WO2021082836A1

WO2021082836A1 - Robot dialogue method, apparatus and device, and computer-readable storage medium

Info

Publication number: WO2021082836A1
Application number: PCT/CN2020/117925
Authority: WO
Inventors: 何柯君; 夏晓松; 覃非
Original assignee: 中国银联股份有限公司
Priority date: 2019-10-30
Filing date: 2020-09-25
Publication date: 2021-05-06
Also published as: CN110955675B; CN110955675A

Abstract

Disclosed are a robot dialogue method, apparatus and device, and a computer-readable storage medium. The robot dialogue method comprises: acquiring input information of a user; determining a target scenario instance matching the input information; querying, in a preset resource warehouse, response information for the input information in the target scenario instance; and displaying the response information. According to the embodiments of the present application, the process of a robot searching for response information can be simplified, thereby improving robot dialogue efficiency and accuracy.

Description

Robot dialogue method, device, equipment and computer readable storage medium

Cross-references to related applications

This application claims the priority of the Chinese patent application 201911044588.X entitled "Robot dialogue method, device, equipment and computer-readable storage medium" filed on October 30, 2019. The entire content of this application is incorporated by reference. In this article.

Technical field

This application relates to the technical field of robot dialogue, and in particular to a method, device, equipment and computer-readable storage medium for robot dialogue.

Background technique

A robot with a dialogue function is essentially a computer program that communicates with people through voice or text. As one of the important applications of artificial intelligence, robots with dialogue functions have been widely used in finance, education, medical care, e-commerce, smart devices and other fields. Excellent robots can significantly improve people's work efficiency, saving enterprises more manpower and time costs.

Robots in related technologies rely on a knowledge base to achieve dialogue. The knowledge base is divided into three parts: matching rules, question and answer knowledge base, and response template.

In related technologies, each time the user’s input information is obtained, the input information is first analyzed through natural language technology to determine the user’s intention, and then the knowledge base’s matching rules are used to match the most similar problem to the input information, and then according to The response template gives the corresponding answer. In the matching process, natural language processing, machine learning, text mining and other technologies are used to optimize the matching method to improve the accuracy of the robot's response.

It can be seen that the accuracy of the existing robot response depends on the realized natural language analysis technology, knowledge domain model and matching rules. In the scenes used by enterprises, due to the large stock of knowledge of enterprises and the complex and changeable user scenarios, The accuracy and efficiency of conversation reply is low.

Summary of the invention

The embodiments of the present application provide a robot dialogue method, device, equipment, and computer-readable storage medium, which can simplify the process of the robot searching for response information and improve the efficiency and accuracy of the robot dialogue.

In the first aspect, an embodiment of the present application provides a robot dialogue method, which includes:

Obtain user input information;

Determine the target scene instance matched by the input information;

Query the response information of the input information in the target scene instance in the preset resource warehouse;

Show the response information.

In the second aspect, an embodiment of the present application provides a robot dialogue device, which includes:

The control module is used to obtain the user's input information; call the scene instance in the scene library to determine the target scene instance matched by the input information; display the response information returned by the execution unit;

The scene library is used to store preset scene instances;

The execution unit is configured to query response information of the input information in the target scene instance in a preset resource warehouse; and return the response information to the control module;

The resource warehouse is used to store the input information contained in each scenario instance and the corresponding response information.

In a third aspect, an embodiment of the present application provides a robot dialogue device, which includes a processor and a memory storing computer program instructions;

The processor implements the robot dialogue method described in the first aspect of the embodiments of the present application when the processor executes the computer program instructions.

In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium. The computer-readable storage medium stores computer program instructions. When the computer program instructions are executed by a processor, the robot described in the first aspect of the embodiments of the present application is implemented. Dialogue method.

The embodiment of the application is provided with scene instances corresponding to different dialogue scenes. After the user's input information is successfully matched with the pre-stored scene instance, the query and reply of response information are performed according to the matched target scene instance. It can be seen that this embodiment limits the search range of the response information, thereby reducing the amount of data corresponding to the search range, and also simplifies the process of searching for the response information, thereby improving the efficiency of the robot conversation. Moreover, after the search process is simplified, the possibility of searching errors is reduced, thereby improving the accuracy of the conversation reply.

Description of the drawings

The features, advantages, and technical effects of exemplary embodiments of the present application will be described below with reference to the accompanying drawings.

FIG. 1 is a schematic flowchart of an embodiment of the robot dialogue method of the present application;

2 is a schematic flowchart of another embodiment of the robot dialogue method of the present application;

FIG. 3 is a schematic flowchart of another embodiment of the robot dialogue method of the present application;

FIG. 4 is a schematic structural diagram of an embodiment of the robot dialogue device of the present application;

FIG. 5 is a schematic structural diagram of another embodiment of the robot dialogue device of the present application;

Fig. 6 is a schematic diagram of the hardware structure of an embodiment of the robot dialogue device of the present application.

Detailed ways

The implementation of the present application will be described in further detail below with reference to the accompanying drawings and embodiments. The detailed description and drawings of the following embodiments are used to exemplarily illustrate the principle of the present application, but cannot be used to limit the scope of the present application, that is, the present application is not limited to the described embodiments.

In the description of this application, it should be noted that relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply these entities. Or there is any such actual relationship or sequence between operations. Moreover, the terms "include", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes those that are not explicitly listed Other elements of, or also include elements inherent to this process, method, article or equipment. If there are no more restrictions, the element defined by the sentence "including..." does not exclude the existence of other same elements in the process, method, article, or equipment that includes the element.

Robots with conversational functions can significantly improve the efficiency of data center operations. In recent years, robots with conversational functions have begun to be used in application scenarios such as intelligent customer service and Chatbot-based Operations (ChatOps).

The key to the realization of dialogue-enabled robotics technology is to build a knowledge base that meets the needs. Two core functional components need to be designed around the knowledge base, namely, Natural Language Understanding (NLU) and Natural Language Expression (Natural Language Generation, NLG). ). The realization algorithm model of robot dialogue in related technologies is mainly divided into two types: retrieval model and generative model:

The embodiments of the application are applied to a retrieval model, and the retrieval model requires that the knowledge base be divided into three parts: matching rules, question and answer knowledge base, and response template. After obtaining the user's input information, first analyze the user's intentions through natural language technology, and then use the matching rules of the knowledge base to match the closest question to the question library, and then give the corresponding answer according to the response template. In the matching process, natural language processing, machine learning, text mining and other technologies are used to optimize the matching method to improve the accuracy of the robot's response.

The retrieval model is more suitable for scenarios where the scope of use is clear and the dialogue intent is clear. The answer of the robot based on the retrieval model is of high quality, and there will be no grammatical errors, unclear semantics, and confusion in the dialogue response. Therefore, the robot dialogue operation oriented to the enterprise scene is mainly realized based on the retrieval model.

In order to provide a robot with a dialogue function that can be applied in an enterprise scenario, the embodiment of the present application achieves the purpose of improving the efficiency and accuracy of the dialogue by dividing specific scenes for the dialogue of the robot.

The following first introduces the robot dialogue method provided by the embodiment of the present application, and the method is applied to a robot with a dialogue function. Fig. 1 shows a schematic flowchart of an embodiment of the robot dialogue method of the present application. As shown in Figure 1, the robot dialogue method may include:

S101. Obtain user input information;

In the process of the user's dialogue with the robot, the human-computer interaction between the user and the robot may be voice interaction, or realized through a touch screen or keyboard. Of course, this application does not limit specific human-computer interaction methods.

S102: Determine a target scene instance matched by the input information;

Since in the enterprise scenario, the interaction between the user and the robot should have a clear intention to use, a similar conversational intention of the user should be summarized, that is, a scenario instance of the robot. The possible input intentions of the user can be divided based on the scene through the scene instance, so that it is convenient to provide a personalized response for each type of scene in the subsequent, and can make the robot have better responsiveness, and can adapt to the needs of complex and changeable application scenarios. . Moreover, due to the independence between the scene instances, this design ensures that the new scene instance will not cause query interference to the existing scene instance after the addition of the new scene instance, and realizes the rapid expansion of the robot dialogue range.

S103: Query the response information of the input information in the target scene instance in the preset resource warehouse;

S104. Display response information.

Wherein, the display method here may be display on the screen, or voice broadcast, or display through a user terminal connected to the robot, etc. The application does not limit the specific display method.

The embodiment of the present application is provided with scene instances corresponding to different dialogue scenes. After the user's input information is successfully matched with the pre-stored scene instance, the query and display of response information are performed according to the matched template scene instance. It can be seen that this embodiment limits the search range of the response information, thereby reducing the amount of data corresponding to the search range, and also simplifies the search process of the response information, thereby improving the efficiency of the robot conversation. Moreover, after the search process is simplified, the possibility of searching errors is reduced, thereby improving the accuracy of the conversation reply.

On the basis of the foregoing embodiments, the present application also provides other embodiments, as shown in FIG. 2, which shows a schematic flowchart of another embodiment of the robot dialogue method of the present application. In S102 of these embodiments, the method for determining the target scene instance matched by the input information may include:

S102a. Judge whether there is a scene instance of the most recent match, if so, go to S102b; if not, go to S102c;

S102b: Determine whether the input information is valid input in the last matched scene instance, if so, use the last matched scene instance as the target scene instance; go to S103;

S102c: Among the scene instances included in the preset scene library, determine the target scene instance matching the input information, and save the target scene instance as the most recently matched scene instance, and then proceed to S103.

In this implementation, after each successful matching of the scene instance, the successfully matched scene instance will be saved, for example, it can be saved through a cache, so that after the next user input information, if the user's input information matches the most recently matched scene instance , It is no longer necessary to query and match the scene instance, but can directly query the response information under the most recently matched scene instance, that is, the embodiment of the present application can realize the state maintenance of multiple rounds of conversations and reduce the number of conversation query scene instances. The amount of data at the same time improves the efficiency of session processing. In addition, through this embodiment, the user can switch between scene instances and realize multiple rounds of dialogue between the user and the robot.

Among them, the above valid input refers to that the input information meets the input requirements of the most recently matched scene instance. Specifically, the process of judging whether the input information is valid input can be: judging whether the input information meets the preset input rules of the last matched scene instance; or it can also be: matching the input information with the last matched scene instance If it matches, it means it is a valid input.

Further, in some embodiments, the method of saving the successfully matched scene instance may be: using the successfully matched scene instance to update the previously saved scene instance of the most recent match. In this way, at most, only one matching scene instance, that is, one target scene instance, can be saved at the same time. Of course, in other embodiments, the update may not be performed, but the scene instance that was successfully matched this time is directly saved separately. In this case, there may be a situation in which multiple matched scene instances are saved at the same time. After each input information is received, it is only necessary to determine whether the input information is valid input in the last saved scene instance. The specific method used above is not limited in this application.

Based on the foregoing embodiments, in other embodiments of the present application, S102 may further include:

In the case that the input information is invalid in the scene instance of the last match, among the scene instances contained in the preset scene library, determine the target scene instance that matches the input information, and save the target scene instance as the most recent Examples of matching scenarios

Based on the foregoing embodiment, in S102c of some other embodiments of the present application, among the scene instances included in the preset scene library, the method of determining the target scene instance matching the input information specifically includes: The preset matching order is sequentially matched with the scene instances contained in the scene library. If the matching is successful, the matching operation is stopped. The preset matching order is: first match the professional scene instance, then match the retrieval scene instance, and finally match the session Scene instance.

The specific process of the above matching process is:

In each professional scene instance contained in the scene library, determine whether there is a target scene instance matching the input information;

If there is a target scene instance matching the input information, proceed to S103, if there is no target scene instance matching the input information, it is determined whether the retrieval scene instance contained in the scene library is a target scene instance matching the input information;

If the retrieval scene instance is a target scene instance, then proceed to S103; if the retrieval scene instance is not a target scene instance, it is determined whether the conversation scene instance contained in the scene library is a target scene instance matching the input information.

Wherein, the matching between the input information and the scene instance in this embodiment refers to the complete match between the key text feature in the input information and the text feature of the scene instance. That is, the matching process is as follows: extract the key text features in the input information; compare the key text features with the text features of the scene instance currently undergoing the matching process, if the two are exactly the same, the matching is successful; otherwise, the matching fails.

In the process of a conversation between a robot oriented to an enterprise scene and a user, the conversation can be divided into three types: professional conversation, retrieval conversation, and conversation scenario conversation. Among them, the search type dialogue includes some queries for professional knowledge, such as Frequently Asked Questions (FAQ) and so on. Conversational conversations include natural communication conversations, such as daily greetings. The content of professional type conversations includes other conversations besides the above two, mainly some content related to the enterprise, such as the position of the company employee, the employee's business scope, the company's business scope, and the company's product list. Since the professional scene has a wide range of more complex conversational requirements, the professional scene includes multiple professional scene instances, while the retrieval scene and the conversation scene only include one scene instance.

Since the scope of professional conversations is the widest, for the purpose of improving efficiency, it is first necessary to determine whether the user's input information meets the professional scene instance. If it does not meet the professional scene instance, then judge whether the input information meets the retrieval scene instance. Retrieve the scene instance, and then judge whether the input information meets the conversation scene instance.

In addition, in some embodiments of the present application, all the scene instances in the scene library may be formed into a work sequence in a certain order to facilitate the traversal of the scene instances during the scene process. Among them, the priority order of the scene instances is the professional scene instance, the retrieval scene instance, and the conversation scene instance. Multiple professional scene instances can be sorted according to the preset weight and creation time of the scene instance. Of course, this application does not limit this. This method can facilitate the traversal process of scene instances and find matching scene instances as soon as possible.

It should be noted that in some embodiments of the present application, the robot can distinguish different users, and can simultaneously establish multiple conversation flows for different users. Among them, the way to distinguish different users may be voiceprint recognition, or may also be distinguished according to the user's login account, which is not limited in this application. However, in the above manner, for the conversation of the same user, at most one matching professional scene instance is saved, that is, for the same user, only one conversation process can be established at the same time. If the robot cannot distinguish between different users, there can only be one conversation flow at the same time, and the input information obtained by the robot can come from different users.

In some cases, the user’s input information may not be entered strictly in accordance with the input rules or input fields. Therefore, it cannot be completely matched with the professional scene instance. However, because the user’s input information actually belongs to the scope of the professional scene instance, even If it is matched with the retrieval scene instance or the conversation scene instance, the matching will not succeed, which will lead to the failure of the response and reduce the usability of the robot.

In order to solve the above technical problems, in other embodiments of the present application, in the case that there is no target scene instance matching the input information in the professional scene instance, before matching the input information with the retrieval scene instance, the method may also include :

Determine whether there is a recommended professional scene instance that can be used as a target scene instance in the scene library to determine whether there is a recommended professional scene instance in the scene library;

In the case where there is a recommended professional scene instance, the recommended professional scene instance is taken as the target scene instance.

In this implementation, when there is no matching professional scene instance, the robot recommends the professional scene instance suitable for the user's input information, so that when the user's input information is not rigorous enough, the appropriate professional scene can be selected as much as possible The instance responds to it to improve the accuracy of the robot conversation.

Further, in some embodiments of the present application, the process of judging whether there is a recommended professional scene instance in the scene library may be:

Extract key text features in the input information;

Determining the degree of matching between the key text feature and the text feature of each professional scene instance contained in the scene library;

The professional scene instance whose matching degree reaches the preset matching degree threshold is taken as the recommended professional scene instance.

In this embodiment, the professional scene instance where the matching degree of the text feature and the key text feature in the input information exceeds the preset matching degree threshold is recommended, because the meaning of the input information is basically determined by the key text in the input information. The characteristics are determined, therefore, the professional scene instance that can reach a certain degree of matching with the key text characteristics is likely to be the content that the user wants to query. Through the above method, it is possible to find a suitable professional scene instance for recommendation as much as possible, thereby improving the accuracy of the robot's response.

Based on the aforementioned embodiment, since the recommended professional scene instance is determined according to whether the user's input information is similar to the professional scene instance, but because the user's input information is not accurate, there may be multiple recommended professions found at the same time. In the case of the scene instance, and the recommended professional scene instance may also be different from the user's needs, if the recommended professional scene instance is directly used by default, it may cause the wrong response message to be returned to the user.

Therefore, in order to solve the above-mentioned technical problems, in some embodiments of the present application, when there are recommended professional scene examples, the method further includes:

Show the prompt information about whether to enter the recommended professional scene instance;

In the case of receiving the entry instruction input by the user, the recommended professional scene instance is taken as the target scene instance.

In these embodiments, by prompting the recommended professional scene instances, in the case where there are multiple recommended professional scene instances, the user can select one of them as the recommended professional scene instance according to their actual needs. In the case that there is only one recommended professional scene instance, the user can also determine whether the recommended professional scene instance meets his own needs. Therefore, in these embodiments, the recommended professional scene instance is not directly used, but the user selects the professional scene instance according to his own needs. The user experience is better and the reliability of the robot dialogue is higher.

In addition, in some embodiments of the present application, when the input information is invalid input in the last matched scene instance, it is not necessary to sequentially match the input information with the professional scene instance, but directly execute the above-mentioned selection recommendation. Operation of professional scene examples. This method is because in the continuous conversation process, if the scene instance to which the current conversation belongs changes, you can basically determine the matching scene instance by directly searching for the recommended professional scene instance, retrieving the scene instance, and the conversation scene instance, thereby reducing the scene instance. Matching time, improve the efficiency of dialogue.

In some embodiments of the present application, each pre-stored scenario instance includes a preset execution rule, and the execution rule includes each state node corresponding to the user's input information and the execution operation corresponding to each state node. These state nodes can include any of the following combinations: input information enters the scene instance for the first time, waiting for input information, input information is invalid, input information is valid, the scene instance exits due to unresponsiveness beyond a preset time, and the scene implementation is actively ended. Of course, other state nodes can also be included. The execution rules corresponding to different scene instances can be the same or different. The execution rules corresponding to specific scene instances can be determined according to actual conditions, such as personalized design according to user intentions. Not limited.

In the enterprise scenario, many information is set with query permissions, and only authorized users can query. Based on this, in some other embodiments of the present application, the robot conversation method may also include:

In the case where the input information is an instance of entering the target scene for the first time, extract the user authority corresponding to the input information;

According to the user's authority, determine whether it is allowed to query the response information of the input information in the target scenario instance through the resource warehouse.

In these embodiments, the input information entering the target scene instance for the first time refers to: the current target scene instance is obtained by matching the current input information, rather than the last matched scene instance, that is, after the target scene instance is successfully matched, The response message has not been queried in this scenario instance. Since each scenario instance corresponds to part of the information in the resource warehouse, this part of information may not be inquired by everyone. Therefore, for the scenario instance corresponding to this part of the information that requires permission to query, enter the target for the first time after entering the information After the scenario instance, you need to first determine the user authority corresponding to the current input information. If the user has the authority, he can find the corresponding response information from the resource warehouse and reply. This method can meet the permission setting requirements of different information in the enterprise scenario, and enrich the scope of the robot's dialogue scenario.

Wherein, the method for extracting the user authority corresponding to the input information may be: obtaining the user identifier carried in the input information, and querying the user authority corresponding to the user identifier in the pre-stored authority list. Among them, the user ID can be a user name or account number, etc., and the acquisition method can be to receive the user ID input by the user through a touch screen or keyboard, or it can also be a comparison of the voiceprint information input by the voice to obtain the user ID. Not limited.

After S104 in some embodiments of the present application, the robot conversation method further includes:

The saved scene event status is updated according to the response information, and the scene event status includes the key features of the response information. Among them, the control logic for the state of the scene event can be implemented by methods such as a finite state machine or a decision tree.

In part of the conversation, there is continuity between the previous and subsequent conversations. For example, if the user's input information is "query business type", the robot's response information is "type 1: business A; type 2: business B; please enter the query serial number", the subsequent user input will be based on the aforementioned robot's response Message, reply "1" or "2". In this case, in order to ensure that the robot's response information can respond based on continuity, it is necessary to save the key features of the previous response information. For example, you can save the information of "1=Business A, 2=Business B", so that after continuing to receive user input information, you can query the response information according to the input information and the key features saved before, realizing multiple rounds of conversations The purpose of coherence.

Based on the foregoing embodiment, in step S103 of some embodiments of the present application, the process of querying the input information in the preset resource warehouse for response information in the target scenario instance may include:

Determine the information to be queried according to the input information and the saved scene event status;

Query the response information of the information to be queried in the target scenario instance in the resource warehouse.

Among them, the information to be queried refers to the information that needs to be queried determined according to the state of the scene event saved in the previous session and the input information in the current round of the session. For example, if the status of the saved scene event is "1=Business A, 2=Business B", and the input information is "1", the information to be queried is the content of Business A. In this embodiment, the response information is determined by the saved scene event state and input information, so that the robot can conduct a continuous dialogue with the user, and meet the purpose of the user's multiple rounds of dialogue.

According to the foregoing embodiments, in some embodiments of the present application, the identification information of the target scene instance and the status of the scene event are saved, in order to achieve the continuous execution of multiple rounds of conversations and reduce the number of repeated matching scene instances. However, due to the robot There will be conversations with different users. If the conversation information of the previous user is always saved, it will not only take up a large storage space, but also affect the conversation processing of subsequent users.

Based on this, in some other embodiments of the present application, the robot conversation method further includes: deleting the target scene instance corresponding to the case that the user’s input information is not received within a preset time from the moment the scene event state is updated. All the saved information of the, the saved information includes the scene event status and the identification information of the target scene instance.

In this embodiment, the time is started at the time when the state of the scene event is updated each time (that is, the time when a dialogue is completed). If the user's input information is not received within the preset time, it is likely to indicate that the user has stopped the dialogue with the robot Therefore, at this time, the related information about the previous user dialogue (that is, the saved scene event state and the identification information of the target scene instance) can be deleted. This method can reduce the storage space occupied by excessive invalid saved information during the dialogue between the robot and the user, and reduce the influence between the dialogues of different users.

In addition, because the boundaries of the existing robot’s dialogue capabilities are limited by the knowledge base, the traditional robot knowledge base is constructed based on combed knowledge, and cannot call the Application Programming Interface (API) interface and data source of the external platform. , The robot is not flexible enough in the dialogue process, and its ability to adapt to scene changes is poor.

In order to solve the above technical problems, in the embodiment of the present application, the resource warehouse may include at least two of an offline database, a preset knowledge base, a tool script, and an open API.

Among them, some pre-edited offline data can be stored in the offline database, such as the names of the employees of the enterprise and the names of the businesses that the employees are responsible for. The preset knowledge base can include explanations of some professional terms and knowledge content related to the business of the enterprise. Tool scripts can include some corporate business-related script names, script content, and script download links. API is an application program interface. The robot can call the corresponding application level through the open API, or access the corresponding platform for data acquisition. Therefore, through the above methods, the functional boundaries of traditional task-based robots are broken, the data sources of the resource warehouse are greatly expanded, and the flexibility of the robot in the dialogue process and the expansion ability to adapt to scene changes are improved.

Based on the foregoing embodiments, the following is a robot dialogue flow in a specific embodiment of the application. Fig. 3 shows a schematic flowchart of another embodiment of the robot dialogue method of the present application. The dialogue flow between the user and the robot can include:

S201. Obtain user input information;

S202: Judge whether there is a scene instance of the most recent match saved, if there is, go to S203; if not, go to S204;

S203: Judge whether the input information is valid input in the last matched scene instance, if it is valid input, go to S208; if it is invalid input, go to S204;

S204. In each professional scene instance contained in the scene library, determine whether there is a target scene instance matching the input information, if it exists, go to S208; if it does not exist, go to S205;

S205: Determine whether there is a recommended professional scene instance that can be used as a target scene instance in the scene library, if it exists, go to S208; if it does not exist, go to S206;

S206: Determine whether the retrieval scene instance contained in the scene library is a target scene instance matching the input information, if yes, go to S208; if not, go to S207;

S207: Determine whether the conversation scene instance included in the scene library is a target scene instance matching the input information, and if the matching is successful, enter S208;

S208. Save the target scene instance as the most recently matched scene instance, and query the input information in the resource warehouse for response information under the target scene instance;

S209. Display response information, and update the saved scene event state according to the response information.

To sum up, the embodiments of the present application are provided with scene instances corresponding to different dialogue scenes. After the user's input information is successfully matched with the scene instances in the preset scene library, the response information is processed according to the target scene instance obtained by the match. Inquiries and responses. It can be seen that this embodiment limits the search range of the response information, thereby reducing the amount of data corresponding to the search range, and simplifies the process of searching for the response information, thereby improving the efficiency of the robot conversation. Moreover, after the search process is simplified, the possibility of searching errors is reduced, thereby improving the accuracy of the conversation reply. In addition, the embodiment of the present application saves the most recently matched scene instance, so that it may no longer be necessary to match the scene instance after receiving the user's input information next time, thereby further improving the efficiency of the robot conversation. And by setting recommended professional scene instances, it is possible to reduce the situation where matching scene instances cannot be found, and to improve the reliability of the robot conversation.

To facilitate understanding, the following is the content of the dialogue between the user and the robot provided in a specific embodiment of this application. Assuming that the knowledge base of a robot (for example, Chatbot) contains natural conversational materials and professional knowledge related to operations, the robot has two professional scene examples: querying the system and querying operation and maintenance events.

User: hello

Chatbot: Welcome to use the operating robot, what can I do for you?

User: Zhang San

Chatbot: Zhang San, Process and Quality Assurance Room of Information Center, Tel: 021-38929999-6100, Email: zhangsan@unionpay.com

User: Zhang San's system

Chatbot: Excuse me, are you inquiring about the business system that Zhang San is responsible for?

User: Yes

Chatbot: The systems that Zhang San is responsible for include: small and micro merchant service platforms, and QRC, a two-dimensional code system.

User: How is the ICBC incident handled by Zhang San?

Chatbot: Do you want to switch to the "Operation and Maintenance Event Query" window?

User: Yes

Chatbot: Checking the permission is passed, it is checking for you...

Chatbot: There are 2 events that meet the conditions of handler = Zhang San and associated ICBC in this week, please enter the query serial number:

[1] INC000107-0113:39 dealt with the transaction fluctuation of ICBC Guangdong Branch, the response was 14

[2] INC00207-0108:00 has been closed ICBC debit transaction abnormalities occur, no transaction sent

User: 1

Chatbot: INC000107-0113:39 to deal with the transaction wave of ICBC Guangdong Branch. The cause of the event: monitoring rules, the incident handler should use Zhang San from one room, with a complete emergency handling manual, 8 similar incidents in the past three months, incident risk Assessment level: slight.

[The user does not input in the follow-up, one minute later]

Chatbot: Exited the operation and maintenance event query window. I can continue to serve you.

Among them, the specific realization process of the above dialogue is:

[User input: hello]

After the robot obtains the user input information "Hello", it checks that there is no currently saved scene instance of the most recent match. It traverses the work sequence of matching professional scene instances without success, there is no recommended professional scene instance, and the search scene instance is not matched. Finally, Call the conversation scenario instance to process the input information, the input information matches the natural conversation "Hello" in the conversation scenario instance successfully, and a response message of "Hello" is generated, "Welcome to the operation robot, what can I do for you?" Reply.

【User input: Zhang San】

After the robot obtains the user input information "Zhang San", it checks that there is no currently saved scene instance of the most recent match, the work sequence of traversing the matching professional scene instance is unsuccessful, and there is no recommended professional scene instance, and the scene instance matches the input information. Success, in line with the text characteristics of the address book in the retrieval scenario instance. The retrieval scenario instance obtains the address book information of "Zhang San" by querying the address book, and generates response information to return to the user.

[User input: Zhang San’s system]

After the robot obtains the user input information "Zhang San's system", it checks that there is no currently saved scene instance of the most recent match, and the work sequence that traverses the matching professional scene instance is unsuccessful. The "Zhang San's system" input by the user partially matches the "query" The text feature of "System" uses "Inquiry System" as an example of a recommended professional scene. A prompt "Is this to inquire about the business system that Zhang San is responsible for operation and maintenance?" is returned to the user.

[User input: yes]

After the robot obtains the user input information "Yes", it checks the current recommended professional scene instance, confirms the user's intention to enter the scene instance according to the user input information, updates the saved target scene instance to "query system" and deletes the professional scene recommendation In the state, the scene instance of the "query system" is called to process the user's input information.

The "Query System" professional scene example extracts the key text feature "Operation and Maintenance System Person in Charge = Zhang San" according to the user input information "Zhang San's system" saved in the scene event state, and queries the knowledge base to obtain the query result "Zhang San" Responsible systems include: small and micro merchant service platform, QR code system QRC.” According to the query results, response messages are generated and returned to users. At the same time, the scene event status of the professional scene instance of the "query system" is updated as: "the person in charge of historical query operation and maintenance = Zhang San".

[User input: How is the ICBC incident handled by Zhang San?]

After the robot obtains the user's input information "How is the ICBC incident handled by Zhang San", it checks the dialogue status. The currently saved professional scene instance is the "query system", and calls the professional scene instance to process the user's input information.

The "query system" professional scene instance judges the input information to be invalid according to the user's input information, and then checks whether there is a recommended professional scene instance according to the input information "How is the ICBC incident handled by Zhang San" (this embodiment is based on the aforementioned In the case that the input information is invalid in the matched scene instance, directly retrieve the recommended professional scene instance scheme), and find that the text characteristics of the scene instance matching the "query operation and maintenance event" match the recommended conditions, and generate a prompt "May I ask" Do you want to switch to the "Operation and Maintenance Event Query" window?" Return to the user.

[User input: yes]

After the robot obtains the user input information "Yes", it checks that there are currently recommended professional scene instances, confirms the user's intention to enter the scene instance based on the user input information, updates the saved target scene instance to "query operation and maintenance events" and deletes the profession In the recommended state of the scene, the scene instance of the "query operation and maintenance event" is called to process the user's input information.

The "query operation and maintenance event" professional scenario example first analyzes the state node of the user's input information that meets the "first entry" state, and invokes the execution operation of the state node to extract the user authority corresponding to the input information. After the user permission check is passed, a prompt "Check permission passed, querying for you..." is generated and returned to the user.

At the same time, the follow-up query function is called to query: "Query operation and maintenance events" professional scene example according to the input information "ICBC event handled by Zhang San", extract the key text feature "event label = ICBC, event handler = Zhang San", and Query the external database to obtain the query results of the two events, and generate response information based on the query results to return to the user.

At the same time, update the robot's scene event status "event to be confirmed by the user, 1=INC001, 2=INC002".

【User input: 1】

After the robot obtains the user input information "1", it checks that there is currently a matching professional scene instance "query operation and maintenance event", and calls the professional scene instance to process the user input.

The professional scene instance of "Query Operation and Maintenance Events" calls the scene event status, and it is found that there is a user's confirmation status "Awaiting user confirmation event, 1=INC001, 2=INC002". According to the user's input information "1", the information to be queried "INC001" is obtained ". Query an external database to obtain the summary information of the INC001 event, generate a response message and return it to the user.

At the same time, the scene event status is updated to "query time = INC001".

[The user does not input in the follow-up, one minute later]

The robot continuously monitors the life cycle of the matched professional scene instance saved by the user. The life cycle of the "query operation and maintenance event" is 1 minute. If the user has no subsequent input within 1 minute, the scene instance will be logged out after 1 minute, that is, delete "Query operation and maintenance event" corresponds to all the saved information, and generate a prompt "Exited the operation and maintenance event query window. I can continue to serve you" to inform the user that the currently matched professional scene instance has been cancelled.

Based on the robot dialogue method given in the foregoing embodiment, an embodiment of the present application also provides a robot dialogue device. FIG. 4 shows a schematic structural diagram of an embodiment of the robot dialogue device of the present application. The robot dialogue device 300 may include:

The scene library 320 is used to store preset scene instances;

The resource warehouse 340 is used to store various input information and corresponding response information under each scene instance; the resource warehouse 340 is a resource collection of knowledge, data, and calling functions that the robot depends on;

The control module 310 is used to obtain the user's input information; call the scene instance in the scene library 320 to determine the target scene instance matched by the input information; display the response information returned by the execution unit 330;

The execution unit 330 is used to query the response information of the input information in the target scene instance in the preset resource warehouse 340; and return the response information to the control module 310, where the execution unit 330 is a functional adapter for the robot to operate the resource warehouse 340, The resources in each resource warehouse 340, such as knowledge bases, APIs, etc., need to be adapted using a separate execution unit 330. The execution unit 330 provides operation support for the scene instance to process user input.

The embodiment of the present application is provided with scene instances corresponding to different dialogue scenes. After the user's input information is successfully matched with the scene instances in the preset scene library, the query and display of response information are performed according to the matched target scene instances. It can be seen that this embodiment limits the search range of the response information, thereby reducing the amount of data corresponding to the search range, and also simplifies the search process of the response information, thereby improving the efficiency of the robot conversation. Moreover, after the search process is simplified, the possibility of searching errors is reduced, thereby improving the accuracy of the conversation reply. Among them, the control module 310 implements matching and invocation of scene instances through the functional interface provided by the scene library 320.

In some embodiments of the present application, the scene library 320 may further include: composing all scene instances into a working sequence in a certain order, so that the control module 310 can traverse the scene instances. Among them, the priority order of the scene instances is professional scene instance, retrieval scene instance, and conversation scene instance. Multiple professional scene instances can be sorted according to the preset weight and creation time of the scene instance. Of course, this application does not limit this. This method can facilitate the traversal process of the control module 310, and find a matching scene instance as soon as possible.

In some embodiments of the present application, the scene library 320 may further include: under the control of the control module 310, the cache module 350 saves the identification information of the successfully matched scene instance.

In some other embodiments of the present application, the control module 310 further includes:

The life cycle management unit is used to control the scene library 320 to delete all the saved information corresponding to the target scene instance in the cache module 350 when the user's input information is not received within a preset time from the time the scene event state is updated, The saved information includes the status of the scene event and the identification information of the target scene instance.

In this embodiment, the time is started every time the scene event status is updated, that is, the time when a dialogue is completed. If the user's input information is not received within the preset time, it is likely that the user has stopped the dialogue with the robot. Therefore, at this time, the related information about the conversation with the previous user can be deleted, and the related information about the conversation with the previous user is the saved scene event state and the identification information of the target scene instance. This method can reduce the storage space occupied by excessive invalid saved information during the dialogue between the robot and the user, and reduce the influence between the dialogues of different users.

In addition, the scene library 320 can also be used to perform operations such as creation, deletion, and management of scene instances under the control of the control module 310.

In some embodiments of the present application, FIG. 5 shows a schematic structural diagram of another embodiment of the robot dialogue device of the present application. The robot dialogue device 300 may also include a cache module 350: for storing successfully matched scene instances. And, in these embodiments, the control module 310 may include:

The input information obtaining unit is used to obtain the input information of the user;

The scene determination unit is configured to use the most recently matched scene instance as the target when the most recently matched scene instance is stored and the input information is valid input in the most recently matched scene instance Scene instance; in the case that the most recently matched scene instance is not saved, the matching unit is triggered.

The matching unit is used to determine the target scene instance matching the input information among the scene instances included in the preset scene library, and save the target scene instance as the most recently matched scene instance.

The response unit is used to respond according to the response information returned by the execution unit 330.

In this implementation, each time the scene instance is successfully matched, the scene instance is saved, which can be specifically cached, so that after the user inputs information the next time, if the user's input information matches the saved scene instance of the most recent match, then It is no longer necessary to query and match scene instances, but can directly query response information under the most recently matched scene instance, that is, the embodiment of the present application can realize the state maintenance of multiple rounds of conversations, and reduce the time when the conversation is inquired about the scene instance. The amount of data improves the efficiency of session processing.

Based on the foregoing embodiments, in other embodiments of the present application, the scene determination unit may also be used to trigger the matching unit when the input information is an invalid input in the last matched scene instance.

Based on the foregoing embodiments, in other embodiments of the present application, the matching unit may be used to: match the input information with the scene instances contained in the preset scene library 320 in a preset matching order, and if the matching is successful Next, the successfully matched scene instance is obtained, and the above matching operation is stopped. The preset matching sequence is: first match the professional scene instance, then match the retrieval scene instance, and finally match the conversation scene instance.

Based on the foregoing embodiments, in other embodiments of the present application, the matching unit may be used to: obtain key text features in the input information; compare the key text features with the text features of the scene instance currently undergoing the matching process, if If the two are exactly the same, the match is successful; otherwise, the match fails.

That is, the control module 310 can perform preliminary processing of the input information, including decomposition, filtering, and extraction of key text features of the input information.

In a specific embodiment, the matching unit specifically includes:

The professional matching unit is used to determine whether there is a target scene instance matching the input information in each professional scene instance included in the scene library 320, if it exists, trigger the execution unit 330 if it does not exist, trigger the search matching unit;

The search matching unit is used to determine whether the search scene instance contained in the scene library 320 is a target scene instance matching the input information; if so, trigger the execution unit 330; if not, trigger the conversation matching unit;

The conversation matching unit is used to determine whether the conversation scene instance contained in the scene library is a target scene instance matching the input information, and if so, trigger the execution unit 330.

In other embodiments of the present application, the matching unit may further include a recommended matching unit, and the professional matching unit triggers the recommended matching unit when none of the matching succeeds. The recommendation matching unit may be used to determine whether there is a recommended professional scene instance in the scene library 320; if there is a recommended professional scene instance, use the recommended professional scene instance as the target scene instance.

In this implementation, when there is no matching professional scene instance, the robot recommends the professional scene instance suitable for the user's input information, so that when the user's input information is not rigorous enough, the appropriate scene can be selected as much as possible The instance responds to it to improve the accuracy of the robot conversation.

Further, the recommended matching unit of some embodiments of the present application may specifically include: extracting key text features in the input information; determining the degree of matching between the key text features and the text features of each professional scene instance contained in the scene library 320; The professional scene instance whose matching degree reaches the preset matching degree threshold is taken as the recommended professional scene instance.

In some embodiments of the present application, when there is a recommended professional scene instance, the recommended matching unit may further include:

The prompt unit is used to display the prompt information of whether to enter the recommended professional scene instance; in the case of receiving the entry instruction input by the user, the recommended professional scene instance is taken as the target scene instance.

In addition, in some embodiments of the present application, the scene determination unit may also be used to trigger the recommended matching unit when the input information is invalid in the scene instance of the last match. Thereby reducing the matching time of scene instances and improving the efficiency of dialogue.

In order to facilitate the specific management of the information response in the scenario instance, FIG. 5 shows a schematic structural diagram of a robot dialogue device provided by another embodiment of the present application; in some embodiments of the present application, the robot dialogue device 300 further includes:

The scene instance module 360 is used to control the execution unit 330 to obtain response information according to the user's input information and the current target scene instance.

Further, in other embodiments of the present application, the scenario instance module 360 may further include:

The authority determining unit is used to extract the user authority corresponding to the input information when the input information is entering the target scene instance for the first time; according to the user authority, determine whether to allow the execution unit 330 to query the response information of the input information in the target scene instance.

In these embodiments, the input information entering the target scene instance for the first time refers to: the current target scene instance is obtained by matching according to the current input information, rather than the last matched scene instance, that is, the current target scene instance is matched after the match is successful. , The response message has not been queried in this scenario instance. Since each scenario instance corresponds to part of the information in the resource warehouse 340, this part of information may not be inquired by everyone with permission. Therefore, for the scenario instance corresponding to this part of the information that requires permission to query, enter the information for the first time. After the target scenario instance, the user authority corresponding to the current input information needs to be determined first. If the user has the authority, the corresponding response information can be searched from the resource warehouse 340 to reply. This method can meet the permission setting requirements of different information in the enterprise scenario, and enrich the scope of the robot's dialogue scenario.

In some embodiments of the present application, the scenario instance module 360 may further include:

The state control unit is used to update the state of the scene event stored in the cache module 350 according to the response information. The state of the scene event includes the key features of the response information.

The cache module 350 can also be used to save scene state information.

Because in part of the conversation, there is continuity before and after the conversation. In this case, in order to ensure that the robot's response information can respond based on consistency, it is necessary to save the key features of the response information previously responded, so that after continuing to receive user input information, it can be based on the input information and The key features saved before are used to query the response information and achieve the goal of continuous conversations in multiple rounds.

Based on the foregoing embodiment, in the control module 310 of some embodiments of the present application, the scene instance module 360 can also be used to: determine the information to be queried according to the saved scene event state and input information; and query the information to be queried through the execution unit 330 Response information under the target scenario instance.

In some embodiments of the present application, the resource warehouse 340 may include at least two of an offline database, a preset knowledge base, a tool script, and an open API.

Among them, some pre-edited offline data can be stored in the offline database, such as the names of the employees of the enterprise and the names of the businesses that the employees are responsible for. The preset knowledge base can include explanations of some professional terms and knowledge content related to the business of the enterprise. Tool scripts can include some corporate business-related script names, script content, and script download links. API is an application program interface. The robot can call the corresponding application level through the open API, or access the corresponding platform for data acquisition. Therefore, through the above method, the functional boundary of the traditional task-based robot is broken, the data source of the resource warehouse 340 is greatly expanded, and the flexibility of the robot in the dialogue process and the expansion ability to adapt to scene changes are improved.

The robot dialogue method and device in the embodiments of the present application can be implemented by a robot dialogue device. Fig. 6 shows a schematic diagram of the hardware structure of an embodiment of the robot dialogue device of the present application.

As shown in FIG. 6, the robot dialogue device 400 includes an input device 401, an input interface 402, a central processing unit 403, a memory 404, an output interface 405, and an output device 406. The input interface 402, the central processing unit 403, the memory 404, and the output interface 405 are connected to each other through the bus 410, and the input device 401 and the output device 406 are respectively connected to the bus 410 through the input interface 402 and the output interface 405, and then communicate with the robot device 400 other components are connected.

Specifically, the input device 401 receives input information from the outside, and transmits the input information to the central processing unit 403 through the input interface 402; the central processing unit 403 processes the input information based on the computer executable instructions stored in the memory 404 to generate output Information, the output information is temporarily or permanently stored in the memory 404, and then the output information is transmitted to the output device 406 through the output interface 405; the output device 406 outputs the output information to the outside of the robot dialogue device 400 for the user to use.

That is to say, the robot dialogue device shown in FIG. 6 can also be implemented as including: a memory storing computer-executable instructions; and a processor, which can implement what is described in the embodiments of the present application when executing the computer-executable instructions Robot dialogue method and device.

The embodiments of the present application also provide a computer-readable storage medium on which computer program instructions are stored; when the computer program instructions are executed by a processor, the robot dialogue method provided in the embodiments of the present application is implemented. Examples of computer-readable storage media may be non-transitory computer-readable storage media, including ROM, RAM, magnetic disks, or optical disks.

The functional blocks shown in the above structural block diagram can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, it can be, for example, an electronic circuit, an application specific integrated circuit (ASIC), appropriate firmware, a plug-in, a function card, and so on. When implemented in software, the elements of this application are programs or code segments used to perform required tasks. The program or code segment may be stored in a machine-readable medium, or transmitted on a transmission medium or a communication link through a data signal carried in a carrier wave. "Machine-readable medium" may include any medium that can store or transmit information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, and so on. The code segment can be downloaded via a computer network such as the Internet, an intranet, and so on.

It should also be noted that the exemplary embodiments mentioned in this application describe some methods or systems based on a series of steps or devices. However, the present application is not limited to the order of the above steps, that is, the steps may be performed in the order mentioned in the embodiments, or may be different from the order in the embodiments, or several steps may be performed at the same time.

The above are only specific implementations of the present application, and those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working process of the above-described system, module and unit can be referred to in the foregoing method embodiment. The corresponding process will not be repeated here. It should be understood that the scope of protection of this application is not limited to this, and any person skilled in the art can easily think of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should cover Within the scope of protection of this application.

Claims

A robot dialogue method, including:

Obtain user input information;

Determine the target scene instance matched by the input information;

Query the response information of the input information in the target scene instance in the preset resource warehouse;

Show the response information.
The method according to claim 1, wherein the determining the target scene instance matched by the input information comprises:

In a case where the most recently matched scene instance is stored and the input information is valid input in the most recently matched scene instance, the most recently matched scene instance is used as the target scene instance;

In the case that the most recently matched scene instance is not saved, among the scene instances contained in the preset scene library, determine the target scene instance that matches the input information, and save the target scene instance as the The most recently matched scene instance.
The method according to claim 2, wherein the determining a target scene instance matching the input information among the scene instances included in the preset scene library comprises:

The input information is sequentially matched with the scene instances contained in the scene library according to a preset matching order, and if the matching is successful, the matching operation is stopped, wherein the preset matching order is: first match the professional scene instance , After matching to retrieve the scene instance, and finally matching the conversation scene instance.
The method according to claim 3, wherein, in the case where there is no target scene instance matching the input information in the professional scene instance, before matching the input information with the retrieval scene instance, Also includes:

Judging whether there is a recommended professional scene instance in the scene library;

In the case where the recommended professional scene instance exists, the recommended professional scene instance is used as the target scene instance.
The method according to claim 4, wherein said determining whether there is a recommended professional scene instance in the scene library comprises:

Extract key text features in the input information;

Determining the degree of matching between the key text feature and the text feature of each professional scene instance contained in the scene library;

The professional scene instance whose matching degree reaches the preset matching degree threshold is taken as the recommended professional scene instance.
The method according to claim 5, wherein, when the recommended professional scene instance exists, the method further comprises:

Show the prompt information of whether to enter the recommended professional scene instance;

In the case of receiving the entry instruction input by the user, the recommended professional scene instance is used as the target scene instance.
The method according to any one of claims 1-6, further comprising:

In a case where the input information is entering the target scene instance for the first time, extract the user authority corresponding to the input information;

According to the user authority, it is determined whether it is allowed to query the response information of the input information in the target scene instance through the resource warehouse.
The method according to any one of claims 1-6, wherein after the displaying the response information, the method further comprises:

The saved scene event state is updated according to the response information, and the scene event state includes the key feature of the response information.
The method according to claim 8, wherein the querying the input information in the preset resource warehouse for response information in the target scene instance comprises:

Determine the information to be queried according to the input information and the saved state of the scene event;

Query the response information of the information to be queried in the target scene instance in the resource warehouse.
The method according to claim 8, further comprising:

In the case that the user's input information is not received within the preset time from the time the scene event state is updated, delete all the saved information corresponding to the target scene instance, and the saved information includes the scene event state and The identification information of the target scene instance.
The method according to claim 1, wherein the resource warehouse includes at least two of an offline database, a preset knowledge base, a tool script, and an open application program calling interface (API).
A robot dialogue device includes:

Scene library, used to save preset scene instances;

Resource warehouse, used to store the input information contained in each scenario instance and its corresponding response information;

The control module is used to obtain the user's input information; call the scene instance in the scene library to determine the target scene instance matched by the input information; display the response information returned by the execution unit;

The execution unit is configured to query response information of the input information in the target scene instance in the resource warehouse; and return the response information to the control module.
A robot dialogue device, comprising: a processor and a memory storing computer program instructions, the processor implements the robot dialogue method according to any one of claims 1-11 when the processor executes the computer program instructions.
A computer-readable storage medium having computer program instructions stored thereon, and when the computer program instructions are executed by a processor, the robot dialogue method according to any one of claims 1-11 is realized.