CN112199477B

CN112199477B - Dialogue management scheme and dialogue management corpus construction method

Info

Publication number: CN112199477B
Application number: CN202010923864.6A
Authority: CN
Inventors: 鄂海红; 宋美娜; 韦帅丽; 李峻迪
Original assignee: Beijing University of Posts and Telecommunications
Current assignee: Beijing University of Posts and Telecommunications
Priority date: 2020-09-04
Filing date: 2020-09-04
Publication date: 2023-03-14
Anticipated expiration: 2040-09-04
Also published as: CN112199477A

Abstract

The application discloses a method for constructing a dialogue management scheme and a dialogue management corpus, wherein the method comprises the following steps: constructing a rule-based dialog flow tree according to dialog scenes, definition intents, word slots and/or actions; traversing the dialogue flow tree, and converting the dialogue flow tree into a format of dialogue management training corpora to construct a dialogue management model adaptive to the dialogue flow tree; in the development stage, the interaction with a user simulator or an artificial intelligence trainer is carried out, after the human-computer conversation experience achieves a preset effect, real interaction data of a user are collected through online operation, the feedback of the user is introduced, and a conversation management model is continuously and iteratively enhanced to realize conversation management. Compared with a pure skill development mode in the related technology, the development mode of the conversation flow has the characteristics of clearer design thought, more visual expression, more efficient development and the like, so that the conversation management scheme and the construction method of the conversation management corpus can be used as a rule for integrating expert knowledge and can solve the problem of difficult generation or construction of the conversation management corpus.

Description

Dialogue management scheme and dialogue management corpus construction method

Technical Field

The application relates to the technical field of man-machine conversation and natural language understanding, in particular to a conversation management scheme and a construction method of conversation management corpora.

Background

From the beginning of artificial intelligence research, people are dedicated to developing intelligent man-machine conversation systems to serve people and replace a part of workers. The human-computer conversation is widely applied to various fields and applications, and particularly task-oriented task-based human-computer conversation systems are included in personal assistants such as siri of apple, cortana of microsoft, baidu Mi and Ali honey, wearable equipment, intelligent home, intelligent customer service and the like. The high-quality task type conversation system can provide services for people quickly, efficiently and uninterruptedly, so that the service experience is improved, and the enterprise is helped to improve the service quality and effectively reduce the labor cost.

The first generation of dialogue system is mainly based on rules, and adopts a way of compiling rules into manual combing business logic to carry out dialogue matching and management. As shown in fig. 1, a series of conditional relations need to be written and logically combined, and the rule-based method has the advantages of transparent internal logic and easy analysis and debugging. However, the method is highly dependent on the artificial knowledge of experts, and has the advantages of large workload, low coverage rate, and poor flexibility and expansibility. Meanwhile, a rule-based man-machine conversation system is completely dominated by the system, the system question and the limited set of paths answered by a user are difficult to avoid, and when a conversation path deviating from a business pre-designed is adopted, the conversation system cannot flexibly process the deviation, and the service experience is poor.

With the development of artificial intelligence, machine learning and deep learning are applied to a human-machine interactive system. The structure of the task-based dialog system based on the model is shown in fig. 2 and mainly comprises three parts: semantic Understanding (NLU), dialog Management (DM), and Language Generation (NLG). Semantic understanding (NLU): the method has the main functions of processing sentences input by a user or speech recognition results and extracting the dialogue intention of the user and the information word slot transmitted by the user. Dialog Management (DM): the Dialog management is divided into two sub-modules, dialog State Tracking (DST) and Dialog Policy Learning (DPL), and its main function is to update the State of the system according to the result of NLU and generate corresponding system action. Natural Language Generation (NLG): and the system action output by the DM is converted into text, and the system action is expressed in a text form.

The model-based dialogue system has strong generalization capability and learning capability, namely, the dialogue capability is continuously enhanced as the dialogues are accumulated. However, the model effect is highly dependent on the corpus, especially the initial corpus is insufficient, the model effect is poor, and the conversation capability is difficult to meet the minimum service requirement. This type of dialog management material is shaped as in fig. 3, and is not a natural language entered by the user, but rather a dialog flow containing intents and word slots and historical states, thus requiring manual labeling of structured data. Under the conditions of more intentions and more semantic slots, the dialogue state is various, the flow jump is extremely complex, and the situation of writing confusion is easy to occur. In addition, the model-based dialogue management has the advantages that complete black boxes are operated inside, the dialogue is not controlled, and expert knowledge cannot be reasonably utilized.

Content of application

The present invention is directed to solving, at least to some extent, one of the technical problems in the related art.

Therefore, a first objective of the present invention is to provide a dialog management scheme and a method for constructing dialog management corpus, so as to quickly implement a man-machine dialog system, and compared with a pure skill development model in the related art, a development model using dialog flow has the characteristics of clearer design thought, more intuitive expression, more efficient development, and the like, so that the dialog management corpus can be used as a rule for incorporating expert knowledge, and the problem of difficult generation or construction of the dialog management corpus can be solved.

A second object of the present invention is to provide an electronic device.

A third object of the invention is to propose a computer-readable storage medium.

In order to achieve the above object, an embodiment of the first aspect of the present application provides a dialog management scheme and a dialog management corpus construction method, including the following steps:

constructing a rule-based dialog flow tree according to dialog scenes, definition intents, word slots and/or actions;

traversing the dialogue flow tree, and converting the dialogue flow tree into a format of dialogue management training corpora to construct a dialogue management model adaptive to the dialogue flow tree;

in the development stage, the interaction with a user simulator or an artificial intelligence trainer is carried out, after the human-computer conversation experience achieves a preset effect, real interaction data of a user are collected through online operation, the feedback of the user is introduced, and the conversation management model is continuously and iteratively enhanced to realize conversation management.

In addition, the dialog management scheme and the dialog management corpus construction method according to the above embodiment of the present invention may further have the following additional technical features:

optionally, the nodes of the dialog flow tree include at least one intent node, at least one word slot node, at least one action node, at least one entry node, and/or at least one pop-up form node.

Alternatively, the different nodes are connected down in sequence starting from the ingress node, so that each path from the ingress node is a possible conversational flow.

Optionally, wherein,

if the conversation is initiated by the user, the intention node which represents the input of the user is connected with the entrance node;

if the conversation is initiated by the conversation system, the action node which comprises the logic of initiating the conversation is connected with the entrance node.

Optionally, a developer-defined action node is connected below the intention node, wherein the action node is connected with a plurality of or single action nodes, and when the action nodes are connected with a plurality of action nodes, the logic and the operation defined by each action node are executed in sequence according to the connection order.

Optionally, wherein,

if the conversation flow is finished, the nodes do not need to be connected after the action nodes;

if the conversation flow is not finished, the intention node is connected behind the action node.

Optionally, if a plurality of word slots are mounted on an intention, when the dialog is conducted to the node of the path, a form taking the mounted word slots as items is formed, and the missing word slots to be filled are clarified in sequence according to the mounting sequence until all the word slots to be filled are clarified.

Optionally, wherein,

if the user turns to other intentions in the form filling process, performing corresponding action aiming at the new intentions, and then remaining the form for continuous clarification;

and if the form needs to be popped out and transferred to other intentions, connecting the popped-out form node under the target intention node.

To achieve the above object, an embodiment of a second aspect of the present application provides an electronic device, including: at least one processor; and a memory communicatively coupled to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being configured to perform a dialog management scheme and a dialog management corpus construction method as described in the above embodiments.

In order to achieve the above object, a third aspect of the present application provides a computer-readable storage medium storing computer instructions for causing a computer to execute the dialog management scheme and the dialog management corpus building method according to the above embodiments.

Therefore, a rule-based dialog flow tree can be constructed according to dialog scenes, definition intents, word slots and/or actions, and then the dialog rules are converted into a format of a dialog management training corpus to be used for training a model, so that a preliminary model of dialog management is obtained. And then, interactive learning is carried out on the model, the capability of the dialogue system is improved in a manual correction mode, and finally online operation is carried out, so that a man-machine dialogue system is quickly realized.

Additional aspects and advantages of the present application will be set forth in part in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the present application.

Drawings

The above and/or additional aspects and advantages of the present application will become apparent and readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings of which:

fig. 1 is a schematic diagram of a rule-based dialog in the related art;

FIG. 2 is a diagram illustrating a structure of a task based dialog system according to a related art model;

FIG. 3 is an exemplary diagram of a dialog management corpus;

FIG. 4 is a flowchart illustrating a dialog management scheme and a method for constructing dialog management corpora according to an embodiment of the present application;

FIG. 5 is a flow diagram of session management according to one embodiment of the present application;

FIG. 6 is an exemplary diagram of a ticket booking dialog flow design according to one embodiment of the present application;

fig. 7 is a schematic structural diagram of an electronic device according to an embodiment of the present application

Detailed Description

Reference will now be made in detail to the embodiments of the present application, examples of which are illustrated in the accompanying drawings, wherein like reference numerals refer to the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are illustrative and intended to explain the present application and should not be construed as limiting the present application.

A dialog management scheme and a dialog management corpus construction method according to an embodiment of the present invention will be described below with reference to the accompanying drawings.

Specifically, fig. 4 is a schematic flowchart of a dialog management scheme and a dialog management corpus construction method provided in an embodiment of the present application.

As shown in fig. 4, the method for constructing a dialogue management scheme and a dialogue management corpus includes the following steps:

in step S401, a rule-based dialog flow tree is built from dialog scenes, definition intents, word slots, and/or actions.

It is to be understood that, for the dialog flow tree of the embodiment of the present application, the smallest units that constitute a dialog flow are nodes.

Optionally, in some embodiments, the nodes of the dialog flow tree include at least one intent node, at least one word slot node, at least one action node, at least one entry node, and/or at least one pop-up form node.

It can be understood that the conversation flow is a tree structure, the process logic of human-computer interaction can be clearly depicted, the method is equivalent to the visualization of the thinking of a human-computer chat process, the basic elements forming the conversation flow are nodes, and the conversation flow tree can be quickly constructed through conversation process design and node arrangement. Wherein, the node can include: at least one intent node, at least one word slot node, at least one action node, at least one entry node, and/or at least one pop-up form node.

Specifically, the intent (intent) node: representing the intent of the wheel to speak the user input may be represented by a parallelogram.

Word slot node: the word slots are parameters needed to complete the task, and include word fill slots and word non-fill slots, which are represented by solid circles and dotted circles, respectively. When the word slots are clarified, the missing word filling slots are traced according to the sequence of the word slot arrangement.

Action (action) node: representing the actions performed by the dialog system, the internal logic of each action node may be customized. The action definition generally performs some logical processing and then returns some replies to the user, which may include operations such as judgment, query, http request, reply to the user, perform tasks, and the like. Indicated by a rectangle.

Entry node (start): the root node of the dialog flow tree is a schematic node and is represented by an ellipse. The entry nodes merely serve as identification means, i.e. identify the beginning of a dialog flow, and each dialog flow tree must have one and only one entry node.

Skip form node (stop): in the dialog flow tree structure, if a plurality of word slots are mounted on an intention node, a form is generated by default, and the mounted missing necessary word slots are clarified in sequence one by one. If the user turns to other intentions in the clarification process, the developer decides whether to continue to leave the form or jump out of the form at this time; if the form is popped out, the form node is required to be identified when the form is going to be changed to a new purpose.

It should be noted that, when designing the dialog flow structure, extra attention needs to be paid to readability and extensibility of the structure. Therefore, the dialog flow structure of the embodiments of the present application places decision logic in the action node definition. If the judgment nodes are provided separately, one judgment node is added to all places with judgment logic, and each judgment node is subjected to branch bifurcation. Thus, the whole dialog flow tree structure is huge and complicated, and readability becomes poor.

Moreover, this approach is poorly scalable, and if the decision logic is modified, the entire tree structure changes. However, if the decision logic is written in the definition of action node, no matter the decision logic is changed or other operations are added, the tree structure is not changed, and only the definition of the relevant action node needs to be adjusted. In addition, the judgment logic is written in the node definition, so that the judgment logic can be combined with an external knowledge base more conveniently and more flexibly. Therefore, the design has better readability, flexibility and expansibility.

In step S402, the dialog flow tree is traversed and converted into a dialog management corpus format to construct a dialog management model adapted to the dialog flow tree capability.

In step S403, in the development phase, after the human-computer interaction experience reaches a preset effect, the real interaction data of the user is collected through online operation, and the feedback of the user is introduced, so as to continuously and iteratively enhance the dialog management model, thereby implementing the dialog management. Wherein all dialogs with the AI assistant during interactive learning can be later derived as NLUs and dialog training examples and appended to the original training data samples, thereby continuously enhancing dialog capabilities. And after the man-machine conversation experience achieves the initial effect, the online operation can be carried out, the real interactive data of the user are collected, meanwhile, the feedback of the user is conveniently introduced through some UI designs, and the model is continuously iteratively enhanced.

In summary, in the embodiment of the present application, the rules and the models are combined for session management, a rule-based session flow tree may be constructed according to a session scene, a definition intention, a word slot, and/or an action, and then the session rules are converted into a format of a session management training corpus for training the models, so as to obtain a preliminary model of session management. And then, performing interactive learning on the model, improving the capability of a dialogue system in a manual correction mode, and finally performing online operation. And in the online service process, collecting real user interaction data and iteratively training the model.

That is to say, as shown in fig. 5, in the embodiment of the present application, an executable process is constructed by designing a dialog flow and reasonably organizing and arranging the nodes, a dialog rule can be quickly and efficiently constructed, and the dialog rule can be converted into a structured dialog management model corpus by traversing a tree, so as to quickly implement a man-machine dialog system.

In order to further understand the dialog flow tree, the construction rules of the dialog flow tree are explained in detail below.

Alternatively, in some embodiments, the different nodes are connected down in sequence starting from the ingress node, so that each path from the ingress node is a possible conversational flow.

Optionally, in some embodiments, if a dialog is initiated for the user, an intention node representing user input is connected to the entry node; if the dialog is initiated by the dialog system, an action node containing the logic of initiating the dialog is connected to the entry node.

Optionally, in some embodiments, the intention node is connected with action nodes defined by developers, wherein the action nodes are connected with a plurality of or single action nodes in series, and when the action nodes are connected with a plurality of action nodes in series, the logic and operations defined by each action node are executed in turn according to the connection order.

Optionally, in some embodiments, if the dialog flow ends, after the node is acted, the node does not need to be connected; if the conversation flow is not completed, an intention node is connected to the action node.

Optionally, in some embodiments, if multiple word slots are mounted on an intention, when the dialog proceeds to the node of the path, a form is formed with the mounted word slots as items, and the missing word slots are clarified in turn according to the mounting order until all the word slots are clarified.

Optionally, in some embodiments, if the user turns to another intention during the single fill-in process, the corresponding action is performed on the new intention, and the form is left for clarification; and if the drop-out form is required to be transferred to other intentions, connecting the drop-out form nodes under the target intention node.

Optionally, in some embodiments, it is allowed to create sub-processes, the building rules of the sub-processes are consistent with the main process, and the built sub-processes can be connected to any place of the main process as a node with an ID.

For example, taking ticket booking as an example, as shown in tables 1 and 2, table 1 is the intentions and word slots defined in the scene, and table 1 is the actions defined in the scene.

TABLE 1

TABLE 2

Specifically, as shown in fig. 6, when the user initiates a session with a street intent to the dialog system, the system performs an act _ street action. When the user is the intention to order a ticket, the system will record the word slots contained in the user input and clarify the missing word must-be-filled word slots one by one in the order of word slots. According to the arranging design of the dialog flow tree, if the user chats to chat during the clarification process, the system executes act _ chat and continues to clarify the word slot of the book _ ticket. However, if the user jumps to the query _ weather intent, the system terminates the book _ ticket form and moves to the query _ weather form.

According to the dialogue management scheme and the dialogue management corpus construction method provided by the embodiment of the application, the rule-based dialogue flow tree can be constructed according to dialogue scenes, definition intents, word slots and/or actions, the dialogue rules are converted into the format of the dialogue management corpus and used for training the model to obtain a preliminary model of dialogue management, then the model is subjected to interactive learning, the capacity of a dialogue system is improved in a manual correction mode, and finally online operation is carried out, so that the man-machine dialogue system is quickly realized.

Fig. 7 is a schematic structural diagram of an electronic device according to an embodiment of the present application. The electronic device may include:

a memory 1201, a processor 1202, and a computer program stored on the memory 1201 and executable on the processor 1202.

The processor 1202 implements the dialogue management scheme and the dialogue management corpus construction method provided in the above-described embodiments when executing the program.

Further, the electronic device further includes:

a communication interface 1203 for communication between the memory 1201 and the processor 1202.

A memory 1201 for storing computer programs executable on the processor 1202.

Memory 1201 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk memory.

If the memory 1201, the processor 1202, and the communication interface 1203 are implemented independently, the communication interface 1203, the memory 1201, and the processor 1202 may be connected to each other by a bus and perform communication with each other. The bus may be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, an Extended ISA (EISA) bus, or the like. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, only one thick line is shown in FIG. 7, but this is not intended to represent only one bus or type of bus.

Optionally, in a specific implementation, if the memory 1201, the processor 1202, and the communication interface 1203 are integrated on a chip, the memory 1201, the processor 1202, and the communication interface 1203 may complete mutual communication through an internal interface.

Processor 1202 may be a Central Processing Unit (CPU), or an Application Specific Integrated Circuit (ASIC), or one or more Integrated circuits configured to implement embodiments of the present Application.

The present embodiment also provides a computer-readable storage medium, on which a computer program is stored, wherein the program, when executed by a processor, implements the dialog management scheme and the dialog management corpus construction method as described above.

In the description herein, reference to the description of the term "one embodiment," "some embodiments," "an example," "a specific example," or "some examples," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the application. In this specification, the schematic representations of the terms used above are not necessarily intended to refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or N embodiments or examples. Moreover, various embodiments or examples and features of various embodiments or examples described in this specification can be combined and combined by one skilled in the art without being mutually inconsistent.

Furthermore, the terms "first", "second" and "first" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or to implicitly indicate the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the feature. In the description of the present application, "N" means at least two, e.g., two, three, etc., unless specifically limited otherwise.

Any process or method descriptions in flow charts or otherwise described herein may be understood as representing modules, segments, or portions of code which include one or more N executable instructions for implementing steps of a custom logic function or process, and alternate implementations are included within the scope of the preferred embodiment of the present application in which functions may be executed out of order from that shown or discussed, including substantially concurrently or in reverse order, depending on the functionality involved, as would be understood by those reasonably skilled in the art of implementing the embodiments of the present application.

The logic and/or steps represented in the flowcharts or otherwise described herein, such as an ordered listing of executable instructions that can be considered to implement logical functions, can be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, processor-containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions. For the purposes of this description, a "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of the computer-readable medium would include the following: an electrical connection (electronic device) having one or N wires, a portable computer diskette (magnetic device), a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium could even be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, via for instance optical scanning of the paper or other medium, then compiled, interpreted or otherwise processed in a suitable manner if necessary, and then stored in a computer memory.

It should be understood that portions of the present application may be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods may be implemented in software or firmware stored in a memory and executed by a suitable instruction execution system. If implemented in hardware, as in another embodiment, any one or combination of the following techniques, which are well known in the art, may be used: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application specific integrated circuit having an appropriate combinational logic gate circuit, a Programmable Gate Array (PGA), a Field Programmable Gate Array (FPGA), or the like.

It will be understood by those skilled in the art that all or part of the steps carried by the method for implementing the above embodiments may be implemented by hardware related to instructions of a program, which may be stored in a computer readable storage medium, and when the program is executed, the program includes one or a combination of the steps of the method embodiments.

In addition, functional units in the embodiments of the present application may be integrated into one processing module, or each unit may exist alone physically, or two or more units are integrated into one module. The integrated module can be realized in a hardware mode, and can also be realized in a software functional module mode. The integrated module, if implemented in the form of a software functional module and sold or used as a stand-alone product, may also be stored in a computer readable storage medium.

The storage medium mentioned above may be a read-only memory, a magnetic or optical disk, etc. While embodiments of the present application have been shown and described above, it will be understood that the above embodiments are exemplary and should not be construed as limiting the present application and that changes, modifications, substitutions and alterations in the above embodiments may be made by those of ordinary skill in the art within the scope of the present application.

Claims

1. A construction method of a dialogue management scheme and a dialogue management corpus is characterized by comprising the following steps:

constructing a rule-based dialog flow tree according to the dialog scene, the definition intention, the word slot and/or the action;

in the development stage, real interactive data of a user are collected through online operation after the human-computer conversation experience achieves a preset effect by interacting with a user simulator or an artificial intelligence trainer, and the feedback of the user is introduced to continuously and iteratively enhance the conversation management model so as to realize conversation management;

wherein the nodes of the dialog flow tree include at least one intention node, at least one word slot node, at least one action node, at least one entry node, and/or at least one pop-up form node;

the dialog flow tree structure places decision logic in the action node definition.

2. A method according to claim 1, characterized by connecting the different nodes down in turn, starting from an ingress node, so that each path from said ingress node is a possible conversational flow.

3. The method of claim 2, wherein,

if the conversation is initiated by the user, an intention node representing the input of the user is connected with the entrance node;

4. The method according to claim 3, wherein the intention node is connected with action nodes defined by a developer, wherein the action nodes are connected with a plurality of or single action nodes in succession, and when the plurality of action nodes are connected in succession, the logic and operations defined by each action node are executed in turn according to the connection order.

5. The method of claim 4, wherein,

if the conversation flow is finished, the action node is not needed to be connected;

6. The method of claim 5, wherein if multiple word slots are mounted on an intention, when the dialog proceeds to the node of the path, a form is formed with the mounted word slots as items, and the missing word slots are clarified in turn in the mounting order until all the word slots are clarified.

7. The method of claim 6, wherein,

and if the form needs to be popped out and transferred to other intentions, connecting the nodes of the popped-out form under the nodes of the target intentions.

8. An electronic device, comprising: a memory, a processor and a computer program stored on the memory and executable on the processor, the processor executing the program to implement the dialogue management scheme and the dialogue management corpus construction method according to any one of claims 1 to 7.

9. A computer-readable storage medium on which a computer program is stored, the program being executed by a processor for implementing the dialogue management scenario and the dialogue management corpus construction method according to any one of claims 1-7.