Semantic recognition and task scheduling method and system for digital employees

Abstract

The invention discloses a semantic recognition and task scheduling method and system for digital employees, and relates to the field of artificial intelligence. The method comprises the following steps: preprocessing a natural language text input by a user to generate a preprocessed text; performing semantic analysis on the preprocessed text by adopting a semantic analysis model, and extracting text key information; performing intention recognition on the text key information by using a deep learning model to generate an intention recognition result; according to the intention recognition result, mapping the user request to a specific business process based on a preset business rule base, and generating a business classification result; and according to a service classification result, dynamically allocating resources required by task execution by adopting a task scheduling algorithm, and adjusting a task scheduling strategy in real time according to monitoring data. According to the scheme, through accurate semantic recognition, intelligent business process mapping and dynamic task scheduling, digital employees can efficiently understand user intentions and quickly and accurately complete task execution, and the intelligent level and the response capability are improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method and system for semantic recognition and task scheduling of digital employees. Background Art

[0002] With the rapid development of artificial intelligence technology, digital employees are playing an increasingly important role in many industries and are widely used in customer service, data processing, smart assistants and other fields. Digital employees can not only simulate certain human work functions to handle repetitive tasks, but also reduce the burden of manpower and improve work efficiency. However, existing digital employees still have some shortcomings in semantic recognition and task scheduling.

[0003] In terms of semantic recognition, although many existing natural language processing (NLP) technologies have been able to achieve basic semantic understanding, in practical applications, digital employees often find it difficult to accurately grasp the complex intentions of users. This is because the semantics expressed by users often have problems such as ambiguity, polysemy, and context dependence. This limitation makes it impossible for digital employees to provide accurate feedback or perform effective tasks when interacting with users. In terms of task scheduling, current algorithms often use static rules or simple priority sorting, which makes it difficult to dynamically adapt to user needs and environmental changes. This results in many digital employees being unable to dynamically allocate tasks according to real-time changing conditions and resource conditions after receiving user requests, resulting in resource waste or task delays. For example, when the priority of a task changes or a new task is introduced, digital employees fail to adjust their scheduling strategies in a timely manner, affecting overall work efficiency.

[0004] Given the shortcomings of existing digital employees in semantic recognition and task scheduling, it is particularly important to develop a method that can accurately recognize semantics and perform intelligent task scheduling. Summary of the invention

[0005] The purpose of the present invention is to provide a method and system for semantic recognition and task scheduling of digital employees, which can improve the ability of digital employees to understand user intentions and execute tasks, and enhance the intelligence level and response speed through efficient semantic recognition and dynamic task scheduling.

[0006] To achieve the above object, the present invention discloses the following technical solution:

[0007] A first aspect of the present invention provides a method for semantic recognition and task scheduling of digital employees, the method comprising the following steps:

[0008] Preprocess the natural language text input by the user to generate preprocessed text;

[0009] Using a semantic analysis model based on a vector library to perform semantic analysis on the preprocessed text to extract key information of the text;

[0010] Using a deep learning model to perform intent recognition on the key information of the text and generate an intent recognition result;

[0011] According to the intention recognition result, based on the preset business rule library, the user request is mapped to a specific business process to generate a corresponding business classification result;

[0012] According to the business classification results, a task scheduling algorithm is used to dynamically allocate resources required for executing tasks, and the task execution process is monitored, and the task scheduling strategy is adjusted in real time according to the monitoring data.

[0013] Optionally, in the above-mentioned semantic recognition and task scheduling method, the preprocessing of the natural language text input by the user includes word segmentation, part-of-speech tagging, grammatical parsing, text cleaning, stem extraction and / or vectorization.

[0014] Optionally, the semantic recognition and task scheduling method described above, wherein the semantic analysis model based on the vector library is used to perform semantic analysis on the preprocessed text to extract key information of the text, comprises:

[0015] Build a vector library based on the pre-collected corpus to generate word vectors or sentence vectors;

[0016] Similarity calculation is performed between the words or sentences in the preprocessed text and the word vectors or sentence vectors in the vector library to identify and extract key information in the preprocessed text.

[0017] Optionally, in the above-mentioned semantic recognition and task scheduling method, the step of using a deep learning model to perform intent recognition on the key information of the text and generating an intent recognition result includes:

[0018] Using a bidirectional long short-term memory network to model the key information of the text, capture forward and backward context information and output a context feature vector;

[0019] Using a convolutional neural network to process the key information of the text, extract local features and output a local feature vector;

[0020] Integrate the context feature vector output by the bidirectional long short-term memory network layer with the local feature vector output by the convolutional neural network layer to form a comprehensive feature representation, and input the comprehensive feature representation to the intent classification layer;

[0021] The comprehensive feature representation is nonlinearly transformed through an activation function to generate an intent recognition result.

[0022] Optionally, in the above-mentioned semantic recognition and task scheduling method, the business rule library includes:

[0023] Business type key features, used to store key feature information corresponding to different business types, including business type keywords, business type key words and / or business type phrases;

[0024] The business type classification standard is used to define classification rules for different business types, including keyword frequency, feature word weight, syntactic structure characteristics, sentiment analysis score, context relevance and / or user characteristics.

[0025] Optionally, the semantic recognition and task scheduling method described above, mapping the user request to a specific business process based on the intent recognition result and a preset business rule library to generate a corresponding business classification result, includes:

[0026] Use the random forest classification algorithm to build multiple decision trees to perform preliminary business classification on the intention recognition results. Each decision tree T i The construction of is based on the following formula:

[0027] T i =buildTree(S,m,d) (1),

[0028] Among them, S represents the sample set, which contains the user's intention recognition results, m represents the number of features randomly selected when building the decision tree, and d represents the maximum depth of the tree;

[0029] The output of each decision tree is combined with the output of other decision trees, and the preliminary classification result is determined by the majority voting mechanism. The classification result C is calculated using the following formula:

[0030] C=majorityVote(T1(x),T2(x),……,T n (x)) (2),

[0031] Where x represents the input vector and n represents the number of decision trees;

[0032] Matching the preliminary classification result with the key features of the business type in the business rule library to determine the business process corresponding to the user request;

[0033] Based on the matched business process and in combination with the business type classification standard, a corresponding business classification result is generated.

[0034] Optionally, in the semantic recognition and task scheduling method, dynamically allocating resources required for executing tasks using a task scheduling algorithm according to the business classification result includes:

[0035] Based on the Q-Learning algorithm, the action value function Q(s, a) is used to evaluate the expected reward of selecting the scheduling action a under the task state s. The action value function is adjusted by the following update rule:

[0036] Q(s t ,a t )←Q(s t ,a t )+α[r t+1 +γmax a′ Q(s t+1 ,a′)-Q(s t ,a t )](3),

[0037] Among them, s t Indicates the current task status, a t Indicates that in the current task state s t The scheduling action taken under t+1 Indicates the execution of scheduling action a t The immediate reward after s, α represents the learning rate, γ represents the discount factor, and a′ represents the state transfer to s t+1 The best action to be chosen later;

[0038] In the current task state t Execute the scheduling action a t After that, according to the immediate reward r t+1 The optimal action a′ after state transfer updates the value of the action value function Q(s, a), optimizes the scheduling strategy, and dynamically allocates the resources required to execute the task based on the optimized scheduling strategy.

[0039] Optionally, in the semantic recognition and task scheduling method, the monitoring task execution process and adjusting the task scheduling strategy in real time according to the monitoring data include:

[0040] Monitor the task execution process and dynamically adjust the task priority through weight factors based on the monitoring data;

[0041] The weight factor is adjusted using an adaptive weight adjustment mechanism:

[0042] Based on the monitoring data and the adjusted priority queue, the task scheduling strategy is adjusted in real time, including changing the execution order of tasks, reallocating resources and / or adjusting the time window of tasks.

[0043] A second aspect of the present invention provides a digital employee semantic recognition and task scheduling system, the system comprising:

[0044] A preprocessing module, used to preprocess the natural language text input by the user to generate preprocessed text;

[0045] An information extraction module, used to perform semantic analysis on the preprocessed text using a semantic analysis model based on a vector library to extract key information of the text;

[0046] An intention recognition module is used to use a deep learning model to perform intention recognition on the key information of the text and generate an intention recognition result;

[0047] A business classification module is used to map the user request to a specific business process based on the intent recognition result and a preset business rule library to generate a corresponding business classification result;

[0048] The strategy adjustment module is used to dynamically allocate resources required for executing tasks using a task scheduling algorithm according to the business classification results, monitor the task execution process, and adjust the task scheduling strategy in real time according to the monitoring data.

[0049] According to a third aspect of the present invention, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, a method for semantic recognition and task scheduling of digital employees as described in the first aspect is implemented.

[0050] The effects provided in the content of the invention are only the effects of the embodiments, not all the effects of the invention. One of the above technical solutions has the following advantages or beneficial effects:

[0051] The application provides a method and system for semantic recognition and task scheduling of digital employees. First, the natural language text input by the user is preprocessed, and combined with a semantic analysis model based on a vector library and a deep learning model, the accurate understanding and intent recognition of the user's semantics is achieved; next, based on the intent recognition result, the user request is matched with the preset business rule library, and the complex user needs are accurately mapped to the specific business process; then, through the task scheduling algorithm, the resources required to execute the task are dynamically allocated according to the business classification results, thereby improving the efficiency of task execution; at the same time, through real-time monitoring of the task execution process and adjustment of the scheduling strategy, it can respond to various changes in execution in a timely manner and adapt to more complex application scenarios. Through precise semantic recognition, intelligent business process mapping and dynamic task scheduling, this solution enables digital employees to efficiently understand user intentions, match appropriate business processes, and quickly and accurately complete task execution, thereby improving the intelligence level and responsiveness of digital employees. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0053] Figure 1 A flowchart of a method for semantic recognition and task scheduling of digital employees according to an embodiment of the present application;

[0054] Figure 2 for Figure 1 A flowchart of a method for implementing step S120;

[0055] Figure 3 for Figure 1 A flowchart of a method for implementing step S130;

[0056] Figure 4 for Figure 1 A flowchart of a method for implementing step S140;

[0057] Figure 5 A schematic diagram of the structure of a digital employee semantic recognition and task scheduling system according to an embodiment of the present application;

[0058] Figure 6 This is a schematic diagram of the structure of an electronic device according to an embodiment of the present application;

[0059] Reference numerals:

[0060] 510-preprocessing module, 520-information extraction module, 530-intention recognition module, 540-service classification module, 550-strategy adjustment module;

[0061] 610 - input unit, 620 - memory, 630 - processor, 640 - output unit. DETAILED DESCRIPTION

[0062] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0063] It should be noted that references to "one embodiment", "embodiment", "example embodiment", etc. in this specification refer to the embodiment being described which may include specific features, structures or characteristics, but not every embodiment must include these specific features, structures or characteristics. In addition, such expressions do not refer to the same embodiment. Furthermore, when describing specific features, structures or characteristics in conjunction with an embodiment, whether or not there is an explicit description, it is indicated that incorporating such features, structures or characteristics into other embodiments is within the knowledge of those skilled in the art.

[0064] In addition, certain words are used in the specification and subsequent claims to refer to specific components or parts. Those with ordinary knowledge in the relevant field should understand that manufacturers can use different nouns or terms to refer to the same component or part. This specification and subsequent claims do not use differences in names as a way to distinguish components or parts, but use differences in the functions of components or parts as the criteria for distinction. "Including" and "including" mentioned throughout the specification and subsequent claims are open-ended terms and should be interpreted as "including but not limited to". In addition, the word "connected" here includes any direct and indirect electrical connection means. Indirect electrical connection means include connection through other devices.

[0065] refer to Figure 1 , Figure 1 A flowchart of a method for semantic recognition and task scheduling of digital employees provided in an embodiment of the present application is shown, and the method includes the following steps:

[0066] S110, preprocessing the natural language text input by the user to generate a preprocessed text;

[0067] In some embodiments, the natural language text input by the user is preprocessed, including word segmentation, part-of-speech tagging, grammatical parsing, text cleaning, stemming and / or vectorization.

[0068] S120, using a semantic analysis model based on a vector library to perform semantic analysis on the preprocessed text to extract key information of the text;

[0069] In some embodiments, the vector library can be constructed based on a pre-collected corpus (such as general language text, professional field literature, conversation data, etc.), and the text information in the corpus is converted into a vector representation in a high-dimensional vector space through a word embedding method.

[0070] S130, using a deep learning model to perform intent recognition on key information of the text and generate an intent recognition result;

[0071] Using deep learning models to identify the intent of key information in text can enable the system to have a higher level of semantic understanding ability and provide more accurate information for subsequent task scheduling and execution.

[0072] S140, according to the intention recognition result, based on the preset business rule library, mapping the user request to a specific business process, and generating a corresponding business classification result;

[0073] In this step, complex user requests are converted into specific business process operations, allowing digital employees to efficiently classify and process tasks in complex business scenarios.

[0074] S150. According to the business classification results, a task scheduling algorithm is used to dynamically allocate resources required for executing tasks, and the task execution process is monitored, and the task scheduling strategy is adjusted in real time according to the monitoring data.

[0075] The method of this embodiment enables digital employees to efficiently understand and process user requests through precise semantic recognition, intelligent business process mapping, and dynamic task scheduling, thereby improving the accuracy of semantic understanding. At the same time, complex user requests are quickly mapped to corresponding business processes using a preset business rule library, thereby optimizing task classification and processing efficiency. In addition, through a dynamic task scheduling algorithm, the system can flexibly allocate and monitor resources, thereby improving resource utilization efficiency. This method enables digital employees to quickly respond to user needs in a complex business environment, reduces manual intervention, improves resource utilization efficiency and the smoothness of task execution, and improves user experience and service quality.

[0076] In some embodiments, in the above step S110, the natural language text input by the user is preprocessed, and the preprocessing process includes but is not limited to:

[0077] Word segmentation, which divides continuous natural language text into separate words or phrases;

[0078] Part-of-speech tagging, marking each word with its part of speech (such as noun, verb, adjective, etc.);

[0079] Syntax parsing, identifying the subject, predicate, object and other components of a sentence, and then constructing a grammatical tree for the sentence;

[0080] Text cleaning to remove irrelevant noise information, such as punctuation, special characters, stop words, etc.

[0081] Stemming, which reduces words to their base form to unify different variations of the same word;

[0082] Vectorization converts the processed text into a numerical form that the computer can understand.

[0083] Through this preprocessing step, the natural language text input by the user can be effectively converted into structured information, providing data support for subsequent semantic analysis and intent recognition.

[0084] In some embodiments, in the above step S120, using a semantic analysis model based on a vector library to perform semantic analysis on the preprocessed text to extract key information of the text further includes:

[0085] S121. Build a vector library based on the pre-collected corpus to generate a word vector or a sentence vector;

[0086] S122, calculating the similarity between the words or sentences in the preprocessed text and the word vectors or sentence vectors in the vector library, and determining the similarity between the words or sentences in the preprocessed text and the known vectors in the vector library;

[0087] S123. Identify and extract key information in the preprocessed text through similarity calculation.

[0088] Through this step, key information with high semantic relevance can be extracted from the preprocessed text, providing accurate input data for subsequent intent recognition and task scheduling, ensuring the efficiency and accuracy of the overall processing flow.

[0089] In some embodiments, in the above step S130, using a deep learning model to perform intent recognition on key text information to generate an intent recognition result further includes:

[0090] S131, using a bidirectional long short-term memory network to model the key information of the text, capture forward and backward context information and output a context feature vector;

[0091] The bidirectional long short-term memory network (Bi-LSTM) processes the input sequence in two directions (forward and backward), which can effectively capture the association information between previous and next words. The generated context feature vector enhances the understanding of the overall meaning of the sentence.

[0092] S132, using a convolutional neural network to process key information of the text, extract local features and output a local feature vector;

[0093] Convolutional Neural Networks (CNNs) can extract local features from text, identify important phrases or patterns, generate and output local feature vectors, and thus more accurately capture the important information and structure in sentences.

[0094] S133, integrating the context feature vector output by the bidirectional long short-term memory network layer and the local feature vector output by the convolutional neural network layer to form a comprehensive feature representation, and inputting the comprehensive feature representation to the intent classification layer;

[0095] By integrating the contextual feature vector and the local feature vector, the semantic features of the text can be fully reflected, providing more accurate information for subsequent intent recognition.

[0096] S134, performing nonlinear transformation on the comprehensive feature representation through an activation function to generate an intent recognition result;

[0097] The activation function (such as ReLU or Sigmoid) is used to perform nonlinear transformation on the comprehensive feature representation, which further enhances the ability to capture complex patterns and improves the accuracy of intent recognition.

[0098] The deep learning technology in this step can achieve efficient and accurate intent recognition, improving the performance of digital employees in understanding user needs and responding capabilities.

[0099] In some embodiments, in the above step S140, the preset business rule library includes:

[0100] Business type key features, used to store key feature information corresponding to different business types, including business type keywords (such as refund, complaint, etc.), business type keywords (such as accounting processing, tax declaration, etc.) and / or business type phrases (such as accounting issues, technical support requests, etc.);

[0101] Business type classification standards are used to define classification rules for different business types, including keyword frequency, feature word weight, syntactic structure characteristics, sentiment analysis scores, context relevance (such as operation details, preventive measures, etc.) and / or user characteristics (such as user preferences, historical requests, etc.).

[0102] In this step, according to the intent recognition results and based on the preset business rule library, the user request is mapped to a specific business process and the corresponding business classification results are generated, including:

[0103] S141. Use the random forest classification algorithm to construct multiple decision trees and perform preliminary business classification on the intent recognition results;

[0104] Specifically, each decision tree T i The construction of is based on the following formula:

[0105] T i =buildTree(S,m,d) (1),

[0106] Among them, S represents the sample set, which contains the user's intention recognition results, m represents the number of features randomly selected when building the decision tree, and d represents the maximum depth of the tree. Each decision tree T i During construction, a subset of features and samples will be randomly extracted to reduce the risk of overfitting. The random forest classification algorithm can effectively learn the relationship between different features and intents, providing data support for subsequent business classification.

[0107] S142, combining the output of each decision tree with the outputs of other decision trees, and determining a preliminary classification result through a majority voting mechanism;

[0108] Each decision tree T i According to the input user intent, the classification result C is obtained through the majority voting mechanism of all decision trees:

[0109] C=majorityVote(T1(x),T2(x),……,T n (x)) (2),

[0110] Here, x represents the input vector and n represents the number of decision trees. This step ensures that even if a single decision tree makes an error in its prediction, the overall model can still produce a relatively accurate classification result. For example, for the input “what should I do if there is no tax on the plate”, different decision trees may output different classifications, but the final majority vote will determine the preliminary classification result.

[0111] S143, matching the preliminary classification result with the key features of the business type in the business rule library to determine the business process corresponding to the user request;

[0112] For example, if "What to do if there is no tax plate" is initially classified as "tax consultation", this classification will be compared with the key features of the relevant business types in the business rule library to confirm the specific business process corresponding to this request.

[0113] S144. Based on the matched business process and in combination with the business type classification standard, generate corresponding business classification results;

[0114] Based on the determined business process and business type classification standards, the system will generate the final business classification results. For example, for the input "What should I do if I don't have a tax disk?", it may be classified as "Tax U shield / certificate problem" under "Tax consultation" so that the user's request can be accurately processed and responded to.

[0115] By combining this step with the classification capabilities of the random forest algorithm, user requests can be efficiently and accurately mapped to specific business processes, improving digital employees' ability to understand and respond to user needs.

[0116] In some embodiments, in the above step S150, according to the service classification result, a task scheduling algorithm is used to dynamically allocate resources required for executing the task, including:

[0117] Based on the Q-Learning algorithm, the action value function Q(s, a) is used to evaluate the expected reward of selecting scheduling action a under task state s. By evaluating the value of selecting a specific action under different task states, the decision of task scheduling is guided. The expression of the action value function is adjusted by the following update rule:

[0118] Q(s t ,a t )←Q(s t ,a t )+α[r t+1 +γmax a′ Q(s t+1 ,a′)-Q(st ,a t )](3),

[0119] Among them, s t Indicates the current task status, such as s t = {task type, task urgency}, used to describe the situation and urgency of the current task; a t Indicates that in the current task state s t The scheduling action taken under t = {scheduling policy}, used to assign tasks to specific resources or adjust task priorities; r t+1 Indicates the execution of scheduling action a t The immediate reward obtained after s is used to reflect the effect of this scheduling decision in the current task state; α represents the learning rate, which is used to control the influence of new information on the Q value update; γ represents the discount factor, which is used to reflect the importance of future rewards to the current decision; a′ represents the time when the state is transferred to s t+1 The best action is then selected.

[0120] In the current task state t Execute the scheduling action a t After that, according to the immediate reward r t+1 The optimal action a′ after the state transition is obtained, and the value of the action value function Q is updated, so that the system can continuously adjust and optimize the scheduling strategy under each state through experience accumulation. For example, if a certain task scheduling brings high immediate rewards (such as shortening the task completion time or improving resource utilization), the corresponding Q value will increase, making the system more inclined to repeat this scheduling strategy under similar states. Conversely, if a certain scheduling leads to low immediate rewards (such as task delays or unreasonable resource allocation), the corresponding Q value will decrease, and the system will try to explore other scheduling strategies. By continuously iteratively updating Q(s, a), the scheduling strategies under different task states can be gradually optimized. After a lot of training and updating, the Q-Learning algorithm can identify which scheduling strategies can bring higher cumulative rewards under various task states. Finally, these optimized scheduling strategies are applied to actual tasks to achieve dynamic resource allocation, ensuring rapid response when tasks are urgent, and efficient use of resources when there are fewer tasks.

[0121] In some embodiments, in the above step S150, monitoring the task execution process and adjusting the task scheduling strategy in real time according to the monitoring data include:

[0122] During the task execution process, continuously monitor the execution status, resource usage and task completion progress of each task, including task completion time, resource consumption and the backlog of unfinished tasks in the task queue;

[0123] According to the monitoring data, the priority of the task is dynamically adjusted. The priority calculation formula is:

[0124] P i =ω e E i +ω r R i (4)

[0125] Among them, P i represents the priority of task i, ω e and ω r represents the weight factor, E i Indicates the degree of urgency, R i Indicates resource demand. The weight factor can be adjusted according to actual conditions to ensure that the system can respond in time when high-priority tasks occur.

[0126] Adopting the adaptive weight adjustment mechanism, the weight factor is adjusted by the following formula:

[0127]

[0128] Among them, β is the adjustment step size, E avg and R avg They are the average urgency and resource requirements of all tasks in the current time period. Through the adaptive weight adjustment mechanism, the system can dynamically adapt to the fluctuations of task urgency and resource requirements and allocate tasks more flexibly;

[0129] Combined with monitoring data and adjusted priority queues, the task scheduling strategy is dynamically updated, including changing the execution order of tasks, reallocating task resources, or adjusting the time window for task allocation, thereby optimizing the overall task execution efficiency.

[0130] In some embodiments, after adjusting the task scheduling strategy, it is also possible to further evaluate whether the adjusted strategy is effective by monitoring the task execution effect (such as task completion rate, resource utilization rate or user satisfaction, etc.), and continuously optimize the scheduling strategy based on the feedback results. For example, if it is found that the new scheduling strategy performs better under a certain state, it will be incorporated into future scheduling decisions. During the monitoring process, if it is found that some resources are overloaded or underutilized, load balancing can be performed in a timely manner to reallocate tasks to underutilized resources to avoid resource waste or task delays, so as to improve the overall system performance. Through the monitoring and adjustment mechanism in the above steps, the scheduling strategy can be continuously optimized during the task execution process to ensure that the task is reasonably and efficiently scheduled under different loads, thereby improving the task completion efficiency and resource utilization rate.

[0131] In this embodiment, through precise semantic recognition and intelligent task scheduling, efficient processing of user requests in complex business scenarios is achieved. The deep learning model is used to improve the ability to understand user intentions, and the dynamically adjusted scheduling strategy ensures the reasonable allocation of resources and the rapid response of tasks. At the same time, the adaptive weight adjustment mechanism can flexibly respond to changes in task urgency and resource requirements, effectively improving the response efficiency and task processing capabilities of the overall system.

[0132] It should be noted that, for the method embodiments, for the sake of simplicity, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the embodiments of the present application are not limited by the described order of actions, because according to the embodiments of the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions involved are not necessarily required by the embodiments of the present application.

[0133] refer to Figure 5 , Figure 5 The following is a schematic diagram of the structure of a digital employee semantic recognition and task scheduling system provided by an embodiment of the present application. The system can be used to implement digital employee semantic recognition and task scheduling. The system described below can correspond to the digital employee semantic recognition and task scheduling method described above. Specifically, the system includes:

[0134] The preprocessing module 510 is used to preprocess the natural language text input by the user to generate a preprocessed text;

[0135] The information extraction module 520 is used to perform semantic analysis on the preprocessed text using a semantic analysis model based on a vector library to extract key information of the text;

[0136] The intention recognition module 530 is used to use a deep learning model to perform intention recognition on key information of the text and generate an intention recognition result;

[0137] The business classification module 540 is used to map the user request to a specific business process based on the intent recognition result and a preset business rule library, and generate a corresponding business classification result;

[0138] The strategy adjustment module 550 is used to dynamically allocate resources required for executing tasks using a task scheduling algorithm according to the business classification results, monitor the task execution process, and adjust the task scheduling strategy in real time according to the monitoring data.

[0139] For the processes not described in detail in the digital employee semantic recognition and task scheduling system in this embodiment, reference may be made to the relevant parts of the above-mentioned digital employee semantic recognition and task scheduling method embodiment, which will not be repeated here.

[0140] In addition, the present application also provides an electronic device, such as Figure 6 As shown, the electronic device may include an input unit 610, a memory 620, a processor 630 and an output unit 640. Among them, the memory 620 stores program instructions that can be run on the processor 630, and the processor 630 calls the program instructions to execute the methods and / or technical solutions based on the aforementioned multiple embodiments. The electronic device can be a mobile terminal device such as a mobile phone or a computer. For the processes not described in detail in the electronic device in this embodiment, reference can be made to the relevant parts in the above-mentioned digital employee semantic recognition and task scheduling method embodiment, which will not be repeated here.

[0141] The above is a detailed introduction to a method and system for semantic recognition and task scheduling of digital employees provided by the present application. The various embodiments in the specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same and similar parts between the embodiments can be referenced to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description. It should be pointed out that for ordinary technicians in this field, without departing from the principles of the present application, several improvements and modifications can be made to the present application, and these improvements and modifications also fall within the scope of protection of the claims of the present application.

[0142] Obviously, those skilled in the art should understand that the above modules or steps of the present application can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, and optionally, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. In this way, the present application is not limited to any specific combination of hardware and software.

[0143] Although the present application has been described in detail above with general descriptions and specific embodiments, it is obvious to those skilled in the art that some modifications or improvements may be made to the present application. Therefore, these modifications or improvements made without departing from the spirit of the present application are within the scope of protection claimed by the present application.

Claims

1. A method for semantic recognition and task scheduling of digital employees, characterized in that: The following steps are involved: Preprocess the natural language text input by the user to generate preprocessed text; Using a semantic analysis model based on a vector library to perform semantic analysis on the preprocessed text to extract key information of the text; Using a deep learning model to perform intent recognition on the key information of the text and generate an intent recognition result; According to the intention recognition result, based on the preset business rule library, the user request is mapped to a specific business process to generate a corresponding business classification result; According to the business classification results, a task scheduling algorithm is used to dynamically allocate resources required for executing tasks, and the task execution process is monitored, and the task scheduling strategy is adjusted in real time according to the monitoring data.

2. The semantic recognition and task scheduling method according to claim 1, characterized in that: The preprocessing of the natural language text input by the user includes word segmentation, part-of-speech tagging, grammatical parsing, text cleaning, stem extraction and / or vectorization.

3. The semantic recognition and task scheduling method according to claim 2, characterized in that: The semantic analysis model based on the vector library is used to perform semantic analysis on the preprocessed text to extract key information of the text, including: Build a vector library based on the pre-collected corpus to generate word vectors or sentence vectors; Similarity calculation is performed between the words or sentences in the preprocessed text and the word vectors or sentence vectors in the vector library to identify and extract key information in the preprocessed text.

4. The semantic recognition and task scheduling method according to claim 3, characterized in that: The using of the deep learning model to perform intent recognition on the key information of the text and generate an intent recognition result includes: Using a bidirectional long short-term memory network to model the key information of the text, capture forward and backward context information and output a context feature vector; Using a convolutional neural network to process the key information of the text, extract local features and output a local feature vector; Integrate the context feature vector output by the bidirectional long short-term memory network layer with the local feature vector output by the convolutional neural network layer to form a comprehensive feature representation, and input the comprehensive feature representation to the intent classification layer; The comprehensive feature representation is nonlinearly transformed through an activation function to generate an intent recognition result.

5. The semantic recognition and task scheduling method according to any one of claims 1 to 4, characterized in that: The business rule library includes: Business type key features, used to store key feature information corresponding to different business types, including business type keywords, business type key words and / or business type phrases; The business type classification standard is used to define classification rules for different business types, including keyword frequency, feature word weight, syntactic structure characteristics, sentiment analysis score, context relevance and / or user characteristics.

6. The semantic recognition and task scheduling method according to claim 5, characterized in that: According to the intention recognition result, based on the preset business rule library, the user request is mapped to a specific business process, and a corresponding business classification result is generated, including: Use the random forest classification algorithm to build multiple decision trees to perform preliminary business classification on the intention recognition results. Each decision tree T i The construction of is based on the following formula: T i =buildTree(S,m,d) (1), Among them, S represents the sample set, which contains the user's intention recognition results, m represents the number of features randomly selected when building the decision tree, and d represents the maximum depth of the tree; The output of each decision tree is combined with the output of other decision trees, and the preliminary classification result is determined by the majority voting mechanism. The classification result C is calculated using the following formula: C=majorityVote(T1(x),T2(x),……,T n (x)) (2), Where x represents the input vector and n represents the number of decision trees; Matching the preliminary classification result with the key features of the business type in the business rule library to determine the business process corresponding to the user request; Based on the matched business process and in combination with the business type classification standard, a corresponding business classification result is generated.

7. The semantic recognition and task scheduling method according to claim 6, characterized in that: The dynamically allocating resources required for executing tasks using a task scheduling algorithm according to the business classification result includes: Based on the Q-Learning algorithm, the action value function Q(s, a) is used to evaluate the expected reward of selecting the scheduling action a under the task state s. The action value function is adjusted by the following update rule: Q(s t ,a t )←Q(s t ,a t )+α[r t+1 +γmax a′ Q(s t+1 ,a′)-Q(s t ,a t )](3), Among them, s t Indicates the current task status, a t Indicates that in the current task state s t The scheduling action taken under t+1 Indicates the execution of scheduling action a t The immediate reward after s, α represents the learning rate, γ represents the discount factor, and a′ represents the state transfer to s t+1 The best action to be chosen later; In the current task state t Execute the scheduling action a t After that, according to the immediate reward r t+1 The optimal action a′ after state transfer updates the value of the action value function Q(s, a), optimizes the scheduling strategy, and dynamically allocates the resources required to execute the task based on the optimized scheduling strategy.

8. The semantic recognition and task scheduling method according to claim 7, characterized in that: The monitoring task execution process adjusts the task scheduling strategy in real time according to the monitoring data, including: Monitor the task execution process and dynamically adjust the task priority through weight factors based on the monitoring data; The weight factor is adjusted using an adaptive weight adjustment mechanism: Based on the monitoring data and the adjusted priority queue, the task scheduling strategy is adjusted in real time, including changing the execution order of tasks, reallocating resources and / or adjusting the time window of tasks.

9. A digital employee semantic recognition and task scheduling system, characterized in that: The system comprises: A preprocessing module, used to preprocess the natural language text input by the user to generate preprocessed text; An information extraction module, used to perform semantic analysis on the preprocessed text using a semantic analysis model based on a vector library to extract key information of the text; An intention recognition module is used to use a deep learning model to perform intention recognition on the key information of the text and generate an intention recognition result; A business classification module is used to map the user request to a specific business process based on the intent recognition result and a preset business rule library to generate a corresponding business classification result; The strategy adjustment module is used to dynamically allocate resources required for executing tasks using a task scheduling algorithm according to the business classification results, monitor the task execution process, and adjust the task scheduling strategy in real time according to the monitoring data.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, a method for semantic recognition and task scheduling of a digital employee according to any one of claims 1 to 8 is implemented.

Images