WO2008119270A1

WO2008119270A1 - A chatting robot system and a method, a device for chatting automatically

Info

Publication number: WO2008119270A1
Application number: PCT/CN2008/070217
Authority: WO
Inventors: Haisong Yang; Yunfeng Liu; Zhiyuan Liu; Rongling Yu
Original assignee: Tencent Technology (Shenzhen) Company Limited
Priority date: 2007-03-30
Filing date: 2008-01-30
Publication date: 2008-10-09
Also published as: CN101076060A

Abstract

A chatting robot system includes a distribution server and at least two robot servers, wherein the distribution server is used for receiving users' sentences, and distributing the users' sentences to the corresponding robot server according to load balancing principle; and the robot servers are used for responding the users' messages and transmitting automatically feedback replies to the users. The invention also discloses a method and a device for chatting automatically. After implementing the embodiment of the invention, lots of users can be supported, expandability of the system also can be increased, and feedback replies has individuation. Additionally, the embodiment of the invention make the robot server to understand the natrual language and exea chatting robot system and a method, a device for chatting automaticallycute logic consequence more expediently.

Description

Chat robot system and automatic chat method and device

Technical field

The present invention relates to the field of artificial intelligence, and more particularly to a chat robot system and an automatic chat method and apparatus. Background of the invention

In recent years, communication methods between people have become increasingly abundant. Instant messaging tools, mobile phone short messages and other means of communication are becoming more popular. Based on these communication means, in addition to the communication between people, it also makes communication between human and artificial intelligence systems possible.

The chat robot system is an artificial intelligence system that can communicate with people through natural language at any time by means of communication. In addition to the chat function, the chat robot system can also have a large number of value-added services, such as weather queries, map queries, life information queries, calculators, dictionaries, etc., and even play games with people. The chat robot system is essentially an automated question answering system. The automatic question answering system is based on natural language understanding technology and involves many disciplines such as computational linguistics, information science and artificial intelligence. It is one of the hotspots of computer application research. Natural language understanding is an important research direction in the field of artificial intelligence. It enables computers to understand and apply human natural language, understand user's conversation content or query intent, and realize effective communication based on natural language between human and computer.

The chat robot utilizes natural language processing technology, knowledge base and real-time updated information resources to complete the analysis and processing of user problems on the one hand and the correct answer generation on the other hand.

At present, there is a chat robot system in the prior art, which comprises a communication module, a query server, a game server, an artificial intelligence server and a corresponding database. In this chat robot system, the database is used as a way of recording knowledge points, and the user performs various conversations with the chat robot through an instant messaging platform or a short message platform. However, the prior art chat robot system architecture lacks an information distribution mechanism, and encounters large difficulties in supporting a large number of users, and thus the scalability of such a system architecture is poor. Summary of the invention

The embodiment of the invention provides a chat robot system to meet a large number of users and improve scalability.

The embodiment of the invention provides a method and a device for implementing a chat robot system to meet a large number of users and improve scalability.

The technical solution of the embodiment of the present invention is implemented as follows:

A chat robot system, the system comprising a distribution server and at least two robot servers, wherein:

a distribution server, configured to receive user statements and distribute the user statements to the corresponding robot server according to a load balancing principle;

A robot server for automatically feeding back responses to the user in response to a user statement.

An automatic chat method, the method comprising:

A. The distribution server receives the user statement and distributes the user statement to the corresponding robot server according to the load balancing principle;

B. The robot server automatically returns a reply to the user in response to the user statement.

A robot server, comprising: a user sentence receiving unit, configured to receive a user statement distributed by a distribution server according to a load balancing principle, and send the user statement to a statement object identification unit;

a statement purpose identification unit, configured to identify whether the user statement is a formatted statement or an unformatted statement;

a formatting language execution unit, configured to execute the formatting statement when the statement purpose recognition unit recognizes that the user statement is a formatted statement; a natural language understanding unit, configured to perform natural language understanding on the user statement to determine a statement attribute when the statement purpose recognition unit recognizes that the user statement is an unformatted statement;

a user attribute management unit, configured to manage attribute information of the user;

The inference engine unit is configured to load the knowledge base, and perform pattern matching in the knowledge base according to the statement attribute and the user attribute information to determine a matching feedback reply.

As can be seen from the above technical solution, the system proposed by the embodiment of the present invention includes a distribution server and at least two robot servers, wherein the distribution server receives the user statement and distributes the user statement to the corresponding robot server according to the load balancing principle; The response is automatically fed back to the user in response to the user statement. Therefore, after the embodiment of the present invention is applied, the distribution server controls the distribution of the user statement according to the load balancing principle. The number of the distribution servers may be multiple, and each distribution server may be connected to multiple robot servers, and each of the robot servers is independent in function. Therefore, the embodiment of the present invention has high scalability, and is particularly suitable for a large number of users. BRIEF DESCRIPTION OF THE DRAWINGS

1 is a schematic structural diagram of a chat robot system according to an embodiment of the present invention; FIG. 2 is a schematic structural diagram of a robot server according to an embodiment of the present invention; FIG. 3 is an exemplary diagram of an automatic chat method according to an embodiment of the present invention; 4 is a schematic structural diagram of a chat robot system according to an embodiment of the present invention; FIG. 5 is a schematic flow chart of an automatic chat method according to an embodiment of the present invention. Mode for carrying out the invention

The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

In the embodiment of the present invention: the distribution server receives the user statement, and then according to load balancing The principle distributes the user statement to the corresponding robot server. The robot server then automatically responds to the user in response to the user statement, so that the load of the robot server is balanced, meets the needs of a large number of users, and improves scalability.

Further, the embodiment of the present invention uses a knowledge base file instead of a database as a carrier of knowledge points. The knowledge base file is more suitable for reasoning and generating personalized responses in a specially designed format, so that the chat robot is more intelligent.

FIG. 1 is a schematic diagram showing an exemplary structure of a chat robot system according to an embodiment of the present invention. As shown in FIG. 1, the system includes a distribution server 101 and at least two robot servers.

102, where:

a distribution server 101, configured to receive a user statement, and distribute the user statement to the corresponding robot server 102 according to a load balancing principle;

The robot server 102 is configured to automatically feed back a response to the user in response to the user statement. Preferably, the number of the distribution servers 101 is at least two, thereby supporting more chat bots, and preferably the system further includes a redirect server 103;

The redirecting server 103 is configured to receive a user session, and redirect the user session to the distribution server 101 that matches the attribute information according to the attribute information of the user;

The attribute information of the user may include the user's IP address and/or the client account number. Thus, the redirect server 103 allocates the distribution server in consideration of the user's IP address and/or client account number. For example, the user session can be redirected to a distribution server that is closer to the IP address, or the user session can be redirected to a distribution server corresponding to the client account. At this time, the distribution server 101 matching the attribute information is configured to receive the user statement of the user session, and distribute the user statement to the corresponding robot server 102 according to the load balancing principle.

For another optional function of the above redirect server, the number of the distribution server 101 is at least two, the system further includes a redirect server 103; The redirect server 103 at this time is configured to receive a user session, and distribute the user statement to the corresponding distribution server 101 according to the load balancing principle;

The distribution server 101 is configured to receive a user statement of the user session from the redirect server 103, and further distribute the user statement to the corresponding robot server 102 according to the load balancing principle.

The distribution server 101 can receive the user statement by any one of the following communication modes or any combination of more than one type: receiving the user statement by means of instant messaging; receiving the user statement by means of email; receiving the user by short message mode Statement.

Similarly, the robot server 102 can automatically feedback the user to the user through any one of the following communication modes or any combination of more than one: receiving the user sentence by means of instant messaging; receiving the user statement by means of email; The way to receive user statements.

2 is an exemplary specific structure of the robot server of FIG. 1. As shown in FIG. 2, the robot server 102 includes:

The statement purpose identification unit 201 is configured to identify whether the user statement is a formatted statement or a non-formatted statement;

The formatting language execution unit 202 is configured to execute the formatting statement when the statement purpose identification unit 201 recognizes that the user statement is a formatting statement;

In Fig. 1, the formatting language execution unit 202 is not explicitly shown. In essence, the formatting language execution unit may include: any one of an information query module, a dialog education module, and a game module, or any combination of more than one, wherein:

An information query module, configured to determine a feedback reply of the formatted statement of the information query class; a dialog education module, configured to determine a feedback reply of the formatted statement of the dialog education class; a game module, configured to determine feedback of the formatted statement of the game class reply.

For the single view, only the information query module 2021 and the game module 2022 are shown in FIG. The information query module 2021 is connected to the utility information database 209 for searching the utility information database 209 to determine a feedback reply of the formatted statement of the information query class.

The bot server 102 further includes a natural language understanding unit 203 for performing natural language understanding on the user statement to determine the statement attribute when the statement object recognition unit 201 recognizes that the user statement is an unformatted statement;

The bot server 102 may further include a user attribute management unit 204 for managing attribute information of the user;

The robot server 102 may further include an inference engine unit 205 for loading the knowledge base 206, and performing pattern matching in the knowledge base 206 according to the statement attribute and the user attribute information to determine a matching feedback reply. The attribute information of the user may include: any one of the user's name, the gender of the user, the city where the user is located, the user's preference, or any combination of more than one.

The natural language understanding unit 203 may perform any one of the following operations or any combination of more than one to determine a statement attribute for the user sentence: a word segmentation, a question stem extraction, a question type judgment, and a topic judgment.

The bot server may further include an alternate inference engine unit 207 for loading the updated knowledge base 208 when the knowledge base is updated, and converting the function with the inference engine unit 205 after loading the updated knowledge base 208.

The chat bot system in FIG. 1 may further include a learning server, and the knowledge base 206 in FIG. 2 may be located in the learning server;

The learning server is used to record user conversations and convert the user conversation into a knowledge base file format and save it in the knowledge base 206.

The chat bot system may further include an audit server. An audit server that audits user conversations based on pre-set knowledge review rules and sends only the audited user conversations to the learning server. For example, to prevent writing yellow in the knowledge base 206 For information such as unhealthy information, some filtering keywords may be preset in the auditing server. When the user dialogue includes a filtering keyword, the auditing server determines that the knowledge does not comply with the knowledge review rule, and does not send the user conversation to the learning server; otherwise, The auditing server determines that the knowledge conforms to the knowledge review rules and sends the user session to the learning server.

The knowledge base file format may include a question part and a reply part, the question part is saved in the form of a question trunk, and the reply part is saved in a complete reservation manner.

The knowledge base 206 is an important part of the chat bot system, storing a large amount of knowledge in a form suitable for reasoning and in the form of a question-and-answer statement pair. When the natural language sentence entered by the user matches a certain sentence in the knowledge base, the corresponding response is returned to the user.

To facilitate reasoning, the knowledge base 206 contains a number of attributes. The following is a description of the properties of the "Questions" section and the "Response" section.

The "question, part, stored in the knowledge base 206 is processed by natural language and saved in the manner of the question backbone. There are various rules for question stem extraction, such as replacing synonyms, such as "email" and "email" ", "Capital," and "Beijing" were replaced by a unified representation. For another example, only words that best express the semantics of the question are retained. For example, "Where is the Beijing Railway Station?", the main sentence is "Beijing Railway Station."

In addition, the "question, which is stored in the knowledge base 206, also includes the type of the question. For example, "Where is the Beijing Railway Station?" ,, this question, the type of question that is obtained through natural language processing is "inquiry place". Combining the "question stem" and "question type" of the "problem", the two attributes can represent a variety of different expressions of a problem. "Where is the Beijing Railway Station?" Other expressions of ", where is the capital railway station?" "How does the Beijing Railway Station go?, the main questions are "Beijing Railway Station", the question type is "question type" are "inquiry place" ". So these questions can return the same response.

The task of extracting the question type and the question stem when generating the knowledge base is completed by the learning server. The natural language understanding unit 203 of the robot server processes the online user input question using the same rules, and also generates the "question stem" and "question type" of the online user question for matching in the knowledge base. In addition, some questions "question trunks, which can contain wildcards, are used to extend the scope of the "question trunk". There are also questionable topics (such as entertainment, military, etc.) that are also an attribute of "problems."

The "response" part saved in the knowledge base 206 is not saved in the same way as the "problem" part, but a complete response is retained. A "problem" in the knowledge base corresponds to multiple "responses", one from user education. "Answer" will include an "Education User ID" attribute. The inference bow of the robot server After matching the "problem", the engine unit selects one of the multiple responses corresponding to the "problem" and returns it to the user. When selecting, the user's own education response is preferentially selected. A response may also contain some replacements that require the inference engine unit of the robot server to replace before returning a final response, which may represent some user attribute or other information. For example: A response "I know you are from XX." The "XX" part of the response ("XX" needs to be in a different format to represent different meanings, the specific format is omitted here) needs to be replaced with the information in the user attribute, the user The attribute query is completed by the user attribute management unit of the robot server. For example, if the city attribute of the user is "Shenzhen", the inference engine unit of the robot server replaces the "XX" part with "Shenzhen" in the final response generated, and finally The robot returns a response "I know you are from Shenzhen."

The chat bot system may further include a control server,

The control server is configured to send a knowledge base update switching instruction to the robot server 102 and the learning server; and a learning server, configured to convert the accumulated user dialogue into a knowledge base file format to update knowledge after receiving the knowledge base update switching instruction Library

The standby inference engine unit in the robot server is configured to load the updated knowledge base after receiving the knowledge base update switching instruction, and convert the function after the updated knowledge base and the inference engine unit. The control server periodically sends a knowledge base update switching instruction to the robot server and the learning server; or

The control server sends a knowledge base update switching instruction to the robot server and the learning server when the user session accumulates to a predetermined level.

The embodiment of the invention also discloses an automatic chat method. Fig. 3 is a diagram showing an exemplary structure of a robot server according to an embodiment of the present invention.

As shown in Figure 3, the method includes:

Step 301: The distribution server receives the user statement, and distributes the user statement to the corresponding robot server according to the load balancing principle;

Step 302: The bot server automatically returns a reply to the user in response to the user statement. In the above process, before the step 301, the user session request may be received, and the user session is redirected to the distribution server matching the attribute information according to the attribute information of the user; and then the distribution server receives the user statement of the user session. And distribute the user statements to the corresponding robot server according to the load balancing principle.

Optionally, the user session may be received before step 301, and the user statement is distributed to the corresponding distribution server according to the load balancing principle; the distribution server receives the user statement of the user session, and then distributes the user statement to the corresponding according to the load balancing principle. Robot server.

4 is a schematic diagram showing an exemplary structure of a chat robot system according to an embodiment of the present invention. In the embodiment shown in FIG. 4, the chat bot system proposed by the embodiment of the present invention includes five types of servers in addition to the necessary front-end communication system (instant messaging platform, short message platform, etc.): a redirecting (Redirect) server, Dispatch server, robot server, learning server, and control server.

When a new session from the current communication system starts, the front-end communication system first informs the redirect server, and the redirect server is based on the IP address of the client in the front-end communication system or The account used by the client determines which distribution server is used as the processing entry for this session.

After determining which distribution server is used as the processing entry, all communication in this session is no longer passed through the redirect server. All data sent from the front-end communication system in this session is directly sent to the determined distribution server until this time. The session ends. The redirect server completes the redirection task of the session in the entire chat bot, and considers the load balancing of each distribution server in the redirection process.

The distribution server is a general portal for session processing of the chat bot system, and is used for receiving all user statements of the session sent from the front-end communication system such as the instant messaging platform, the short message platform, etc., and distributing the user statements to each according to a certain mechanism. Robot server, this mechanism makes the load of each robot server approximately balanced. In order to facilitate expansion, there may be multiple distribution servers, and each distribution server may correspond to multiple robot servers.

The robot server is a server in the chat robot system that processes and responds to user statements. The robot server in a chat robot system may be one or more. User statements may be chat statements, information query questions, conversational education, other formatting commands, and the like. The robot server parses the user statement, determines its destination category, and performs different processing, and then returns the response directly to the front-end communication system such as the instant messaging platform or the short message platform.

As an exemplary analysis in the figure, the robot server may include a statement purpose recognition unit, a natural language understanding unit, an inference engine unit, a user attribute management unit, a formatting language execution unit, and the like, wherein the formatting language execution unit may include a game module and information. Query modules, etc.

The statement purpose recognition unit of the robot server recognizes the target category of the user statement, and the judgment statement is a formatted statement or an unformatted language statement. If it is a formatted statement, the discriminating is a dialog education formatted statement, an information query formatted statement, Game formatting commands, or other formatting commands (such as setting user properties). The natural language understanding unit of the robot server performs natural language processing on unformatted user sentences, and completes processing including word segmentation, question stem extraction, question type judgment, topic judgment, and the like.

The inference engine unit of the robot server loads the knowledge base file generated by the learning server, generates the natural language response based on the user sentence attribute output by the natural language understanding unit, and the user attribute extracted by the user attribute management unit, and completes the processing of the user chat sentence. .

To update the knowledge base, a robot server can contain two inference engine units, one for online real-time processing, generate natural language responses, and the other is a backup inference engine unit for loading recently updated knowledge bases.

The user attribute management unit of the robot server performs storage and query reading of user attributes. These attributes include the user's name, gender, city, and so on. The inference engine unit returns a personalized response based on these attributes. These attributes are voluntarily submitted by the user and are submitted in a number of ways, including by submitting some formatting commands to the chat bot.

The game module of the robot server processes the game formatting commands, and according to the game logic, advances the interactive game.

The information query module of the robot server implements practical information query, including weather query, map query, postal code query, mobile phone attribution query, life information query, dictionary query, calculator and other value-added service functions.

The learning server is responsible for recording the user education dialogue, which is saved in the user education database, and is left to be edited for manual review. After the manual review, these conversations will be converted into knowledge base files by the learning server. This knowledge base file format is logical reasoning. And specially designed. In addition to user education, the knowledge in the knowledge base includes other sources such as self-editing knowledge and knowledge deposited by the knowledge community. These different sources of knowledge are aggregated and merged into the knowledge base during the conversion process.

The knowledge base contains question and answer topic attributes, context attributes, and users who educate the Q&A

ID attribute (if the previous paragraph is an instant messaging platform, then the ID is the user using instant messenger The account ID) and other attributes for reasoning, where the user ID attribute is used to implement personalization during the chat process. In addition, the knowledge base also contains some semantic wildcards to extend the scope of the applicable questions for the response. Finally, such a knowledge base file is sent to each robot server under the command of the control server.

The control server is responsible for sending the knowledge base update switching instruction to the robot server and the learning server. The timing of the update switching instruction may be timed, or may be when the new user education session received by the learning server accumulates to a certain extent. After the command is issued, the learning server converts the question and answer dialog in the user education database into a knowledge base file and sends it to each robot server. Then the control server sends instructions to the robot server, and the backup inference engine unit in the robot server loads the latest knowledge base, and then The roles of the backup inference engine and the online inference engine are exchanged to complete the replacement of the knowledge base.

Before a new session begins, the redirect server first determines which distribution server is the total entry for the session (considering the load balancing of each distribution server), and then all the users of the session from the previous communication system. The statement is first passed to the distribution server, which passes the user statement to one of the robot servers (when the distribution server chooses to forward to which robot server, the robot server load balancing is also considered), after the robot server finishes processing the user statement The response is returned directly to the front-end communication system.

The main operational flow of the chat bot is the processing flow of the user's sentence during system operation. The process is described in detail below.

First, a session is established when a user starts chatting with the chat bot. The front-end communication system retains the redirect server address of the chat bot system. When the new session starts, the front-end communication system first informs the redirect server, and the redirect server determines the session. Which distribution server is handed over as a processing entry, and the front-end communication system is notified, after which all user statements transmitted from the front-end communication system of this session are handed over to the redirect server. When the redirect server decides which distribution server to process, it considers the load balancing or service characteristics of each distribution server, such as the account ID (such as the account of the instant messaging tool) used by the user when using the front-end communication tool. The number segment is used for distribution.

The following describes the flow of a chat bot to process a user question in this session after a session is established.

First, the distribution server of the chat bot first obtains the user sentence from the front-end communication system, and the distribution server transmits the user statement to one of the bot servers (when the distribution server chooses to forward to which bot server, the load balancing of each bot server is considered).

The robot server calls the statement destination identification unit to judge the intent of the user statement, and then performs different processing according to the target category, and the robot server returns the response directly to the front-end communication system after the user statement finishes processing.

FIG. 5 is a schematic flow chart of an automatic chat method according to an embodiment of the present invention. As shown in Figure 5, the method includes:

Step 501 to step 503: The distribution server receives the user statement from the front-end communication system, and the distribution server uses the load balancing mechanism to forward the user statement to a certain robot server, and the machine server determines the intention of the user statement;

Step 504: Determine whether the user statement is an unformatted language statement or a formatted statement, if it is a non-formatting language statement, execute step 505, if it is a formatting language statement, execute step 509;

Step 505 to step 506: performing segmentation, question stem extraction, question type judgment, topic judgment and the like on the user sentence, and determining whether the user's intention is to perform some information inquiry through natural language according to the question type and the topic judgment result. If yes, go to step 510, otherwise go to step 507;

Step 507 to step 508: The user attribute management unit of the robot server reads the user attribute to the inference engine unit, and the inference engine unit searches for a match from the knowledge base according to the user statement. a knowledge point, and generate a personalized response according to the topic attribute, the question type attribute, the user attribute and the like, and then execute step 514 and end the process;

Step 509: determining that the formatted statement is an information query, a dialogue education or a game command, if it is an information query, performing step 510; if it is a dialogue education, executing step 512; if it is a game command, executing step 511;

Step 510: The information query module of the robot server parses the user query intent, and searches from the utility information database, and takes the search result as a response, and then executes step 514 and ends the process;

Step 511: The game module of the robot server parses the user's game command, advances the game according to the game logic, and responds to the current progress of the game.

Step 512~Step 513: The robot server forwards the dialogue of the user education to the learning server; the learning server records it in the user education database; the robot server responds with a thank-you user education, and then executes step 514 and ends the process;

Step 514: The bot server returns the response directly to the front-end communication system, and the user finally receives the chat bot response.

As shown in FIG. 5, in step 502, the distribution server adopts a distribution mechanism that considers load balancing or service characteristics, for example, adopts a principle similar to redirect server redirection, according to the account ID used by the user when using the front-end communication tool ( The number segment of the instant messaging tool's account number is used for distribution.

At step 504, in order to facilitate identification, the formatted statement may be agreed with the user in advance. For example, "tq Shenzhen" means to query Shenzhen weather, "Q: Are you a robot? A: Yes, I am very smart." Indicates user education dialogue. The statement object class identification unit of the robot server recognizes the format statement according to these pre-agreed formats.

At step 505, the natural language understanding unit of the robot server adopts natural language processing technology to perform lexical analysis, grammar analysis, and semantic division on natural language sentences input by the user. Analysis and other processing.

The reason for the decision at step 506 is: Some users are accustomed to querying information in a natural language rather than a formatted statement, such as the user's statement "What is the weather in Shenzhen tomorrow?" In this case, the chat robot is processed through natural language. After the user's true intention is identified, the process is transferred to the information query module to process the information query.

At step 508, in order to make the chat robot personalized, the same user question can generate different responses, and the inference engine unit of the robot server searches for the semantic matching knowledge points from the knowledge base according to the user statement, and according to the topic attribute, A question type attribute, a user attribute, and the like generate a personalized response.

At step 512, if it is a dialog education formatted statement, the user education dialogue is forwarded to the learning server, and a response to the user education is returned directly to the user. The learning server records the user education conversation along with the user's ID—in the user education database.

As new user education conversations are added to the knowledge base, and knowledge accumulated by the knowledge community continues to be added to the knowledge base, the learning server continuously generates new knowledge bases, and the robot server constantly updates the knowledge base. The above process is completed by the control server by sending an instruction.

In summary, in the embodiment of the present invention, the beneficial effects of the chat robot system proposed by the embodiment of the present invention include:

(1) The chat bot proposed in the embodiment of the present invention may include a plurality of distribution servers, each of which may be connected to a plurality of bot servers, each of which is functionally independent. After the embodiment of the present invention is applied, the distribution server controls the distribution of the user statement according to the load balancing principle. The number of the distribution servers may be multiple, and each distribution server may be connected to multiple robot servers, and each of the robot servers is functionally independent, and thus the present invention is implemented. The example is very scalable and is especially suitable for a large number of users. Moreover, the embodiment of the present invention preferably further includes a redirect server. The redirect server decides which distribution server is used as the processing entry by one session, and then uses the distribution server as the total portal and distributor of all the user statements of the chat robot in one session, so that the robot server load is basically balanced and the system is expanded. Sexuality makes the chat bot system more suitable for supporting a large number of users.

(2) The embodiment of the present invention uses a knowledge base file instead of a database as a carrier of knowledge points. This knowledge base file is more suitable for reasoning and generating personalized responses in a specially designed format, which makes the chat robot more intelligent.

(3) The embodiment of the present invention generates a personalized response according to information such as a topic, a question type, a user attribute, and the like, so that the robot is more realistic in imitating the human language.

(4) The embodiment of the present invention supports information query (including natural language query) and human-computer interactive game, so that the user can get more services through the chat robot.

The above is only the preferred embodiment of the present invention and is not intended to limit the scope of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and scope of the present invention are intended to be included within the scope of the present invention.

Claims

Claim

A chat robot system, characterized in that the system comprises a distribution server and at least two robot servers, wherein:

The chat bot system according to claim 1, wherein the number of the distribution servers is at least two, and the system further comprises a redirect server;

a redirecting server, configured to receive a user session, and redirect the user session to a distribution server that matches the attribute information according to the attribute information of the user;

The distribution server matching the attribute information is configured to receive a user statement of the user session, and distribute the user statement to the corresponding robot server according to the load balancing principle.

The chat bot system according to claim 2, wherein the attribute information of the user includes an IP address of the user and/or a client account number.

a redirecting server, configured to receive a user session, and distribute the user statement to a corresponding distribution server according to a load balancing principle;

The distribution server is configured to receive the user statement of the user session, and further distribute the user statement to the corresponding robot server according to the load balancing principle.

The chat robot system according to claim 1, wherein the robot server comprises:

a formatting language execution unit for identifying that the user statement is The formatting statement is executed when the statement is formatted;

a natural language understanding unit, configured to perform natural language understanding on the user statement to determine a statement attribute when the statement purpose recognition unit recognizes that the user statement is an unformatted statement;

The chat bot system according to claim 5, wherein the attribute information of the user includes: any one of a user's name, a user's gender, a user's city, a user's hobby, or any combination of more than one. .

The chat bot system according to claim 5, wherein the natural language understanding unit is configured to perform any one of the following operations or any combination of more than one on the user statement to determine a statement attribute:

Participle;

Question stem extraction;

Question type judgment;

Topic judgment.

The chat robot system according to claim 5, wherein said robot server further comprises a standby inference engine unit,

The standby inference engine unit is configured to load an updated knowledge base when the knowledge base is updated, and convert the function with the inference unit after loading the updated knowledge base.

The chat bot system according to claim 5, wherein the system further comprises a learning server, wherein the knowledge base is located in the learning server;

The learning server is configured to record a user conversation, and convert the user conversation into a knowledge base file format and save it in the knowledge base.

10. The chat robot system according to claim 9, wherein the system Further including an audit server,

The audit server is configured to audit the user dialogue according to a preset knowledge review rule, and only send the audited user dialog to the learning server.

11. The chat bot system according to claim 9, wherein the knowledge library file format includes a question portion and a response portion,

The problem part is saved in the way of the question trunk, and the response part is saved in a complete reservation.

The chat bot system according to claim 11, wherein the formatting language execution unit comprises: any one of an information query module, a dialog education module, and a game module, or any combination of more than one, wherein:

The information query module is configured to determine a feedback reply of the formatted statement of the information query class; the dialog education module is configured to determine a feedback reply of the formatted statement of the dialogue education class; and the game module is configured to determine the game class Feedback reply to the formatted statement.

The chat bot system according to claim 1, wherein the distribution server is configured to receive a user statement by any one of the following communication methods or any combination of more than one:

Receiving user statements by means of instant messaging;

Receiving user statements by email;

Receive user statements by short message.

The chat robot system according to claim 1, wherein the robot server is configured to automatically feed back a response to the user by any one of the following communication modes or any combination of more than one:

Receiving user statements by means of instant messaging;

Receiving user statements by email;

Receive user statements by short message.

15. The chat robot system according to claim 9, wherein the system Further including a control server,

The control server is configured to send a knowledge base update switching instruction to the robot server and the learning server;

The learning server is configured to convert the accumulated user dialogue into a knowledge base file format to update the knowledge base after receiving the knowledge base update switching instruction;

An alternate inference engine unit in the robot server is configured to load the updated knowledge base after receiving the knowledge base update switching instruction, and convert the function after the updated knowledge base and the inference engine unit.

The chat robot system according to claim 15, wherein the control server periodically transmits a knowledge base update switching instruction to the robot server and the learning server; or the control server sends the robot session to the robot server when the user dialogue is accumulated to a predetermined degree. The learning server sends a knowledge base update switching instruction.

17. An automatic chat method, the method comprising:

The distribution server receives the user statement and distributes the user statement to the corresponding robot server according to the load balancing principle;

The bot server automatically feeds back the replies to the user in response to the user statement.

The automatic chat method according to claim 17, wherein the method further comprises: receiving a user session request, and redirecting the user session to the user session according to the attribute information of the user before the distribution server receives the user statement The distribution server whose attribute information matches.

The automatic chat method according to claim 17, wherein the method further comprises: receiving a user session before the distribution server receives the user statement, and distributing the user statement to the corresponding distribution server according to the load balancing principle.

The automatic chat method according to any one of claims 17 to 19, wherein the user statement is a chat statement, an information query question, a dialogue education statement or a format Command.

21. A robot server, comprising:

a user statement receiving unit, configured to receive a user statement distributed by the distribution server according to a load balancing principle, and send the user statement to a statement destination identification unit;

a formatting language execution unit, configured to execute the formatting statement when the statement purpose recognition unit recognizes that the user statement is a formatted statement;

The robot server according to claim 21, wherein the natural language understanding unit is configured to perform any one of the following operations or any combination of more than one on the user statement to determine a statement attribute:

Participle;

Question stem extraction;

Question type judgment;

Topic judgment.