intelligent agent system.

1 is an overall configuration diagram of an artificial intelligent agent system according to an embodiment of the present invention.

2 is a flowchart illustrating the operation of an artificial intelligent agent according to an embodiment of the present invention.

3 is a view for explaining the operation of the task scheduler.

4 is a view for explaining the operation of the user agent system and the remote agent system according to an embodiment of the present invention.

FIG. 5A is a diagram for describing a first model of the remote agent system and the user agent system of FIG. 4.

FIG. 5B is a diagram for describing a second model of the remote agent system and the user agent system of FIG. 4.

FIG. 6 is a diagram for describing a job scheduler according to an exemplary embodiment of the present invention and has an external interface capable of directly manipulating a schedule.

7A to 7B are diagrams for explaining an example of the graphical user interface (GUI) of the PIMS scheduler of FIG. 6 and the input window, respectively.

8A is a diagram for explaining registration of an external execution module according to an embodiment of the present invention;

8B is a diagram for describing an execution procedure of an external execution module according to an embodiment of the present invention.

9 is a view for explaining the operation of the user knowledge base unit according to an embodiment of the present invention.

10 is a view for explaining the object structure of the user knowledge base according to an embodiment of the present invention.

11A is a view for explaining an operation flow of a meta search engine module according to an embodiment of the present invention;

11B is a view for explaining the flow of active operation of the agent collaboration meta-search engine according to an embodiment of the present invention.

FIG. 11C is a diagram illustrating a manual operation flow of an agent collaboration meta search engine according to an embodiment of the present invention. FIG.

12 is a diagram for explaining an agent collaboration module according to an embodiment of the present invention.

FIG. 13A is a view for explaining a check processing procedure of existence / activity between collaboration modules of FIG. 12;

FIG. 13B is a view for explaining an agent list request processing procedure of the facilitator of FIG. 11;

FIG. 13C is a view for explaining an agent software upgrade request processing procedure of the facilitator of FIG. 12.

14 is a view for explaining a web page displayed on the user-side PC according to an embodiment of the present invention.

15A is a diagram for explaining new agent registration according to an embodiment of the present invention;

15B is a view for explaining the use of an agent according to an embodiment of the present invention;

15C is a diagram for explaining the selection of an agent server after a predetermined time elapses according to an embodiment of the present invention;

15D is a diagram for explaining inter-agent collaboration according to an embodiment of the present invention.

<Description of the symbols for the main parts of the drawings>

100: agent interface unit 110: agent interface module

120: virtual agent interface module 200: natural language processing unit

210: syntax analysis module 220: verb subcategorization dictionary module

230: noun category dictionary module 240: meaning network module

250: Korean English Dictionary module 300: Task Scheduler

400: agent collaboration unit 500: knowledge base unit

600: various module unit 610: execution module

620: knowledge base storage module 630: knowledge base request module

640: meta search request module 650: PIMS schedule module

660: external execution module 700: meta search unit

710: meta search module 720: profile storage module

730: user search profile module 740: index word extraction module

750: information search engine module 800: chat unit

Artificial intelligent agent system according to one feature for realizing the above object of the present invention,

Receives a command from the user in at least one of sentences or voices, requests conversion into a predetermined format through the first path, and outputs the converted command according to the response through the second path. An agent interface unit outputting information according to the response to the user side;

A natural language processor configured to convert the command provided through the first path into a predetermined format, and to feed the converted command back to the agent interface unit;

A task scheduler configured to perform a search schedule task according to a user's converted command input through the second path, and to provide information according to the response to the agent interface unit;

A knowledge base unit which stores information provided from the user and adapts the user environment by storing the user's profile based on the stored information; And

It provides a command input via the task scheduler to the external one or more artificial intelligent Internet agent system side to request data, and provides the data according to the response to the agent interface unit, the external artificial intelligence Internet agent system When a request for data is input from the data base, the knowledge base is checked, the command is processed and transmitted to the external artificial intelligence internet agent system, and the agent collaboration unit maintains an agent profile.

The artificial intelligence agent system may include: a meta search unit requesting a search to at least one information search engine when the converted command is search request information, and outputting search information according to a response thereof; And

When the data is input from the user via the task scheduler, the data is stored in the knowledge base, and when the search request is input from the task scheduler, the knowledge base unit is searched first. The module may be provided to the unit, and if there is no corresponding material, the meta search unit may be started to request a search, and the module unit may further provide search information according to the response to the task scheduler unit.

In addition, the driving method of the artificial intelligent agent system according to one feature for realizing another object of the present invention,

(a) checking whether a command of at least one type of a sentence or a voice is input from a user and converting the command into a functional grammar through a natural language processing process when the command is input;

(b) checking whether the converted command is an internal execution command, and in the case of the internal execution command, controlling driving of a corresponding stored program; And

(c) checking whether the converted command is a knowledge base-related command, and if the command is a knowledge base-related command, perform a knowledge base processing or a knowledge base search and convert the result into at least one of a text and a voice. The output is made to include.

Here, the method of driving the artificial intelligence agent system, (d) if the converted command is a knowledge base-related command to perform an agent collaboration operation and output the result to the user in at least one form of text or voice It may be made to include more.

In addition, the method of driving an artificial intelligent agent system includes (e) checking whether the converted command is a meta-search search command and, if the meta-search search command is performed, performs a meta-search to search results. Outputting to the user side in the form of at least one of a text and a voice; And (f) performing the agent collaboration operation and outputting the result to the user in the form of at least one of a text and a voice.

The artificial intelligent agent system may further include (g) checking whether the converted command is a schedule management command and, in the case of the schedule management command, performing a schedule management operation.

In addition, the artificial intelligent agent system driving method includes: (h) checking whether the converted command is a command for an external execution module, and performing an external execution operation in the case of an external execution module command; And (i) if the converted command is not an external execution module command, performing a chat with a user and an agent.

In addition, a computer-readable recording medium storing a method for driving an artificial intelligent agent system according to another aspect for realizing another object of the present invention,

(a) checking whether a command of at least one type of a sentence or a voice is input from a user and converting the command into a functional grammar through a natural language processing process when the command is input;

(b) checking whether the converted command is an internal execution command, and in the case of the internal execution command, controlling driving of a corresponding stored program; And

(c) checking whether the converted command is a knowledge base-related command, and if the command is a knowledge base-related command, perform a knowledge base processing or a knowledge base search and convert the result into at least one of a text and a voice. The output is made to include.

According to such an artificial intelligent agent system, a driving method thereof, and a computer-readable recording medium storing the same, it becomes a live program by adding various knowledge to a stationary computer program. Searching the base, searching the Internet, and sharing information with other agent systems can automatically solve the task you want.

Then, embodiments will be described so that those skilled in the art can easily implement the present invention.

1 is an overall configuration diagram of an artificial intelligent agent system according to an embodiment of the present invention.

Referring to FIG. 1, the artificial intelligent agent system according to an embodiment of the present invention may include an agent interface unit 100, a natural language processing unit 200, a job scheduler unit 300, an agent collaboration unit 400, and a knowledge base unit 500. ), Various modules 600, a meta search unit 700, and a chat unit 800.

The agent interface unit 100 includes an agent interface module (AIF) 110 and a virtual agent interface module (VAIF) 120, and a command or user input according to a user's key operation through the agent interface module 110. The command input through the voice of the voice is provided to the natural language processing unit 200 via the virtual agent interface module 120 to receive a command converted into a predetermined function grammar (FG) adapted to the agent program.

In addition, the agent interface unit 100 may provide a command converted into a functional grammar to the job scheduler 300, and if there is a response, the agent interface unit 100 may be provided via the virtual agent interface module 120 to receive the agent interface module 110. Output through In this case, the response to the command may be, for example, a response to a query by the user, and the response path may also be provided through the task scheduler 300 or the response may be directly provided.

In addition, the agent interface module 110 may display a display window form for inputting a text command together with a predetermined character, an instant message window, a memo pad message window, and the like on a user computer screen.

In addition, the agent interface module 110 may be online with the virtual agent interface of the separated agent program managed by the user of the agent program. This connection allows the user to execute agent operations from a PC at his office to a PC at home.

The natural language processor 200 includes a syntax interpreter module 210, a verb subcategorization dictionary module 220, a noun category dictionary module 230, a semantic network module 240, and a Korean-English band dictionary module 250. The sentence or voice input from the interface module 120 is converted into a predetermined format, that is, a functional grammar, and fed back to the virtual agent interface module 120. Here, the noun category dictionary module 230 is a dictionary in which a noun category categorized by the verb subcategories dictionary module 220 is tagged, and speeds up access by a hashing algorithm.

Next, an example of an input / output program of the syntax analysis module 210 will be described as follows.

_{-------------------------------------------------- -------------------------------------------------- ----------}

_{enu DurType}

_{

_{EH = 1, ED1, ED2, EW1, EW2, EM1, EM2, EY}

_}

_{Class ccmd}

_{

_{int cmd; // command code}

_{int verb; // verb syntax code when cmd = 'p'}

_{char obj [3] [80]; // list of objects corresponding to command code}

_{CCmd * ptr; // obj is another CCmd pointer when the value of obj is '<ptr>'}

_{// example: (letknow (search 'diablo'))}

_{// cmd = 'letknow', obj [0] = '<ptr>'}

_{// ptr = CCmd pointer of dynamically allocated (search 'diablo')}

_}

_{class CCmdResult}

_{

_{BOOL bHasTimeClause; // TRUE if time clause is included in statement, FALSE other}

_{Cstring reststr; // sentence except time clause}

_{CTime begine, end; // start time, end time}

_{DurType dtype; // repeat cycle code}

_{int m, d, w, h; // month, day, week day, hour according to repeat cycle code}

_{CTypePtrList <CCmd, CCmd *> cmdchain;}

_{CString from, to;}

_}

_{CString Word;}

_{BYTE Tag;}

_{WORD stops, ends // position in input sentence}

_};

_{CTypePtrArray <CMorpheme, CMorpheme *> CMorphemeArray;}

_{-------------------------------------------------- -------------------------------------------------- ----------}

The task scheduler 300 is a core module that manages all of the agent's tasks. The task scheduler 300 is started from the agent interface unit 100 and operates in the background separately from the agent interface unit 100. All commands centered on the task scheduler are mainly delivered in the functional syntax (FG) format, and the result is text, X.M.L, file-path, U.R. L (URL) and the like, a detailed description thereof will be described in Figure 2 below.

Next, the functional grammar (FG) used in the present invention will be described.

The functional grammar system is as follows.

_{-------------------------------------------------- -------------------------------------------------- ----------}

_{S: = [<TMCLAUSE>,] <cmdclause>}

_{cmdclause: = '(' <cmd> <obj> * [from (<loc>)] [to (loc)] ')' | '(' <cmd> [<cmd> *] ')'}

_{cmd: = copy | move | delete | execute | open | change | create | send | take | give | search | notify | want | is | has | locate | p | s | word | what | when | who | where | query | click | defrule |}

_{obj: = that | '?' | <noun> | <cmdclause>}

_{loc: = <noun> | <cmdclause>}

_{noun: = <NNP> <wc> | <NNCG> <wc>}

_{wc: = 'GM' | 'PM' | 'GT' | 'PT' | 'GL' | 'PL'}

_{-------------------------------------------------- -------------------------------------------------- ----------}

Here, that is the <obj> previously specified as a tubular form, me is a keyword indicating a user, you is a keyword indicating an agent, and '?' Is an object not specified in the sentence. The <obj> may be attached to the part-of-speech subcategory of obj. For example, 'home' should be written as a common noun 'home: G' and 'Kim Chul-soo' as a proper noun as 'Kim Chul-soo: P'. Is indicated. The part-of-speech classification code uses the part-of-speech classification code of the nouns that the verb receives from the verb subcategorization dictionary. For example, human nouns are 'GM', thing nouns are 'PT', 'GT', and place nouns are 'GL' and 'PL'.

_{-------------------------------------------------- -------------------------------------------------- ----------}

_{tmclause: = '(' tclause <start-tmd> [end-tmd] [duration] ')'}

_{start-tmd, end tmd: = <date> <time>}

_{duration: = <hourly> | <dayly> | <weekly> | <monthly> | <yearly>}

_{hourly: = EH <n hour> // every 3 hours}

_{dayly: = ED1 <day_no> | // every 3 days}

_{ED2 <day_no> <hourly> // every 2 days at 2 pm ...}

_{weekly: = EW1 <week_no> | // every Tuesday ...}

_{EW2 <week_no> <hourly> // every Tuesday at 10am ...}

_{monthly: = EM1 <day_no> | // every 20 days ...}

_{EM2 <day_no> <hourly> // at 10am on the 20th of every month ...}

_{yearly: = EY <mon_no> / <day_no> // on December 25 of every month ...}

_{date: = [0-9] {2,4} / <mon_no> <day_no>}

_{time: = [0-9] {, 2}: [0-9] {, 2}: [0-9] {, 2}}

_{week_no: = [1-7]}

_{day_no: = [1-31]}

_{mon_no: = [1-12]}

_{-------------------------------------------------- -------------------------------------------------- ----------}

Next, the command syntax form of the functional grammar, which is a feature of the present invention, will be briefly described.

First, the command syntax format may be configured as follows, and a detailed description thereof will be omitted.

_{(copy <obj> [from (<loc>)] [to (<loc>)])}

_{(move <obj> [from (<loc>)] [to (<loc>)])}

_{(delete <obj> [from (<loc>)])}

_{(execute <obj> [from (<loc>)])}

_{(open <obj> [from (<loc>)] [to (<loc>)])}

_{(close [<obj>])}

_{(quit [<obj>])}

_{(change <obj> [from (<loc>)] [to (<loc>)])}

_{(create <obj> [from (<loc>)])}

_{(notify <obj>)}

_{(send <obj> [to (<loc>)])}

_{(take <obj> [from (<loc>)])}

_{(search <obj> [from (<loc>)])}

_{(take <obj> [from (<loc>)])}

_{(search <obj> [from (<loc>)])}

_{(compress <obj> from (<loc>))}

_{(uncompress <obj> to (<loc>))}

In addition, the knowledge base definition / predicate extended syntax format may be configured as follows, and a detailed description thereof will be omitted.

_{(is <obj1> <obj2>)}

_{(has <obj1> <obj2>)}

_{(locate <obj1> from <loc>)}

_{(p <verv> <obj> *)}

_{(word <noun> *)}

In addition, the syntax related to the query can be configured as follows, and a detailed description thereof will be omitted.

_{(what <obj>)}

_{(when <obj>)}

_{(who <obj>)}

_{(where <obj>)}

_{(query <obj>)}

In addition, the other syntax extension syntax format can be configured as follows, detailed description thereof will be omitted.

_{(click <obj>)}

_{(add-grammar <keyword> *, <cmd-code>)}

_{(defrule <conditional element> => <execution block>)}

In response to FIG. 1, the agent collaboration unit 400 requests a query by accessing a remote agent system connected online when there is a query request together with a cooperative command with an agent program remotely deployed from the job scheduler 300. The information according to the response is outputted in a text or voice form to the user requesting the query via the job scheduler 300 and the agent interface unit 100. Of course, it is obvious that the response to the query may be directly output to the agent interface unit 100 without passing through the job scheduler 300.

The knowledge base unit 500 stores information that adapts to user characteristics. For example, the user may provide various information to a knowledge base (KB) of the user, or through an Internet search or a remote agent system. You can receive and store various information through collaboration with. In other words, the knowledge base 500 according to the present invention is a program that initially set the agent in the computer environment, even if the same time is elapsed, and as the user's query or data requests increase the various characteristics that are linked to the user characteristics Information can be stored.

The various module units 600 may execute the execution module 610, the knowledge base storage module 620, and the knowledge base request module 630 as a work instruction is input from the agent interface unit 100 via the job scheduler 300. , The meta search request module 640, the schedule module 650 of the personal information management system (PIMS), and the external execution module 660 are performed in this order. At this time, the received sentence is processed by the task scheduler 300 first depending on the time section. For example, if there is a time clause in a sentence, it is registered with the task scheduler along with the command. If it is an execution command, the execution command is executed at the corresponding time. If meta search is used, the search precision parameter of the search engine is adjusted. If there is no time clause, it is executed immediately.

The meta search unit 700 includes a meta search module 710, a profile storage module 720, a user search profile module 730, an index word extraction module 740, and an information search engine module 750. As a meta search execution request is input from the module unit 600, a request is made to an Internet search site, a search engine, and a search robot, and the search result information according to the response is stored in the knowledge base unit 500.

Chat unit 800 is used when the user does not find the information requested by the knowledge base unit 500, the agent collaboration unit 400, the meta search unit 700, the artificial intelligence according to the user and the present invention The desired information is stored in the knowledge base unit 500 through a chat with an agent system.

As described above, the artificial intelligent agent system according to the present invention can be functionally separated into an interface agent, an information agent, and a work agent.

In other words, the interface agent receives and analyzes the user's query and shows the result through interaction with the user, and acquires and models the data including the user's habits and preferences.

In addition, the information agent accesses the information sources of different traits scattered on the Internet backbone network to find the desired information and monitor the change of the information.

In addition, the task agent actually performs the task requested by the user with the knowledge of the area of the given task and the capability information of other task agents or information agents.

2 is a flowchart illustrating a method of driving an artificial intelligent agent according to an embodiment of the present invention.

1 and 2, first, it is checked whether a command is input from a user (step S100), and if a command is input, the command is converted into a functional grammar through natural language processing (step S110).

Then, it is checked whether the command converted into the functional grammar is an internal execution command (step S120), and in the case of the internal execution command, the driving of the corresponding program stored in the computer system in which the agent program is embedded (step S122). The process returns to step S100.

If it is not an internal execution command in step S120, it is checked whether the command converted into the functional grammar is a knowledge base related command (step S130).

In the case of the knowledge base related command in step S130, the knowledge base processing and searching are performed, and the agent collaboration module is activated to perform the agent collaboration operation (step S132), and the respective results are presented to the user in the form of text or voice. Output (step 134). Here, the knowledge base processing operation is an operation for educating an agent system by providing predetermined information or data to a knowledge base managed by an agent program, and the knowledge base searching operation is an operation for obtaining specific information from an already stored knowledge base. to be. In addition, the agent collaboration operation is a series of operations that request information search from other agent systems connected online when a specific data is not retrieved from the knowledge base stored in the user computer system. Of course, when the corresponding data is not searched after the knowledge base search operation is performed, the agent collaboration operation may be performed, or the agent collaboration operation may be performed simultaneously with the knowledge base search operation.

If it is not a knowledge base related command in step S130, it is checked whether or not the command converted into the functional grammar is a meta search command (step S140). In the case of a meta search command, the meta search module is started. The agent collaboration module is activated and the result is provided to the user side (steps S142 and S144).

If it is not the search method of the meta method in step S140, it is checked whether or not the command converted into the functional grammar is a schedule management command (step S150). In the case of the schedule management command, a schedule management operation is performed (step S152). .

If it is not a schedule management command in step S150, it is checked whether it is a command for an external execution module (step S160), and if it is an external execution module command, an external execution operation is performed by starting an external extension module (step S162). .

If it is not an external execution module command in step S160, a chat is performed between the user and the agent system (step S170).

As described above, the artificial intelligence agent system according to the present invention includes an artificial intelligence concept having natural language interpretation ability, intelligent information processing ability, and machine learning ability, user helper ability, and ability to efficiently manage information necessary for the user; It is a synthesis of agent concept with automatic problem solving / helping ability. It can automatically solve the task that user wants through voice, sentence recognition, natural language processing, etc. It can help you work with and act as an agent to automate and intelligent your information collection and information processing.

Hereinafter, the artificial intelligence agent system according to the present invention will be described in more detail.

3 is a view for explaining the operation of the task scheduler, it is logically connected to the agent interface unit 100, agent collaboration unit 400, various module unit 600, all the commands are mainly functional syntax (FG) The result is sent and received through text, X.M.L (xml), file path (URL), etc.

Then, a summary of communication with the agent interface unit 100 and the agent collaboration unit 400 will be described in Table 1 below.

Job Scheduler Unit (TSC) Agent Interface Unit (AIF) _{(add-grammar <keyword> * <string-id>) ▶ (aif gesture <no>) ▶ (aif tts <vtype>) ▶ (aif character <id> <name> <path>) ▶ (Result: xml) ◀ (joblist) ◀ (jobctrl <lcommand> <ljob-id>) ◀ (vconnect <ip>) ◀ (vdisconnect) ◁ (ac-msg <id> <name> <msg>) (ac-msg <id> <name> <msg>) ▷ ◁ (ac-config-request) ◁ (ac-config <item> <value>) ◁ (ac-agent-list) ◁ (ac-agent-col-config <id> <kb> <ir> <file>) ◁ (ac-agent-files-request <id>) ◁ (ac-agent-file-get-request <id> <file-paqth>) ◁ (add-server <agent-id>) ◁ (aif character <id> <name> <path> <ip>)}

Table 2 below describes a summary of communication between the job scheduler 300 and the agent collaboration unit 400.

Job Scheduler Unit (TSC) Agent Collaboration Department (ACM) _{(ac-kb-request <FG-cmd>) ▶ Result: text, xml (ac-ir-request <kw-list> <hm-kw-list> <begin-n> <n> <kin-domain-flag> <depth>) ▶ Result: binary (ac-msg <id> <name> <msg>) ▶ ◀ (ac-msg <id> <name> <msg>) (ac-config-request) ▶ Result: xml (ac-config <item> <value>) ▶ (ac-agent-list) ▶ Result: xml (ac-agent-col-config <id> <kb> <ir> <file>) ▶ (ac-agent-files-request <id>) ▶ Result: xml (ac-agent-file-get-request <id> <file-path>) ▶ Result: text (ac-char-request <id> <name> <path> <ip>) ▶ Result: text ◀ (ac-remote-req <id> <FG cmd> *) Result: text ◀ (ac-mse-status) (add-server <agent-ip>) ▶ ◀ (mod-update <file-url>)}

The task scheduler 300 transmits the new functional grammar (FG) to the syntax analysis module 210 of the natural language processing unit 200 via the agent interface unit 100. When a new module is added to the task scheduler 300, a new function grammar in response to a keyword is transmitted to the syntax interpretation module 210.

The job scheduler 300 receives the function grammar FG from the agent interface unit 100, interprets the command, and inserts the command into the job queue. When the command is fetched from the work queue and the module is executed as a thread, the work scheduler 300 monitors the executed thread and waits for the end of the work. When the command is finished, the command processing result is transmitted to the agent interface unit 100.

Table 3 below describes the structure of the work queue described above.

Property Type Remark Id Int _{Job ID} Sdt Ctime _{Job execution time} Edt Ctime _{Job end time (if statement contains time clause)} Cdt Ctime _{Current running time} Thandle HANDLE _{Thread or Precess handle} prio int _Priority Cmd Csting _{Work command} Stat BYTE _{Work status 0 = end of job 1 = work in progress 2 = Waiting for job (if there are multiple commands, all jobs must finish before continuing) 3 = suspend operation}

In addition, the job queue is a first-in, first-out (FIFO) structure of the job is left as a file in the form of a history, the history is for example two days of the last one is automatically deleted. Registered job queues are meta-search commands and agent collaboration. When a file is opened or executed, it must also have a process handle for that command.

4 is a view for explaining the operation of the user agent system and the remote agent system according to an embodiment of the present invention.

Referring to FIG. 4, when information retrieval (IR) is performed in a collaborative mode, a primary meta search result is obtained through a meta search engine, and then divided into n number of remote agents, and each remote agent system Send it in the form of requesting collaboration. When the collaboration is successfully completed and receives the respective results from the remote agent system, it is collected and transmitted to the agent interface unit 100.

When a time clause is specified in the function syntax (FG) received from the agent interface unit 100 and operates in the cooperative mode, the routing hop count of the remote agent system, the response minimum response time, Determine the depth of page searches. Here, the hop count is the total number of paths that the agent system takes.

FIG. 5A illustrates a first model of the remote agent system and the user agent system of FIG. 4, and FIG. 5B illustrates a second model of the remote agent system and the user agent system of FIG. 4. to be.

Referring to FIG. 5A, when a first model of a remote agent system and a user agent system has a time section of a function grammar (FG) received from the agent interface unit 100, and operates in a cooperative mode, the first model of the remote agent system and the user agent system may be remotely connected according to time. Determine the maximum number of hops, the minimum response time, and the depth of page retrieval for the agent system.

The first model has an advantage of not burdening the agent system on the routing path, but has a disadvantage in that a rush phenomenon may occur in the user agent system.

On the contrary, in the case of FIG. 5B, the burden on the user agent system is low. However, the remote agent system on the routing path has a disadvantage.

FIG. 6 is a diagram for describing a job scheduler according to an exemplary embodiment of the present invention and has an external interface capable of directly manipulating a schedule.

Referring to FIG. 6, the task scheduler 300 according to an embodiment of the present invention includes a personal information management (PIMS) schedule interface module 310, a PIMS scheduler module 320 having a predetermined plug-in function, and a PIMS scheduler. Graphical user interface module 330 of the user, including a storage module 340 for storing the user's schedule information, for example, every five minutes to communicate with the personal information management (PIMS) scheduler (external module) regularly Check the schedule. The PIMS scheduler interface 310 receives a time clause set as a function grammar from the agent interface unit 100. The input / output function grammar is as follows.

_{-------------------------------------------------- -------------------------------------------------- ----------}

_{-> (tcluse <start date> [<end date> [<duration>]])}

_{<-Ok: 0}

_{-> (? tcluse today)}

_{<-3; Total number of schedules today}

_{<-2000/7/10 12:00-13:20 Lunch Appointment}

_{<-2000/7/10 12:00-??: ?? Evening appointment}

_{<-2000/7/10 12:00-??: ?? Date}

_{-> (? tclase'2000 / 7 / 10''2000 / 7/20 '); Schedule data from 7/10 to 7/20}

_{-------------------------------------------------- -------------------------------------------------- ----------}

The PIMS scheduler module 320 is an O.X. (OCX) that performs a kind of plug-in function, and both input and output are variable types.

In general, modules require omitted execution parameters. In this case, the job scheduler 300 may wait for an input value at the input of the agent interface unit 100, which is called a job scheduler demand input mode.

As described below, when the agent interface unit 100 receives the command 'find Randy Rose's mp3 from tomorrow afternoon', and expands it to a functional grammar (assuming 2000/7/10 today), (tclause '2000/7/11 13:00'), (search (has mp3 'Randy Rose')) ', and the meta search module determines how long to search Randy Rose's mp3 in the tclause. To do this, a query must be requested from the agent interface to complete the tclause.

_{-------------------------------------------------- -------------------------------------------------- ----------}

_{User: find Randy Rose's mp3 by tomorrow afternoon}

_{Agent: When do you want to find it? <-demand input mode}

_{User: Until July 12 (tclause '2000/7/12 23:29')}

_{Agent: Yes, sir.}

_{-------------------------------------------------- -------------------------------------------------- ----------}

7A to 7B are diagrams for explaining an example of the graphical user interface (GUI) of the PIMS scheduler of FIG. 6 and the input window, respectively.

FIG. 8A is a diagram for describing registration of an external execution module according to an embodiment of the present invention, and FIG. 8B is a diagram for explaining an execution procedure of an external execution module according to an embodiment of the present invention. The external execution module must take care of the time clause of the function grammar (FG). Upon notification of the agent collaboration module (ACM), an unload / load of an external execution module (DLL) is performed. In addition, the external execution module is provided with all the original text, functional grammar (FG), and morphological analysis results given by the agent interface unit 100. The control of the external execution module is periodically executed in the task scheduler 300 to process a function grammar (FG), a user notification, and the like.

Referring to FIG. 8A, first, the agent collaboration unit (ACM) 400 periodically asks the remote server if there is an update (step A1), and if the remote server has an updated module, the agent collaboration unit (ACM) ( 400) (step A2). The Agent Coupling Unit (ACM) 400 then stores the corresponding module received from the remote server on the user's computer (step A3) and notifies the task scheduler unit (TSC) 300 that the new module has been received (step A3). A4), the task scheduler unit (TSC) 300 informs the external execution module controller 670 of the file name of the new module (step A5).

The external execution module control module 670 registers the new module 680 (step A6), and the new module notifies the TSC of the add-grammar to be used (step A7).

Subsequently, the task scheduler 300 transmits the grammar definition to the agent interface unit 100 to the agent interface unit 100 (step A8), and the agent interface unit 100 is a natural language. The processing unit 200 informs the new grammar (step A9).

Referring to FIG. 8B, first, when a user inputs a sentence (step B1), the agent interface unit 100 transmits the user sentence to the natural language processing unit 200 (step B2), and the natural language processing unit 200 is a new module. If a specific keyword registered as (add-grammar) is found, it is redefined as a grammar used by the new module and transmitted to the agent interface unit 100 (step B3).

The agent interface unit 100 transmits the functional grammar (FG), the stemming analysis result, and the original text to the job scheduler unit (TSC) 300 (step B4).

The task scheduler 300 finds a command related to the external execution module from the corresponding instruction set of the function grammar (FG) and provides it to the external execution module control module 670 (step B5). 670 passes the functional grammar (FG), stemming results and original text to the new module 680 (step B6).

9 is a view for explaining the operation of the user knowledge base unit according to an embodiment of the present invention.

Referring to FIGS. 1 and 9, the operation of the user knowledge base unit 500 will be described. First, a knowledge base (KB) for preserving knowledge information using a memory cache 530 and an external disk 550 may be described. A function is called by the execution module 610 of the various module units 600. The knowledge base classification speeds up the access of the command object CMNObj by the object name ObjName through the hashing table 520 by the object name and ID. The knowledge base maintained in the memory cache 530 uses about 1 MB of memory cache and the rest are read from the external disk 550 whenever necessary.

Each object has an ID (ObjID). More specifically, if the MSB is set in the ID, the normal noun or unique noun ID used in the semantic network, that is, the semantic network module 240 generated from the noun category dictionary module 230. Each time an object is added, it can be increased up to 2,247,748 million.

In addition, the string hash table 520 has a hash table 520 for all objects in the memory cache 530, and the cache manager 540 maintains the cache values of the frequently used objects together with counters to store the cache in the memory cache ( 530 and the external disk 550.

A query of the knowledge base unit 500 according to an embodiment of the present invention will be described with reference to FIG. 9.

Request an object and find it in a hash table. Then, if there is no corresponding object through the cache manager 540, it is taken from the external hard disk 550, loaded into the cache, and returned from the cache.

Referring to FIG. 9, the storage of the knowledge base unit 500 according to an embodiment of the present invention is described. A request is made to create an object, the cache manager records an object in a cache and a disk, and the cache manager records an object use. To clear the least frequently used objects from the cache.

Next, the object will be described in more detail with reference to FIG. 10.

10 is a view for explaining the object structure of the user knowledge base according to an embodiment of the present invention, the program is as follows.

_{-------------------------------------------------- -------------------------------------------------- ----------}

_{class CMNDerivedFormType}

_{

_{DWORD id; // object}

_}

_{class CMNHasType}

_{

_{WORD hasid; // serial}

_{DWORD id; // if it from concept network}

_{CString name; // if it has no name}

_}

_{class CMNVerbType}

_{

_{WORD verbid; // Particle type}

_{char ptype; // related object}

_};

_{typedf CTypePtrArray <CMNDerivedFromType, CMNDerivedFromType *> CMNDerivedArray;}

_{typedf CTypePtrArray <CMNHasType, CMNHasType *> CMNHasArray;}

_{typedf CTypePtrArray <CMNVerbType, CMNVerbType *> CMNVerbArray;}

_{class CMNObj: public CObject}

_{

_{DWORD id;}

_{CString name;}

_{CMNDerivedArray instance; // is instance of}

_{CMNHasArray has;}

_{CMNVerbArray verb;}

_}

_{typedf CTypePtrArray <CMNObj, CMNObj *> CMNDerivedArray;}

_{#define MAX_CASHED_OBJ_IN_MEMORY512}

_{-------------------------------------------------- -------------------------------------------------- ----------}

Referring to FIG. 10, the object ID is a list ('is' list) that performs a defining function, such as "Internet is media", and a list ('has', which performs a location function, such as "Korea is in Asia"). list) and a 'possess' list that performs a possessive function, such as "hands are part of the body."

More specifically, the definition of is is

_{(is <obj1> <obj2>)}

In this case, <obj1> cannot have a common noun, and when <obj2> comes with a normal noun, it simply establishes a link relationship with the noun. When <obj2> has a proper noun, <obj2> is newly created. Create and set a link with that obj. If <obj1> is a common noun and <obj2> is a proper noun, the positions are interchanged. The return message of (is <obj1> <obj2>) is <obj1>.

For example, "Internet is media" is (is "Internet" media "), and" Rose's name is xx "(has' Rose '(has' rose' (is' name'xx) '))).

Also, locate of

_{(locate <obj1> <obj2>)}

It can be configured as follows, and its return message is <obj2>.

For example, "Korea is in Asia" can be expressed as "locate" Korea "Asia".

Also, owned has,

_{(has <obj1> <obj2>)}

<Obj1> is the subject and is created if the subject does not exist in the knowledge base. If <obj2> is a normal noun, only a link is set. If <obj2> is a proper noun, a new <obj2> is created and a link is established with the obj. The return message is <obj2>.

For example, "hands are part of the body" can be represented as (has 'body' hands '), and "rose's rose will be missed" (has' mija '(has' withdrawal 'rose') )

In the above embodiments, the basic forms of definition, location, and ownership have been briefly described, but it is obvious that compound statements or predicate expansion forms are possible.

FIG. 11A is a diagram for describing an operation flow of a meta search engine module according to an embodiment of the present invention, and FIG. 11B is a diagram for describing a flow of active operation of an agent collaboration meta search engine according to an embodiment of the present invention. 11C is a view for explaining the flow of manual operation of an agent collaboration meta search engine according to an embodiment of the present invention.

Referring to FIG. 11A, a search keyword is first transmitted to a meta search module.

The search engine then obtains a list of URLs and enters the search keywords into a Helmholtz Machine, a type of search engine. In this example, inputting a search keyword into the Helmholtz machine is described as an example. However, in consideration of the usefulness of the current work, the Bayesian network, the latent variable model, and the like may be selected. Such a search engine performs document classification from sites visited by a user by clicking on a document searched by a meta search engine or by a user simply visiting through web browsing and learning this as an internal variable of the search engine itself. In addition, when the meta search engine receives the search keyword, the keyword string having a similar correlation is output from the search keyword, and the frequency of appearance of the site and keyword string visited by the meta search engine is investigated. Reflect on the poem.

Then visit the site based on the URL list to get the HTML of the site and remove the tag. In this case, the collaboration controller simply bypasses it.

The noun index is then extracted from the untagged HTML.

In addition, by inputting a search keyword inputted to the Helmholtz machine, the frequency of the extracted index word list is evaluated, and the score is assigned to each URL, and the HTML of the site is obtained until all the URLs of the URL list are visited. After that, repeat the step of removing tags.

Then, based on the click profile, the search results are rearranged in order of the most frequently visited sites.

Referring to FIG. 11B, when a search keyword is first transmitted to a meta search engine module, the meta search engine module obtains a URL list through various search engines and inputs a search document into a Helmholtz machine.

Visit the site based on the list of URLs, get the site's HTML, and remove the tag.

The URL list is then divided by the number of remote agent systems participating in the collaboration. For example, if the number of remote cooperative agent systems is 10 and the list of URLs is 1,000, the number is divided by 100, which is called an AC URL List, and the first 100 of them is called a local URL list.

The collaborative information retrieval (IR) controller then divides the agent collaborative (AC) URL list into the agent collaborative module, and then visits the site with a local URL list (100 in the present invention), extracts indexes and assigns a ranking. .

The meta search engine waits for a result from the remote agent system that sends the AC URL List, and applies the click propile ranking by adding the ranking result and the result obtained after the wait.

Referring to FIG. 11C, first, an information retrieval (IR) command (ac-ir-request <kw-list) via an agent collaboration unit (ACM), a job scheduler unit (TSC), an information retrieval (IR) module, and a meta retrieval engine. > <hm-kw-list> <begin-n> <n> <in-domain-flag> <depth>) and search engines with values <kw-list>, <begin-n>, and <n> Get a list of URLs from.

The collaborative information retrieval (IR) controller visits the site with the above-mentioned URL list and <hm-kw-list>, performs indexing and ranking application, and repeats the operation until all entries in the URL list are generated.

The list of ranked URLs is then sent to the agent collaboration module.

12 is a diagram for explaining an agent collaboration module according to an embodiment of the present invention.

Referring to FIG. 12, an agent collaboration module according to an embodiment of the present invention operates as a thread, and receives a request from a online remote agent system to process a corresponding command with a listener 410. It includes a requester (420) for transmitting a request to the remote agent side, and a facilitator (Facilitator) 430 that operates as a thread and maintains an agent profile DB.

The structure of the agent profile DB is shown in Table 4 below.

Field type remark bLiveCheck BOOL _{Whether the last live check was successful} nAgentType WORD _{Bit value 1 = agent profile received from server 2 = profile received from the remote agent system 4 = A collaboration request has been made from a remote agent system 8 = A collaboration request has been made to a remote agent system 0x20 = agent list has been sent to remote agent system 0x40 = flag indicating that the remote agent system is an agent server 0x80 = flag indicating that this is a seed server} TmCheck CTime _{Last (or last live check) date and time, the last time used by the facilitator or requester} tmStable CTime _{This time is increased if the facilitator has performed a live check before the live check has been successful. That is, the remote agent system is operating stably without being down this time.} tmNotify CTime _{The time the last notification was sent if the remote agent system sent it. Judging this time determines whether to send notifications to the remote agent system.} cntUsage DWORD _{Indicates how often the requester uses this remote agent system.} dwID DWORD _{Unique Adi from Remote Agent System} StrName CString _{Username of remote agent} Strip CString _{IP address} nRespTime WORD _{Response time (msec) at live check of facilitator}

Listener (Lstener) responds to the request of the remote agent system, the request of the remote agent system can be divided into three broad.

The first is a request to send a response when a remote agent system connects to announce its existence, to measure the response time of the remote agent system, and to add the remote agent system to its agent list.

The second is a request that the remote agent system sends when it requests an agent list.

Third, a request for sending a query to the remote agent system through the requester and receiving a response to the query.

The requestor requests a query from the remote agent system side, at which time the remote agent systems in the remote agent list have the fastest response time.

The facilitator plays a key role in the performance of the agent collaboration module, which maintains the list of agents with the fastest network access for smooth communication with remote agent systems when in collaboration mode. In addition, the facilitator checks periodically whether the agent software has been upgraded to the server or stable agent system, and regularly checks the presence / activity of the remote agent systems registered in its agent list (Live / Presence Check). Requesting and receiving responses from counterpart remote agent systems to prioritize agent systems according to their response time. When the agent system is running, this work is performed regularly, and the agent system which always responds among the remote agent systems and has a short response time is selected as its own server. The agent system selected as the server is used for later agent upgrade.

More specifically, the user-side agent system notifies its presence to the seed server, waits for receiving a response, records the response time in the agent profile DB, and is executed once when the agent system is first run.

It also requests an agent list from the seed server, which is executed only once when the seed server is first run after the agent system has been installed from the target. It then requests an agent list for the remote agent system in its agent profile DB.

It also requires the seed server to update the module, and the facilitator manages the version number of every module. The version information is used to compare the version between the remote agent system and the user agent system. When the version of the remote agent system is low, the requester thread that transmits the updated data module from the user agent software to the remote agent system is executed. In this case, the agent collaboration module sends a module up data signal to the task scheduler when the file is targeted for module update by intercepting a signal from the agent collaboration module that receives a response to a normal file request and notifies it to the task scheduler.

It also performs regular live checks on all remote agent systems in its profile database. At this point, the live check is performed when the agent system is playing and the behavior is exactly the same as the presence notifier, but the difference is that the presence notifies only the seed server.

FIG. 13A is a view for explaining a check processing procedure of existence / activity between collaboration modules of FIG. 12.

Referring to FIG. 13A, when the user agent system outputs a signal for checking existence / activity on the remote agent system side (step S411), the remote agent system side provides a response to the user agent system side (step S413). ). Here, the remote agent system may be an agent system installed and operated on a general user side PC similarly to the user side agent system, or may be a predetermined Internet website that initially provides a service of an agent program.

If the user agent system side records and stores in the agent profile DB based on the response message of presence / activity from the remote agent system (step S415), and transmits the current notification recognition message to the remote agent system side (step S417), the remote agent On the system side, this is recorded and stored in the agent profile DB (step S419).

FIG. 13B is a diagram for describing an agent list request processing procedure of the facilitator of FIG. 11.

Referring to FIG. 13B, when the agent list is requested from the user agent system to the remote agent system side (step S421), the remote agent system transmits the list to the user agent system side (step S423), and the agent profile DB is held. Record and store (step S425).

Subsequently, the user agent system records and stores the agent list provided to the remote agent system in the agent profile DB held by the user agent system (step S427).

FIG. 13C is a view for explaining an agent software upgrade request processing procedure of the facilitator of FIG. 12.

Referring to FIG. 13C, as the module update is checked in the user agent system (step S431), the remote agent system side transmits the updated module list to the user agent system side (step S432).

If the user agent system side requests a module (file) from the remote agent system side (step S433), the file transfer start header is transmitted to the user agent system side (step S434).

Subsequently, when the user agent system side transmits a file reception preparation message to the remote agent system side (step S435), the remote agent system side transmits the file (updated file) to the user agent system side (step S436).

The user agent system side receives the file (step S437) and updates the agent program through the facilitator upgrade module (step S438).

14 is a view for explaining a web page displayed on the user-side PC according to an embodiment of the present invention.

Referring to FIG. 14, the artificial intelligence agent system according to the present invention may be driven on a general web page including a feature browser, a site, and a search display window. In particular, the feature browser shows an agent result and a web page display window, the site and search display window enables viewing of general web surfing and search results, and a suggestion result display window allowing viewing of the results arranged by the agent system. .

In addition, the guide, which is a type of agent dialog that can be used with the speech recognition engine to input natural language, can use characters similar or identical to the help agent used in Microsoft's Outlook.

15A is a diagram for explaining new agent system registration according to an embodiment of the present invention, FIG. 15B is a diagram for explaining the use of an agent according to an embodiment of the present invention, and FIG. 15C is according to an embodiment of the present invention. FIG. 15D is a view for explaining agent server selection after a predetermined time elapses, and FIG. 15D is a diagram for explaining cooperation between agent systems according to an embodiment of the present invention.

Referring to FIG. 15A, by first downloading and installing an agent program from an agent server, a newly registered first agent system notifies its presence to the agent server side and requests a list known to the agent server (step S11). The agent server also has a list of previously registered second, third to n-th agent systems, and adds the first agent system to it. Here, the agent server stores a predetermined program for the artificial intelligence agent system and provides the corresponding program when a download request of the agent program is input from the general client side. The agent program provided at this time provides an initialized knowledge base that can be stored in common in each agent system. This initialized knowledge base is installed on each user's PC, and as the user trains the agent program, an agent system that is completely different, that is, linked to the characteristics of the user, is created after a certain time.

Subsequently, the agent server delivers the first agent system newly registering the second, third to n-th agent systems from its own list (step S12). At this time, the first agent system knows the existence of the second, third to n-th agent system.

Subsequently, the first agent system connects to the second, third to n-th agent systems in order to inform the second, third to n-th agent systems of their existence and requests an agent list (step S13).

Then, the second, third to n-th agent system adds the first agent system to its own agent list and sends its own list in response to the request of the first agent system (step S14).

Referring to Fig. 15B, first the first agent system informs the server of its existence (only for one week) (step S21). It does not request the agent list from the server.

Subsequently, the first agent system checks agent systems registered in its agent list and updates the ranking (step S22). At this time, the ranking is made by measuring the response time for the request in the agent list of the first agent system. It is also used to designate its own collaboration partner when the first agent system becomes collaborative in the future.

Subsequently, the second, third to n-th agent systems that receive the list request from the first agent system transmit their own lists to the first agent system (step S23). In addition, the response time from the first agent system is measured to determine the rank of the agent list.

Referring to FIG. 15C, first the first agent system no longer informs the server of its existence or requests a list, but instead announces its existence to a second agent system registered in its agent list and displays the list. Request (step S31).

Subsequently, the second agent system knows the existence of the first agent system, measures the response time, and transmits the list (step S32).

As in the steps S31 to S32 described above, the first agent system notifies its presence to the third and fourth agent systems, requests a list, measures the response time, and then transmits the list to the first agent system. (Steps S33, S34, S35, S36).

The first agent system then selects one of the agent systems that has the most stable response to its agent list and registers it as its secondary server. The agent system, which becomes the secondary server, is used by the first agent system to announce the presence of the body. Here, stable response means always responding to a request of the first agent system.

Referring to FIG. 15D, first, the first agent system selects the agent systems that responded most quickly from the agent list known to the agent, and in the example of the present invention, selects the second agent system and the third agent system to send an inquiry (step). S41, S42), the agent system knowing the answer to the query, in the example of the present invention, the third agent system immediately sends a response to the first agent system side.

In addition, the second agent system that does not know the answer to the query re-sends the question to the agent systems that have responded the fastest in their agent list, that is, the fifth and sixth agent systems (steps S41-1, S41-). 2). This is called routing.

The number of routing hops between agent systems that do not know the answer to the query is limited. If all agent systems maintain 100 of the nearest (or fastest) agent system in their agent list and limit the number of hops to 4, then theoretically 100 ⁴ = 1 billion agents will participate in agent collaboration. Here, the routing hop count is the total number of paths passed by the agent system. In the above example, the paths to the first agent system, the second agent system, the sixth agent system, and the seventh agent system have a hop number of three, The route to the first, second, and sixth agent systems has two hops.

The first agent system, which has received a response through the various paths described above, adds to the agent list its agent system that has sent a response to its query.

As described above, when a query by the user is input to the agent system side, toss the query to other agent systems to obtain an answer to the query, even if the number of hops is limited to 4 on the first agent system side. It sends a query to 100 million agent systems and performs the operation. Of course, if the number of hops mentioned in the present invention is limited to 4, it is theoretical to send a query to 100 ⁴ = 1 billion agent systems. In addition, the agent list maintained by the second agent system may overlap with the agent list maintained by the first agent system.

Next, the functions of the artificial intelligence agent system according to the present invention will be described.

The artificial intelligence agent system according to the present invention operates a character that reacts in various ways according to a user's command. The character is a user interface element that can be user-characterized, and according to a user's command, a character may simultaneously provide a response to it in voice and text, and may give various accessories to the character as the information managed by the agent program increases.

In addition, the artificial intelligence agent system according to the present invention does not need help in the manual. That is, you can ask the agent system installed on your computer without having to go through the manual and call and e-mail to learn how to use the software.

In addition, the artificial intelligent agent system according to the present invention analyzes all the sentences input by the user one by one through natural language processing, the agent system remembers, the stored contents are preserved in the form of knowledge, and are always reused. In addition, individual agent systems will have different knowledge (which reflects the characteristics of the user), which will be shared across the network. The amount of knowledge accumulated by a user may not be very high, but the amount of knowledge that tens or hundreds of thousands of agent systems accumulate through a network may be enormous. And because this knowledge is shared in real time, you don't have to waste time clicking through the results retrieved by the search engine to get one or two pieces of information.

In addition, the artificial intelligent agent system according to the present invention remembers the agent system even if the site has never been visited even if the site has not been bookmarked. In other words, every site visited and every word entered is remembered by the agent system. Through these acquired memories, the agent system gradually learns what individual users like, frequently visited sites, and what they want and need. This means that as the frequency of use of the agent system increases, the agent system will exist as another individual person in the computer or on the network that reflects the characteristics of the individual user as it is. The agent software that the user first downloaded and installed is the same, but over time, the agent system can appear in cyberspace with different personalities, with different knowledge for each user.

In addition, commercially available distributed computer environments, such as peer-to-peer networks, do not exist or are used for extremely limited purposes, and these peer-to-peer networks simply share files. However, the artificial intelligence agent system according to the present invention will share knowledge with secondary file sharing. For example, assuming that it takes 10 hours to render a picture, and 100,000 copies are made to create 3D animations, it can be done within 100 hours using the P2P network, which normally takes 100,000 * 10 hours. Computer resources that cannot be utilized in a distributed environment for each individual can be efficiently utilized through an agent P2P network.

In addition, the artificial intelligent agent system according to the present invention may have various external modules. In other words, an external module that brings the news that corresponds to the user's interest among the news articles once every morning to reflect the user's taste, an external module that monitors the stock price and automatically sells / bushes the stock at the time of buying / buying, According to the present invention, various modules such as an external module which collects only the articles posted on the frequently-visited web bulletin board and shows them in the morning, as well as an external module which notifies the product when a product which may be of interest to the market appear in the market according to the present invention It can be mounted on artificial intelligence agents.

Although the above has been described with reference to a preferred embodiment of the present invention, those skilled in the art will be able to variously modify and change the present invention without departing from the spirit and scope of the invention as set forth in the claims below. It will be appreciated.

As described above, by using the intelligent agent system according to the present invention, it becomes a live program while adding knowledge to a stationary program, and can automatically solve a task desired by the user on behalf of the user.

In other words, in a flood of information, users are being indiscriminately attacked, whether they want it or not, and the agent system is forced to categorize the information needed by the user and the rest of the information in this myriad of information. Appeared. Using such an agent system, the user can find out exactly what he / she wants by judging what the user wants, having a conversation between themselves, searching for the agent system that he needs, and moving around the network.

In addition, in order to start a program stored in the user's computer, the user side has had the trouble of searching for a file in which the program is stored, but inputs a command for commanding a start or voice recognition function to the display window of the intelligent agent system according to the present invention. In the case of a built-in agent system, a program embedded in its own computer can be started by means of a predetermined microphone or the like.

Claims (15)

Receives a command from the user in at least one of sentences or voices, requests conversion into a predetermined format through the first path, and outputs the converted command according to the response through the second path. An agent interface unit outputting information according to the response to the user side; A natural language processor configured to convert the command provided through the first path into a predetermined format, and to feed the converted command back to the agent interface unit; A task scheduler configured to perform a search schedule task according to a user's converted command input through the second path, and to provide information according to the response to the agent interface unit; A knowledge base unit which stores information provided from the user and adapts the user environment by storing the user's profile based on the stored information; And It provides a command input via the task scheduler to the external one or more artificial intelligent Internet agent system side to request data, and provides the data according to the response to the agent interface unit, the external artificial intelligence Internet agent system When a request for data is input from the agent, an agent collaboration unit which checks the knowledge base, processes the corresponding command, transmits the command to the external artificial intelligence internet agent system, and maintains an agent profile. Artificial intelligent Internet agent system comprising a. The method of claim 1, A meta search unit for requesting at least one information search engine to search if the converted command is search request information and outputting search information according to a response thereof; And When the data is input from the user via the task scheduler, the data is stored in the knowledge base, and when the search request is input from the task scheduler, the knowledge base unit is searched first. Various module units which provide the information to the job scheduler if the corresponding data is not present, and the meta search unit is started and requested to be searched. Artificial intelligent Internet agent system further comprises. The method of claim 1, The artificial intelligence Internet agent system, characterized in that it further comprises a chat unit for storing the desired information in the knowledge base through communication with the user. The method of claim 3, wherein the chat unit, The artificial intelligence Internet agent system, characterized in that used when the user does not find the information requested by the knowledge base unit, agent collaboration unit, meta search unit. The method of claim 1, wherein the natural language processing unit, Syntax analysis module; Verb subcategorization dictionary module; A noun category dictionary module for tagging a noun category required by the verb subcategorization dictionary module; Semantic network module; And Korean English Band Dictionary Module Artificial intelligent Internet agent system comprising a. (a) checking whether a command of at least one type of a sentence or a voice is input from a user and converting the command into a functional grammar through a natural language processing process when the command is input; (b) checking whether the converted command is an internal execution command, and in the case of the internal execution command, controlling driving of a corresponding stored program; And (c) checking whether the converted command is a knowledge base-related command, and if the command is a knowledge base-related command, perform a knowledge base processing or a knowledge base search and convert the result into at least one of a text and a voice. Output to Method of driving an artificial intelligent agent comprising a. The method of claim 6, (d) if the converted command is a knowledge base related command, performing an agent collaboration operation and outputting the result to the user in at least one of a text and a voice; Method of driving an artificial intelligent agent further comprising a. The method of claim 6, (e) Checking whether the converted command is a meta-search search command, and if the meta-search search command is performed, the meta-search is performed to form a result of the query in at least one of a text and a voice. Outputting to the user side; And (f) performing an agent collaboration operation and outputting the result to the user in the form of at least one of a text and a voice; Method of driving an artificial intelligent agent further comprising a. The method of claim 6, (g) checking whether the converted command is a schedule management command and performing a schedule management operation when the converted command is a schedule management command. Method of driving an artificial intelligent agent further comprising a. The method of claim 6, (h) checking whether the converted command is a command for an external execution module, and performing an external execution operation in the case of an external execution module command; And (i) if the converted command is not an external execution module command, chatting with a user and an agent Method of driving an artificial intelligent agent further comprising a. (a) checking whether a command of at least one type of a sentence or a voice is input from a user and converting the command into a functional grammar through a natural language processing process when the command is input; (b) checking whether the converted command is an internal execution command, and in the case of the internal execution command, controlling driving of a corresponding stored program; And (c) checking whether the converted command is a knowledge base-related command, and if the command is a knowledge base-related command, perform a knowledge base processing or a knowledge base search and convert the result into at least one of a text and a voice. Output to A computer-readable recording medium having a built-in artificial intelligence agent program comprising a. The method of claim 11, (d) if the converted command is a knowledge base related command, performing an agent collaboration operation and outputting the result to the user in at least one of a text and a voice; A computer-readable recording medium having an artificial intelligence agent program further comprising a. The method of claim 11, (e) Checking whether the converted command is a meta-search search command, and if the meta-search search command is performed, the meta-search is performed to form a result of the query in at least one of a text and a voice. Outputting to the user side; And (f) performing an agent collaboration operation and outputting the result to the user in the form of at least one of a text and a voice; A computer-readable recording medium having an artificial intelligence agent program further comprising a. The method of claim 11, (g) checking whether the converted command is a schedule management command and performing a schedule management operation when the converted command is a schedule management command. A computer-readable recording medium having an artificial intelligence agent program further comprising a. The method of claim 11, (h) checking whether the converted command is a command for an external execution module, and performing an external execution operation in the case of an external execution module command; And (i) if the converted command is not an external execution module command, chatting with a user and an agent A computer-readable recording medium having an artificial intelligence agent program further comprising a.