Automatic generation method of task flow based on intention recognition

Abstract

The present invention discloses a method for automatically generating task flows based on intent recognition. This method, based on a text processing method for intent recognition, performs word segmentation on user input text and extracts slot information containing intent. Through the combination of natural language processing technology based on intent recognition, intelligent analysis algorithms, and an adaptive task scheduling mechanism, the method significantly improves the accuracy of task matching and scheduling efficiency, reduces the need for manual intervention, and can dynamically adjust task execution strategies based on resource conditions. Furthermore, an anomaly detection mechanism can identify potential problems in real time during the task scheduling process and automatically adjust the task execution path, effectively avoiding resource waste and task delays.

Description

Technical Field

[0001] The present invention relates to the fields of computer information technology, artificial intelligence and intelligent decision-making, and in particular to a method for automatically generating a task flow based on intent recognition. Background Art

[0002] With the rapid development of artificial intelligence technology, especially in task management and process automation, understanding user natural language instructions and converting them into executable task processes has become a critical issue that needs to be addressed. Traditional task management systems often rely on structured user input, requiring users to master specific instructions and formats to effectively interact with the system. This requires users to expend additional effort to learn and use the system, increasing the learning curve and reducing the user experience. By combining intent-based natural language understanding technology with automatic task process generation methods, user intent is converted into specific task steps, each task link is identified, and the corresponding execution process is automatically generated. This can help enterprises build more intelligent, flexible, and easy-to-use task management systems.

[0003] Intent recognition-based automatic task flow generation methods have broad application prospects in modern enterprise work environments, particularly in rapidly changing markets and complex operational environments. They can perceive and adapt to various external changes in real time. By deeply parsing and understanding user input, they identify implicit needs and goals, generate a series of executable task steps, and dynamically adjust and optimize the task flow through intelligent decision-making algorithms. This typically involves multiple technical steps, including semantic understanding, intent recognition, task mapping, process design, and decision-making optimization. Intent recognition utilizes methods such as word embedding, semantic matching, and semantic annotation to provide the system with accurate user intent. Automatic task flow generation, through rule engines and process modeling, translates the identified intent into a specific task execution process. During execution, intelligent decision-making techniques dynamically adjust task steps to ensure that the execution logic conforms to the enterprise's operational specifications and address diverse business scenarios and emergencies. By combining and optimizing these technologies, automatic task flow generation systems enable dynamic management of automatic task flow generation through a graphical interface or programming interface. They can dynamically create new process templates or adjust existing process structures based on specific user needs, not limited to built-in standard processes. Highly intelligent task management can not only improve enterprise production efficiency, but also significantly reduce human intervention and errors, ensure the smooth progress of business processes, and greatly enhance the intelligence level of task management and resource scheduling. Summary of the Invention

[0004] The present invention aims to provide a method for automatically generating task flows based on intent recognition, so as to improve the efficiency and accuracy of task management and resource scheduling.

[0005] The purpose of the present invention is achieved through the following technical solutions:

[0006] The method and system for automatically generating task processes based on intention recognition include the following steps:

[0007] Step S1: Receive text input from the user, segment the text, extract slot information containing intent, and generate a corresponding natural language response;

[0008] Step S2: Perform word segmentation on the text information in the task pool and construct an inverted index to obtain a set of candidate task information containing intent;

[0009] Step S3: Summarize the semantic information of the user input text information and the candidate task information, evaluate the task matching degree of each candidate task and sort them in descending order to obtain a to-do task set containing intent information;

[0010] Step S4: Based on the task set with the highest task matching degree and the existing resource situation, an adaptive task scheduling method is constructed, anomaly detection values ​​are introduced, anomaly feedback is output, and a task flow is generated;

[0011] Step S5: Receive the task process scoring data given by the user, construct the perplexity scoring function, obtain feedback data and update the task status, and store the feedback information in the database.

[0012] Furthermore, the step S1 specifically includes:

[0013] Step S101: receiving input text from the user terminal, performing word segmentation on the text information T, and obtaining text information T = {t1, t2, ..., t m}, where m is the number of words in the text information T;

[0014] Step S102: extracting slot information related to the user's intention from the segmented text information T, and performing named entity annotation on the text information T input by the user;

[0015] Step S103: Generate a corresponding natural language response for the extracted slot information according to predefined rules, and return it to the user end through the automatic generation system.

[0016] Furthermore, the step S2 specifically includes:

[0017] Step S201: Segment all text information in the task pool P to obtain p i ={t i1 ,t i2 ,…,t iq}, where i represents the i-th task information, and q represents the task information pi The number of words in

[0018] Step S202: traverse all task information in the task pool P, and update the index of each word in each task information, and build an inverted index Q (t ij )={p i |t ij ∈p i}, returns the word containing the word t ij The set of all task information, where j represents the task information p i The jth word in ;

[0019] Step S203: Based on the text information T input by the user, search the task information containing the word from the task pool P through the inverted index to generate a candidate task information set D = {d1, d2, ..., d s}, where s represents the number of segmentations in the candidate task information.

[0020] Furthermore, the step S3 specifically includes:

[0021] Step S301: Aggregate the semantic information of the text information T and all words in the candidate task information set D using the average pooling method to obtain feature vectors W containing intent information. T and The average pooling formula is as follows:

[0022]

[0023] in, is the kth word t in the text information T k The corresponding word vector, is the lth word d in the candidate task information set D l The corresponding word vector;

[0024] Step S302: Based on the feature vector W of the text information T and the candidate task information set D T and Calculate the task matching degree of each candidate task in the candidate task set D

[0025] Step S303: According to task matching For each candidate task d l Arrange in descending order to obtain the sorted candidate task set R = {r1, r2, ..., r s}, the calculation formula for descending order is as follows:

[0026]

[0027] Step S304: Based on the sorted candidate task set R, recommend the first w tasks with the highest task matching degree, and generate a to-do task set U containing intention information = {u1, u2, ..., u w}.

[0028] Furthermore, in step S302, the task matching degree of the candidate task The calculation method is:

[0029] (1) Obtain the feature vector W of the text information T and the candidate task information set D T and Construct task matching function:

[0030]

[0031] Among them, W' T is the feature vector W of the text information T T The transpose of

[0032] (2) Calculate the original matching degree of the candidate task based on the task matching function And normalize the matching degree of all candidate tasks to obtain the task matching degree of the candidate tasks The calculation method is:

[0033]

[0034] (3) Output the task matching degree of candidate tasks

[0035] Furthermore, the step S4 specifically includes:

[0036] Step S401: constructing an adaptive task scheduling method based on the task contents of the to-do task set U and the usage of existing resources to automatically generate a task plan U';

[0037] Step S402: Record user task information in real time and monitor task execution status according to the task sequence and resource allocation in the task plan, and store the task execution status in the database;

[0038] Step S403: According to the task execution status, an abnormality detection value is introduced to analyze whether the task execution status is abnormal. The abnormality detection value judgment method is as follows:

[0039]

[0040] Among them, u λ The λth subtask planned for the current task;

[0041] Step S404: Output abnormality feedback.

[0042] Furthermore, the specific process of the adaptive task scheduling method in step S401 is as follows:

[0043] (1) According to the task content of the to-do task set U, combined with each subtask u λ The urgency and resource requirements of the subtask are evaluated and its priority is evaluated. The subtask priority calculation formula is as follows:

[0044]

[0045] Among them, α1 and α2 are weight coefficients, V(u λ ) is the subtask u λ The degree of relationship with other tasks, H(u λ ) is the subtask u λ The remaining execution time;

[0046] (2) Sort by subtask priority and calculate subtask u λ Resource utilization Φ(u λ ), the calculation formula is as follows:

[0047]

[0048] Where ζ is the full-time equivalent number, The maximum allowed task execution time;

[0049] (3) Make a judgment: If Φ(u λ )≤ε, then generate the task plan U' according to the atomic task priority, otherwise downgrade the current subtask priority and reorder it, where ε is the known resource utilization;

[0050] (4) Output the generated task plan U'.

[0051] Furthermore, the task execution status in step S403 is defined as:

[0052] (1) The task is executed normally (A(u λ )=0): Continue to complete the current subtask;

[0053] (2) Insufficient existing resources (A(u λ )=1): You can apply for extended resources or wait for resources to be released and try again;

[0054] (3) Task timeout (A(u λ )=2): You can choose to retry the task or extend the task duration;

[0055] (4) Execution failed (A(u λ )=3): Repair or retry;

[0056] (5) Data error (A(uλ )=4): Check input data and repair data format;

[0057] (6) Dependency failure (A(u λ )=5): Re-execute the predecessor subtask;

[0058] (7) Network anomaly (A(u λ )=6): Check the network connection and try again after it is restored.

[0059] Furthermore, the step S5 specifically includes:

[0060] Step S501: The user terminal provides scoring data for evaluating the automatic generation effect of the current task process, including the accuracy of the task process, resource utilization efficiency and exception handling effect;

[0061] Step S502: constructing a perplexity scoring function based on the scoring data and generating feedback data;

[0062] Step S503: Update and save the current task status, and store the feedback information in the database.

[0063] Furthermore, the perplexity scoring function in step S502 is:

[0064]

[0065] in, is the normalized result of the n-th item score, F n is the scoring result of the nth item, σ is the highest score of the scoring item, and C is the total number of scoring items.

[0066] The beneficial effects of the present invention include:

[0067] (1) The text processing method based on intent recognition of the present invention accurately understands user needs by segmenting the text input by the user and extracting slot information containing intent. It then automatically generates a set of recommended tasks based on the semantic matching degree of candidate task information. This method improves the intelligent level of task recommendation, can accurately match user needs, reduce manual intervention and errors, and improve overall task scheduling efficiency.

[0068] (2) Traditional task flow automatic generation systems rely on simple task matching, resulting in inaccurate task recommendations and reduced user experience. However, task flow automatic generation systems based on user intent information can more accurately match and sort tasks, ensuring that the generated tasks meet the user's real needs and improving the intelligence level of task processing. In addition, the system can continuously optimize the task flow generation results based on the user's intent, making it more intelligent in task flow generation and resource optimization, which can effectively improve user work efficiency.

[0069] (3) By analyzing the matching degree between candidate tasks and existing resources, the system can implement adaptive task scheduling, thus ensuring that task allocation is more scientific and reasonable. This method can dynamically evaluate the resource requirements of each task based on the complexity, priority, and type of required resources, and match them with existing resources. At the same time, the matching mechanism can effectively avoid resource waste or overload, ensure that the computing and storage resources in the system can be fully utilized, achieve optimal resource configuration, and improve task execution efficiency;

[0070] (4) In the process of task scheduling, the real-time introduction of anomaly detection methods can not only detect faults or anomalies that have occurred, but also predict and warn of possible anomalies, and automatically adjust the scheduling strategy based on the anomaly feedback information; in addition, the anomaly detection mechanism can identify problems in a timely manner during task execution, discover potential regularity problems and adjust task scheduling, reducing potential risks in system operation and improving the accuracy of future task scheduling.

[0071] (5) The perplexity score is based on user feedback information and task completion status, and can accurately identify potential problems and deficiencies in the system that are difficult to identify. These feedback data will be updated to the database in real time to ensure that the system can self-optimize through the continuous accumulation of historical data. By continuously learning and analyzing historical feedback information, the system can identify the task recommendations and scheduling strategies that best meet user needs, thereby improving the accuracy of task matching and the response speed of the system. As the amount of data increases, the system can not only predict user needs more accurately, but also flexibly adjust task priorities and scheduling methods to reduce delays and improve efficiency. It can not only improve the user experience, but also ensure that the system maintains efficient operation under different load conditions, and has greater adaptability and intelligence.

[0072] (6) The present invention significantly reduces manual intervention through automated task recommendation and scheduling processes. Furthermore, during user-system interaction, the system can make intelligent decisions based on real-time task requirements and resource status, automatically assign the most appropriate tasks, and dynamically adjust task priorities and scheduling order. This adaptive task scheduling method ensures efficient and smooth task execution, reduces operational complexity, and enhances user experience.

[0073] (7) The present invention can be applied to the automatic generation of complex task processes of AI digital humans. It adopts intention recognition technology, has a high degree of simulation of human interaction and decision-making capabilities, and can accurately identify the potential intentions expressed by users. The system intelligently adjusts the task sequence and execution strategy based on the different needs of users, past interaction history, and real-time changes in tasks. During the task execution process, the system can track the progress of each link and provide timely feedback to ensure the high-quality completion of the task. The automatic generation method of task processes based on intention recognition enables AI digital humans to flexibly adapt to user needs and personalized characteristics, provide a more accurate and natural interactive experience, improve user satisfaction, and promote the further development of AI digital human technology.

[0074] Other advantages, objects, and features of the present invention will be described in part in the following description and, in part, will be apparent to those skilled in the art upon examination of the following or may be learned from practice of the present invention. The objects and other advantages of the present invention may be realized and obtained through the following description and the preceding claims. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be further described in detail below with reference to the accompanying drawings, in which:

[0076] Figure 1 It is the overall workflow diagram of the present invention;

[0077] Figure 2 Automatically generate a system interface diagram for the task flow of the present invention;

[0078] Figure 3 A schematic diagram of slot information containing intent extracted for the present invention;

[0079] Figure 4 Flowchart for task matching calculation and to-do task set generation in the present invention

[0080] Figure 5 A task plan diagram generated by the task flow automatic generation system of the present invention;

[0081] Figure 6This is a flow chart of the abnormal detection value judgment method of the present invention;

[0082] Figure 7 This is a schematic diagram of task status update and feedback storage in the present invention. DETAILED DESCRIPTION

[0083] The preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the preferred embodiments are only for illustrating the present invention, and are not intended to limit the scope of protection of the present invention.

[0084] There are many problems with traditional task process automatic generation systems, including but not limited to defects such as low task matching accuracy, low resource utilization, and poor reliability of task execution. In order to solve the above problems, the present invention uses natural language processing technology based on intent recognition, intelligent analysis algorithms, and adaptive task scheduling mechanisms. By combining natural language processing with intelligent scheduling, it greatly improves the accuracy of task matching and scheduling efficiency, reduces the need for manual intervention, and can dynamically adjust task execution strategies according to resource conditions. At the same time, the anomaly detection mechanism can identify potential problems in real time during the task scheduling process and automatically adjust the task execution path, effectively avoiding resource waste and task delays. The method has strong adaptability and can be widely used in various task scheduling systems, especially in scenarios that require efficient resource management and precise task matching, such as intelligent customer service, enterprise resource management, automated production and other fields.

[0085] The overall workflow of the method for automatically generating task flows based on intention recognition in this embodiment is shown in the following figure: Figure 1 As shown, the following steps are included:

[0086] Step S1: Receive text input from the user, segment the text, extract slot information containing intent, and generate a corresponding natural language response;

[0087] Step S2: Perform word segmentation on the text information in the task pool and construct an inverted index to obtain a set of candidate task information containing intent;

[0088] Step S3: Summarize the semantic information of the user input text information and the candidate task information, evaluate the task matching degree of each candidate task and sort them in descending order to obtain a to-do task set containing intent information;

[0089] Step S4: Based on the task set with the highest task matching degree and the existing resource situation, an adaptive task scheduling method is constructed, anomaly detection values ​​are introduced, anomaly feedback is output, and a task flow is generated;

[0090] Step S5: Receive the task process scoring data given by the user, construct the perplexity scoring function, obtain feedback data and update the task status, and store the feedback information in the database.

[0091] The specific steps of the above method will be further described below through a specific embodiment.

[0092] In this embodiment, step S1 specifically includes the following steps: the task process automatically generates a system interface such as Figure 2 As shown:

[0093] Step S101: receiving input text from the user terminal, performing word segmentation on the text information T, and obtaining text information T = {t1, t2, ..., t m}, where m is the number of words in the text information T;

[0094] Step S102: extracting slot information related to the user's intention from the segmented text information T, and performing named entity annotation on the text information T input by the user;

[0095] Step S103: Generate a corresponding natural language response for the extracted slot information according to predefined rules, and return it to the user end through the automatic generation system.

[0096] In this embodiment, for step S1, taking the user input text "latest work dynamics" as an example, word segmentation processing is performed based on the text information T to obtain text information T = {"latest", "work", "dynamics"}, and relevant slot information containing intention is extracted, such as Figure 3 As shown in the figure, the named entity annotation results obtained in markdown format are:

[0097] ##Users request to view the latest work dynamics, and the task flow automatically generates a system response request

[0098] **service_request{"task":"work","action":"dynamic"}

[0099] --action_service

[0100] Among them, the "##" symbol is the starting identifier of a conversation, the "**" symbol is the identifier of the user's intention, and the "--" symbol is the natural language response generated by the automatic generation system.

[0101] Step S2 specifically includes the following steps:

[0102] Step S201: Segment all text information in the task pool P to obtain p i ={t i1 ,t i2 ,…,t iq}, where i represents the i-th task information, and q represents the task information p i The number of words in

[0103] Step S202: traverse all task information in the task pool P, and update the index of each word in each task information, and build an inverted index Q (t ij )={p i |t ij ∈p i}, returns the word containing the word t ij The set of all task information, where j represents the task information p i The jth word in ;

[0104] Step S203: Based on the text information T input by the user, search the task information containing the word from the task pool P through the inverted index to generate a candidate task information set D = {d1, d2, ..., d s}, where s represents the number of segmentations in the candidate task information.

[0105] In this embodiment, for step S2, taking the task pool P = {"work record", "to-do work", "project approval", "industry dynamics", "work report"} as an example, the text information in the task pool P is segmented to obtain p1 = {"work", "record"}, p2 = {"to-do", "work"}, p3 = {"project", "approval"}, p4 = {"industry", "dynamics"}, p5 = {"work", "report"}, and an inverted index is constructed based on the segmentation:

[0106] "work" = {p1,p2,p5};

[0107] "record" = {p1};

[0108] "To do" = {p2};

[0109] "project" = {p3};

[0110] "approval" = {p3};

[0111] "industry" = {p4};

[0112] "dynamic" = {p4};

[0113] "Report" = {p5};

[0114] According to the text information T = {"latest", "work", "dynamic"} input by the user, a candidate task information set D = {p1, p2, p4, p5} is generated.

[0115] Step S3 specifically includes the following steps. The overall process of task matching calculation and to-do task set generation is as follows: Figure 4 As shown:

[0116] Step S301: Aggregate the semantic information of the text information T and all words in the candidate task information set D using the average pooling method to obtain feature vectors W containing intent information. T and The average pooling formula is as follows:

[0117]

[0118] in, is the kth word t in the text information T k The corresponding word vector, is the lth word d in the candidate task information set D l The corresponding word vector;

[0119] In step S301 of this embodiment, the word vector of each word is obtained through the pre-trained word embedding model. Assume that in the text information T, the word vector of the segmented word is The text information T is averaged and pooled to obtain the feature vector W. T :

[0120]

[0121] Assume that in the candidate task information set D,

[0122] The candidate task information set D is average pooled to obtain the feature vector

[0123]

[0124] Step S302: Based on the feature vector W of the text information T and the candidate task information set D T and Calculate the task matching degree of each candidate task in the candidate task set D

[0125] In step S302, the task matching degree of the candidate task The calculation method is:

[0126] (1) Obtain the feature vector W of the text information T and the candidate task information set D T and Construct task matching function:

[0127]

[0128] Among them, W' T is the feature vector W of the text information T T The transpose of

[0129] (2) Calculate the original matching degree of the candidate task based on the task matching function And normalize the matching degree of all candidate tasks to obtain the task matching degree of the candidate tasks The calculation method is:

[0130]

[0131] (3) Output the task matching degree of candidate tasks

[0132] In step S302 of this embodiment, the calculation result of the task matching function is:

[0133]

[0134] The original matching degree of the candidate task The calculation result is:

[0135]

[0136] After normalization, the task matching degree of the candidate task The calculation result is:

[0137]

[0138] Step S303: According to task matching For each candidate task d l Arrange in descending order to obtain the sorted candidate task set R = {r1, r2, ..., r s}, the calculation formula for descending order is as follows:

[0139]

[0140] In step S303 of this embodiment, the candidate task set R containing intent information arranged in descending order = {"industry dynamics", "work report", "work record", "to-be-done tasks"}.

[0141] Step S304: Based on the sorted candidate task set R, recommend the first w tasks with the highest task matching degree, and generate a to-do task set U containing intention information = {u1, u2, ..., u w}.

[0142] In step S304 of this embodiment, the known task flow automatically generates the system recommendation of the top three tasks with the highest task matching degree, and the to-do task set U = {u1 = "industry dynamics", u2 = "work report", u3 = "work record"}.

[0143] Step S4 specifically includes the following steps:

[0144] Step S401: constructing an adaptive task scheduling method based on the task contents of the to-do task set U and the usage of existing resources to automatically generate a task plan U';

[0145] In step S401, the specific process of the adaptive task scheduling method is as follows:

[0146] (1) According to the task content of the to-do task set U, combined with each subtask u λ The urgency and resource requirements of the subtask are evaluated and its priority is evaluated. The subtask priority calculation formula is as follows:

[0147]

[0148] Among them, α1 and α2 are weight coefficients, V(u λ ) is the subtask u λ The degree of relationship with other tasks, H(u λ ) is the subtask u λ The remaining execution time;

[0149] (2) Sort by subtask priority and calculate subtask u λ Resource utilization Φ(u λ ), the calculation formula is as follows:

[0150]

[0151] Where ζ is the full-time equivalent number, The maximum execution time allowed for the task;

[0152] (3) Make a judgment: If Φ(u λ )≤ε, then generate the task plan U' according to the atomic task priority, otherwise downgrade the current subtask priority and reorder it, where ε is the known resource utilization;

[0153] (4) Output the generated task plan U'.

[0154] In step S401 of this embodiment, assuming that the weight coefficients of the subtask priority calculation formula are α1=0.6 and α2=0.4, and the relationship between subtasks u1 (industry dynamics), u2 (work report), and u3 (work report) and other tasks are V(u1)=0.85, V(u2)=0.75, and V(u3)=0.8 respectively; the remaining execution time of subtasks u1, u2, and u3 is H(u1)=3 (hours), H(u2)=2 (hours), and H(u3)=4 (hours), respectively, then the calculation result of each subtask priority is:

[0155]

[0156] Sort by the priority of each task, from high to low, and the priority order is u1, u2, u3;

[0157] Assume that the full-time equivalent number ζ = 1, that is, each employee works full-time, and the maximum allowed task execution time is Then each task u λ Resource utilization Φ(u λ ) is calculated as:

[0158]

[0159] Given that resource utilization ε = 0.65, subtasks u1 and u2 need to be downgraded. The task plan U' generated by the automatic task flow generation system is {"work record", "industry dynamics", "work report"}. The specific generated content is as follows: Figure 5 shown.

[0160] Step S402: Record user task information in real time and monitor task execution status according to the task sequence and resource allocation in the task plan, and store the task execution status in the database;

[0161] Step S403: According to the task execution status, an abnormality detection value is introduced to analyze whether the task execution status is abnormal. The abnormality detection value judgment method is as follows:

[0162]

[0163] Among them, u λ The λth subtask planned for the current task;

[0164] In step S403, the task execution status is defined and the operations performed are:

[0165] (1) The task is executed normally (A(u λ )=0): Continue to complete the current subtask;

[0166] (2) Insufficient existing resources (A(u λ )=1): You can apply for extended resources or wait for resources to be released and try again;

[0167] (3) Task timeout (A(u λ )=2): You can choose to retry the task or extend the task duration;

[0168] (4) Execution failed (A(u λ )=3): Repair or retry;

[0169] (5) Data error (A(u λ)=4): Check input data and repair data format;

[0170] (6) Dependency failure (A(u λ )=5): Re-execute the predecessor subtask;

[0171] (7) Network anomaly (A(u λ )=6): Check the network connection and try again after it is restored.

[0172] In step S403 of this embodiment, Figure 6 As shown, taking A(u1)=0, A(u2)=1, A(u3)=5 as an example, through abnormality detection, it is judged that subtask u1 is executed normally; subtask u2 cannot continue to execute due to insufficient resources; subtask u3 is recommended to re-execute its predecessor subtask before execution.

[0173] Step S404: Output abnormality feedback.

[0174] Step S5 specifically includes the following steps:

[0175] Step S501: The user terminal provides scoring data for evaluating the automatic generation effect of the current task process, including the accuracy of the task process, resource utilization efficiency and exception handling effect;

[0176] Step S502: Based on the scoring data, a perplexity scoring function is constructed to generate feedback data;

[0177] In step S502, the perplexity scoring function is:

[0178]

[0179] in, is the normalized result of the n-th item score, F n is the scoring result of the nth item, σ is the highest score of the scoring item, and C is the total number of scoring items.

[0180] In step S502 of this embodiment, assuming that the highest score for the scoring item is 5 points, the user terminal scores the accuracy of the task flow, resource utilization efficiency, and exception handling effect with 4 points, 3 points, and 5 points, respectively, for the automatic generation effect of the current task flow. The normalized result of each score is:

[0181]

[0182] The perplexity score results are:

[0183]

[0184] Step S503: Update and save the current task status, and store the feedback information in the database. Figure 7 shown.

[0185] In this embodiment, the system receives text input from the user, captures the user's intention through word segmentation and slot extraction, performs word segmentation on the text information in the task pool, builds an inverted index, and quickly screens out possible candidate tasks. Through deep semantic analysis and dynamic resource scheduling, the task plan is automatically generated, and abnormal situations are monitored during the task, timely feedback and iterative optimization are carried out, and the feedback information is stored in the database for continuous improvement. In addition, through a multi-step, multi-technology integration approach, it is possible to efficiently and accurately understand the user's intention information, generate and optimize the task process, have good scalability, adapt to the needs of task generation of different types and scales, and provide a more flexible, reliable and efficient task management solution, which is of great significance to improving task management efficiency and user satisfaction.

[0186] It should be appreciated that embodiments of the present invention can be implemented or practiced by computer hardware, a combination of hardware and software, or by computer instructions stored in a non-transitory computer-readable memory. The methods can be implemented in a computer program using standard programming techniques, including a non-transitory computer-readable storage medium configured with a computer program, wherein the storage medium so configured causes the computer to operate in a specific and predefined manner according to the methods and drawings described in the specific embodiments. Each program can be implemented in a high-level procedural or object-oriented programming language to communicate with the computer system. However, if desired, the program can be implemented in assembly or machine language. In any case, the language can be a compiled or interpreted language. In addition, the program can be run on a programmed application-specific integrated circuit for this purpose.

[0187] Furthermore, the operations of the processes described herein may be performed in any suitable order unless otherwise indicated herein or otherwise clearly contradicted by the context. The processes described herein (or variations and / or combinations thereof) may be performed under the control of one or more computer systems configured with executable instructions and may be implemented as code (e.g., executable instructions, one or more computer programs, or one or more applications) that is executed collectively on one or more processors, by hardware, or a combination thereof. The computer program includes a plurality of instructions that can be executed by one or more processors.

[0188] Further, the methods can be implemented in any type of computing platform that is operably connected to a suitable computer, including but not limited to a personal computer, a minicomputer, a mainframe, a workstation, a network or distributed computing environment, a separate or integrated computer platform, or in communication with a charged particle tool or other imaging device, etc. Various aspects of the present invention can be implemented as machine-readable code stored on a non-transitory storage medium or device, whether removable or integrated into a computing platform, such as a hard disk, an optical read and / or write storage medium, RAM, ROM, etc., so that it can be read by a programmable computer, and when the storage medium or device is read by the computer, it can be used to configure and operate the computer to perform the processes described herein. In addition, the machine-readable code, or portions thereof, can be transmitted over a wired or wireless network. When such media includes instructions or programs that implement the steps described above in conjunction with a microprocessor or other data processor, the invention described herein includes these and other different types of non-transitory computer-readable storage media. When programmed according to the methods and techniques of the present invention, the present invention also includes the computer itself.

[0189] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solutions, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for automatically generating task flows based on intention recognition, characterized by: The method comprises the following steps: Step S1: Receive text input from the user, segment the text, extract slot information containing intent, and generate a corresponding natural language response; Step S2: Perform word segmentation on the text information in the task pool and construct an inverted index to obtain a set of candidate task information containing intent; Step S2 specifically includes: Step S201: Segment all text information in the task pool P to obtain p i ={t i1 , t i2, …, t iq }, where i represents the i-th task information, and q represents the task information p i The number of words in Step S202: traverse all task information in the task pool P, and update the index of each word in each task information, and build an inverted index Q (t ij )={p i | t ij ∈p i }, returns the word containing the word t ij The set of all task information, where j represents the task information p i The jth word in ; Step S203: Based on the text information T input by the user, search the task information containing the word from the task pool P through the inverted index to generate a candidate task information set D = {d1, d2, …, d s }, where s represents the number of segmentations in the candidate task information; Step S3: Based on the user input text information and the semantic information of the candidate task information, the task matching degree of each candidate task is evaluated and sorted in descending order to obtain a to-do task set containing the intent information; Step S3 specifically includes: Step S301: The semantic information of all words in the text information T and the candidate task information set D are aggregated using the average pooling method to obtain feature vectors containing intent information. , the average pooling formula is as follows: in, is the kth word t in the text information T k The corresponding word vector, is the first Participle The corresponding word vector; Step S302: Based on the feature vector of the text information T and the candidate task information set D , calculate the task matching degree of each candidate task in the candidate task set D ; Step S303: According to task matching , for each candidate task Arrange in descending order to obtain the sorted candidate task set R={r1, r2, …, r s }, the calculation formula for descending order is as follows: Step S304: Based on the sorted candidate task set R, recommend the first w tasks with the highest task matching degree, and generate a to-do task set U={u1, u2, …, u w }; Step S4: Based on the set of pending tasks and the existing resource situation, an adaptive task scheduling method is constructed, an abnormality detection value is introduced, and feedback information is output; Step S4 specifically includes: Step S401: constructing an adaptive task scheduling method based on the task contents of the to-do task set U and the usage of existing resources to automatically generate a task plan U'; Step S402: Record user task information in real time and monitor task execution status according to the task sequence and resource allocation in the task plan, and store the task execution status in the database; Step S403: According to the task execution status, an abnormality detection value is introduced to analyze whether the task execution status is abnormal. The abnormality detection value judgment method is as follows: Among them, u λ The λth subtask planned for the current task; Step S404: output abnormal feedback; Step S5: Receive task process scoring data from the user, construct a perplexity scoring function, obtain feedback data and update the task status, and store the feedback information in a database. Step S5 specifically includes: Step S501: The user terminal provides scoring data for evaluating the automatic generation effect of the current task process, including the accuracy of the task process, resource utilization efficiency and exception handling effect; Step S502: constructing a perplexity scoring function based on the scoring data and generating feedback data; Step S503: Update and save the current task status, and store the feedback information in the database.

2. The method for automatically generating task flows based on intention recognition according to claim 1, characterized in that: The step S1 specifically includes: Step S101: receiving input text from the user terminal, performing word segmentation on the text information T, and obtaining text information T = {t1, t2, …, t m }, where m is the number of words in the text information T; Step S102: extracting slot information related to the user's intention from the segmented text information T, and performing named entity annotation on the input text information T; Step S103: Generate a corresponding natural language response for the extracted slot information according to predefined rules, and return it to the user end through the automatic generation system.

3. The method for automatically generating task flows based on intention recognition according to claim 1, characterized in that: In step S302, the task matching degree of the candidate task The calculation method is: (1) Obtain the feature vectors of the text information T and the candidate task information set D , build the task matching function: in, is the feature vector W of the text information T T The transpose of (2) Calculate the original matching degree of the candidate task based on the task matching function , and normalize the matching degree of all candidate tasks to obtain the task matching degree of the candidate tasks , the calculation method is: (3) Output the task matching degree of candidate tasks .

4. The method for automatically generating task flows based on intention recognition according to claim 1, characterized in that: The specific process of the adaptive task scheduling method in step S401 is as follows: (1) According to the task content of the to-do task set U, combined with each subtask The urgency and resource requirements of the subtask are evaluated and its priority is evaluated. The subtask priority calculation formula is as follows: Among them, α1 and α2 are weight coefficients, For subtasks The degree of relationship with other tasks, For subtasks The remaining execution time; (2) Sort by subtask priority and calculate subtasks Resource utilization , the calculation formula is as follows: in is the full-time equivalent number, The maximum execution time allowed for the task; (3) Make a judgment: If , then generate the task plan U' according to the atomic task priority, otherwise downgrade the current subtask priority and reorder it, where ε is the known resource utilization; (4) Output the generated task plan U'.

5. The method for automatically generating task flows based on intention recognition according to claim 1, characterized in that: The task execution status definition and execution operations in step S403 are: (1) The task is executed normally (A(u λ )=0): Continue to complete the current subtask; (2) Insufficient existing resources (A(u λ )=1): You can apply for additional resources or wait for resources to be released and try again; (3) Task timeout (A(u λ )=2): You can choose to retry the task or extend the task duration; (4) Execution failed (A(u λ )=3): Repair or retry; (5) Data error (A(u λ )=4): Check input data and fix data format; (6) Dependency failure (A(u λ )=5): Re-execute the predecessor subtask; (7) Network anomaly (A(u λ )=6): Check the network connection and try again after it is restored.

6. The method for automatically generating task flows based on intention recognition according to claim 1, characterized in that: The perplexity scoring function in step S502 is: in, is the normalized result of the n-th item score, F n is the scoring result of the nth item, σ is the highest score of the scoring item, and C is the total number of scoring items.

Images