Student transaction allocation and service evaluation method and system based on multi-modal hosting

Abstract

The invention provides a student transaction allocation and service evaluation method and system based on multi-modal hosting, and relates to the technical field of intelligent transaction scheduling and service evaluation, and the method comprises the steps: extracting semantic features in a student transaction request, fusing the semantic features with transaction type information to generate a transaction feature vector, constructing a transaction time sequence association graph, and obtaining a transaction time sequence association graph; identifying a processing dependency relationship, and forming a transaction family and a processing priority sequence thereof; combining skill vectors of the processing personnel, and realizing an optimal matching path by utilizing self-adaptive dynamic programming; and finally, dynamic adjustment and feedback updating are performed according to the matching result and the service score, so that the transaction processing efficiency and the service matching accuracy can be effectively improved.

Description

Technical Field

[0001] The present invention relates to intelligent transaction scheduling and service evaluation technologies, and in particular to a method and system for student affair allocation and service evaluation based on multi-modal hosting. Background Art

[0002] The current management modes of student hosting services generally rely on manual intervention or allocation strategies based on fixed rules, lacking the ability to deeply understand transaction content and dynamically adjust, and it is difficult to adapt to the personalized and diverse hosting transaction requests of students. Facing complex and changeable transaction types and processing requirements, traditional systems are difficult to efficiently extract transaction semantic features, unable to achieve structured modeling of the complete information of transactions, affecting the response efficiency and processing accuracy of the overall service.

[0003] Existing technologies generally adopt static matching or simple priority mechanisms in the process of transaction scheduling and allocation, failing to fully consider the potential dependency relationships and resource competition situations among transactions, resulting in problems such as conflicts, repetitions, or delays in task execution. At the same time, the allocation of the capabilities of processing personnel also fails to accurately match the transaction characteristics, lacking a matching mechanism based on multi-stage optimization, and it is difficult to effectively balance service quality and resource utilization rate, hindering the refined upgrade of the digital governance of hosting services.

[0004] With the development of multi-modal information fusion, graph structure modeling, and intelligent scheduling algorithms, there is an urgent need to construct a comprehensive transaction management method that can integrate transaction semantic features, dependency relationships, processing capabilities, and historical performance, support the automatic clustering, intelligent scheduling, and service evaluation closed-loop feedback of student affairs, so as to improve the intelligent level and service quality of hosting transaction processing. Therefore, there is an urgent need to propose a method for student affair allocation and service evaluation based on multi-modal hosting. Summary of the Invention

[0005] Embodiments of the present invention provide a method and system for student affair allocation and service evaluation based on multi-modal hosting, which can solve the problems in the prior art.

[0006] In the first aspect of the embodiments of the present invention, A method for student affair allocation and service evaluation based on multi-modal hosting is provided, including: Receiving student affair request data, extracting a transaction description from the student affair request data and performing semantic analysis to extract semantic features, and fusing the semantic features with a transaction type identifier to generate a transaction feature vector containing complete transaction information; Construct a transaction time - series association graph based on transaction feature vectors, calculate the feature propagation between nodes in the transaction time - series association graph through recursive depth feature extraction, generate a processing dependency probability matrix, construct a transaction clustering evaluation function based on the processing dependency probability matrix, generate transaction families through iterative optimization clustering of the transaction clustering evaluation function, and generate a processing priority sequence for the transaction families according to the constraint strength in the processing dependency probability matrix; Construct a skill vector based on the professional skills of the processing personnel, input the skill vector and the transaction feature vectors in the transaction family into the adaptive dynamic programming algorithm, obtain the optimal matching path through multi - stage decision optimization, and generate a comprehensive matching score according to the optimal matching path score and the historical service score; Assign the transaction family and the processing priority sequence to the target processing personnel with the highest comprehensive matching score, and update the professional skill vector and the historical service score of the target processing personnel according to the processing evaluation score of the transaction family for subsequent matching calculations of transaction assignment.

[0007] In an alternative embodiment, Extract the transaction description from the student transaction request data and perform semantic analysis to extract semantic features, and fuse the semantic features with the transaction type identifier to generate a transaction feature vector containing complete transaction information, including: Extract the transaction description from the student transaction request data, decompose the transaction description at multiple levels according to core information, attribute information, and environmental information, extract transaction target and requirement features from the core information, extract transaction timeliness and urgency features from the attribute information, and extract processing scenario and resource requirement features from the environmental information to obtain multi - level transaction features; Perform a correlation analysis on the multi - level transaction features, calculate the feature recombination coefficient based on the information integrity and information redundancy between features, and combine transaction features with a complementarity greater than a preset complementarity threshold according to the feature recombination coefficient to generate transaction semantic features; Establish a mapping rule library for transaction type identifiers, store the corresponding relationship and confidence level between transaction type identifiers and transaction semantic features in the mapping rule library, perform feature matching in the mapping rule library according to the transaction semantic features to obtain the probability distribution of transaction type identifiers, and combine the transaction semantic features with the transaction type identifier with the highest confidence level to generate a transaction feature vector containing complete transaction information.

[0008] In an alternative embodiment, Construct a transaction time - series association graph based on transaction feature vectors, calculate the feature propagation between nodes in the transaction time - series association graph through recursive depth feature extraction, and generate a processing dependency probability matrix, including: Construct a transaction time - series association graph based on transaction feature vectors, where nodes represent transactions and the connecting edges between nodes represent the time - series associations between transactions; Establish a conflict transaction marking library based on historical processing data. The conflict transaction marking library records the types of transaction combinations that generate resource competition. Match the nodes in the transaction time - series association graph with the conflict transaction marking library to identify node pairs with resource competition and generate a set of competing nodes; For the node pairs in the set of competing nodes, extract the resource requirement information of each node, calculate the resource overlap degree between the node pairs, and determine the resource competition intensity as the competition edge weight between the competing node pairs according to the resource overlap degree; For the remaining node pairs in the transaction time - series association graph, extract the time interval information and transaction feature vectors between the nodes, and calculate the time - series similarity between the node pairs as the association edge weight between the remaining node pairs; Combine the competition edge weight and the association edge weight to form complete edge weight information. Use the complete edge weight information to guide the recursive propagation of node features, update the node features to obtain the propagated node features, and calculate the processing dependency probability between nodes based on the propagated node features to generate a processing dependency probability matrix.

[0009] In an alternative embodiment, Construct a transaction clustering evaluation function based on the processing dependency probability matrix. Generate transaction families by iteratively optimizing the clustering of the transaction clustering evaluation function, and generate a processing priority sequence for the transaction families according to the constraint strength in the processing dependency probability matrix, including: Construct a transaction clustering evaluation function based on the processing dependency probability matrix. The transaction clustering evaluation function includes an intra - family processing dependency constraint term and an inter - family processing dependency constraint term; Extract the historical execution records from the processing dependency probability matrix, calculate the execution duration fluctuation coefficient of the transactions according to the historical execution records, construct an execution stability matrix and input it into the transaction clustering evaluation function to perform initial clustering on the transactions to obtain an initial set of transaction families; Calculate the difference in the execution duration fluctuation coefficients of the transaction pairs in the initial set of transaction families. Determine the processing dependency constraint weight and the fluctuation compensation coefficient of the transaction pairs in the transaction clustering evaluation function according to the difference in the execution duration fluctuation coefficients, and perform iterative optimization on the set of transaction families. When the evaluation value converges, obtain the final set of transaction families; Select transactions with an execution duration fluctuation coefficient less than a preset fluctuation threshold from the final set of transaction families to construct a benchmark execution sequence. Calculate the interference degree of the remaining transaction pairs with respect to the benchmark execution sequence according to the constraint strength in the processing dependency probability matrix, determine the optimal insertion positions of the remaining transactions, and generate a transaction family processing priority sequence that satisfies the processing dependency constraints according to the benchmark execution sequence and the optimal insertion positions.

[0010] In an alternative embodiment, Generating a transaction family processing priority sequence that satisfies processing dependency constraints based on the reference execution sequence and the optimal insertion position includes: Construct an initial constraint edge set based on the reference execution sequence, calculate the transfer constraint strength between nodes in the initial constraint edge set using the shortest path algorithm, and supplement the transfer constraints with transfer constraint strength greater than the preset strength threshold to the initial constraint edge set to obtain a complete constraint edge set; Perform hierarchical partitioning on the complete constraint edge set based on the in-degree and out-degree of the nodes to obtain a constraint graph that includes inter-layer constraint relationships and intra-layer constraint relationships; Obtain the deviation value between the actual execution duration and the expected duration of the transaction, the number of constraint violations, and the degree of competition, calculate the performance score of the transaction based on the deviation value, the number of violations, and the degree of competition, perform weighted combination of the performance score and the basic priority to obtain the updated transaction priority, and determine the set of affected nodes affected by the updated priority in the constraint graph; Adjust the priorities of the nodes in the set of affected nodes in sequence, calculate the length of the dependency chain between nodes based on the constraint graph, extract the optimized nodes with the dependency chain length greater than the preset length threshold, reorder the optimized nodes to obtain multiple reordering schemes, and select the reordering scheme with the minimum processing delay as the target reordering scheme; Compare the target reordering scheme with the historical priority sequence, and when the improvement amplitude exceeds the preset improvement threshold, update the transaction family processing priority sequence using the target reordering scheme.

[0011] In an alternative embodiment, Construct a skill vector based on the professional skills of the processing personnel, input the skill vector and the transaction feature vector in the transaction family into the adaptive dynamic programming algorithm, obtain the optimal matching path through multi-stage decision optimization, and generate a comprehensive matching score based on the optimal matching path score and the historical service score, including: Obtain the professional qualification characteristics, work experience characteristics, and skill coverage characteristics of the processing personnel, and generate a professional skill vector of the processing personnel through weighted combination based on preset weights; Calculate the matching degree between the professional skill vector and the transaction feature vector in the transaction family to obtain a basic matching value, calculate the satisfaction degree of the skill items for the transaction features based on the basic matching value, and generate an initial state value of the state space; Construct a skill transfer matrix to represent the complementary relationship between skills, calculate the state transition probability based on the skill transfer matrix, and combine the state transition probability with the initial state value to construct the state space; Obtain the service completion quality and processing timeliness rate of the processing personnel, calculate the historical service score, adjust the state value of the state space based on the historical service score to construct a state value function, perform policy iteration calculation on the state value function to obtain a state value sequence, determine the optimal state transition path based on the state value sequence, and generate an optimal matching path; Calculate the state transition benefit of the optimal matching path to obtain a path score, and generate a comprehensive matching score by combining the path score and the historical service score according to the weights determined by the transaction characteristics.

[0012] In an alternative embodiment, Performing policy iteration calculation on the state value function to obtain a state value sequence, determining the optimal state transition path based on the state value sequence, and generating an optimal matching path includes: Obtain the immediate benefit and discount factor of the state, construct the Bellman equation, update the current state value by calculating the expected value after state transition to obtain the initial state value sequence, calculate the state transition probability under the initial matching policy based on the initial state value sequence, adjust the temperature parameter according to the value improvement amplitude of the historical matching data, and update the state transition probability based on the temperature parameter; Starting from the current state, calculate the state values of all transferable states according to the updated state transition probability, select the transfer state with the largest state value and add it to the matching path, and determine whether the matching path meets any of the conditions of the target state, path length limit, or state value benefit threshold. When it is satisfied, mark the matching path as a candidate matching path; Calculate the state values that can be transferred in multiple steps in the future for each state node in the candidate matching path, select the transfer state with the largest multi-step state value to update the candidate matching path, and obtain the optimized matching path; Compare the optimized matching path with the historical optimal matching path. When the state value is larger, update the historical optimal matching path, set the lower limit of the state value based on the historical optimal matching path, and remove the branches below the lower limit of the state value to obtain the final optimal matching path.

[0013] In the second aspect of the embodiments of the present invention, Provide a student affair allocation and service evaluation system based on multi-modal hosting, including: A first unit, configured to receive student affair request data, extract a transaction description from the student affair request data, perform semantic analysis to extract semantic features, and perform feature fusion on the semantic features and the transaction type identifier to generate a transaction feature vector containing complete transaction information; A second unit, configured to receive student affair request data, extract affair descriptions from the student affair request data, perform semantic analysis to extract semantic features, fuse the semantic features with affair type identifiers, and generate affair feature vectors containing complete affair information; A third unit, configured to construct a skill vector based on the professional skills of handlers, input the skill vector and the affair feature vectors in an affair family into an adaptive dynamic programming algorithm, obtain an optimal matching path through multi-stage decision optimization, and generate a comprehensive matching score according to the optimal matching path score and historical service scores; A fourth unit, configured to assign the affair family and the processing priority sequence to the target handler with the highest comprehensive matching score, update the professional skill vector and historical service scores of the target handler according to the processing evaluation scores of the affair family, for subsequent matching calculations of affair assignments.

[0014] In a third aspect of the embodiments of the present invention, There is provided an electronic device, including: A processor; A memory for storing instructions executable by the processor; Wherein, the processor is configured to call the instructions stored in the memory to execute the method described above.

[0015] In a fourth aspect of the embodiments of the present invention, There is provided a computer-readable storage medium, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, the method described above is implemented.

[0016] In this embodiment, intelligent recognition, dynamic clustering, precise allocation, and continuous optimization of student affairs can be achieved. By constructing affair feature vectors and temporal association graphs, the system can not only identify potential dependency relationships between affairs, but also evaluate resource competition and processing priorities, solving the problems of frequent affair processing conflicts and fragmented processes in the traditional mode. Using the adaptive dynamic programming algorithm to match handlers with affair families enables the allocation decision to take into account both ability fitness and historical service performance, significantly improving the professionalism and stability of affair processing. Compared with the existing methods that rely on manual judgment or static rules, the present application provides a scheduling strategy that integrates semantic understanding, graph structure analysis, and path optimization, breaking through the contradiction between affair complexity and service personalization, realizing the transformation of student affair processing from extensive to intelligent, and improving the overall response efficiency and user satisfaction of the trusteeship service system. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a flowchart of the method for student affair allocation and service evaluation based on multi-modal trusteeship according to the embodiments of the present invention; Figure 2This is the heat map of the processing dependency probability matrix in the embodiments of the present invention; Figure 3 This is the comparison graph of the convergence process of the state value function policy iteration in the embodiments of the present invention; Figure 4 This is the schematic structural diagram of the student affairs allocation and service evaluation system based on multimodal hosting in the embodiments of the present invention. Specific Embodiments

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0019] The technical solutions of the present invention will be described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments.

[0020] Figure 1 This is the flow schematic diagram of the method for student affairs allocation and service evaluation based on multimodal hosting in the embodiments of the present invention, as Figure 1 shown, the method includes: Receiving student affairs request data, extracting the transaction description from the student affairs request data, performing semantic analysis to extract semantic features, and fusing the semantic features with the transaction type identifier to generate a transaction feature vector containing complete transaction information; Constructing a transaction time series association graph based on the transaction feature vector, calculating the feature propagation between nodes in the transaction time series association graph through recursive depth feature extraction, generating a processing dependency probability matrix, constructing a transaction clustering evaluation function based on the processing dependency probability matrix, iteratively optimizing the clustering of the transaction clustering evaluation function to generate transaction families, and generating a processing priority sequence for the transaction families according to the constraint strength in the processing dependency probability matrix; Constructing a skill vector based on the professional skills of the processing personnel, inputting the skill vector and the transaction feature vectors in the transaction families into the adaptive dynamic programming algorithm, obtaining the optimal matching path through multi-stage decision optimization, and generating a comprehensive matching score according to the optimal matching path score and the historical service score; Allocating the transaction families and the processing priority sequence to the target processing personnel with the highest comprehensive matching score, and updating the professional skill vector and the historical service score of the target processing personnel according to the processing evaluation score of the transaction families for subsequent matching calculations in transaction allocation.

[0021] In an alternative embodiment, Extract the transaction description from the student affairs request data, perform semantic analysis to extract semantic features, fuse the semantic features with the transaction type identifier, and generate a transaction feature vector containing complete transaction information, including: Extract the transaction description from the student affairs request data, decompose the transaction description at multiple levels according to core information, attribute information, and environmental information, extract transaction target and requirement features from the core information, extract transaction timeliness and urgency features from the attribute information, and extract processing scenario and resource requirement features from the environmental information to obtain multi-level transaction features; Perform a correlation analysis on the multi-level transaction features, calculate a feature recombination coefficient based on the information completeness and information redundancy between features, and combine transaction features with a complementarity greater than a preset complementarity threshold according to the feature recombination coefficient to generate transaction semantic features; Establish a mapping rule library for transaction type identifiers. The mapping rule library stores the corresponding relationship and confidence level between transaction type identifiers and transaction semantic features. Perform feature matching in the mapping rule library according to the transaction semantic features to obtain the probability distribution of transaction type identifiers, and combine the transaction semantic features with the transaction type identifier with the highest confidence level to generate a transaction feature vector containing complete transaction information.

[0022] Exemplarily, first, it is necessary to extract transaction description information from the transaction request data submitted by students. The transaction request data can include multimodal data such as text descriptions, pictures, and audio. The transaction description usually comes from the free text information submitted by students or parents on the platform, and the content covers the purpose of the transaction application, specific demands, time requirements, and execution background, etc. After preliminary word segmentation and syntactic parsing of the transaction description, each piece of text is divided according to content, and the information is divided into three categories: core information, attribute information, and environmental information. The core information mainly includes the request intention and target verb phrase, such as "apply to change the trusteeship course", "feedback on the course quality problem", or "apply for early pick-up and drop-off", etc.; the attribute information indicates the time range and urgency of the request, such as "process before tomorrow afternoon" or "must be completed today", etc.; the environmental information reflects the background and constraint conditions of the transaction, such as "due to the parent's temporary business trip and no one to pick up and drop off" or "the student is not adapted to the current course content", etc.

[0023] Feature extraction is performed on the above three types of information respectively. For the core information, the verb-object structure is determined through syntactic dependency analysis to identify specific transaction objectives and their corresponding demands, thereby forming transaction objective features and demand features; for attribute information, the timeliness level of the transaction is extracted by identifying the description time limit, time range, and qualifier. For example, "immediately" and "today" are classified as high timeliness, and "before next week" is classified as medium timeliness; at the same time, if the text contains words such as "must" and "urgent", it is correspondingly marked as a transaction with a high urgency level. Through the analysis of location words, causal conjunctions, and background description words in the environmental information, the scene elements and resource dependencies related to transaction processing are extracted. For example, information such as "lack of caregivers", "lack of stationery materials", and "need to arrange psychological counseling" is abstracted into resource demand features and processing environment features.

[0024] After the multi-level transaction features extracted are stored in a unified structured manner, all features are compared pairwise to construct a feature association network. In this process, the information redundancy and complementary relationship between different transaction features are dynamically analyzed. If there is an obvious complementary relationship between the contents described by two features, for example, one describes "class time conflict" and the other describes "hoping to switch the trusteeship course", it is regarded as a high complementary relationship. The system assigns a combination priority to each group of features. According to the set complementary threshold, only the feature pairs with a complementary degree greater than this threshold are retained and combined into new transaction semantic feature units. This combination fuses the context information through the semantic information reconstruction method to maintain semantic consistency and integrity. For example, "needing to apply for course adjustment due to family changes" is formed by fusing the two original features of "family changes" and "course adjustment demand".

[0025] A mapping rule library of transaction type identifiers and semantic features is established. In the rule library, each mapping rule contains a set of semantic features, the corresponding transaction type identifier, and its confidence level. For example, in a certain rule, when the semantic features include "lack of caregivers" and "urgently need to arrange temporary trusteeship", the corresponding transaction type is "urgent trusteeship arrangement" with a high confidence level. The mapping rule library is constructed through historical data training and manual rule setting methods, covering common transaction scenarios in the trusteeship service, including various types such as regular adjustment transactions, emergency replacement transactions, feedback and suggestion transactions, and course evaluation transactions. After receiving the structured transaction semantic features, the system automatically searches in the rule library for the transaction type identifier that best matches the feature set and returns a list of matching results sorted by confidence level. The confidence level is comprehensively evaluated based on the feature occurrence frequency, matching accuracy, and rule weight.

[0026] Select the item with the highest confidence level from the matching results as the type identifier of this transaction, and jointly encode it with the semantic features of this transaction to generate a complete transaction feature vector.

[0027] The eigenvector contains the complete information of the transaction, retaining both the original semantic features and endowing clear type attributes. This multi-level feature extraction and fusion method can make full use of the information in multi-modal data to improve the accuracy of transaction understanding. At the same time, through the dynamic update of the rule base, the system can continuously optimize the feature mapping effect and improve the accuracy of transaction classification.

[0028] In terms of calculating the information integrity, a calculation method based on semantic similarity is adopted. First, a feature word vector space is established, and each feature is represented in vector form. By calculating the cosine similarity between vectors, the semantic correlation degree between features is obtained. When the semantic similarity between two features is low but the information complementarity is strong, it indicates that the combination of these two features can provide a more complete transaction description.

[0029] The calculation of information redundancy is based on the overlap analysis of feature content. By extracting the keyword items in the features, the repeated occurrence frequency of the items is calculated. When two features contain a large number of identical or similar word items, it indicates the existence of information redundancy. According to the preset redundancy threshold, the feature content that needs to be merged or deleted is screened out.

[0030] The calculation of the feature recombination coefficient comprehensively considers two factors: information integrity and redundancy. When the complementarity between features is strong and the redundancy is low, the recombination coefficient is high, indicating that these features are suitable for combination. For example, there is a strong complementary relationship between the maintenance target feature and the timeliness feature, and after combination, it can better express the urgency of the transaction.

[0031] The mapping rule base of transaction type identification adopts a dynamic update mechanism. After each transaction is processed, the system will optimize the rule base according to the actual processing results. If a certain feature template has a high matching degree with the actual situation, the confidence level of the corresponding rule will be increased; otherwise, the confidence level will be decreased. This adaptive learning mechanism enables the rule base to be continuously improved and the accuracy of feature mapping to be improved.

[0032] In the feature fusion stage, the system adopts a weighted combination method to generate the final transaction eigenvector. Different types of features will be assigned different weights. For example, the weight of core information is usually higher than that of environmental information. The setting of weights is based on the importance and reliability of features and can be optimized and adjusted through historical data analysis.

[0033] In this embodiment, by decomposing the transaction description at multiple levels, the core objectives, attribute features, and environmental conditions are systematically divided, enhancing the ability to capture and distinguish multi-dimensional information of transactions. Combining the correlation analysis and recombination between features, the expression of semantic features is optimized, redundant information interference is reduced, and the accuracy and integrity of transaction semantics are improved. The mapping rule library is used to achieve the dynamic matching and confidence evaluation of transaction type identifiers and semantic features, effectively improving the accuracy and reliability of transaction classification. Finally, the generated transaction feature vector completely reflects the semantic information and type attributes of the transaction, facilitating subsequent intelligent scheduling, resource allocation, and decision support, and significantly enhancing the intelligent level and response efficiency of the student affairs processing system.

[0034] In an alternative embodiment, Based on the transaction feature vector, a transaction time-series association graph is constructed. By recursively extracting deep features, the feature propagation between nodes in the transaction time-series association graph is calculated, and the generated processing dependency probability matrix includes: Based on the transaction feature vector, a transaction time-series association graph is constructed, where the nodes represent transactions, and the connecting edges between the nodes represent the time-series associations between transactions; A conflict transaction marking library is established based on historical processing data. The conflict transaction marking library records the types of transaction combinations that generate resource competition. The nodes in the transaction time-series association graph are matched with the conflict transaction marking library to identify node pairs with resource competition and generate a set of competing nodes; For the node pairs in the set of competing nodes, the resource requirement information of each node is extracted, the resource overlap degree between the node pairs is calculated, and the resource competition intensity is determined according to the resource overlap degree as the competition edge weight between the competing node pairs; For the remaining node pairs in the transaction time-series association graph, the time interval information and transaction feature vector between the nodes are extracted, and the time-series similarity between the node pairs is calculated as the association edge weight between the remaining node pairs; The competition edge weight and the association edge weight are combined to form complete edge weight information. The complete edge weight information is used to guide the recursive propagation of node features, update the node features to obtain the propagated node features, and calculate the processing dependency probability between nodes based on the propagated node features to generate a processing dependency probability matrix.

[0035] Exemplarily, after receiving multiple student affairs encoded by feature vectors, a transaction time-series association graph is constructed based on these transaction feature vectors. Each node in this graph represents an independent student affair, and the request content and context attributes are represented by the transaction feature vectors carried inside the node. The edges between transactions are used to represent the chronological order formed during the execution of historical transactions. During the graph construction process, the system establishes an initial edge set according to the transaction submission time order. At the same time, in combination with the actual task chain relationships existing in the hosting service scenario, such as "applying for course replacement" needs to be executed after "course scheduling is completed", directional connections are established between transaction nodes with clear chronological dependencies to form a complete transaction time-series association graph.

[0036] Based on historical processing records, a conflict transaction marking library is established. This library contains combinations of transaction types that cause conflicts due to overlapping resource allocations, such as "course replacement application" and "course arrangement review", "temporary hosting application" and "teacher schedule adjustment", etc. After the transaction time-series association graph is constructed, all node pairs in the graph are retrieved and compared with the conflict transaction marking library in turn to identify node pairs with resource competition. The identified transaction node pairs are included in the competition node set, and the resource characteristics of each pair of transactions in this set are compared.

[0037] The resource requirement information of each transaction node comes from the resource fields contained in its feature vector. Keywords such as teacher requirements, venue occupancy, and time period usage are extracted from it, and a unified resource vector is constructed. For any node pair, compare the content of their resource vectors, identify the resource items that are reused, count the number and intensity of overlapping items, and generate a resource overlap degree evaluation for this node pair. The higher the resource overlap degree value, the stronger the competition for platform resources by this transaction combination. Based on this, the edge weight of this competition edge is set. The edge weight value range is set by configuration parameters. For example, complete overlap is recorded as the maximum weight, and partial overlap is assigned proportionally.

[0038] In addition to the competition nodes, calculate the time correlation between the remaining node pairs in the graph. Extract the submission time, expected processing completion time, and service coverage time period of two transactions to obtain time interval information from them. Combining the semantic dimensions of the transaction feature vectors, comprehensively calculate the transaction target similarity, time period overlap situation, and processing window consistency to form a time-series similarity score for the node pair. The time-series similarity takes into account the proximity in time and the similarity of transaction characteristics. The time-series similarity serves as the association edge weight between these non-competition node pairs. The competition edge weights and association edge weights are uniformly processed to form complete edge weight information. These weight information guide the propagation process of node features in the graph. Feature propagation is carried out recursively. Each node will receive feature information from adjacent nodes and update the features according to the edge weights.

[0039] During the feature propagation process, the edge weights determine the intensity of feature transfer. Among the node pairs with larger competing edge weights, the feature transfer will be inhibited, indicating that these transactions need to stagger their processing times. Among the node pairs with larger associated edge weights, the feature transfer is more sufficient, which is conducive to the centralized processing of similar transactions.

[0040] Based on the propagated node features, the system calculates the processing dependence probability between nodes. The processing dependence probability reflects the processing order constraint relationship between transactions. The higher the dependence probability, the more likely the processing result of the previous transaction will affect the processing process of the subsequent transaction. Finally, the system organizes the processing dependence probabilities between all node pairs into a matrix form to generate a processing dependence probability matrix. This processing dependence probability matrix provides an important basis for subsequent transaction clustering and priority sorting. The probability values in the matrix consider both the processing constraints brought by resource competition and reflect the temporal correlation between transactions, and can guide the system to generate a reasonable transaction processing plan.

[0041] In this embodiment, by constructing a transaction time sequence association graph and introducing a historical conflict transaction mark library, it is possible to effectively discover transaction combinations that may have resource competition and improve the conflict avoidance ability during transaction concurrent processing. Further, by combining the resource overlap degree and the time sequence similarity to construct a differentiated edge weight, the association relationship modeling between transactions is more hierarchical and targeted. Through the feature recursive propagation mechanism guided by the complete edge weight, the evolution process of the dependence relationship between transactions can be dynamically captured, and then a highly credible processing dependence probability matrix can be generated, providing accurate dependence information support for transaction clustering, priority sorting, and scheduling optimization, and significantly improving the intelligence and response efficiency of the overall system.

[0042] In an alternative embodiment, Construct a transaction clustering evaluation function based on the processing dependence probability matrix, and generate transaction families through iterative optimization clustering of the transaction clustering evaluation function, and generate a processing priority sequence for the transaction families according to the constraint strength in the processing dependence probability matrix, including: Construct a transaction clustering evaluation function based on the processing dependence probability matrix, and the transaction clustering evaluation function includes an intra-family processing dependence constraint term and an inter-family processing dependence constraint term; Extract the historical execution records in the processing dependence probability matrix, calculate the execution duration fluctuation coefficient of the transactions according to the historical execution records, construct an execution stability matrix and input it into the transaction clustering evaluation function to perform initial clustering on the transactions to obtain an initial set of transaction families; Calculate the difference in the execution duration fluctuation coefficients of the transaction pairs in the initial set of transaction families, determine the processing dependence constraint weight and the fluctuation compensation coefficient of the transaction pairs in the transaction clustering evaluation function according to the difference in the execution duration fluctuation coefficients, and perform iterative optimization on the set of transaction families. When the evaluation value converges, obtain the final set of transaction families; Select transactions with an execution duration fluctuation coefficient less than a preset fluctuation threshold from the set of final transaction families to construct a benchmark execution sequence, calculate the interference degree of the remaining transactions on the benchmark execution sequence according to the constraint strength in the processing dependency probability matrix, determine the optimal insertion positions of the remaining transactions, and generate a transaction family processing priority sequence that satisfies the processing dependency constraints based on the benchmark execution sequence and the optimal insertion positions.

[0043] Exemplarily, first, it is necessary to construct a transaction clustering evaluation function according to the transaction processing dependency probability matrix. This evaluation function consists of two main parts: the intra-family processing dependency constraint term and the inter-family processing dependency constraint term. The intra-family processing dependency constraint term is used to measure the strength of the processing dependency relationship between transactions within the same transaction family, and the inter-family processing dependency constraint term is used to measure the strength of the processing dependency relationship between transactions in different transaction families. Each element in the processing dependency probability matrix represents the processing dependency probability between the corresponding two transactions, and the value range is from 0 to 1. 0 means that the two transactions are completely independent without a dependency relationship, and 1 means that the two transactions have a mandatory processing dependency order. Taking 10 transactions as an example, in the generated processing dependency probability matrix, the processing dependency probability between transaction 1 and transaction 2 is 0.8, indicating that when transaction 1 is executed, there is an 80% probability that it needs to rely on the processing result of transaction 2.

[0044] Next, extract the historical execution record information from the processing dependency probability matrix, and calculate the fluctuation coefficient according to the historical execution duration of each transaction. The execution duration fluctuation coefficient reflects the degree of transaction execution stability, and the smaller the fluctuation coefficient, the more stable the execution duration. The specific calculation method is to calculate the ratio of the standard deviation of the execution duration of each transaction in the historical execution records to the average value. For example, for a certain transaction in the past 100 executions, the average execution duration is 10 seconds and the standard deviation is 2 seconds, then its execution duration fluctuation coefficient is 0.2. Construct an execution stability matrix based on the calculated execution duration fluctuation coefficients of each transaction.

[0045] Input the execution stability matrix into the transaction clustering evaluation function, and use the spectral clustering method to perform initial clustering on the transactions to obtain an initial set of transaction families. The spectral clustering method first constructs a transaction similarity matrix, then calculates the eigenvalues and eigenvectors of the matrix, and finally divides the transactions into different families based on the eigenvectors. The construction of the similarity matrix comprehensively considers two factors: the processing dependency probability and the execution stability. Taking the above 10 transactions as an example, through spectral clustering, 3 initial transaction families may be obtained, containing 3, 4, and 3 transactions respectively.

[0046] For each pair of transactions in the initial set of transaction families, calculate the difference in the execution duration fluctuation coefficients between them. The larger the difference in the fluctuation coefficients, the greater the difference in the execution stability of the two transactions. Based on the difference in the fluctuation coefficients, determine the processing dependency constraint weight and the fluctuation compensation coefficient in the clustering evaluation function for each pair of transactions. The processing dependency constraint weight is inversely proportional to the difference in the fluctuation coefficients, that is, the smaller the constraint weight for a pair of transactions with a greater difference in execution stability. The fluctuation compensation coefficient is directly proportional to the difference in the fluctuation coefficients and is used to balance the impact brought by the difference in execution stability. For example, when the difference in the fluctuation coefficients of a pair of transactions is 0.3, its processing dependency constraint weight can be set to 0.7, and the fluctuation compensation coefficient can be set to 0.3.

[0047] Based on the updated constraint weights and fluctuation compensation coefficients, perform iterative optimization on the set of transaction families. In each iteration, first calculate the evaluation value of the current transaction family partition, then try to migrate transactions between different families, and select the migration plan that can reduce the evaluation value. The calculation of the evaluation value needs to consider the processing dependency constraints of the transactions within the family, the processing dependency constraints between the transactions of different families, and the fluctuation compensation of the execution stability. When the change amplitude of the evaluation value in consecutive multiple iterations is less than the preset threshold, it is considered that the evaluation value converges, and the final set of transaction families is obtained.

[0048] In the final set of transaction families, select the transactions with execution duration fluctuation coefficients less than the preset fluctuation threshold to construct the benchmark execution sequence. Assume that the set fluctuation threshold is 0.15, then the transactions with fluctuation coefficients less than this threshold will be preferentially considered for inclusion in the benchmark execution sequence. The execution durations of these transactions are relatively stable and are suitable as the benchmark for task scheduling. For the remaining transactions not included in the benchmark execution sequence, it is necessary to calculate their interference degrees on the benchmark execution sequence according to the constraint strength in the processing dependency probability matrix.

[0049] The calculation of the interference degree considers the processing dependency probabilities between the transaction to be inserted and each transaction in the benchmark sequence. The specific calculation method is to perform a weighted sum of the processing dependency probabilities between the transaction to be inserted and each transaction in the benchmark sequence, and the weights gradually decrease as the position of the transaction in the benchmark sequence gets further back. For example, the processing dependency probabilities of a certain transaction to be inserted with the first 3 transactions in the benchmark sequence are 0.6, 0.4, and 0.2 respectively, and using position weights of 0.5, 0.3, and 0.2, the calculated interference degree is 0.44.

[0050] Based on the calculated interference degree, determine the optimal insertion position for each transaction to be inserted. The selection of the optimal insertion position needs to balance two objectives: one is to minimize the interference to the benchmark execution sequence as much as possible, and the other is to ensure the satisfaction degree of the processing dependency constraints. The specific determination method is to try to insert the transaction to be processed at each position in the benchmark sequence, calculate the overall interference degree and the constraint violation degree after insertion, and select the position with the smallest weighted sum of the two as the optimal insertion position.

[0051] Finally, according to the reference execution sequence and the determined optimal insertion positions, a transaction family processing priority sequence that satisfies the processing dependency constraints is generated. The generation process adopts an iterative approach, where each time the transaction with the least interference degree is selected from the transactions to be inserted and inserted into the corresponding optimal position. For example, a certain transaction family contains 8 transactions, among which 3 transactions form the reference execution sequence, and the remaining 5 transactions determine the insertion positions in ascending order of interference degree, finally obtaining a complete processing priority sequence.

[0052] Through the above detailed steps, the transaction clustering evaluation and priority sequence generation based on the processing dependency probability matrix are realized. This method fully considers the processing dependency relationships and execution stability characteristics among transactions, obtains a reasonable division of transaction families through iterative optimization, and generates a processing priority sequence that satisfies the constraints based on the reference sequence and interference degree analysis. The entire implementation process has strong operability and practical value.

[0053] In this embodiment, an accurate balance can be achieved between the transaction processing dependency relationship and execution stability, significantly improving the scientificity and execution efficiency of transaction scheduling. By combining the processing dependency probability with the historical execution fluctuation characteristics, the constructed transaction clustering evaluation function can effectively identify the internal coupling relationships and external conflict risks among transactions, ensuring that the clustering results have high relevance and stability. By introducing a fluctuation compensation mechanism, the dynamic adjustment of execution uncertainty is realized, improving the convergence effect of clustering iteration. The finally generated transaction family processing priority sequence takes into account both the dependency strength and execution stability, effectively reducing the risks of transaction conflicts, resource congestion, and processing delays, and improving the intelligent level and operation performance of the overall scheduling system.

[0054] Figure 2 This is the heat map of the processing dependency probability matrix in the embodiment of the present invention, as Figure 2 shown. This figure shows the strength of the mutual dependency relationships among 8 transactions (T1 to T8). This matrix is the basis for constructing the transaction clustering evaluation function, where the matrix element value represents the dependency probability of the row transaction on the column transaction. From the figure, multiple high-dependency relationship groups can be clearly observed: there are strong dependencies of 0.78 - 0.85 between T1 and T2, T4; there are strong associations of 0.75 - 0.92 between T2 and T3, T6; the dependency probabilities between T3 and T4, T5, T8 are between 0.64 - 0.81; the dependency probability between T4 and T5 is as high as 0.88; the dependency probabilities between T5 and T6, T7 are between 0.61 - 0.79; the dependency between T6 and T7 is 0.87; the dependency between T7 and T8 is 0.83; there are dependencies of 0.76 and 0.74 between T8 and T1, T6 respectively. This complex dependency relationship network needs to be reasonably divided by the clustering evaluation function proposed in this technical solution. Compared with the traditional K-means algorithm or hierarchical clustering method, this solution can more accurately capture the processing dependency constraints among transactions.

[0055] In an alternative embodiment, generating a transaction family processing priority sequence that satisfies processing dependency constraints based on the reference execution sequence and the optimal insertion position includes: constructing an initial constraint edge set based on the reference execution sequence, calculating the transfer constraint strength between nodes in the initial constraint edge set using the shortest path algorithm, and supplementing the transfer constraints with transfer constraint strength greater than a preset strength threshold to the initial constraint edge set to obtain a complete constraint edge set; performing a hierarchical partitioning on the complete constraint edge set based on the in-degree and out-degree of nodes to obtain a constraint graph including inter-layer constraint relationships and intra-layer constraint relationships; obtaining the deviation value between the actual execution duration and the expected duration of a transaction, the number of constraint violations, and the degree of competition, calculating the performance score of the transaction based on the deviation value, the number of violations, and the degree of competition, performing a weighted combination of the performance score and the basic priority to obtain an updated transaction priority, and determining an influence node set affected by the updated priority in the constraint graph; sequentially adjusting the priorities of each node in the influence node set, calculating the dependency chain length between nodes based on the constraint graph, extracting optimized nodes with a dependency chain length greater than a preset length threshold, reordering the optimized nodes to obtain multiple reordering schemes, and selecting the reordering scheme with the minimum processing delay as the target reordering scheme; comparing the target reordering scheme with the historical priority sequence, and when the improvement amplitude exceeds a preset improvement threshold, updating the transaction family processing priority sequence using the target reordering scheme.

[0056] Exemplarily, in the process of generating the transaction family processing priority sequence, it is first necessary to construct an initial constraint edge set based on the reference execution sequence. These transactions form a reference execution sequence according to the submission time and urgency. First, identify the constraint relationships between directly adjacent transactions and construct an initial constraint edge set.

[0057] Calculate the transfer constraint strength between nodes in the initial constraint edge set using the shortest path algorithm. The transfer constraint strength represents the strength of the indirect dependency relationship between transactions and is calculated by accumulating the direct constraint strengths on the path. When the strength of a certain transfer constraint exceeds a preset strength threshold, the system adds this transfer constraint to the initial constraint edge set. In this way, the system obtains a complete constraint edge set including direct constraints and important transfer constraints. When performing a hierarchical partitioning on the complete constraint edge set, the system analyzes based on the in-degree and out-degree of nodes. The in-degree represents the number of constraint edges pointing to the node, and the out-degree represents the number of constraint edges emitted from the node. By comparing the in-degree and out-degree of nodes, the system can determine the dependency hierarchy relationship of transactions.

[0058] Real-time obtain the execution status data of transactions, including the deviation value between the actual execution duration and the expected duration. For example, for a certain repair work expected to be completed in two hours but actually taking three hours, the deviation value is one hour. The number of constraint violations records the number of times a transaction fails to execute according to the constraint requirements, such as not being completed within the specified time or affecting the normal execution of other transactions. The degree of competition reflects the intensity of resource contention between a transaction and other transactions. Based on these data, the system calculates the performance score of the transaction. The performance score considers three dimensions: time deviation, violation situation, and resource competition, and is calculated with weighting according to the importance of each dimension. The lower the performance score, the worse the execution effect of the transaction, and its priority needs to be adjusted appropriately.

[0059] Weightedly combine the performance score with the basic priority of the transaction to obtain the updated priority. The basic priority is mainly determined by the urgency and importance of the transaction, while the performance score reflects the actual execution effect of the transaction. The combined priority can more comprehensively reflect the processing requirements of the transaction.

[0060] When the priority of a certain transaction changes, the system needs to determine the set of affected nodes in the constraint graph. These nodes include other transactions that have direct constraint relationships or strong transitive constraint relationships with this transaction. Adjust the priorities of each node in the set of affected nodes in sequence, and calculate the length of the dependency chain between the nodes based on the constraint graph. The length of the dependency chain represents the number of constraint edges passed from one node to another. When the length of the dependency chain exceeds the preset length threshold, these nodes are marked as optimized nodes and need to be optimized with emphasis. When reordering the optimized nodes, multiple alternative reordering schemes are generated. Each scheme needs to satisfy the dependency relationships in the constraint graph, while considering resource utilization efficiency and processing time. By simulating and calculating the processing delays that each scheme may cause, select the scheme with the minimum delay as the target reordering scheme.

[0061] Finally, compare the target reordering scheme with the historical priority sequence. Evaluate the optimization effect of the new scheme by calculating the improvement degrees of indicators such as processing time and resource utilization rate. When the improvement amplitude exceeds the preset improvement threshold, adopt the new reordering scheme to update the processing priority sequence of the transaction family.

[0062] In the prior art, transaction priority sorting usually relies on static rules or simple processing order settings, making it difficult to dynamically reflect the complex dependencies and differences in execution performance among transactions, and prone to priority conflicts, resource congestion, and low scheduling efficiency. In this application, an initial constraint edge set is constructed through a benchmark execution sequence and an optimal insertion position, and transitive constraint strength is introduced for complementation to make the expression of dependencies among transactions more complete. Further, a constraint graph is formed through hierarchical partitioning, and combined with the actual execution performance and historical default behaviors of transactions, the transaction priorities are dynamically adjusted, and a set of affected nodes is constructed to achieve fine control of local priorities. On this basis, a mechanism for determining the length of the dependency chain and multi-scheme reordering is designed to optimize the processing path from a global perspective, and an improvement threshold is set to control the priority update conditions to avoid system instability caused by excessive adjustment. Starting from enhancing the adaptability and scheduling stability of priority sorting, this solution significantly enhances the system's response ability to complex dependencies and dynamic behaviors of transactions, and effectively improves the problems of rigid priorities and lack of dynamic optimization ability in the prior art.

[0063] In an alternative embodiment, A skill vector is constructed based on the professional skills of the processor, and the skill vector and the transaction feature vector in the transaction family are input into the adaptive dynamic programming algorithm. Through multi-stage decision optimization, an optimal matching path is obtained, and a comprehensive matching score is generated according to the optimal matching path score and the historical service score, including: Obtain the professional qualification characteristics, work experience characteristics, and skill coverage characteristics of the processor, and generate the professional skill vector of the processor through weighted combination based on preset weights; Calculate the matching degree between the professional skill vector and the transaction feature vector in the transaction family to obtain a basic matching value, calculate the satisfaction degree of the skill item for the transaction feature based on the basic matching value, and generate an initial state value of the state space; Construct a skill transfer matrix to represent the complementary relationship between skills, calculate the state transition probability based on the skill transfer matrix, and combine the state transition probability with the initial state value to construct a state space; Obtain the service completion quality and processing timeliness rate of the processor, calculate the historical service score, adjust the state value of the state space based on the historical service score to construct a state value function, perform policy iteration calculation on the state value function to obtain a state value sequence, determine the optimal state transition path based on the state value sequence, and generate an optimal matching path; Calculate the state transition benefit of the optimal matching path to obtain a path score, and combine the path score and the historical service score according to the weights determined by the transaction characteristics to generate a comprehensive matching score.

[0064] Exemplarily, the basic information of the handler is structurally extracted to obtain their professional qualification characteristics, work experience characteristics, and skill coverage characteristics in the managed service scenario. The professional qualification characteristics include categories such as whether they hold a teacher qualification certificate, a psychological counseling certificate, and an interest-based teaching qualification. The work experience characteristics include the cumulative service years, the number of service positions in previous years, and the diversity of service types. The skill coverage characteristics cover the current course teaching ability, activity organization ability, emergency response ability, and communication and coordination ability. Weight parameters are preset for various characteristics. For example, the weight of qualification integrity is higher than that of the number of positions, and the weight of service years is slightly higher than the coverage of activity types. After all characteristics are standardized, the professional skill vector of the handler is generated according to the set weights. The vector dimension structure is consistent with the transaction feature vector to ensure comparability between the two.

[0065] For the currently to-be-allocated transaction family, the feature vectors of each transaction in it are extracted and matched one by one with the professional skill vector of the handler. The consistency degree of the corresponding dimensions between the vectors is calculated to obtain the basic matching value between each pair of transactions and the handler. The higher the basic matching value, the greater the execution potential of the handler in this transaction. Further analyze the coverage of each skill item in the transaction feature vector, and set the skill satisfaction score based on whether the handler has this ability and the ability level, forming the initial value of each state. This initial value is used to represent the static matching effect without considering skill combination and migration.

[0066] Based on the multi-dimensional skill label system, a skill migration matrix is constructed, which reflects the complementary and substitutable relationships between different skills. For example, "classroom management ability" can partially replace "discipline control ability", and "course explanation ability" and "student emotion guidance" have a synergistic effect. Migration probability values are assigned to each pair of skills. Through this matrix, the transfer paths between the initial states are constructed. The higher the migration probability, the more logically feasible the skill combination path is in the current task processing scenario. The state transfer path represents the possibility that the handler can continue to process another transaction after completing a certain transaction.

[0067] Combining the above migration probability and the initial state value, a complete state space is constructed. Each state represents the adaptation degree of processing a certain transaction. The transfer between states is guided by the skill migration relationship, and it enters the dynamic programming calculation stage. Obtain the historical service completion quality data of the handler, including the post-service evaluation score, the transaction processing feedback content and quantitative scoring. The processing timeliness data includes information such as the average time from receiving the transaction to completion and the overdue rate. These data are comprehensively normalized to form the historical service score. Using this score as a regulatory factor, the state values in the state space are weighted and adjusted to enhance the value representation of high-scoring handlers at certain transaction nodes, so that the historical ability is effectively reflected.

[0068] After the state value function is constructed, policy iteration update operations are performed. All reachable state paths are traversed, and the state value sequence is updated. During the iteration, the transfer path with the highest return is selected as the optimal policy for each step. After a fixed number of rounds or when the convergence condition is met, the optimal state transfer path is finally generated. This path represents the optimal solution for selecting the task processing order and matching objects from multiple transactions. Each node in the path represents the execution goal of a transaction, and each edge represents the dynamic adaptation of capacity migration.

[0069] After obtaining the optimal path, calculate the return brought by each state transfer in the path. The return is jointly determined by the matching degree, the fluency of skill transfer, and the processing experience. The sum of the returns of all state transfers on the path constitutes the path score. Taking the path score as the main reference value, it is combined with the historical service rating of the processing personnel again, and combined with the weight factors set by the current transaction. For example, some transactions prioritize processing quality, and some transactions prioritize processing speed. The two are weighted and calculated according to different transaction dimensions to generate a comprehensive matching score.

[0070] Taking the information of a processing personnel as an example, he has a teacher qualification certificate and a psychological counseling qualification, with 5 years of work experience, has participated in basic trusteeship and interest course organization, the average historical service evaluation is 4.8 points (out of 5), and the average response time is 30 minutes, which is lower than the system average. The transaction family includes three tasks: "Student temporary trusteeship application", "Course content adjustment", and "Student feedback of abnormal emotions". The system generates skill vectors and compares them with transaction characteristics one by one, and the obtained basic matching values are 0.78, 0.69, and 0.91 respectively. Combining with the skill transfer matrix, the path is preferably "Emotion abnormality processing → Temporary trusteeship → Course content adjustment". According to this path, the calculated score is 0.85, and the historical score is 0.92. Considering the weight of 0.7 for the "Student emotion" transaction prioritizing quality and the weight of 0.3 for the "Course adjustment" transaction prioritizing efficiency, the comprehensive matching score of this processing personnel for this transaction family is calculated to be 0.88, and the sorting priority is significant.

[0071] Support multiple processing personnel to participate in the matching score calculation in parallel. Finally, use the comprehensive matching scores of each processing personnel for each transaction as the basis for the allocation decision. When the ratio of the number of processing personnel to the transaction volume is unbalanced, the system performs batch matching according to the optimal path to achieve intelligent scheduling and allocation of multiple transactions, avoiding problems such as resource waste and low execution efficiency, and improving the response quality of the trusteeship service and the satisfaction of students. The overall mechanism supports business environments such as dynamic addition of processing personnel, real-time addition of transactions, and real-time adjustment of policy parameters, and has good scalability and stability.

[0072] In this embodiment, by constructing a matching mechanism between multi-dimensional transaction features and the skills of processing personnel, the intelligence and precision of transaction allocation are realized. This not only improves the efficiency of student affair processing but also significantly enhances the rationality of transaction matching and service quality. The feature propagation and dependency analysis mechanism can identify the sequential logic and resource conflicts between transactions, effectively avoiding situations of chaotic processing order or resource scheduling conflicts. The comprehensive matching score mechanism introduces historical service capabilities into the decision-making process, realizes the quantitative assessment and dynamic feedback of the service performance of processing personnel during the transaction allocation process, thereby supporting the priority scheduling of high-quality personnel and ensuring that student affairs receive timely, professional, and personalized responses.

[0073] In an alternative embodiment, Performing policy iteration calculation on the state value function to obtain a state value sequence, and determining an optimal state transition path based on the state value sequence, the generation of the optimal matching path includes: Obtaining the immediate reward of the state and the discount factor, constructing the Bellman equation, updating the current state value by calculating the expected value after state transition to obtain an initial state value sequence, calculating the state transition probability under the initial matching policy based on the initial state value sequence, adjusting the temperature parameter according to the value improvement amplitude of historical matching data, and updating the state transition probability based on the temperature parameter; Starting from the current state, calculating the state values of all transferable states according to the updated state transition probability, selecting the transfer state with the largest state value and adding it to the matching path, and determining whether the matching path meets any of the conditions of the target state, path length limit, or state value gain threshold. When it is satisfied, marking the matching path as a candidate matching path; Calculating the state values of the states that can be transferred in multiple steps in the future for each state node in the candidate matching path, selecting the transfer state with the largest multi-step state value to update the candidate matching path, and obtaining an optimized matching path; Comparing the optimized matching path with the historical optimal matching path, updating the historical optimal matching path when the state value is larger, setting a lower limit of the state value based on the historical optimal matching path, and removing the branches below the lower limit of the state value to obtain the final optimal matching path.

[0074] Exemplarily, based on the previously constructed state space, an initial state value sequence is generated for each handler in a specific transaction assignment context. The required parameters include the immediate reward of each state and the historical feedback impact factor. The immediate reward is given according to the current suitability of the handler for a certain student affair, which is used to reflect the direct effectiveness of the handler in this task. The historical feedback impact factor is calculated by combining evaluation results such as historical processing success rate, parent satisfaction, and affair resolution time, which represents the discounted impact of the long-term performance of this handler when dealing with similar tasks. Based on this, a recurrence function for value update is constructed. Relying on the expected reward after transaction transfer, the current state value is iteratively corrected, and a complete initial state value sequence is gradually formed. Each state value in this sequence reflects the cumulative advantage of the current processing path at a certain stage.

[0075] Based on the initial state value sequence, a state transition probability distribution under the initial matching strategy is constructed. This probability determines the preferred jump direction between different task nodes at each stage. The initial value of the transition probability is determined by the relative difference between state values. To improve the dynamic adaptability of the matching path, the system introduces a temperature parameter to control the sensitivity of the jump probability. The size of the temperature parameter is adjusted according to the value change amplitude during the iteration of the matching strategy in the historical transaction matching data. If the value improvement brought by the strategy update in the historical matching is significant, the temperature parameter is reduced to increase the bias towards the optimal jump path. If the historical improvement amplitude is small, the temperature parameter is increased to increase the diversity of path exploration. Based on the updated state transition probability, starting from the current transaction state, it gradually jumps to the next stage, and the transfer target node with the maximum state value in the current stage is selected and included in the matching path.

[0076] During the matching path generation process, it is continuously judged whether the current path meets the set termination conditions. The termination conditions include whether all target affairs have been covered, whether the path length reaches the system set limit, or whether the current cumulative value exceeds the specified reward threshold. When any of these conditions is met, the current path is marked as a candidate matching path. This path reflects the optimal intermediate result of the current strategy in terms of feasibility, integrity, and efficiency.

[0077] Based on the candidate path, the multi-step state evolution trend in the future stage is further considered. Starting from each state node in the path, the state values that may be reached under multiple rounds of state transfer are predicted. By comparing the cumulative effects of state values in all reachable paths, the transfer path with the optimal future value is selected as the optimization direction for the current node, and the state jump scheme in the candidate path is gradually updated to form an optimized matching path. The optimized path not only based on the current state performance but also incorporates the evolution potential in the future stage, realizing a comprehensive evaluation of the long-term matching effect.

[0078] Compare the value of the optimized matching path with the optimal matching path stored in the historical record. If the cumulative state value of the current path is greater, replace the original path in the historical record as the latest optimal matching path. During the replacement process, simultaneously update the lower bound of the state value at the path start point, which is used as the filtering threshold in the subsequent path construction process. For path branches with state values lower than this threshold, exclude them from further matching calculations to ensure a balance between the accuracy and efficiency of the path calculation process.

[0079] Exemplarily, a certain operator A has three skills: trusteeship course arrangement, emergency handling, and psychological counseling. Their historical service quality score is 4.6 points, and the average transaction response time is 25 minutes. The current transaction family to be allocated includes three items: temporary trusteeship application, student behavior anomaly report, and course content adjustment. Map the three transactions to three state nodes respectively. The immediate benefit values generated by matching the operator skill vector with the three transaction feature vectors are 0.82, 0.76, and 0.88 respectively. Combine the historical scores to generate the initial state values of 0.78, 0.74, and 0.84 respectively. In the first round of policy iteration, it is judged that the course content adjustment has the highest value and is set as the first-hop path node. Construct a path candidate sequence according to the state transition probability as "course adjustment → temporary trusteeship → behavior anomaly". In future multi-step predictions, it is found that the combination of "behavior anomaly → course adjustment → temporary trusteeship" obtains a higher cumulative value. Accordingly, correct the matching path, and judge that the cumulative value of this path is higher than the historical optimal record, and update it as the current optimal path. Finally, output the optimal path and accurately allocate the transactions to operator A accordingly, realizing the stage scheduling of processing tasks and the rational utilization of resources. During the whole process, the system does not need to exhaustively search all state combinations, only retains high-value path branches and dynamically adjusts the policy direction to ensure that the allocation algorithm has stable convergence and practical feasibility in large-scale transaction scenarios. The technical solution not only improves the quality of transaction allocation, but also realizes the prediction optimization of the processing process and the avoidance of resource risks, effectively enhancing the intelligent decision-making ability of the multi-modal trusteeship platform in the transaction scheduling link.

[0080] Figure 3 This is a comparison chart of the state value function policy iteration convergence process in the embodiment of the present invention, as Figure 3As shown, this figure presents a comparison of the proposed technical solution with traditional Q-learning and SARSA algorithms during the convergence process of the state value function. It can be clearly seen from the figure that the convergence speed of the state value function of the proposed technical solution is significantly faster than that of traditional methods. It reaches a state value of 0.85 after only 40 iterations, while Q-learning and SARSA reach only 0.60 and 0.48 respectively under the same number of iterations. After 100 iterations, the state value of the proposed technical solution reaches 1.05, which is 46% and 81% higher than 0.72 of Q-learning and 0.58 of SARSA respectively. This advantage stems from the dynamic adjustment of the temperature parameter based on the improvement amplitude of the historical matching data value and the innovative multi-step state value look-ahead calculation in the proposed technical solution. The traditional Q-learning algorithm tends to greedily select states with higher immediate rewards, while SARSA is more conservative and unable to effectively balance exploration and exploitation. The proposed technical solution achieves faster and more efficient state value convergence through the improved calculation of the Bellman equation and the policy iteration of the state value sequence.

[0081] Existing technologies generally adopt static rule matching or fixed-weight scoring methods during the student affairs allocation process, making it difficult to dynamically adapt to the differences in the capabilities of processing personnel and the changes in the characteristics of affairs. This leads to untimely transaction responses, low allocation efficiency, and a lack of the ability to optimize the potential dependencies and processing paths between transactions in a multi-transaction concurrent scenario. This application constructs a state space and a state value function, introduces a policy iteration calculation and a path optimization mechanism, and realizes the dynamic programming of the processing path and the global optimal search. The improvement starting point is to break through the limitation of single matching decision-making, introduce multi-stage value judgment, and enable the skill transfer ability of processing personnel and the execution order of affairs to jointly participate in the decision-making process. In path generation and optimization, by introducing dynamic adjustment means such as temperature parameters and state value lower bounds, efficient path screening and value-oriented matching control are achieved, effectively avoiding problems such as large path randomness and non-convergence of policy iteration in existing solutions. Finally, significant improvements in the matching accuracy, policy stability, and resource utilization efficiency of the transaction processing path are realized, making the hosting service allocation more intelligent, targeted, and sustainable.

[0082] In another alternative implementation, the overall configuration of the hosting mode can be carried out by the education management unit, including maintaining the basic information of all schools, grades, classes, and students within the region. After completing the input of the basic school organizational structure and student information, a unified interface is provided to support the import operation, and at the same time, subsequent manual adjustment and editing are allowed to achieve the dynamic update and centralized management of student data.

[0083] Administrators customize the service content configuration under the "1+X" model according to local actual needs, including the specific categories, service hours, teaching teachers, enrollment targets, service forms, and class hours of basic trusteeship projects and extended trusteeship projects. In the trusteeship mode setting, it is possible to flexibly set whether to open single-choice and multiple-choice items, whether to allow over-enrollment, and even the specific service activity content for each day of the week can be accurately configured. Taking a school as an example, the basic project settings include two items: daily homework guidance and physical activities, and the extended projects include hard-tipped calligraphy, scientific experiments, and music appreciation. An upper limit on the number of applicants is set for each trusteeship service. After the number of students applying reaches the upper limit, the system automatically prohibits further applications and prompts that the quota is full.

[0084] After each school completes the release of trusteeship courses, parent users can submit registration information through the unified platform. The platform automatically identifies the school, grade, and class to which the student belongs based on the student ID number filled in by the parent, and automatically fills in the student information maintained on file for the user to confirm before submission. After the registration information is submitted, the system enters the review stage according to the preset process. The review levels and permissions are uniformly configured by the Education Bureau, and can be designated as class teachers, school administrators, or Education Bureau administrators. The system supports an automatic review function, which makes judgments based on historical registration data, student status, and project capacity. If manual review is required, relevant review personnel can log in to the platform to view the applications awaiting review and perform approval or rejection operations based on the matching of student information and courses. The review results are synchronized to the parent end in real time.

[0085] After the registration review is passed, if the system has configured a payment function, the student's parents need to complete the payment operation. The trusteeship project supports binding fee templates and matching fees according to the project. For example, the basic trusteeship is set at 200 yuan per month, and the extended course such as the calligraphy class is set at 300 yuan per session. The system automatically generates a payment list after the review is passed. Payments can be completed through a unified payment interface and are associated with the financial module. The fee information can be queried and exported by school, class, and student dimensions to form an audit-traceable record.

[0086] During the service process, information on students' attendance, leave, early departure, and other behaviors is continuously collected. The school management end inputs the sign-in data daily by the teaching teacher. The daily service execution situation is summarized, and automatic warnings are issued for abnormal information (such as not signing in for three consecutive days). After each trusteeship period ends, the platform opens an evaluation function. Students and parents can evaluate multiple dimensions such as course content, teaching teachers, service environment, and time slot arrangement in the system, supporting a combination mode of star ratings and text evaluations. The background can set the opening and closing times of the evaluation. The evaluation results are automatically archived in the data center to provide a decision-making basis for optimizing later services.

[0087] All data is automatically classified in the background to form multi-dimensional reports, including registration statistics, project saturation analysis, evaluation satisfaction analysis, payment completion rate analysis, etc. For example, in a certain district in a certain month, a total of 60 classes of trusteeship courses were offered, with a total of 1,320 students registered. Among them, 1,214 students completed the payment after registration, and the payment rate reached 92%. Among them, 61% of the students participated in expansion projects, and the average score of course satisfaction was 4.7 points (out of 5 points), and the evaluation coverage rate reached 89%.

[0088] Support the access of off-campus institutions, including institution name, legal person, contact information, teaching type, etc. After the institution registers on the platform, it submits a cooperation application for review. After the review is passed, the institution can release cooperation courses and set optional schools, grade segments, teaching times, etc. When students choose courses from off-campus institutions, they need to be additionally reviewed and approved, and an independent payment item is generated.

[0089] For the teacher side, school administrators enter the information of teaching teachers, supporting batch import by template and individual manual addition. The teaching hours of teachers per semester are automatically counted by the system, and the course arrangement is configured and displayed through the course schedule management function. Managers can view the teaching teacher's schedule and attendance records according to the dimensions of school, class, and course to ensure that the execution of teaching tasks is controllable.

[0090] At the same time, permissions are configured for different roles, and the visible range and operation range of the interface are controlled through atomic-level function switches. Each account can be assigned multiple roles. For example, an administrator of a certain school has both the permission to review student information and the permission to manage courses. In the background system, the binding relationship between roles and function modules is clearly displayed and can be dynamically adjusted and entrusted for management. For example, when there are personnel changes during the school holidays in a certain school, the trusteeship review permission can be temporarily entrusted to other accounts for execution.

[0091] In terms of the evaluation mechanism, the evaluation results of each course and teacher are transmitted back to assist in optimizing the course configuration for the next semester. For example, a certain expansion course, "Science Experiment Course", received a "very satisfied" evaluation ratio as high as 95% this semester. In the written evaluations of students, "strong interest" and "many hands-on experiences" were frequently mentioned. This course can be preferentially recommended and extended to more schools in the next semester.

[0092] Finally, key operation indicators are presented through the data cockpit. Core data such as the total number of registrations in the region, the total amount of trusteeship payments, service coverage rate by school, registration proportion of different types of trusteeship projects, and evaluation satisfaction of each project are displayed on the management side. At the same time, indicators such as the number of registrations, course evaluations, and course quantities of community trusteeship projects are separately counted to facilitate targeted resource investment and optimized allocation.

[0093] Figure 4This is a schematic structural diagram of a student affairs allocation and service evaluation system based on multimodal hosting, as Figure 4 shown. The system includes: A first unit, configured to receive student affairs request data, extract a transaction description from the student affairs request data, perform semantic analysis to extract semantic features, and fuse the semantic features with a transaction type identifier to generate a transaction feature vector containing complete transaction information; A second unit, configured to receive student affairs request data, extract a transaction description from the student affairs request data, perform semantic analysis to extract semantic features, and fuse the semantic features with a transaction type identifier to generate a transaction feature vector containing complete transaction information; A third unit, configured to construct a skill vector based on the professional skills of the processing personnel, input the skill vector and the transaction feature vectors in the transaction family into an adaptive dynamic programming algorithm, obtain an optimal matching path through multi-stage decision optimization, and generate a comprehensive matching score according to the optimal matching path score and the historical service score; A fourth unit, configured to assign the transaction family and the processing priority sequence to the target processing personnel with the highest comprehensive matching score, and update the professional skill vector and the historical service score of the target processing personnel according to the processing evaluation score of the transaction family for subsequent matching calculations in transaction allocation.

[0094] In the third aspect of the embodiments of the present invention, a kind of electronic device is provided, including: a processor; a memory for storing instructions executable by the processor; wherein, the processor is configured to call the instructions stored in the memory to execute the method described above.

[0095] In the fourth aspect of the embodiments of the present invention, a computer-readable storage medium is provided, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, the method described above is implemented.

[0096] The present invention can be a method, a device, a system, and / or a computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions thereon for performing various aspects of the present invention.

[0097] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for student affairs allocation and service evaluation based on multi-modal hosting, characterized in that Including: Receiving student affair request data, extracting affair descriptions from the student affair request data, conducting semantic analysis to extract semantic features, fusing the semantic features with affair type identifiers to generate affair feature vectors containing complete affair information; Constructing an affair time-series association graph based on the affair feature vectors, calculating the feature propagation between nodes in the affair time-series association graph through recursive deep feature extraction to generate a processing dependency probability matrix, constructing an affair clustering evaluation function based on the processing dependency probability matrix, generating affair families through iterative optimization clustering of the affair clustering evaluation function, and generating a processing priority sequence for the affair families according to the constraint strength in the processing dependency probability matrix; Constructing a skill vector based on the professional skills of processors, inputting the skill vector and the affair feature vectors in the affair families into an adaptive dynamic programming algorithm, obtaining an optimal matching path through multi-stage decision optimization, and generating a comprehensive matching score based on the optimal matching path score and historical service scores; Assigning the affair families and the processing priority sequence to the target processor with the highest comprehensive matching score, and updating the professional skill vector and historical service score of the target processor according to the processing evaluation scores of the affair families for subsequent matching calculations in affair assignment.

2. The method according to claim 1, wherein Extracting affair descriptions from the student affair request data, conducting semantic analysis to extract semantic features, and fusing the semantic features with affair type identifiers to generate affair feature vectors containing complete affair information includes: Extracting affair descriptions from the student affair request data, decomposing the affair descriptions at multiple levels according to core information, attribute information, and environmental information, extracting affair target and requirement features from the core information, extracting affair timeliness and urgency features from the attribute information, and extracting processing scenario and resource requirement features from the environmental information to obtain multi-level affair features; Conducting correlation analysis on the multi-level affair features, calculating a feature recombination coefficient based on the information integrity and information redundancy between features, and combining affair features with a complementarity greater than a preset complementarity threshold according to the feature recombination coefficient to generate affair semantic features; Establishing a mapping rule library for affair type identifiers, storing the corresponding relationships and confidence levels between affair type identifiers and affair semantic features in the mapping rule library, performing feature matching in the mapping rule library according to the affair semantic features to obtain the probability distribution of affair type identifiers, and combining the affair semantic features with the affair type identifier with the highest confidence level to generate affair feature vectors containing complete affair information.

3. The method according to claim 1, characterized in that, Constructing an affair time-series association graph based on the affair feature vectors, calculating the feature propagation between nodes in the affair time-series association graph through recursive deep feature extraction to generate a processing dependency probability matrix includes: Constructing an affair time-series association graph based on the affair feature vectors, where nodes represent affairs and the connecting edges between nodes represent time-series associations between affairs; Establishing a conflict affair marking library based on historical processing data, the conflict affair marking library recording the types of affair combinations that generate resource competition, matching the nodes in the affair time-series association graph with the conflict affair marking library, identifying node pairs with resource competition and generating a set of competing nodes; For node pairs in the set of competing nodes, extract the resource requirement information of each node, calculate the resource overlap degree between node pairs, and determine the resource competition intensity based on the resource overlap degree as the weight of the competition edge between competing node pairs; For the remaining node pairs in the transaction time sequence association graph, extract the time interval information and transaction feature vectors between nodes, and calculate the time sequence similarity between node pairs as the weight of the association edge between the remaining node pairs; Combine the competition edge weight and the association edge weight to form the complete edge weight information, use the complete edge weight information to guide the recursive propagation of node features, update the node features to obtain the propagated node features, and calculate the processing dependence probability between nodes based on the propagated node features to generate a processing dependence probability matrix.

4. The method according to claim 1, characterized in that Construct a transaction clustering evaluation function based on the processing dependence probability matrix, generate transaction families by iteratively optimizing the clustering of the transaction clustering evaluation function, and generate a processing priority sequence for the transaction families according to the constraint intensity in the processing dependence probability matrix, including: Construct a transaction clustering evaluation function based on the processing dependence probability matrix, and the transaction clustering evaluation function includes an intra-family processing dependence constraint term and an inter-family processing dependence constraint term; Extract the historical execution records in the processing dependence probability matrix, calculate the execution duration fluctuation coefficient of the transaction according to the historical execution records, construct an execution stability matrix and input it into the transaction clustering evaluation function to perform initial clustering on the transaction to obtain an initial set of transaction families; Calculate the difference in the execution duration fluctuation coefficients of transaction pairs in the initial set of transaction families, determine the processing dependence constraint weight and the fluctuation compensation coefficient of the transaction pairs in the transaction clustering evaluation function according to the difference in the execution duration fluctuation coefficients, and perform iterative optimization on the set of transaction families. When the evaluation value converges, obtain the final set of transaction families; Select transactions with an execution duration fluctuation coefficient less than a preset fluctuation threshold from the final set of transaction families to construct a benchmark execution sequence, calculate the interference degree of the remaining transaction pairs to the benchmark execution sequence according to the constraint intensity in the processing dependence probability matrix, determine the optimal insertion position of the remaining transactions, and generate a processing priority sequence of transaction families that satisfies the processing dependence constraint according to the benchmark execution sequence and the optimal insertion position.

5. The method according to claim 4, wherein Generating a processing priority sequence of transaction families that satisfies the processing dependence constraint according to the benchmark execution sequence and the optimal insertion position includes: Construct an initial constraint edge set based on the benchmark execution sequence, use the shortest path algorithm to calculate the transfer constraint intensity between nodes in the initial constraint edge set, and supplement the transfer constraints with a transfer constraint intensity greater than a preset intensity threshold to the initial constraint edge set to obtain a complete constraint edge set; Perform hierarchical partitioning on the complete constraint edge set based on the in-degree and out-degree of nodes to obtain a constraint graph including inter-layer constraint relationships and intra-layer constraint relationships; Obtain the deviation value between the actual execution duration and the expected duration of the transaction, the number of constraint violations, and the competition degree, calculate the performance score of the transaction based on the deviation value, the number of violations, and the competition degree, perform weighted combination of the performance score and the basic priority to obtain the updated transaction priority, and determine the set of affected nodes affected by the updated priority in the constraint graph; Adjust the priorities of the nodes in the set of influence nodes in sequence, calculate the length of the dependency chain between nodes based on the constraint graph, extract the optimized nodes whose dependency chain length is greater than the preset length threshold, reorder the optimized nodes to obtain multiple reordering schemes, and select the reordering scheme with the minimum processing delay as the target reordering scheme; Compare the target reordering scheme with the historical priority sequence, and when the improvement amplitude exceeds the preset improvement threshold, update the transaction family processing priority sequence with the target reordering scheme.

6. The method according to claim 1, characterized in that Construct a skill vector based on the professional skills of the processing personnel, input the skill vector and the transaction feature vector in the transaction family into the adaptive dynamic programming algorithm, optimize through multi-stage decision-making to obtain the optimal matching path, and generate a comprehensive matching score based on the optimal matching path score and the historical service score, including: Obtain the professional qualification characteristics, work experience characteristics and skill coverage characteristics of the processing personnel, and generate the professional skill vector of the processing personnel through weighted combination based on the preset weights; Calculate the matching degree between the professional skill vector and the transaction feature vector in the transaction family to obtain the basic matching value, calculate the satisfaction degree of the skill item for the transaction feature based on the basic matching value, and generate the initial state value of the state space; Construct a skill transfer matrix to represent the complementary relationship between skills, calculate the state transition probability based on the skill transfer matrix, and combine the state transition probability with the initial state value to construct the state space; Obtain the service completion quality and processing timeliness rate of the processing personnel, calculate the historical service score, adjust the state value of the state space based on the historical service score to construct the state value function, perform policy iteration calculation on the state value function to obtain the state value sequence, determine the optimal state transition path based on the state value sequence, and generate the optimal matching path; Calculate the state transition benefit of the optimal matching path to obtain the path score, and combine the path score and the historical service score according to the weights determined by the transaction characteristics to generate the comprehensive matching score.

7. The method according to claim 6, wherein Perform policy iteration calculation on the state value function to obtain the state value sequence, determine the optimal state transition path based on the state value sequence, and generate the optimal matching path, including: Obtain the immediate benefit and discount factor of the state, construct the Bellman equation, update the current state value by calculating the expected value after state transition to obtain the initial state value sequence, calculate the state transition probability under the initial matching strategy based on the initial state value sequence, adjust the temperature parameter according to the value improvement amplitude of the historical matching data, and update the state transition probability based on the temperature parameter; Starting from the current state, calculate the state values of all transferable states according to the updated state transition probability, select the transfer state with the maximum state value and add it to the matching path, and determine whether the matching path meets any of the conditions of the target state, path length limit or state value benefit threshold. If it meets, mark the matching path as the candidate matching path; Calculate the state values that the candidate matching path can transfer in the future in multiple steps for each state node, select the transfer state with the maximum multi-step state value to update the candidate matching path, and obtain the optimized matching path; Compare the optimized matching path with the historical optimal matching path. When the state value is greater, update the historical optimal matching path. Set the lower bound of the state value based on the historical optimal matching path, and remove the branches below the lower bound of the state value to obtain the final optimal matching path.

8. A student affairs allocation and service evaluation system based on multimodal hosting, which is used to implement the method described in any one of the foregoing claims 1-7, and is characterized in that, Comprising: A first unit, configured to receive student affair request data, extract a transaction description from the student affair request data, perform semantic analysis to extract semantic features, and perform feature fusion on the semantic features and a transaction type identifier to generate a transaction feature vector containing complete transaction information; A second unit, configured to receive student affair request data, extract a transaction description from the student affair request data, perform semantic analysis to extract semantic features, and perform feature fusion on the semantic features and a transaction type identifier to generate a transaction feature vector containing complete transaction information; A third unit, configured to construct a skill vector based on the professional skills of a handler, input the skill vector and the transaction feature vectors in a transaction family into an adaptive dynamic programming algorithm, obtain an optimal matching path through multi-stage decision optimization, and generate a comprehensive matching score according to the optimal matching path score and the historical service score; A fourth unit, configured to assign a transaction family and a processing priority sequence to a target handler with the highest comprehensive matching score, and update the professional skill vector and the historical service score of the target handler according to the processing evaluation score of the transaction family for subsequent matching calculations of transaction assignment.

9. An electronic device, characterized in that, Comprising: A processor; A memory for storing instructions executable by the processor; Wherein, the processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, the method according to any one of claims 1 to 7 is implemented.

Images