Session management methods, devices, terminal equipment, and storage media

Abstract

This invention discloses a conversation management method, apparatus, terminal device, and storage medium. The method involves acquiring text information input by a user; performing slot and intent recognition on the text information to obtain extracted slots and intents; performing conversation management on the extracted slots and intents to obtain operation instructions, wherein the conversation management includes one or more of slot confirmation, intent confirmation, and post-intent verification; parsing the operation instructions into corresponding operation content, and displaying the operation content on the front end for the user to view. This invention improves the processing capabilities of voice assistant systems.

Description

Technical Field

[0001] This invention relates to the field of voice interaction technology, and in particular to a session management method, apparatus, terminal device, and storage medium. Background Technology

[0002] With the rapid development of mobile internet and artificial intelligence technologies, collaborative office systems have gradually matured, evolving from PCs to mobile devices. However, as office scenarios become more complex and diverse, the functions within these systems are increasing, and existing intelligent voice assistant systems are insufficient in their processing capabilities when faced with tasks involving multiple rounds of information processing.

[0003] Therefore, it is necessary to propose a solution to improve the processing capabilities of voice assistant systems.

[0004] The above content is only used to help understand the technical solution of the present invention and does not represent an admission that the above content is prior art. Summary of the Invention

[0005] The main objective of this invention is to provide a conversation management method, apparatus, terminal device, and storage medium, which aims to improve the processing capabilities of a voice assistant system.

[0006] To achieve the above objectives, the present invention provides a session management method, the session management method comprising:

[0007] Obtain text information input by the user;

[0008] The text information is subjected to slot and intent recognition to obtain the extracted slots and intents;

[0009] The extracted slots and intents are subjected to session management to obtain operation instructions, wherein the session management includes one or more of slot confirmation, intent confirmation, and post-intent verification;

[0010] The operation instructions are parsed into corresponding operation content, and the operation content is displayed on the front end for the user to view.

[0011] Optionally, the step of performing slot and intent recognition on the text information to obtain the extracted slots and intents includes:

[0012] Based on a preset dictionary list, slot replacement is performed on the text information to obtain an expression list;

[0013] The template database is searched sequentially according to the list of expressions to obtain the search results;

[0014] Determine whether a directly matching input template exists in the search results;

[0015] If a directly matching input template exists in the search results, the intent corresponding to the input template is returned.

[0016] Optionally, after the step of determining whether a directly matching input template exists in the search results, the method further includes:

[0017] If no directly matching input template is found in the search results, then it is determined whether a regular expression matching the text information exists in the preset regular expression database.

[0018] If a regular expression exists in the preset regular expression database that matches the text information, then the intent corresponding to the regular expression is returned.

[0019] Optionally, before the step of performing session management on the extracted slots and intents to obtain operation instructions, the method further includes:

[0020] Determine whether the user is interacting for the first time or has completed the previous intent flow;

[0021] If the user is interacting for the first time or has completed the previous intent process, then enter the start state and identify whether the extracted intent is a single intent;

[0022] If the extracted intent is not the single intent, then all identified intent options are returned to the user to confirm the specific intent to be executed and proceed to the specific intent.

[0023] If the extracted intent is a single intent, then proceed to the specific intent corresponding to the extracted intent.

[0024] Optionally, after determining whether the user is interacting for the first time or has completed the previous intent flow step, the method further includes:

[0025] If the user is not interacting for the first time or has not completed the previous intent flow, that is, the user is in the intent flow, then it is determined whether the extracted intent is consistent with the current intent.

[0026] If the extracted intent is inconsistent with the current intent, then the intent transfer state is entered, that is, a confirmation instruction is returned asking whether the intent needs to be switched, so that the user can decide whether to switch intents.

[0027] Optionally, the step of performing session management on the extracted slots and intents to obtain operation instructions includes:

[0028] Enter the specific intent corresponding to the extracted intent, load the execution flow of the specific intent, and determine whether the necessary slots in the execution flow have been filled.

[0029] If the necessary slots in the execution process are not filled, the process enters the slot inquiry state, that is, it returns the query instruction for the missing slots to confirm the missing slots with the user and obtain the operation to be executed.

[0030] If the necessary slots in the execution process have been filled, then it is determined whether to perform a second confirmation on the necessary slots and / or the specific intent according to the configuration, and perform a post-intent verification to obtain the operation to be executed;

[0031] The operation to be performed is converted into an operation instruction.

[0032] Optionally, the step of determining whether to perform secondary confirmation of the necessary slots and / or the specific intent based on the configuration, and performing post-intent verification to obtain the operation to be executed includes:

[0033] Determine whether to perform a second confirmation of the necessary slots based on the configuration;

[0034] If a second confirmation is required for the necessary slot, a confirmation instruction corresponding to the necessary slot is returned for the user to confirm.

[0035] Determine whether to conduct a second confirmation of the specific intent based on the configuration;

[0036] If a second confirmation of the specific intent is required, the confirmation statement corresponding to the specific intent is returned, and the necessary slots are packaged for the user to perform a second confirmation to obtain a confirmation result.

[0037] Based on the confirmation result, determine whether the specific intent has a subsequent intent;

[0038] If the specific intent has a subsequent intent, then re-examine whether the necessary slots of the subsequent intent have been filled, until an intent whose necessary slots have not been filled is encountered and / or all the subsequent intents have been checked, and the operation to be executed is obtained.

[0039] If the specific intent has no subsequent intent, then the operation to be executed corresponding to the specific intent is obtained.

[0040] Optionally, before the step of parsing the operation instruction into corresponding operation content and displaying the operation content to the front end, the method further includes:

[0041] Define basic instructions so that the instruction parsing module can parse the operation instructions based on the basic instructions. The basic instructions include intent instructions and function instructions. The intent instructions are used to call external resources, and the function instructions are used to convert internal data.

[0042] Furthermore, to achieve the above objectives, the present invention also provides a session management device, the session management device comprising:

[0043] The acquisition module is used to acquire text information input by the user;

[0044] The natural language understanding module is used to perform slot and intent recognition on the text information to obtain the extracted slots and intents;

[0045] The dialogue state tracking module is used to perform dialogue management on the extracted slots and intents to obtain operation instructions. The dialogue management includes one or more of slot confirmation, intent confirmation, and post-intent verification.

[0046] The instruction parsing module is used to parse the operation instruction into corresponding operation content and display the operation content on the front end for the user to view.

[0047] In addition, to achieve the above objectives, the present invention also provides a terminal device, the terminal device including a memory, a processor, and a session management program stored in the memory and executable on the processor, wherein the session management program, when executed by the processor, implements the steps of the session management method as described above.

[0048] In addition, to achieve the above objectives, the present invention also provides a computer-readable storage medium storing a session management program, which, when executed by a processor, implements the steps of the session management method as described above.

[0049] This invention proposes a conversation management method, apparatus, terminal device, and storage medium. The method involves acquiring user-input text information; performing slot and intent recognition on the text information to obtain extracted slots and intents; performing conversation management on the extracted slots and intents to obtain operation instructions, wherein the conversation management includes one or more of slot confirmation, intent confirmation, and post-intent verification; parsing the operation instructions into corresponding operation content, and displaying the operation content on the front end for the user to view. Slot and intent recognition of the text information accurately identifies the user's operation intent and the keyword slot information contained in the input text, thus providing a data foundation for the correct execution of subsequent operations. The conversation management method provides a unified multi-turn conversation management solution for robots that is feature-rich, compatible with multiple systems, and easy to maintain, reusing the single functions of various closed systems. By parsing the operation instructions into corresponding operation content and displaying the operation content on the front end, instructions can be parsed into different forms, empowering intelligent assistants on different platform robots, enhancing the uniformity of office systems, and thus improving the processing capabilities of voice assistant systems. Attached Figure Description

[0050] Figure 1 This is a schematic diagram of the functional modules of the terminal device to which the session management device of the present invention belongs;

[0051] Figure 2 This is a flowchart illustrating an exemplary embodiment of the session management method of the present invention;

[0052] Figure 3 for Figure 2 In this embodiment, step S20 involves performing slot and intent recognition on the text information to obtain the extracted slots and intents.

[0053] Figure 4 for Figure 2 In this embodiment, step S30 performs session management on the extracted slots and intents to obtain operation instructions. The session management includes one or more of the following: slot confirmation, intent confirmation, and post-intent verification.

[0054] Figure 5 This is a schematic diagram of the process for tracking dialogue state in an embodiment of the present invention.

[0055] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0056] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0057] The main solution of this invention is as follows: First, acquire text information input by the user; second, perform slot and intent recognition on the text information to obtain extracted slots and intents; third, perform session management on the extracted slots and intents to obtain operation instructions, wherein the session management includes one or more of slot confirmation, intent confirmation, and post-intent verification; fourth, parse the operation instructions into corresponding operation content and display the operation content on the front end for the user to view. Slot and intent recognition on the text information can accurately identify the user's operation intent and the keyword slot information contained in the input text, thereby providing a data foundation for the correct execution of subsequent operations. Session management of the extracted slots and intents provides a feature-rich, multi-system compatible, and easily maintainable unified multi-turn conversation management solution for robots, reusing the single functions of various closed systems. By parsing the operation instructions into corresponding operation content and displaying the operation content on the front end, instructions can be parsed into different forms, empowering intelligent assistants on different platform robots, enhancing the uniformity of office systems, and thus improving the processing power of the voice assistant system.

[0058] Technical terms involved in the embodiments of this invention:

[0059] NLU (Nature Language Understanding): is a general term for all methods, models, or tasks that support machines in understanding text content.

[0060] Intent: The name of the action the user wants to perform;

[0061] Slots: The data information required to perform this operation.

[0062] In existing technologies, collaborative office systems have gradually matured, evolving from PCs to mobile devices. However, with the increasing complexity and diversity of office scenarios, the number of functions within these systems is constantly growing, leading to a surge in the number of systems on the platform. Faced with a large number of system applications and functions, users need to constantly switch between systems to meet specific work scenario requirements. Furthermore, there are challenges in informing users about new system functions, and users face learning costs associated with these new features. Some platforms have launched office robots to provide a unified entry point; however, these often require customized development for specific office scenarios, resulting in long development cycles for single scenarios. End-to-end approaches using artificial intelligence lack visualized and explainable processes. Additionally, existing voice office assistants often employ closed design approaches when handling tasks, making it difficult for users to switch between new scenarios. After the current scenario ends, the user's historical information from multiple rounds is not fully utilized. In summary, existing collaborative office platforms face the following problems:

[0063] 1. Users need to manually open the application within the office platform to complete the operation. Complex office scenarios may require opening multiple systems.

[0064] 2. The system's application functions are deeply complex and difficult for users to access directly; having multiple systems would impose too high a learning cost on users.

[0065] 3. The entry points provided by the same assistant require customized development and have low reusability;

[0066] 4. The assistant's ability to process multiple rounds of information is insufficient.

[0067] This invention provides a solution based on a system design centered around conversation management technology in an assistant system. This includes parsing key information from user input text, an interpretable and unified task flow, modular operation, the ability to extend and integrate new functions through standard configuration specifications, and multi-turn information processing. It primarily addresses issues in voice assistant systems such as long development cycles for customized solutions, inconsistent conversation flows leading to high maintenance costs, complex and poorly interpretable processes, and insufficient multi-turn information processing capabilities.

[0068] The main function is to implement the conversation management part of the office assistant. Based on the needs of different office scenarios, the intent and slots in the text data are extracted and multiple rounds of conversation management are completed to guide users to supplement information or complete the system functions that users want to perform.

[0069] The overall processing flow in this invention is as follows:

[0070] 1) The intent and slots of the user's input text information are identified through the NLU (Natural Language Understanding) module;

[0071] 2) Manage multi-turn conversation information by tracking the extracted intents and slots through dialogue state tracking;

[0072] 3) The command parsing module parses the operation list returned by the session management and completes the specified operation.

[0073] Compared with the prior art, the present invention has the following characteristics:

[0074] 1) Expandable intent and scene recognition, allowing users to define rules to quickly verify the feasibility of solutions;

[0075] 2) One-click configuration supports quick integration of third-party system applications with the voice assistant to achieve conversation management;

[0076] 3) Instructions are interpretable, and configurations can be transformed into visual processes to explain the task flow;

[0077] 4) Multi-turn semantic understanding provides multi-turn task processing capabilities at the system level.

[0078] Specifically, refer to Figure 1 , Figure 1 This is a schematic diagram of the functional modules of the terminal device to which the session management device of the present invention belongs. The session management device can be a device independent of the terminal device, capable of session management, and can be implemented on the terminal device in hardware or software form. The terminal device can be a smart mobile terminal with data processing capabilities, such as a mobile phone or tablet computer, or it can be a fixed terminal device or server with data processing capabilities.

[0079] In this embodiment, the terminal device to which the session management device belongs includes at least an output module 110, a processor 120, a memory 130, and a communication module 140.

[0080] The memory 130 stores the operating system and session management program. The session management device can store information such as the acquired user-input text information, the extracted slots and intents obtained by performing slot and intent recognition on the text information, the session management of the extracted slots and intents, the sent operation commands, and the operation content corresponding to the operation commands obtained by the command parsing module in the memory 130. The output module 110 may be a display screen, etc. The communication module 140 may include a WIFI module, a mobile communication module, and a Bluetooth module, etc., and communicates with external devices or servers through the communication module 140.

[0081] When the session management program in memory 130 is executed by the processor, it performs the following steps:

[0082] Obtain text information input by the user;

[0083] The text information is subjected to slot and intent recognition to obtain the extracted slots and intents;

[0084] The extracted slots and intents are subjected to session management to obtain operation instructions, wherein the session management includes one or more of slot confirmation, intent confirmation, and post-intent verification;

[0085] The operation instructions are parsed into corresponding operation content, and the operation content is displayed on the front end for the user to view.

[0086] Furthermore, when the session management program in memory 130 is executed by the processor, it also performs the following steps:

[0087] Based on a preset dictionary list, slot replacement is performed on the text information to obtain an expression list;

[0088] The template database is searched sequentially according to the list of expressions to obtain the search results;

[0089] Determine whether a directly matching input template exists in the search results;

[0090] If a directly matching input template exists in the search results, the intent corresponding to the input template is returned.

[0091] Furthermore, when the session management program in memory 130 is executed by the processor, it also performs the following steps:

[0092] If no directly matching input template is found in the search results, then it is determined whether a regular expression matching the text information exists in the preset regular expression database.

[0093] If a regular expression exists in the preset regular expression database that matches the text information, then the intent corresponding to the regular expression is returned.

[0094] Furthermore, when the session management program in memory 130 is executed by the processor, it also performs the following steps:

[0095] Determine whether the user is interacting for the first time or has completed the previous intent flow;

[0096] If the user is interacting for the first time or has completed the previous intent process, then enter the start state and identify whether the extracted intent is a single intent;

[0097] If the extracted intent is not the single intent, then all identified intent options are returned to the user to confirm the specific intent to be executed and proceed to the specific intent.

[0098] If the extracted intent is a single intent, then proceed to the specific intent corresponding to the extracted intent.

[0099] Furthermore, when the session management program in memory 130 is executed by the processor, it also performs the following steps:

[0100] If the user is not interacting for the first time or has not completed the previous intent flow, that is, the user is in the intent flow, then it is determined whether the extracted intent is consistent with the current intent.

[0101] If the extracted intent is inconsistent with the current intent, then the intent transfer state is entered, that is, a confirmation instruction is returned asking whether the intent needs to be switched, so that the user can decide whether to switch intents.

[0102] Furthermore, when the session management program in memory 130 is executed by the processor, it also performs the following steps:

[0103] Enter the specific intent corresponding to the extracted intent, load the execution flow of the specific intent, and determine whether the necessary slots in the execution flow have been filled.

[0104] If the necessary slots in the execution process are not filled, the process enters the slot inquiry state, that is, it returns the query instruction for the missing slots to confirm the missing slots with the user and obtain the operation to be executed.

[0105] If the necessary slots in the execution process have been filled, then it is determined whether to perform a second confirmation on the necessary slots and / or the specific intent according to the configuration, and perform a post-intent verification to obtain the operation to be executed;

[0106] The operation to be performed is converted into an operation instruction.

[0107] Furthermore, when the session management program in memory 130 is executed by the processor, it also performs the following steps:

[0108] Determine whether to perform a second confirmation of the necessary slots based on the configuration;

[0109] If a second confirmation is required for the necessary slot, a confirmation instruction corresponding to the necessary slot is returned for the user to confirm.

[0110] Determine whether to conduct a second confirmation of the specific intent based on the configuration;

[0111] If a second confirmation of the specific intent is required, the confirmation statement corresponding to the specific intent is returned, and the necessary slots are packaged for the user to perform a second confirmation to obtain a confirmation result.

[0112] Based on the confirmation result, determine whether the specific intent has a subsequent intent;

[0113] If the specific intent has a subsequent intent, then re-examine whether the necessary slots of the subsequent intent have been filled, until an intent whose necessary slots have not been filled is encountered and / or all the subsequent intents have been checked, and the operation to be executed is obtained.

[0114] If the specific intent has no subsequent intent, then the operation to be executed corresponding to the specific intent is obtained.

[0115] Furthermore, when the session management program in memory 130 is executed by the processor, it also performs the following steps:

[0116] Define basic instructions so that the instruction parsing module can parse the operation instructions based on the basic instructions. The basic instructions include intent instructions and function instructions. The intent instructions are used to call external resources, and the function instructions are used to convert internal data.

[0117] This embodiment, through the above-described scheme, specifically involves acquiring user-input text information; performing slot and intent recognition on the text information to obtain extracted slots and intents; performing session management on the extracted slots and intents to obtain operation instructions, wherein the session management includes one or more of slot confirmation, intent confirmation, and post-intent verification; parsing the operation instructions into corresponding operation content, and displaying the operation content on the front end for the user to view. Slot and intent recognition of the text information can accurately identify the user's operation intent and the keyword slot information contained in the input text, thereby providing a data foundation for the correct execution of subsequent operations. Session management of the extracted slots and intents provides a feature-rich, multi-system compatible, and easily maintainable unified multi-turn conversation management solution for robots, reusing the single functions of various closed systems. By parsing the operation instructions into corresponding operation content and displaying the operation content on the front end, instructions can be parsed into different forms, empowering intelligent assistants on different platform robots, enhancing the uniformity of office systems, and thus improving the processing capabilities of the voice assistant system.

[0118] Based on, but not limited to, the terminal device architecture described above, embodiments of the method of the present invention are proposed.

[0119] The execution subject of the method in this embodiment can be a session management device or a terminal device, etc. This embodiment takes a session management device as an example.

[0120] Reference Figure 2 , Figure 2 This is a flowchart illustrating an exemplary embodiment of the session management method of the present invention. The session management method includes:

[0121] Step S10: Obtain the text information input by the user;

[0122] During the interaction between the user and the system, the user needs to input the desired operation as text information into the system so that the NLU (Natural Language Understanding) module in the system can identify the intent and slots in the text information.

[0123] Step S20: Perform slot and intent recognition on the text information to obtain the extracted slots and intents;

[0124] The NLU module primarily ensures the correct recognition of the user's intent and the keyword slot information contained in the input text, thus providing a data foundation for the correct execution of subsequent operations. Custom external NLU modules are supported; upon entering a module, the external interface is called and its results are provided to subsequent processes. If no external NLU module is configured, the system's built-in NLU functionality is used.

[0125] After obtaining the text input from the user, the system first searches the dictionary list, performs slot replacement on the text, and enumerates the possible user-asked expressions. Then, it searches the template database for each expression, checking if there is a directly matching input template. If so, it returns the corresponding intent directly; otherwise, it continues to match the original text against the regular expression database, checking if there is a directly matching regular expression. If so, it returns the corresponding intent directly; otherwise, it returns nothing.

[0126] Step S30: Perform session management on the extracted slots and intents to obtain operation instructions, wherein the session management includes one or more of slot confirmation, intent confirmation, and post-intent verification;

[0127] Dialogue state tracking primarily manages and validates the dialogue state of historical and currently collected slot information, providing a list of instructions for the robot to display and respond to. The robot needs to provide different responses depending on the dialogue state, which mainly includes the start state, intent transition state, intent confirmation state, follow-up slot state, slot confirmation state, instruction confirmation state, and subsequent intent validation. After all states have been validated sequentially, the conversation management system transmits the necessary operations as instructions to the instruction parsing module for parsing.

[0128] In addition, the slots in the dialogue state are all obtained from the system's slot pool, which saves the user's historical slots for the implementation of multi-turn interactions.

[0129] Step S40: The operation instruction is parsed into corresponding operation content, and the operation content is displayed on the front end for the user to view.

[0130] Before the steps of parsing the operation instruction into corresponding operation content and displaying the operation content on the front end for the user to view, the method further includes:

[0131] Define basic instructions so that the instruction parsing module can parse the operation instructions based on the basic instructions. The basic instructions include intent instructions and function instructions. The intent instructions are used to call external resources, and the function instructions are used to convert internal data.

[0132] The system defines several basic types of instructions:

[0133] a.INTENT (Intent Command): Primarily responsible for calling external resources. Here, the access party configures its system interface call address, and the command parsing module completes the call and obtains the returned data.

[0134] b. FUNC (Function Command): Primarily responsible for internal data transformation, processing data obtained from various data sources for subsequent intent command invocation or front-end display. It mainly consists of:

[0135] a) MAP (Data Transformation): Extracts data from the return of a specified intent and fills it into the input slots of another intent. The transformed slots are synchronously stored in the slot pool for unified management.

[0136] b) TEXT (Text Display): Adds a line of text information to the information displayed on the front end.

[0137] c) VIEW (Data Display): If the data to be displayed requires charts, the data will be packaged into a predefined format and returned to the front end for display.

[0138] d) BLOCK (Wait for interaction): When this operation is detected, parsing will stop, the parsed display information will be handed over to the front end for display, and the system will wait for the user's next input.

[0139] Command parsing is primarily used to complete the retrieval of external data resources and to translate the dialogue state tracking returned by the commands. After parsing, the system-defined commands are converted into content formats such as text or charts that can be displayed on the front end.

[0140] The instruction parser will iterate through the returned instruction list in sequence and process it according to the operation type defined above until the parsing is completed or the BLOCK operation is reached.

[0141] In this embodiment, text information input by the user is acquired; slot and intent recognition is performed on the text information to obtain extracted slots and intents; the extracted slots and intents are then subjected to session management to obtain operation instructions, wherein the session management includes one or more of slot confirmation, intent confirmation, and post-intent verification; the operation instructions are parsed into corresponding operation content, and the operation content is displayed on the front end for the user to view. Slot and intent recognition of the text information can accurately identify the user's operation intent and the keyword slot information contained in the input text, thereby providing a data foundation for the correct execution of subsequent operations. Session management of the extracted slots and intents provides a feature-rich, multi-system compatible, and easy-to-maintain unified multi-turn conversation management solution for robots, reusing the single functions of various closed systems. By parsing the operation instructions into corresponding operation content and displaying the operation content on the front end, instructions can be parsed into different forms, empowering intelligent assistants on different platform robots, enhancing the uniformity of the office system, and thus improving the processing capabilities of the voice assistant system.

[0142] Reference Figure 3 , Figure 3 for Figure 2 In this embodiment, step S20 involves slot and intent recognition of the text information to obtain the extracted slots and intents. This embodiment is based on the above... Figure 2 In the embodiment shown, step S20 includes:

[0143] Step S201: Based on a preset dictionary list, perform slot replacement on the text information to obtain an expression list;

[0144] The dictionary list is searched, the text information is slotted, and the possible user query expressions are enumerated. For example, the text "Find Xiao Li in the Technology Department" contains "Technology Department" in the department dictionary and "Xiao Li" in the name dictionary. Therefore, the possible expression list is ["Find {Department}Xiao Li", "Find Technology Department {Name}", "Find {Department}{Name}"].

[0145] Step S202: Search the template database sequentially according to the expression list to obtain the search results;

[0146] After obtaining the list of expressions, the template database can be searched sequentially according to the expressions in the list to obtain the search results, thereby determining whether there is an input template that matches the expression in the search results.

[0147] Step S203: Determine whether there is a directly matching input template in the search results;

[0148] The search results are checked to see if a directly matching input template exists. If it does, the corresponding user intent is returned directly; otherwise, the process continues to the next step. For example, if "search {department}{name}" is found, the intent corresponding to the template, "search address book", and the slots "department-technical department" and "name-Xiao Li" are returned directly.

[0149] Step S204: If a directly matching input template exists in the search results, the intent corresponding to the input template is returned.

[0150] If no directly matching input template is found in the search results, then it is determined whether a regular expression matching the text information exists in the preset regular expression database.

[0151] If a regular expression exists in the preset regular expression database that matches the text information, then the intent corresponding to the regular expression is returned.

[0152] The original text information is matched against a regular expression database to check if a direct match exists. If a match is found, the intent corresponding to the regular expression is returned directly; otherwise, the process continues to the next step. For example, if the text "I want to find someone" did not match in the previous step, but the database contains the expression "I want to find .*", the intent corresponding to that expression, "Find contacts," is returned directly without slots. If all of the above matches, the user intent from the previous round is returned; otherwise, nothing is returned.

[0153] This embodiment, through the above-described scheme, specifically involves performing slot replacement on the text information based on a preset dictionary list to obtain an expression list; sequentially searching a template database according to the expression list to obtain search results; determining whether a directly matching input template exists in the search results; and if a directly matching input template exists, returning the intent corresponding to the input template. The NLU (Natural Language Understanding) module identifies the intent and slots in the user's input text information, ensuring accurate recognition of the user's operational intent and the keyword slot information contained in the input text, thus providing a data foundation for subsequent correct execution of operations. Simultaneously, it enables interaction with the user through natural language, reducing the user's system learning cost and thereby improving the user experience and satisfaction.

[0154] Reference Figure 4 , Figure 4 for Figure 2 In this embodiment, step S30 performs session management on the extracted slots and intents to obtain operation instructions, which is illustrated in the following flowchart. This embodiment is based on the above. Figure 2 In the embodiment shown, the method further includes the following steps before step S30:

[0155] Determine whether the user is interacting for the first time or has completed the previous intent flow;

[0156] If the user is interacting for the first time or has completed the previous intent process, then enter the start state and identify whether the extracted intent is a single intent;

[0157] If the extracted intent is not the single intent, then all identified intent options are returned to the user to confirm the specific intent to be executed and proceed to the specific intent.

[0158] If the extracted intent is a single intent, then proceed to the specific intent corresponding to the extracted intent.

[0159] The chatbot enters the "Start" state when a user interacts with it for the first time or when a previous intent is completed. It transitions to other states when a specific intent is identified. If the user's expression is ambiguous, the session management system identifies multiple intents that meet the criteria and returns all intent options to the user to confirm the final intent to be executed. Otherwise, it continues to verify the follow-up question slot status.

[0160] In addition, if the user is not interacting for the first time or has not completed the previous intent flow, that is, the user is in the intent flow, then it is determined whether the extracted intent is consistent with the current intent.

[0161] If the extracted intent is inconsistent with the current intent, then the intent transfer state is entered, that is, a confirmation instruction is returned asking whether the intent needs to be switched, so that the user can decide whether to switch intents.

[0162] If the user is already in a certain intent flow, when the user changes the current intent, the system will enter an intent transition state, return a confirmation instruction, and confirm with the user whether they need to switch intents. Otherwise, it will continue to the next state for verification.

[0163] Step S301: Enter the specific intent corresponding to the extracted intent, load the execution flow of the specific intent, and determine whether the necessary slots in the execution flow have been filled.

[0164] Upon entering a specific intent, the execution flow for that intent is loaded. It needs to determine if the necessary slots for intent execution have been filled. If not, a query for missing slots is returned to the user for confirmation. Otherwise, the slot status is checked and confirmed again. After extracting the intent and slots from the text data, multiple rounds of session management are completed to guide the user to supplement information or perform the system function they wish to execute.

[0165] Step S302: If the necessary slots in the execution process are not filled, the process enters the slot inquiry state, that is, returns the query instruction for the missing slots to confirm the missing slots with the user and obtain the operation to be executed.

[0166] If there are any missing slots in the execution flow of this intent, the system needs to enter the slot inquiry state, return the query command for the missing slots, and confirm with the user to complete the filling of the missing slots. This will enable accurate identification of the user's operation intent and improve the user experience.

[0167] Step S303: If the necessary slots in the execution process have been filled, determine whether to perform a second confirmation on the necessary slots and / or the specific intent according to the configuration, and perform a post-intent verification to obtain the operation to be executed.

[0168] Determine whether to perform a second confirmation of the necessary slots based on the configuration;

[0169] If a second confirmation is required for the necessary slot, a confirmation instruction corresponding to the necessary slot is returned for the user to confirm.

[0170] Determine whether to conduct a second confirmation of the specific intent based on the configuration;

[0171] If a second confirmation of the specific intent is required, the confirmation statement corresponding to the specific intent is returned, and the necessary slots are packaged for the user to perform a second confirmation to obtain a confirmation result.

[0172] Based on the confirmation result, determine whether the specific intent has a subsequent intent;

[0173] If the specific intent has a subsequent intent, then re-examine whether the necessary slots of the subsequent intent have been filled, until an intent whose necessary slots have not been filled is encountered and / or all the subsequent intents have been checked, and the operation to be executed is obtained.

[0174] If the specific intent has no subsequent intent, then the operation to be executed corresponding to the specific intent is obtained.

[0175] Slots can be configured with a secondary confirmation option. If enabled, a confirmation command corresponding to the slot is returned for user interaction and confirmation. If the user confirms, the process continues to the next step. If the user rejects or modifies the slot, confirmation is performed again. If this configuration is not available, the confirmation command status is checked again. Intents can also be configured with a secondary confirmation option. If enabled, after all necessary slots for the intent are filled, a confirmation statement corresponding to the intent is returned, and all slots are packaged and sent to the user for secondary confirmation. If the user confirms, the process continues to the next step. If the user rejects or modifies the intent slot, confirmation is performed again. If this configuration is not available, the process continues to check the next status. Furthermore, slots in the dialogue state are all obtained from the system's slot pool. The slot pool stores the user's historical slots for multi-turn interactions, and the configuration can be transformed into a visual flowchart to explain the task flow.

[0176] Furthermore, some complex scenarios require the combination of multiple intents to complete. These intents will be executed sequentially according to the order of the instruction list. If multiple intents are configured, after the completion and confirmation of the preceding intent, the fill status of the follow-up question slots of the subsequent intents will be re-checked until an incomplete intent or instruction is encountered and the process ends. After verification, the execution list is returned to the instruction parsing module for execution.

[0177] Step S304: Convert the operation to be performed into an operation instruction.

[0178] The process of returning instructions is in a non-dialogue state. The session management will transmit the operation to be performed in the form of instructions to the instruction parsing module for parsing. The instructions will be parsed into charts or text to be displayed to the user and guide the user to continue interacting.

[0179] This embodiment, through the above-described scheme, specifically involves entering the specific intent corresponding to the extracted intent, loading the execution flow of the specific intent, and determining whether the necessary slots in the execution flow have been filled. If the necessary slots in the execution flow have not been filled, a slot follow-up state is entered, i.e., a query instruction for the missing slot is returned to confirm the missing slot with the user, resulting in the operation to be executed. If the necessary slots in the execution flow have been filled, a secondary confirmation of the necessary slots and / or the specific intent is determined according to the configuration, and a post-intent verification is performed to obtain the operation to be executed. The operation to be executed is then converted into an operation instruction. After extracting the intent and slots from the text data, multi-round conversation management is completed, thereby guiding the user to supplement information or complete the system function operation that the user wants to perform. This conversation management method provides a unified multi-round conversation management scheme for robots that is feature-rich, compatible with multiple systems, and easy to maintain. It reuses the single functions of various closed systems and provides user interaction through voice. The intelligent dialogue state tracking mechanism realizes a highly interpretable system flow, decomposing complex office scenarios into independent modules for management, thereby greatly reducing the development cost of new scenarios.

[0180] Furthermore, embodiments of the present invention also propose a session management device, the session management device comprising:

[0181] The acquisition module is used to acquire text information input by the user;

[0182] The natural language understanding module is used to perform slot and intent recognition on the text information to obtain the extracted slots and intents;

[0183] The dialogue state tracking module is used to perform dialogue management on the extracted slots and intents to obtain operation instructions. The dialogue management includes one or more of slot confirmation, intent confirmation, and post-intent verification.

[0184] The instruction parsing module is used to parse the operation instruction into corresponding operation content and display the operation content on the front end for the user to view.

[0185] The overall processing flow is as follows:

[0186] The NLU (Natural Language Understanding) module is used to identify the intent and slots in the text input by the user.

[0187] The extracted intents and slots are managed through dialogue state tracking to control multi-turn conversation information.

[0188] The command parsing module parses the operation list returned by the session management and completes the specified operation.

[0189] The NLU module primarily ensures the correct recognition of the user's intent and the keyword slot information contained in the input text, thus providing a data foundation for the correct execution of subsequent operations. Custom external NLU modules are supported; upon entering a module, an external interface is called, and its results are provided to subsequent processes. If an external NLU module is not configured, the system's built-in NLU functionality is used, with the specific implementation process as follows:

[0190] The dictionary list is searched, and the text information is slotted to obtain a list of possible user query expressions. For example, the text "Find Xiao Li in the Technical Department" contains "Technical Department" in the department dictionary and "Xiao Li" in the name dictionary. Therefore, the possible expression list is: ["Find {Department}Xiao Li", "Find Technical Department {Name}", "Find {Department}{Name}"].

[0191] The expression is searched sequentially in the template database to check if a directly matching input template exists. If a match is found, it is returned directly; otherwise, the process continues to the next step. For example, if "search {department}{name}" is found, the corresponding intent "search address book" and the slots "department-technical department" and "name-Xiao Li" are returned directly.

[0192] The original text is matched against a regular expression database. If a direct match is found, it is returned directly; otherwise, the process continues to the next step. For example, if the text "I want to find someone" was not matched in the previous step, but the database contains the expression "I want to find.*", the intent corresponding to that expression, "find contacts," is returned directly without slots.

[0193] If all of the above match, return the user intent from the previous round; otherwise, return empty.

[0194] Dialogue state tracking primarily involves managing and validating the dialogue state using historical and currently collected slot information, and providing a list of instructions for the robot to display and respond to. The robot needs to provide different response information depending on the dialogue state it is in.

[0195] Reference Figure 5 , Figure 5 This is a schematic diagram of the dialogue state tracking process in an embodiment of the present invention, such as... Figure 5 As shown, the specific implementation process of dialogue state tracking is as follows:

[0196] 1) Start State: The user enters the start state when interacting with the robot for the first time or when the previous intention is completed. When a specific intention is recognized, the user will move to another state.

[0197] 2) Intent Transition State: When a user is already in an intent flow, and changes their current intent, the system enters the intent transition state, returns a confirmation instruction, and asks the user if they want to switch intents. Otherwise, it continues to the next state.

[0198] 3) Confirm Intent Status: When the user's expression is ambiguous, the session management system identifies multiple intents that meet the conditions and returns all intent options to the user to confirm the specific intent to be executed. Otherwise, it continues to verify the next status.

[0199] 4) Follow up on slot status: Upon entering a specific intent, the execution flow of that intent is loaded to determine whether the necessary slots for intent execution have been filled. If not, a query command for missing slots is returned to confirm with the user. Otherwise, the process continues to verify the next status.

[0200] 5) Confirm Slot Status: Slots can be configured with a secondary confirmation option. If enabled, the corresponding confirmation command for the slot will be returned for user interaction and confirmation. If the user confirms, proceed to the next step; if the user rejects or modifies the slot, confirmation will be requested again. If this configuration is not available, continue to the next status verification.

[0201] 6) Confirm Instruction Status: The intent can be configured with a secondary confirmation option. If enabled, after all necessary slots for the intent are filled, a confirmation statement corresponding to the intent will be returned, and all slots will be packaged and sent to the user for secondary confirmation. If the user confirms, proceed to the next step; if the user rejects or modifies the intent slots, confirmation will be performed again. If this configuration is not available, continue to the next status verification.

[0202] 7) Post-Intent Validation: In some complex scenarios, multiple intents need to be combined to complete a task. These intents will be executed sequentially according to the order of the instruction list. If multiple intents are configured, after the previous intent is completed and the instruction status is confirmed, the fill status of the follow-up question slots of the subsequent intents will be re-checked until an incomplete intent or instruction is encountered and the process ends. After the validation is completed, the execution list is returned to the instruction parsing module for execution.

[0203] 8) Return command: In non-dialogue state, the session management will transmit the operation to be performed as a command to the command parsing module for parsing. The command is parsed into a chart or text and displayed to the user, guiding the user to continue interacting.

[0204] 9) User interaction: The system displays the parsed text or charts of the instructions to the front end, waiting for the next round of user input for interaction.

[0205] In the above, all slots in the dialogue state are obtained from the system's slot pool, which stores the user's historical slots for the implementation of multi-turn interactions.

[0206] The instruction parsing process primarily facilitates the retrieval of external data resources and transforms the dialogue state tracking returned instructions. After parsing, the system-defined instructions are converted into content formats such as text or charts that the front-end can display. The specific process is as follows:

[0207] 1) The system defines several basic types of instructions:

[0208] a.INTENT (Intent Command): Primarily responsible for calling external resources. Here, the access party configures its system interface call address, and the command parsing module completes the call and obtains the returned data.

[0209] b. FUNC (Function Command): Primarily responsible for internal data transformation, processing data obtained from various data sources for subsequent intent command invocation or front-end display. It mainly consists of:

[0210] a) MAP (Data Transformation): Extracts data from the return of a specified intent and fills it into the input slots of another intent. The transformed slots are synchronously stored in the slot pool for unified management.

[0211] b) TEXT (Text Display): Adds a line of text information to the information displayed on the front end.

[0212] c) VIEW (Data Display): If the data to be displayed requires charts, the data will be packaged into a predefined format and returned to the front end for display.

[0213] d) BLOCK (Wait for interaction): When this operation is detected, parsing will stop, the parsed display information will be handed over to the front end for display, and the system will wait for the user's next input.

[0214] The instruction parser will iterate through the returned instruction list in sequence and process it according to the operation type defined above until the parsing is completed or the BLOCK operation is reached.

[0215] This embodiment primarily implements the conversation management component of the office assistant. Based on different office scenario requirements, it extracts intents and slots from text data to complete multi-turn conversation management, guiding users to supplement information or perform desired system functions. This conversation management method provides a feature-rich, multi-system compatible, and easily maintainable unified multi-turn conversation management solution for robots. It reuses single functions from various closed systems and provides user interaction via voice. The intelligent dialogue state tracking mechanism achieves a highly interpretable system flow, breaking down complex office scenarios into independent modules for management, thus significantly reducing the development cost of new scenarios. Interacting with users through natural language reduces the system learning cost for users, thereby improving user experience and satisfaction. Developers of independent systems can also integrate through new conversation intents, increasing the exposure channels for system functions and improving the usage rate of the office system. An independent instruction parsing module can also parse instructions into different forms, empowering the intelligent assistant to different platform robots and enhancing the uniformity of the office system.

[0216] Furthermore, this embodiment of the invention also proposes a terminal device, which includes a memory, a processor, and a session management program stored in the memory and executable on the processor. When the session management program is executed by the processor, it implements the steps of the session management method described above.

[0217] Since this session management program employs all the technical solutions of all the aforementioned embodiments when executed by the processor, it possesses at least all the beneficial effects brought about by all the technical solutions of all the aforementioned embodiments, which will not be elaborated upon here.

[0218] Furthermore, embodiments of the present invention also propose a computer-readable storage medium storing a session management program, which, when executed by a processor, implements the steps of the session management method described above.

[0219] Since this session management program employs all the technical solutions of all the aforementioned embodiments when executed by the processor, it possesses at least all the beneficial effects brought about by all the technical solutions of all the aforementioned embodiments, which will not be elaborated upon here.

[0220] Compared to existing technologies, the conversation management method, apparatus, terminal device, and storage medium proposed in this invention acquire text information input by the user; perform slot and intent recognition on the text information to obtain extracted slots and intents; perform conversation management on the extracted slots and intents to obtain operation instructions, wherein the conversation management includes one or more of slot confirmation, intent confirmation, and post-intent verification; parse the operation instructions into corresponding operation content, and display the operation content on the front end for the user to view. Slot and intent recognition on the text information can accurately identify the user's operation intent and the keyword slot information contained in the input text, thereby providing a data foundation for the correct execution of subsequent operations. The conversation management method for the extracted slots and intents provides a unified multi-turn conversation management scheme for robots that is feature-rich, compatible with multiple systems, and easy to maintain, reusing the single functions of various closed systems. By parsing the operation instructions into corresponding operation content and displaying the operation content on the front end, instructions can be parsed into different forms, empowering intelligent assistants on different platform robots, enhancing the uniformity of office systems, and thus improving the processing capabilities of voice assistant systems.

[0221] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0222] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0223] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, controlled terminal, or network device, etc.) to execute the methods of each embodiment of this application.

[0224] The above are merely preferred embodiments of the present invention and do not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A session management method, characterized in that, The session management method includes the following steps: Obtain text information input by the user; The text information is subjected to slot and intent recognition to obtain the extracted slots and intents; By tracking the dialogue state, the extracted intents and slots are managed through multi-turn conversation information to obtain operation instructions. The conversation management includes one or more of the following: slot confirmation, intent confirmation, and post-intent verification. The dialogue state includes one or more of the following: start state, intent transfer state, intent confirmation state, follow-up slot state, slot confirmation state, instruction confirmation state, and post-intent verification. The operation instructions are parsed into corresponding operation content, and the operation content is displayed on the front end for the user to view; Before the step of parsing the operation instruction into corresponding operation content and displaying the operation content to the front end, the method further includes: Define basic instructions so that the instruction parsing module can parse the operation instructions based on the basic instructions. The basic instructions include intent instructions and function instructions. The intent instructions are used to call external resources, and the function instructions are used to convert internal data. Before the step of managing the extracted intents and slots through dialogue state tracking to obtain operation instructions, the method further includes: Determine whether the user is interacting for the first time or has completed the previous intent flow; If the user is interacting for the first time or has completed the previous intent process, then enter the start state and identify whether the extracted intent is a single intent; If the extracted intent is not the single intent, then all identified intent options are returned to the user to confirm the specific intent to be executed and proceed to the specific intent. If the extracted intent is the single intent, then proceed to the specific intent corresponding to the extracted intent; The step of determining whether the user is interacting for the first time or has completed the previous intent flow also includes: If the user is not interacting for the first time or has not completed the previous intent process, then determine whether the extracted intent is consistent with the current intent. If the extracted intent is inconsistent with the current intent, then an intent transfer state is entered, allowing the user to determine whether to switch intents. The step of managing multi-turn conversation information by tracking the extracted intent and slots through dialogue state to obtain operation instructions includes: Enter the specific intent corresponding to the extracted intent, load the execution flow of the specific intent, and determine whether the necessary slots in the execution flow have been filled. If the necessary slots in the execution process are not filled, the process enters the slot inquiry state to confirm the missing slots with the user and obtain the operation to be executed. If the necessary slots in the execution process have been filled, then it is determined whether to perform a second confirmation on the necessary slots and / or the specific intent according to the configuration, and perform a post-intent verification to obtain the operation to be executed; The operation to be performed is converted into an operation instruction.

2. The session management method as described in claim 1, characterized in that, The step of performing slot and intent recognition on the text information to obtain the extracted slots and intents includes: Based on a preset dictionary list, slot replacement is performed on the text information to obtain an expression list; The template database is searched sequentially according to the list of expressions to obtain the search results; Determine whether a directly matching input template exists in the search results; If a directly matching input template exists in the search results, the intent corresponding to the input template is returned.

3. The session management method as described in claim 2, characterized in that, The step of determining whether a directly matching input template exists in the search results further includes: If no directly matching input template is found in the search results, then it is determined whether a regular expression matching the text information exists in the preset regular expression database. If a regular expression exists in the preset regular expression database that matches the text information, then the intent corresponding to the regular expression is returned.

4. The session management method as described in claim 1, characterized in that, The step of determining whether to perform secondary confirmation of the necessary slots and / or the specific intent based on the configuration, and performing post-intent verification to obtain the operation to be executed includes: Determine whether to perform a second confirmation of the necessary slots based on the configuration; If a second confirmation is required for the necessary slot, a confirmation instruction corresponding to the necessary slot is returned for the user to confirm. Determine whether to conduct a second confirmation of the specific intent based on the configuration; If a second confirmation of the specific intent is required, the confirmation statement corresponding to the specific intent is returned, and the necessary slots are packaged for the user to perform a second confirmation to obtain a confirmation result. Based on the confirmation result, determine whether the specific intent has a subsequent intent; If the specific intent has a subsequent intent, then re-examine whether the necessary slots of the subsequent intent have been filled, until an intent whose necessary slots have not been filled is encountered and / or all the subsequent intents have been checked, and the operation to be executed is obtained. If the specific intent has no subsequent intent, then the operation to be executed corresponding to the specific intent is obtained.

5. A session management device, characterized in that, The session management device includes: The acquisition module is used to acquire text information input by the user; The natural language understanding module is used to perform slot and intent recognition on the text information to obtain the extracted slots and intents; The dialogue state tracking module is used to manage multi-turn conversation information of extracted intents and slots through dialogue state tracking. The conversation management includes one or more of slot confirmation, intent confirmation, and post-intent verification. The dialogue state includes one or more of the following: start state, intent transfer state, intent confirmation state, follow-up slot state, slot confirmation state, instruction confirmation state, and post-intent verification. The instruction parsing module is used to parse operation instructions into corresponding operation content and display the operation content on the front end for the user to view; Before parsing the operation instruction into corresponding operation content and displaying the operation content to the front end, the process further includes: Define basic instructions so that the instruction parsing module can parse the operation instructions based on the basic instructions. The basic instructions include intent instructions and function instructions. The intent instructions are used to call external resources, and the function instructions are used to convert internal data. The step of managing multi-turn conversational information based on extracted intents and slots through dialogue state tracking, before obtaining operation instructions, also includes: Determine whether the user is interacting for the first time or has completed the previous intent flow; If the user is interacting for the first time or has completed the previous intent process, then enter the start state and identify whether the extracted intent is a single intent; If the extracted intent is not the single intent, then all identified intent options are returned to the user to confirm the specific intent to be executed and proceed to the specific intent. If the extracted intent is the single intent, then proceed to the specific intent corresponding to the extracted intent; The process of determining whether the user is interacting for the first time or has completed the previous intent flow also includes: If the user is not interacting for the first time or has not completed the previous intent process, then determine whether the extracted intent is consistent with the current intent. If the extracted intent is inconsistent with the current intent, then an intent transfer state is entered, allowing the user to determine whether to switch intents. The process of managing multi-turn conversation information by tracking the extracted intent and slots through dialogue state, and obtaining operation instructions, includes: Enter the specific intent corresponding to the extracted intent, load the execution flow of the specific intent, and determine whether the necessary slots in the execution flow have been filled. If the necessary slots in the execution process are not filled, the process enters the slot inquiry state to confirm the missing slots with the user and obtain the operation to be executed. If the necessary slots in the execution process have been filled, then it is determined whether to perform a second confirmation on the necessary slots and / or the specific intent according to the configuration, and perform a post-intent verification to obtain the operation to be executed; The operation to be performed is converted into an operation instruction.

6. A terminal device, characterized in that, The terminal device includes a memory, a processor, and a session management program stored in the memory and executable on the processor. When the session management program is executed by the processor, it implements the steps of the session management method as described in any one of claims 1-4.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a session management program, which, when executed by a processor, implements the steps of the session management method as described in any one of claims 1-4.

Images