Intelligent collaborative management method and platform based on artificial intelligence driving

Abstract

The application discloses an intelligent collaborative management method and platform based on artificial intelligence driving. The method comprises the following steps: obtaining a target post to be processed in an exchange platform, and determining a multi-dimensional feature vector of the target post; constructing a graph neural network; analyzing the multi-dimensional feature vector and the graph neural network by using a deep Q network to obtain a target node in the graph neural network for processing the target post; forwarding the target post to the target node, obtaining a processing result of the target node for the target post, and forwarding the processing result to an information receiving area or a target communication address associated with the target post. The application solves the technical problem of low post management efficiency caused by the fact that related technologies cannot efficiently and accurately distribute posts in the exchange platform to objects suitable for processing the posts.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, in particular to an intelligent collaborative management method and platform based on artificial intelligence driving. BACKGROUND

[0002] The enterprise forum platform in the related art is mainly used for employee communication, supports employees to publish posts to share experience, raise questions or discuss business-related topics, and other employees can comment and reply under the post. Taking an enterprise forum of a certain bank as an example, an employee can publish a discussion post about the promotion strategy of a new credit product, and colleagues can express their own opinions in the comment area. However, the related platform is not deeply combined with the task management process. When valuable task demands are generated in the forum, such as tasks to optimize a certain business process according to the discussion, the administrator needs to manually screen the relevant post content, export and organize it, and then distribute it to the corresponding department again. During the task execution process, the forum cannot track the task progress and cannot close the loop of the task, resulting in chaotic task management and difficulty in ensuring the timely and high-quality completion of the task.

[0003] In view of the above problems, no effective solutions have been proposed so far. SUMMARY

[0004] The present application provides an intelligent collaborative management method and platform based on artificial intelligence driving, to at least solve the technical problem of low post management efficiency caused by the fact that the related art cannot efficiently and accurately distribute posts in the communication platform to the objects suitable for processing the posts.

[0005] According to one aspect of the present application, an intelligent collaborative management method driven by artificial intelligence is provided, including: obtaining target posts to be processed in a communication platform, and determining a multidimensional feature vector of the target post, wherein the multidimensional feature vector includes: topic classification features, sentiment features, and keyword coding features; constructing a graph neural network, wherein the nodes in the graph neural network include: department nodes and employee nodes, and the edges in the graph neural network include: inter-departmental collaboration edges, department-employee affiliation edges, and inter-employee collaboration edges, and the department node attributes include at least: the number of posts to be processed, the average post processing time, and the professional field code, and the employee node attributes include at least: the current Status, skill feature vector, historical satisfaction score; use the first deep Q network to analyze the multidimensional feature vector and the graph neural network to obtain the target node in the graph neural network for processing the target post, wherein the status in the first deep Q network includes: the multidimensional feature vector of the current post, the topological structure and node attributes of the graph neural network; the actions in the first deep Q network include: selecting the target department node, selecting the target employee node, and performing multi-hop decision-making along the graph edge; forwarding the target post to the target node, obtaining the processing result of the target node for the target post, and forwarding the processing result to the information receiving area or target communication address associated with the target post.

[0006] Optionally, the reward function of the first deep Q network is determined by the following method: determining a first sub-reward function based on the matching score, wherein the matching score is used to characterize the degree of semantic matching between the post content and the node's professional field, and the matching score is obtained by calculating the similarity between the post keywords and the node's preset tags using natural language processing technology; determining a second sub-reward function based on the time delay, wherein the time delay is used to characterize the time difference from the post creation timestamp to the successful allocation timestamp; determining a third sub-reward function based on the load index, wherein the load index is used to characterize the ratio of the node's current task queue length to the node's maximum processing capacity; determining a fourth sub-reward function based on the satisfaction score, wherein the satisfaction score is used to characterize the feedback rating provided by the user after the post processing is completed; determining the reward function based on the first sub-reward function, the second sub-reward function, the third sub-reward function, and the fourth sub-reward function.

[0007] Optionally, a reward function is determined according to the first sub-reward function, the second sub-reward function, the third sub-reward function, and the fourth sub-reward function, including: obtaining an initial weight coefficient for each sub-reward function; continuously detecting a post allocation success rate indicator, an average processing delay indicator, and a user satisfaction mean indicator; when it is detected that the post allocation success rate indicator decreases, increasing the first initial weight coefficient corresponding to the first sub-reward function and the fourth initial weight coefficient corresponding to the fourth sub-reward function; when it is detected that the average processing delay indicator increases, increasing the second initial weight coefficient corresponding to the second sub-reward function and the third initial weight coefficient corresponding to the third sub-reward function; when it is detected that the user satisfaction mean indicator decreases, increasing the fourth initial weight coefficient; performing a normalization operation after each weight coefficient update so that the sum of the updated weight coefficients is 1, and obtaining the first weight coefficient, the second weight coefficient, the third weight coefficient, and the fourth weight coefficient; determining the reward function of the first deep Q network according to the first sub-reward function, the first weight coefficient, the second sub-reward function, the second weight coefficient, the third sub-reward function, the third weight coefficient, and the fourth sub-reward function, the fourth weight coefficient.

[0008] Optionally, determining a multidimensional feature vector of a target post comprises: extracting a semantic embedding vector of the target post using a pre-trained language model; calculating different probability distributions corresponding to different topic categories through a topic attention layer; normalizing different probability distributions to obtain a first feature component; calculating a vocabulary strength score of the target post based on a financial sentiment dictionary, wherein the financial sentiment dictionary includes a plurality of different preset words for representing positive emotions, neutral emotions, and negative emotions, and the vocabulary strength score is used to quantify the strength of the emotional tendency explicitly expressed in the target post; determining the emotional probability of the target post through a deep learning model, wherein the emotional probability is used to indicate the probability that the target post belongs to positive emotions, neutral emotions, and negative emotions; dynamically weighting the lexical strength score and the emotional probability according to the density of financial terms to obtain the first feature component. Two feature components, where the financial term density is the ratio of the number of financial professional terms appearing in the target post to the total number of words in the target post; common stop words and high-frequency financial function words in the target post are removed, and professional terms whose word frequency-inverse document frequency exceeds a preset threshold are retained; the core words in the professional terms are screened through the centrality measurement index of the term co-occurrence graph, and the third feature component is determined based on the core words, where the term co-occurrence graph is used to represent the relationship of the co-occurrence of professional terms retained after screening in the target post, the nodes of the term co-occurrence graph are the screened professional terms, and the edges of the term co-occurrence graph are the semantic associations between professional terms; the first feature component, the second feature component and the third feature component are spliced, and the splicing result is subjected to dimensionality reduction processing through an orthogonal constrained linear transformation layer to obtain a multidimensional feature vector.

[0009] Optionally, after obtaining the target post to be processed in the communication platform, the method further includes: encoding the target post using a pre-trained language model to obtain a semantic feature vector; determining the normalized click-through rate value of the target post and the weight of the poster's rank; splicing the semantic feature vector, the normalized click-through rate value and the weight of the poster's rank, and determining the splicing result as a state vector; using a second deep Q network to analyze the state vector to determine whether the target post is an essential post, wherein the action space of the second deep Q network includes: marking essential posts and not marking essential posts; the reward function of the second deep Q network includes: compliance score, expert review pass rate, Mislabeling penalty, where the compliance score is the quantitative result of risk scanning of post content through the preset financial compliance rule library. The compliance score is the difference between 1 and the target ratio, and the target ratio is the ratio of the number of occurrences of risk keywords to the total number of keywords; the expert review pass rate is the ratio of the number of essential posts marked and confirmed by experts to the total number of essential posts marked; the mislabeling penalty includes the first cost and the second cost, where the first cost is the product of the post exposure and the preset unit attention cost, and the second cost is the product of the post knowledge value coefficient and the preset attenuation factor. The preset unit attention cost is determined based on the employee's working hours loss when reading the preset content.

[0010] Optionally, if the second deep Q network is used to determine that the target post is an essential post, after obtaining the processing result of the target node for the target post, the processing result is sent to the target object for review, and the review result is obtained, and the review result is forwarded to the information receiving area or target communication address associated with the target post.

[0011] Optionally, the processing result includes at least: target content, processing conclusion type, user privacy level identifier and result sensitivity label; the processing result is forwarded to the information receiving area associated with the target post or the target communication address information receiving area, including: when the processing conclusion type is a first type used to represent a public reply request, the target content is forwarded to the information receiving area; when the processing conclusion type is a second type used to represent personal transaction processing, the target content is forwarded to the target communication address; when the user privacy level identifier is a first identifier used to represent public authority, the target content is forwarded to the information receiving area; when the user privacy level identifier is a second identifier used to represent private authority, the target content is forwarded to the target communication address; when the result sensitivity label is a first label used to represent privacy data, the target content is forwarded to the target communication address.

[0012] According to another aspect of the present application, an intelligent collaborative management platform driven by artificial intelligence is also provided, including: an acquisition unit for acquiring target posts to be processed in the communication platform and determining a multidimensional feature vector of the target post, wherein the multidimensional feature vector includes: topic classification features, sentiment features, and keyword coding features; a construction unit for constructing a graph neural network, wherein the nodes in the graph neural network include: department nodes and employee nodes, and the edges in the graph neural network include: inter-departmental collaboration edges, department-employee affiliation edges, and inter-employee collaboration edges, and the department node attributes include at least: the number of posts to be processed, the average post processing time, and the professional field code, and the employee node attributes include at least: when Previous state, skill feature vector, historical satisfaction score; an analysis unit, used to use the first deep Q network to analyze the multidimensional feature vector and the graph neural network to obtain the target node in the graph neural network for processing the target post, wherein the state in the first deep Q network includes: the multidimensional feature vector of the current post, the topological structure and node attributes of the graph neural network, and the actions in the first deep Q network include: selecting the target department node, selecting the target employee node, and performing multi-hop decision-making along the graph edge; a forwarding unit, used to forward the target post to the target node, obtain the processing result of the target node for the target post, and forward the processing result to the information receiving area or target communication address associated with the target post.

[0013] According to another aspect of the present application, a non-volatile storage medium is also provided, which includes a stored program, wherein when the program runs, the device where the storage medium is located is controlled to execute the above-mentioned artificial intelligence-driven intelligent collaborative management method.

[0014] According to another aspect of the present application, an electronic device is provided, including: a memory and a processor, the processor being used to run a program stored in the memory, wherein the above-mentioned artificial intelligence-driven intelligent collaborative management method is executed when the program is running.

[0015] According to another aspect of the present application, a computer program is also provided, wherein when the computer program is executed by a processor, the above-mentioned intelligent collaborative management method based on artificial intelligence driving is implemented.

[0016] According to another aspect of the present application, a computer program product is provided, which includes a non-volatile computer-readable storage medium, wherein the non-volatile computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned artificial intelligence-driven intelligent collaborative management method is implemented.

[0017] In this application, the target posts to be processed in the communication platform are obtained, and the multidimensional feature vector of the target posts is determined, wherein the multidimensional feature vector includes: topic classification features, sentiment features, and keyword coding features; a graph neural network is constructed, wherein the nodes in the graph neural network include: department nodes and employee nodes, and the edges in the graph neural network include: inter-departmental collaboration edges, department-employee affiliation edges, and inter-employee collaboration edges. The attributes of the department nodes include at least: the number of posts to be processed, the average processing time of the posts, and the professional field coding. The attributes of the employee nodes include at least: current status, skill feature vector, and historical satisfaction score; the first deep Q network is used to analyze the multidimensional feature vector and the graph neural network to obtain the target node in the graph neural network for processing the target posts, wherein the first deep The states in the Q network include: the multidimensional feature vector of the current post, the topological structure and node attributes of the graph neural network. The actions in the first deep Q network include: selecting the target department node, selecting the target employee node, and executing multi-hop decisions along the graph edge; forwarding the target post to the target node, obtaining the processing result of the target node for the target post, and forwarding the processing result to the information receiving area or target communication address associated with the target post. This achieves the purpose of efficiently and accurately allocating posts in the communication platform to the target nodes in the graph neural network, thereby achieving the technical effect of improving the management efficiency of posts, and thus solving the technical problem of low post management efficiency caused by the inability of related technologies to efficiently and accurately allocate posts in the communication platform to objects suitable for processing the posts. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0019] Figure 1 This is a flow chart of an artificial intelligence-driven intelligent collaborative management method according to an embodiment of the present application;

[0020] Figure 2 This is a structural diagram of an artificial intelligence-driven intelligent collaborative management platform according to an embodiment of the present application;

[0021] Figure 3 This is a hardware structure block diagram of a computer terminal based on an artificial intelligence-driven intelligent collaborative management method according to an embodiment of the present application. DETAILED DESCRIPTION

[0022] In order to enable those skilled in the art to better understand the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of this application.

[0023] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in a sequence other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0024] According to an embodiment of the present application, a method embodiment of an intelligent collaborative management method driven by artificial intelligence is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0025] Figure 1 is a flow chart of an intelligent collaborative management method based on artificial intelligence drive according to an embodiment of the present application, such as Figure 1 As shown, the method includes the following steps:

[0026] Step S102: obtaining a target post to be processed in the communication platform, and determining a multi-dimensional feature vector of the target post, wherein the multi-dimensional feature vector includes: topic classification features, sentiment features, and keyword coding features.

[0027] Among them, communication platforms include but are not limited to: employee forums, instant messaging platforms, and document collaboration platforms.

[0028] It's understandable that in order to more accurately understand and process the massive amount of posts on communication platforms, especially to identify those with potential task value or requiring further action, it's necessary to convert each post of particular interest—the target post—into a structured mathematical representation. This mathematical representation is a "multidimensional feature vector," essentially a list or array of numbers, where each position (dimension) represents a quantified feature value for a specific aspect of the post.

[0029] Specifically, the multidimensional feature vector consists of three core components. Topic classification features, which use natural language processing techniques (such as topic models or text classifiers) to automatically determine the business area or content category that the post primarily discusses, such as "credit products," "internal process optimization," "IT system issues," or "customer service feedback," and use specific numerical codes to represent the topic category to which it belongs. Sentiment features, which reflect the emotional tendency or attitude strength expressed by the author of the post, can be a numerical value calculated through a sentiment analysis model, such as a score ranging from -1 (extremely negative) to +1 (extremely positive), or a more fine-grained classification such as the codes corresponding to positive, neutral, and negative sentiment. Keyword encoding features, which involve identifying and quantifying core terms or phrases that appear in a post (such as "slow approval," "system lag," or "recommendation to add feature X"), can be generated through word embedding technology to form a numerical sequence that captures the key concepts mentioned in the post and their importance.

[0030] By calculating and combining the eigenvalues ​​of these three dimensions, a multidimensional feature vector is ultimately formed. This vector becomes the "digital fingerprint" of the post.

[0031] Step S104: construct a graph neural network, wherein the nodes in the graph neural network include: department nodes and employee nodes, the edges in the graph neural network include: inter-department collaboration edges, department-employee affiliation edges and inter-employee collaboration edges, the department node attributes include at least: the number of posts to be processed, the average post processing time, and the professional field code, and the employee node attributes include at least: current status, skill feature vector, and historical satisfaction score.

[0032] Graph neural networks consist of nodes (or vertices) and edges. Nodes represent entities, and edges represent connections between nodes. Nodes and edges can each have their own feature vectors describing their attributes, and the entire graph can also have global information. By iteratively transferring and aggregating feature information about nodes and their neighbors, and learning representations of nodes and graphs, graph neural networks can capture the dependencies and topological structure between nodes in the graph. A skill feature vector is a vector representation of the multifaceted characteristics of a skill. Each dimension of a skill (such as skill type, proficiency, and application scenario) is quantified as a series of numerical values, and a vector composed of these numerical values ​​is used to comprehensively describe a skill.

[0033] In step S104 above, the nodes in the graph neural network represent the basic building blocks of the enterprise and are primarily divided into two categories: department nodes and employee nodes. Department nodes, such as the Credit Department, Risk Control Department, and IT Support Department, represent functional units within the enterprise; employee nodes represent specific individual members. The connections between nodes are represented by edges. Inter-departmental collaboration edges depict the formal or informal collaborations between departments arising from projects or processes (e.g., collaboration between the Credit Department and the Risk Control Department on approval processes); department-employee affiliation edges clearly identify the subordinate relationship between an employee and their department; and inter-employee collaboration edges reflect the actual work networks formed between individual employees based on historical project collaboration, knowledge sharing, or social interactions (e.g., employees from two different departments frequently collaborate to solve customer problems).

[0034] Furthermore, to enable graph neural networks to effectively learn and automatically assign tasks, each node is assigned attributes describing its status and capabilities. For department nodes, these attributes include at least: the number of pending posts, which quantifies the department's current backlog of pending task requests received from the forum and is a key indicator of departmental workload; the average processing time for posts; a professional field code, which, through a vectorized representation, describes the department's core responsibilities and the types of business it specializes in (such as credit approval, system maintenance, and customer complaints), facilitating task matching to the most appropriate department. For employee nodes, these attributes include at least: the current status (e.g., idle, busy, or on vacation); a skill feature vector, a multidimensional vector that encodes the employee's skills and proficiency (e.g., data analysis - advanced, customer communication - intermediate, Python programming - proficient), which serves as the core basis for accurately matching the required skills to tasks; and a historical satisfaction score, calculated based on the quality and timeliness of past task completions or peer reviews, which measures the employee's reliability and performance.

[0035] Step S106: Use the first deep Q network to analyze the multidimensional feature vector and the graph neural network to obtain the target node in the graph neural network for processing the target post, wherein the state in the first deep Q network includes: the multidimensional feature vector of the current post, the topological structure and node attributes of the graph neural network, and the actions in the first deep Q network include: selecting the target department node, selecting the target employee node, and performing multi-hop decision-making along the graph edge.

[0036] Step S106 is used to combine the multidimensional feature vector representing the core information of the post to be processed in step S102 and the graph neural network that depicts the internal structure, capabilities and relationships of the enterprise in step S104, and use reinforcement learning technology (specifically deep Q network, DQN) to automatically and intelligently decide which department or employee should be responsible for handling the task requirements contained in the post.

[0037] In step S106, the state of the first deep Q-network includes the multidimensional feature vector of the current target post. This multidimensional feature vector provides core information about the task requirements, such as the topic of discussion (e.g., "credit approval process optimization"), the sentiment expressed (e.g., "negative," indicating serious process problems"), and key demand terms (e.g., "simplification," "acceleration," and "automation"). The state also includes the current topology of the entire graph neural network and the real-time properties of all nodes within the network. The current topology reflects the collaborative relationships between departments, the affiliations between departments and employees, and the possible collaborative links between employees. The real-time properties of all nodes include the current number of posts to be processed (load), average processing time (efficiency), and expertise (ability match) for each department, as well as the current status of employees (whether they are available), skill vectors (whether they possess the skills required to handle the task), and historical satisfaction scores (reliability and performance). It can be understood that this state comprehensively captures "what the task is" and "what departments and employees can currently do."

[0038] Actions primarily fall into three categories: selecting a target department node (directly assigning a task to a specific department), selecting a target employee node (directly assigning a task to a specific employee), and executing multi-hop decisions along graph edges. Multi-hop decisions are particularly important, simulating the flow of posts (or tasks in reinforcement learning) through graph neural networks. For example, consider the decision "Step 1: Select Department A -> Step 2: Transfer to Department B along the collaboration edge between Department A and Department B -> Step 3: Select Employee C within Department B." Or "Step 1: Select Employee X -> Step 2: Transfer to Employee Y along the collaboration edge between Employee X and Employee Y." This action design allows for flexible handling of situations requiring cross-departmental collaboration or the search for specific experts, significantly enhancing the rationality and adaptability of decision-making.

[0039] Notably, multi-hop decision-making allows the agent to perform continuous, path-like movements along a graph structure composed of department and employee nodes and the various edges connecting them (inter-department collaboration edges, department-employee affiliation edges, and inter-employee collaboration edges), ultimately reaching one or more target nodes. Specifically, each "hop" represents an action chosen by the DQN in its current state. This action is not a final assignment, but rather instructs the system to move to the next adjacent "node" along a specific "edge" in the graph. For example: The first hop action: Starting from a starting point (e.g., department node A, initially associated with a post based on its topic), the agent moves along an inter-department collaboration edge to another department node B. This simulates a task being transferred from department A to department B due to cross-departmental collaboration. The second hop action: After reaching department node B, the agent decides to move from department node B to an employee node C, belonging to that department, along a department-employee affiliation edge. This indicates that the task has been assigned to a specific employee C within department B. The third hop action: The DQN may also, from employee node C, decide to move along an inter-employee collaboration edge to another employee node D (even though D may belong to another department). This simulates the situation where employee C believes that he or she cannot complete the task independently, or believes that employee D is a more suitable expert, and therefore sends a task referral or collaboration request to D.

[0040] This "jump" process continues until the DQN finds the most appropriate final processing node (target department or employee), or until the maximum number of hops is reached. The choice of each "hop" (which edge to take, which adjacent node to reach) depends on the DQN's analysis and evaluation of the current combined state (the post's feature vector and the real-time structure / properties of the entire graph), with the goal of maximizing long-term cumulative reward.

[0041] Step S108: forward the target post to the target node, obtain the processing result of the target node for the target post, and forward the processing result to the information receiving area or target communication address associated with the target post.

[0042] According to the above steps, the target posts to be processed in the communication platform are obtained, and the multidimensional feature vector of the target posts is determined, wherein the multidimensional feature vector includes: topic classification features, sentiment features, and keyword coding features; a graph neural network is constructed, wherein the nodes in the graph neural network include: department nodes and employee nodes, and the edges in the graph neural network include: inter-departmental collaboration edges, department-employee affiliation edges, and inter-employee collaboration edges. The attributes of the department nodes include at least: the number of posts to be processed, the average processing time of the posts, and the professional field code. The attributes of the employee nodes include at least: the current status, the skill feature vector, and the historical satisfaction score; the first deep Q network is used to analyze the multidimensional feature vector and the graph neural network. , obtaining the target node for processing the target post in the graph neural network, wherein the state in the first deep Q network includes: the multidimensional feature vector of the current post, the topological structure and node attributes of the graph neural network, and the actions in the first deep Q network include: selecting the target department node, selecting the target employee node, and performing multi-hop decision-making along the graph edge; forwarding the target post to the target node, obtaining the processing result of the target node for the target post, and forwarding the processing result to the information receiving area or target communication address associated with the target post, thereby achieving the purpose of efficiently and accurately allocating the posts in the communication platform to the target nodes in the graph neural network, thereby achieving the technical effect of improving the management efficiency of the posts.

[0043] The following Figure 1 The steps shown are exemplified and explained.

[0044] According to some optional embodiments of the present application, the reward function of the first deep Q network in step S106 is determined by the following method: determining a first sub-reward function based on the matching score, wherein the matching score is used to characterize the degree of semantic matching between the post content and the node's professional field, and the matching score is obtained by calculating the similarity between the post keywords and the node's preset tags using natural language processing technology; determining a second sub-reward function based on the time delay, wherein the time delay is used to characterize the time difference from the post creation timestamp to the successful allocation timestamp; determining a third sub-reward function based on the load index, wherein the load index is used to characterize the ratio of the node's current task queue length to the node's maximum processing capacity; determining a fourth sub-reward function based on the satisfaction score, wherein the satisfaction score is used to characterize the feedback rating provided by the user after the post processing is completed; determining the reward function based on the first sub-reward function, the second sub-reward function, the third sub-reward function, and the fourth sub-reward function.

[0045] In this embodiment, the first sub-reward function is based on a matching score. The matching score is used to accurately measure the degree of semantic fit between the task requirements contained in the target post (reflected by keyword encoding, topic classification, etc. in its multidimensional feature vector) and the professional capabilities of the potential processing node (department or employee). Specifically, natural language processing techniques (such as word vector similarity calculation and topic model matching analysis) can be used to calculate the similarity score between the keywords or core semantics extracted from the post and the node's preset professional field label (such as the department's "professional field code" or the core capability label in the employee's "skill feature vector"). The higher the matching score, the more expertise the node has in handling the task, and the greater the potential for task completion quality and efficiency.

[0046] Time latency refers to the difference between the timestamp of a forum post creation and the timestamp of the task represented by that post being successfully assigned to the target node. This metric directly reflects the efficiency of the task assignment process. Excessive latency can lead to delayed response times, impacting business operations and employee experience. The load index assesses the busyness and resource availability of the target node (department or employee) at the time of task assignment. The load index is defined as the ratio of the node's current queue length for pending tasks (such as the number of pending posts for a department or the number of tasks currently assigned to an employee) to the node's maximum processing capacity. A higher ratio indicates a more overloaded node, and assigning new tasks at this point may result in processing delays, reduced quality, or even system crashes. The satisfaction score is the final quality feedback after the task is closed. It is typically derived from feedback provided by the post initiator or relevant users after the task is completed (such as a five-star rating or a satisfied / unsatisfied label). The satisfaction score directly reflects whether the task execution results have truly solved the problem and met the needs.

[0047] Specifically, the above-mentioned determination of the reward function based on the first sub-reward function, the second sub-reward function, the third sub-reward function, and the fourth sub-reward function can be achieved by the following method: obtaining the initial weight coefficient of each sub-reward function; continuously detecting the post allocation success rate index, the average processing delay index, and the user satisfaction mean index; when the post allocation success rate index is detected to be decreasing, increasing the first initial weight coefficient corresponding to the first sub-reward function and the fourth initial weight coefficient corresponding to the fourth sub-reward function; when the average processing delay index is detected to be increasing, increasing the second initial weight coefficient corresponding to the second sub-reward function and the third initial weight coefficient corresponding to the third sub-reward function; when the user satisfaction mean index is detected to be decreasing, increasing the fourth initial weight coefficient; performing a normalization operation after each weight coefficient update so that the sum of the updated weight coefficients is 1, and obtaining the first weight coefficient, the second weight coefficient, the third weight coefficient, and the fourth weight coefficient; determining the reward function of the first deep Q network according to the first sub-reward function, the first weight coefficient, the second sub-reward function, the second weight coefficient, the third sub-reward function, the third weight coefficient, and the fourth sub-reward function, the fourth weight coefficient.

[0048] In this embodiment, first, an initial weight coefficient is set for each sub-reward function (for example, based on business experience or preliminary experimental settings: matching coefficient , delay coefficient , load factor , satisfaction coefficient Continuously monitor several core business operation indicators: post assignment success rate indicator (the proportion of posts successfully converted into tasks and assigned), average processing delay indicator (the total average time from post creation to task completion), and user satisfaction mean indicator (the average score of user feedback on task processing results).

[0049] Next, according to the changing trends of these monitoring indicators, the corresponding weight coefficients are dynamically adjusted:

[0050] 1. When the post assignment success rate indicator is detected to be decreasing, it means that there is a problem in the task assignment process. It may be that the task requirements do not match the processor's capabilities (low matching degree), or the processing results fail to satisfy the user (resulting in a decrease in the willingness to convert to subsequent tasks). Therefore, it is necessary to increase the weight coefficient corresponding to the first sub-reward function (matching score) And the weight coefficient corresponding to increasing the fourth sub-reward function (satisfaction score) The purpose of this is to guide DQN to pay more attention to selecting nodes with highly matched professional capabilities and ultimately achieving high satisfaction when making decisions, thereby increasing the probability of tasks being successfully accepted and effectively completed.

[0051] 2. When the average processing delay indicator is detected to be increasing, it indicates that the overall cycle from task submission to completion has become longer and the efficiency has decreased. The delay may be due to the fact that the allocation process itself takes too long (large time delay) or that the task is assigned to an already busy node (high load), causing the task to be backlogged in the queue. Therefore, it is necessary to increase the weight coefficient corresponding to the second sub-reward function (time delay). And the weight coefficient corresponding to increasing the third sub-reward function (load index) This in turn prompts DQN to prioritize nodes that can quickly respond to allocation requests (low time latency) and are currently lightly loaded (low load index), thereby shortening the overall processing cycle.

[0052] 3. When the user satisfaction index is detected to be declining, it reflects that the final quality of the task processing results is poor and the users are dissatisfied. Although there may be many reasons for the decline in satisfaction, increasing the weight of the fourth sub-reward is the most direct response. Therefore, it is necessary to increase the weight coefficient corresponding to the fourth sub-reward function (satisfaction score) .

[0053] Finally, after adjusting one or more weight coefficients according to the above rules, a normalization operation is performed to ensure that the sum of all four updated weight coefficients is strictly equal to 1. The reward function of the first deep Q network is obtained by multiplying these four dynamically adjusted weight coefficients by the corresponding sub-reward function values ​​and then adding them together.

[0054] This dynamically constructed reward function intelligently adjusts the relative importance (weight coefficients) of each dimension. When the allocation success rate is low, it prioritizes matching and satisfaction; when processing latency is high, it prioritizes fast response and load balancing; and when users are dissatisfied, ultimate satisfaction is paramount. This closed-loop feedback mechanism enables DQN to continuously learn and optimize its allocation strategy to meet evolving business needs, ultimately driving the entire intelligent task allocation system toward optimal overall performance.

[0055] According to some other optional embodiments of the present application, determining the multi-dimensional feature vector of the target post can be achieved by the following method: extracting the semantic embedding vector of the target post using a pre-trained language model; calculating different probability distributions corresponding to different topic categories through a topic attention layer; normalizing the different probability distributions to obtain a first feature component; calculating the vocabulary strength score of the target post based on a financial sentiment dictionary, wherein the financial sentiment dictionary includes a plurality of different preset words for representing positive emotions, neutral emotions and negative emotions, and the vocabulary strength score is used to quantify the strength of the emotional tendency explicitly expressed in the target post; determining the emotional probability of the target post through a deep learning model, wherein the emotional probability is used to indicate the probability that the target post belongs to positive emotions, neutral emotions and negative emotions; and obtaining the vocabulary strength score by dynamically weighting the vocabulary strength according to the density of financial terms. The second feature component is obtained by combining the financial term density and sentiment probability, wherein the financial term density is the ratio of the number of financial professional terms appearing in the target post to the total number of words in the target post; the common stop words and high-frequency financial function words in the target post are removed, and the professional terms whose word frequency-inverse document frequency exceeds the preset threshold are retained; the core words in the professional terms are screened by the centrality measurement index of the term co-occurrence graph, and the third feature component is determined based on the core words, wherein the term co-occurrence graph is used to represent the relationship of the co-occurrence of professional terms retained after screening in the target post, the nodes of the term co-occurrence graph are the screened professional terms, and the edges of the term co-occurrence graph are the semantic associations between the professional terms; the first feature component, the second feature component and the third feature component are spliced, and the splicing result is reduced in dimensionality through an orthogonal constrained linear transformation layer to obtain a multidimensional feature vector.

[0056] In this example, the original text of the target post is first encoded using a pre-trained language model (such as BERT or RoBERTa), generating a semantic embedding vector that contains overall semantic information. Next, to capture the core business areas discussed in the post, a topic attention layer is introduced. This layer receives the semantic embedding vector and calculates the probability distribution of the post belonging to different pre-defined topic categories (e.g., "credit products," "risk control," "operations and maintenance," "customer service," "internal processes," etc.). This probability distribution is normalized (ensuring that all probabilities sum to 1) to form the first feature component representing the primary discussion topic of the post, clearly quantifying the topic orientation of the post content.

[0057] Furthermore, a two-pronged strategy is employed for sentiment analysis to improve accuracy in the financial context. First, a post's lexical strength score is calculated based on a specially constructed financial sentiment lexicon (containing a large number of pre-defined financial-related terms labeled as positive, neutral, or negative). This score directly quantifies the strength of the sentiment expressed explicitly in the text by counting and analyzing the occurrence of sentiment words in the post and their strength (possibly taking into account word frequency, position, and modifiers). Second, a deep learning sentiment analysis model is used to predict the probability distribution of whether the post as a whole has positive, neutral, or negative sentiment. For example, a recurrent neural network (RNN) introduces a loop structure within the network, enabling information to be passed along the sequence, thereby capturing sequential information in the text. For a post, the model processes each word (or character) individually. Each word is converted into a fixed-dimensional vector through an embedding layer, which represents the word's semantic information. The RNN then uses the information from previous words to influence the processing of subsequent words, thereby understanding the context and relationship of the text. Through a multi-layered neural network architecture, the model gradually extracts higher-level text features. At the final stage of the model, there is an output layer, which can be a fully connected neural network layer. Its number of output nodes corresponds to the sentiment category, such as positive, neutral, and negative. Each node outputs a probability value. These probabilities are processed through the softmax function so that their sum is 1, thus forming a probability distribution. The probability distribution can intuitively represent the likelihood of a post belonging to each sentiment category. For example, a post may have a 0.7 probability of positive sentiment, a 0.2 probability of neutral sentiment, and a 0.1 probability of negative sentiment. These probabilities can be used to determine the overall sentiment of the post and also reflect the degree of uncertainty in the sentiment classification.

[0058] Notably, to combine the strengths of these two methods and adapt to the specific characteristics of financial text, financial term density (the ratio of the number of financial terminology appearing in a post to the total number of words in the post) is introduced as a dynamic weighting factor. Specifically, when the financial term density is above a threshold, it indicates a highly specialized text and relies more heavily on the lexical strength score derived from the financial sentiment lexicon. When the financial term density is below the threshold, the sentiment probability predicted by the deep learning sentiment analysis model is more favored. These two results are dynamically weighted and fused based on the term density to generate a second feature component that reflects the sentiment orientation and strength of the post.

[0059] Furthermore, to identify the core appeals and professional focus of posts, we focus on extracting key professional terms. Post text is preprocessed to remove common stop words and high-frequency financial function words (such as certain high-frequency but low-information financial connectives). Then, professional terms whose term frequency-inverse document frequency value exceeds a preset threshold are retained. These filtered terms constitute the candidate keyword set. To further identify the most core and representative terms, a term co-occurrence graph is constructed. The nodes in the term co-occurrence graph represent each filtered professional term, and the edges between nodes are established based on the frequency of co-occurrence or semantic affinity (calculated using word embeddings) of the terms within the post context window. Centrality metrics (such as degree centrality, closeness centrality, or eigenvector centrality) are calculated for each term node in the graph. Terms with high centrality are considered to be central to the post's semantic network and more representative of the post's core issues. These filtered core terms are used to form the third feature component, which can be a vector obtained by taking the average or weighted average of their word embeddings (with weights determined by centrality).

[0060] Finally, the first feature component representing thematic orientation, the second feature component representing sentiment intensity, and the third feature component embodying the core professional appeal are concatenated to form a higher-dimensional temporary vector. To compress this high-dimensional temporary vector into a practical, information-dense, and dimensionally controllable multidimensional feature vector while reducing redundancy between features, a linear transformation layer with an orthogonal constraint (e.g., a linear layer whose weight matrices are constrained to be approximately orthogonal) is used for dimensionality reduction. The orthogonal constraint helps preserve as much information as possible from the original feature space and reduces correlation between features during the dimensionality reduction process. The final output is a low-dimensional but highly information-retaining multidimensional feature vector that comprehensively encodes the post's information across the three key dimensions of topic, sentiment, and core professional appeal.

[0061] In some optional embodiments of the present application, after obtaining the target post to be processed in the communication platform, the method further includes: encoding the target post using a pre-trained language model to obtain a semantic feature vector; determining the click-through rate normalization value and the poster's rank weight of the target post; splicing the semantic feature vector, the click-through rate normalization value and the poster's rank weight, and determining the splicing result as a state vector; using a second deep Q network to analyze the state vector to determine whether the target post is an essence post, wherein the action space of the second deep Q network includes: marking essence posts and not marking essence posts; the reward function of the second deep Q network includes: compliance score, expert review score, and expert feedback score. The pass rate and the penalty for mislabeling include: the compliance score is the quantitative result of risk scanning of post content through the preset financial compliance rule base; the compliance score is the difference between 1 and the target ratio; the target ratio is the ratio of the number of occurrences of risk keywords to the total number of keywords; the expert review pass rate is the ratio of the number of essential posts marked and confirmed by experts to the total number of essential posts marked; the penalty for mislabeling includes the first cost and the second cost; the first cost is the product of the post exposure and the preset unit attention cost; the second cost is the product of the post knowledge value coefficient and the preset attenuation factor; the preset unit attention cost is determined based on the employee's working hours loss when reading the preset content.

[0062] This embodiment first uses a pre-trained language model to perform deep semantic encoding on the target post, generating a semantic feature vector containing contextual information. This vector captures the post's core value in terms of business insights and knowledge depth. Simultaneously, two key auxiliary metrics are calculated: a normalized click-through rate (which converts the post's historical clicks into a relative popularity index in the [0,1] range, eliminating the impact of traffic differences across different sections) and the poster's rank weight (a preset weight coefficient based on the poster's rank within the enterprise structure, such as "department director = 0.9, general employee = 0.5," reflecting differences in authority). These three types of information are fused into a comprehensive state vector through a splicing operation, forming a complete input representing the post's value potential.

[0063] Secondly, the state vector is input into the second deep Q network for decision analysis. The action space of the network includes a binary choice: marking the essence post or not marking the essence post. The reward function adopts a triple constraint mechanism to ensure the quality of decision-making: the compliance score is calculated by scanning the preset financial compliance rule library, and the specific formula is 1 - (number of risk keywords / total number of keywords), and the score tends to 1, indicating full compliance, and if it is lower than the threshold, a negative reward is triggered; the expert review pass rate is used as a feedback supervision signal, which is calculated as the ratio of the number of essence posts marked and confirmed by the field experts to the total number of marked posts (for example, if 80 out of 100 posts are approved by experts, the reward is 0.8); the mislabeling penalty quantifies the cost of wrong decisions, including two key cost items: the first cost is the exposure amount of the post x the preset unit attention cost (the unit cost is converted into monetary value according to the time loss of employees reading typical documents), which is used to punish the attention resources occupied by low-quality posts; the second cost is the post knowledge value coefficient x the preset decay factor (the knowledge value coefficient is calculated by backtracking indicators such as the number of knowledge base references, and the decay factor increases exponentially with the delay of the mislabeling discovery time), which is used to punish the implicit loss of missing high-value content.

[0064] Finally, the intelligent decision is realized by dynamically balancing the reward factors, when the compliance score is low, the second deep Q network avoids risky content; when the expert review pass rate decreases, the second deep Q network will tighten the marking standard; the double-cost design of mislabeling penalty ensures a balance between "reducing noise interference" and "avoiding loss of knowledge assets". The second deep Q network learns the optimal strategy in continuous training, that is, only when the comprehensive state vector of the post indicates that it can bring positive net benefits (expected reward > 0), the essence marking is triggered, so as to realize the accurate capture of knowledge value.

[0065] Further, the above embodiment further includes the following steps: if it is determined by the second deep Q network that the target post is an essence post, after obtaining the processing result of the target node for the target post, the processing result is sent to the target object for review, and a review result is obtained, and the review result is forwarded to the information receiving area or the target communication address associated with the target post.

[0066] As some optional embodiments of the present application, the processing result at least includes: target content, processing conclusion type, user privacy level identifier and result sensitivity label.

[0067] Furthermore, forwarding the processing result to the information receiving area associated with the target post or the target communication address information receiving area can be achieved by the following method: when the processing conclusion type is the first type used to indicate a public reply request, forwarding the target content to the information receiving area; when the processing conclusion type is the second type used to indicate personal transaction processing, forwarding the target content to the target communication address; when the user privacy level is identified as the first identifier used to indicate public authority, forwarding the target content to the information receiving area; when the user privacy level is identified as the second identifier used to indicate private authority, forwarding the target content to the target communication address; when the result sensitivity label is the first label used to indicate privacy data, forwarding the target content to the target communication address.

[0068] Figure 2 This is a structural diagram of an artificial intelligence-driven intelligent collaborative management platform according to an embodiment of the present application, such as Figure 2 As shown, the platform includes:

[0069] The acquisition unit 22 is used to acquire the target post to be processed in the communication platform and determine the multidimensional feature vector of the target post, wherein the multidimensional feature vector includes: topic classification features, sentiment features, and keyword coding features.

[0070] Construction unit 24 is used to construct a graph neural network, wherein the nodes in the graph neural network include: department nodes and employee nodes, the edges in the graph neural network include: inter-department collaboration edges, department-employee affiliation edges and inter-employee collaboration edges, the department node attributes include at least: the number of posts to be processed, the average post processing time, and the professional field code, and the employee node attributes include at least: current status, skill feature vector, and historical satisfaction score.

[0071] An analysis unit 26 is configured to analyze the multidimensional feature vector and the graph neural network using the first deep Q network to obtain a target node in the graph neural network for processing a target post, wherein the state in the first deep Q network includes: the multidimensional feature vector of the current post, the topological structure and node attributes of the graph neural network; and the actions in the first deep Q network include: selecting a target department node, selecting a target employee node, and executing multi-hop decisions along the graph edge.

[0072] The forwarding unit 28 is configured to forward the target post to the target node, obtain the processing result of the target node for the target post, and forward the processing result to the information receiving area or target communication address associated with the target post.

[0073] Optionally, the reward function of the first deep Q network is determined by the following method: determining a first sub-reward function based on the matching score, wherein the matching score is used to characterize the degree of semantic matching between the post content and the node's professional field, and the matching score is obtained by calculating the similarity between the post keywords and the node's preset tags using natural language processing technology; determining a second sub-reward function based on the time delay, wherein the time delay is used to characterize the time difference from the post creation timestamp to the successful allocation timestamp; determining a third sub-reward function based on the load index, wherein the load index is used to characterize the ratio of the node's current task queue length to the node's maximum processing capacity; determining a fourth sub-reward function based on the satisfaction score, wherein the satisfaction score is used to characterize the feedback rating provided by the user after the post processing is completed; determining the reward function based on the first sub-reward function, the second sub-reward function, the third sub-reward function, and the fourth sub-reward function.

[0074] Optionally, a reward function is determined based on the first sub-reward function, the second sub-reward function, the third sub-reward function, and the fourth sub-reward function, including the following steps: obtaining an initial weight coefficient for each sub-reward function; continuously detecting a post allocation success rate indicator, an average processing delay indicator, and a user satisfaction mean indicator; when it is detected that the post allocation success rate indicator decreases, increasing the first initial weight coefficient corresponding to the first sub-reward function and the fourth initial weight coefficient corresponding to the fourth sub-reward function; when it is detected that the average processing delay indicator increases, increasing the second initial weight coefficient corresponding to the second sub-reward function and the third initial weight coefficient corresponding to the third sub-reward function; when it is detected that the user satisfaction mean indicator decreases, increasing the fourth initial weight coefficient; performing a normalization operation after each weight coefficient update so that the sum of the updated weight coefficients is 1, and obtaining a first weight coefficient, a second weight coefficient, a third weight coefficient, and a fourth weight coefficient; determining the reward function of the first deep Q network based on the first sub-reward function, the first weight coefficient, the second sub-reward function, the second weight coefficient, the third sub-reward function, the third weight coefficient, and the fourth sub-reward function, the fourth weight coefficient.

[0075] Optionally, determining the multidimensional feature vector of the target post includes the following steps: extracting the semantic embedding vector of the target post using a pre-trained language model; calculating different probability distributions corresponding to different topic categories through a topic attention layer; normalizing the different probability distributions to obtain a first feature component; calculating the vocabulary strength score of the target post based on a financial sentiment dictionary, wherein the financial sentiment dictionary includes a plurality of different preset words for representing positive emotions, neutral emotions, and negative emotions, and the vocabulary strength score is used to quantify the strength of the emotional tendency explicitly expressed in the target post; determining the emotional probability of the target post through a deep learning model, wherein the emotional probability is used to represent the probability that the target post belongs to positive emotions, neutral emotions, and negative emotions; dynamically weighting the lexical strength score and the emotional probability according to the density of financial terms to obtain to the second feature component, where the financial term density is the ratio of the number of financial professional terms appearing in the target post to the total number of words in the target post; remove common stop words and high-frequency financial function words in the target post, and retain professional terms whose word frequency-inverse document frequency exceeds a preset threshold; screen the core words in the professional terms through the centrality measurement index of the term co-occurrence graph, and determine the third feature component based on the core words, where the term co-occurrence graph is used to represent the relationship of the co-occurrence of professional terms retained after screening in the target post, the nodes of the term co-occurrence graph are the screened professional terms, and the edges of the term co-occurrence graph are the semantic associations between the professional terms; splice the first feature component, the second feature component and the third feature component, and perform dimensionality reduction processing on the splicing result through an orthogonal constrained linear transformation layer to obtain a multidimensional feature vector.

[0076] Optionally, after obtaining the target post to be processed in the communication platform, the following steps are also included: using a pre-trained language model to encode the target post to obtain a semantic feature vector; determining the click-through rate normalization value and the poster's rank weight of the target post; splicing the semantic feature vector, the click-through rate normalization value and the poster's rank weight, and determining the splicing result as a state vector; using a second deep Q network to analyze the state vector to determine whether the target post is an essence post, wherein the action space of the second deep Q network includes: marking essence posts and not marking essence posts; the reward function of the second deep Q network includes: compliance score, expert review pass rate , mislabeling penalty, where the compliance score is the quantitative result of risk scanning of post content through the preset financial compliance rule library, the compliance score is the difference between 1 and the target ratio, and the target ratio is the ratio of the number of occurrences of risk keywords to the total number of keywords; the expert review pass rate is the ratio of the number of essential posts marked and confirmed by experts to the total number of essential posts marked; the mislabeling penalty includes the first cost and the second cost, where the first cost is the product of the post exposure and the preset unit attention cost, and the second cost is the product of the post knowledge value coefficient and the preset attenuation factor, and the preset unit attention cost is determined based on the employee's working hours loss when reading the preset content.

[0077] Optionally, if the target post is determined to be an essential post using a second deep Q network, after obtaining the processing result of the target node for the target post, the processing result is sent to the target object for review, and the review result is obtained, and the review result is forwarded to the information receiving area or target communication address associated with the target post.

[0078] Optionally, the processing result includes at least: target content, processing conclusion type, user privacy level identifier, and result sensitivity label. Forwarding the processing result to the information receiving area associated with the target post or the target communication address information receiving area includes the following steps: when the processing conclusion type is the first type used to indicate a public reply request, forwarding the target content to the information receiving area; when the processing conclusion type is the second type used to indicate personal transaction processing, forwarding the target content to the target communication address; when the user privacy level identifier is the first identifier used to indicate public authority, forwarding the target content to the information receiving area; when the user privacy level identifier is the second identifier used to indicate private authority, forwarding the target content to the target communication address; when the result sensitivity label is the first label used to indicate private data, forwarding the target content to the target communication address.

[0079] It should be noted that the above Figure 2 Each module in the embodiment can be a program module (for example, a set of program instructions that implement a specific function) or a hardware module. For the latter, it can be expressed in the following forms, but is not limited to this: each of the above modules is expressed in the form of a processor, or the functions of each of the above modules are implemented by a processor.

[0080] It should be noted that Figure 2 The preferred implementation of the embodiment shown can be found in Figure 1 The relevant description of the illustrated embodiment will not be repeated here.

[0081] Figure 3 The following is a hardware block diagram of a computer terminal for implementing an intelligent collaborative management method driven by artificial intelligence. Figure 3As shown, the computer terminal 30 may include one or more processors 302 (illustrated as 302a, 302b, ..., 302n in the figure) (the processor 302 may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.), a memory 304 for storing data, and a transmission module 306 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the BUS), a network interface, a power supply, and / or a camera. Those skilled in the art will understand that Figure 3 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 3 More or fewer components than shown, or with Figure 3 Different configurations shown.

[0082] It should be noted that the one or more processors 302 and / or other data processing circuits described above may generally be referred to herein as "data processing circuitry." The data processing circuitry may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuitry may be a single, independent processing module, or may be fully or partially integrated into any of the other components of the computer terminal 30. As discussed in the embodiments of the present application, the data processing circuitry functions as a processor control (e.g., the selection of a variable resistor terminal path connected to an interface).

[0083] The memory 304 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the intelligent collaborative management method driven by artificial intelligence in the embodiment of the present application. The processor 302 executes various functional applications and data processing by running the software programs and modules stored in the memory 304, that is, realizing the above-mentioned intelligent collaborative management method driven by artificial intelligence. The memory 304 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 304 may further include a memory remotely located relative to the processor 302, and these remote memories may be connected to the computer terminal 30 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof.

[0084] Transmission module 306 is configured to receive or transmit data via a network. A specific example of the aforementioned network may include a wireless network provided by the communications provider of computer terminal 30. In one embodiment, transmission module 306 includes a network interface controller (NIC), which can be connected to other network devices via a base station to enable communication with the Internet. In another embodiment, transmission module 306 may be a radio frequency (RF) module, configured to communicate with the Internet wirelessly.

[0085] The display may be, for example, a touch screen liquid crystal display (LCD) that enables a user to interact with a user interface of the computer terminal 30 .

[0086] It should be noted that, in some optional embodiments, the above Figure 3 The computer terminal shown may include hardware elements (including circuits), software elements (including computer code stored on a computer-readable medium), or a combination of hardware elements and software elements. Figure 3 This is merely one example of a particular embodiment and is intended to illustrate the types of components that may be present in the computer terminal described above.

[0087] It should be noted that Figure 3 The computer terminal shown is used to execute Figure 1 The intelligent collaborative management method based on artificial intelligence driving is shown in the figure, so the relevant explanations in the execution method of the above command are also applicable to the electronic device and will not be repeated here.

[0088] An embodiment of the present application also provides a non-volatile storage medium, which includes a stored program, wherein when the program runs, the device where the storage medium is located is controlled to execute the above-mentioned artificial intelligence-driven intelligent collaborative management method.

[0089] A program for a non-volatile storage medium to perform the following functions: obtaining a target post to be processed in a communication platform and determining a multidimensional feature vector of the target post, wherein the multidimensional feature vector includes: topic classification features, sentiment features, and keyword encoding features; constructing a graph neural network, wherein the nodes in the graph neural network include: department nodes and employee nodes, and the edges in the graph neural network include: inter-department collaboration edges, department-employee affiliation edges, and inter-employee collaboration edges, and the attributes of the department node include at least: the number of posts to be processed, the average post processing time, and the professional field code, and the attributes of the employee node include at least: the current status, the skill feature vector, and the historical satisfaction score; using a first deep Q network to analyze the multidimensional feature vector and the graph neural network to obtain a target node in the graph neural network for processing the target post, wherein the state in the first deep Q network includes: the multidimensional feature vector of the current post, the topological structure and node attributes of the graph neural network, and the actions in the first deep Q network include: selecting a target department node, selecting a target employee node, and performing multi-hop decisions along the graph edges; forwarding the target post to the target node, obtaining a processing result of the target node for the target post, and forwarding the processing result to an information receiving area or a target communication address associated with the target post.

[0090] An embodiment of the present application also provides an electronic device, including: a memory and a processor, the processor being configured to run a program stored in the memory, wherein the above-mentioned artificial intelligence-driven intelligent collaborative management method is executed when the program is running.

[0091] The processor is used to run a program that performs the following functions: obtaining a target post to be processed in a communication platform and determining a multidimensional feature vector of the target post, wherein the multidimensional feature vector includes: topic classification features, sentiment features, and keyword encoding features; constructing a graph neural network, wherein the nodes in the graph neural network include: department nodes and employee nodes, and the edges in the graph neural network include: inter-departmental collaboration edges, department-employee affiliation edges, and inter-employee collaboration edges, and the department node attributes include at least: the number of posts to be processed, the average post processing time, and the professional field code, and the employee node attributes include at least: current status, skill feature vector, and historical satisfaction score; using a first deep Q network to analyze the multidimensional feature vector and the graph neural network to obtain a target node in the graph neural network for processing the target post, wherein the state in the first deep Q network includes: the multidimensional feature vector of the current post, the topological structure and node attributes of the graph neural network, and the actions in the first deep Q network include: selecting a target department node, selecting a target employee node, and performing multi-hop decisions along the graph edges; forwarding the target post to the target node, obtaining the processing result of the target node for the target post, and forwarding the processing result to an information receiving area or a target communication address associated with the target post.

[0092] The serial numbers of the above embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0093] In the above embodiments of the present application, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0094] In the above-mentioned embodiments of the present application, the collected information is information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data comply with relevant laws, regulations and standards, take necessary protection measures, do not violate public order and good morals, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0095] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only exemplary. For example, the division of the units can be a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0096] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0097] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0098] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the relevant technology, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a number of instructions for enabling a computer device (which can be a personal computer, server or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk, etc. Various media that can store program code.

[0099] The above is only a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. An intelligent collaborative management method based on artificial intelligence, characterized in that: include: Obtaining a target post to be processed in the communication platform, and determining a multidimensional feature vector of the target post, wherein the multidimensional feature vector includes: topic classification features, sentiment features, and keyword encoding features; Constructing a graph neural network, wherein the nodes in the graph neural network include department nodes and employee nodes, the edges in the graph neural network include inter-department collaboration edges, department-employee affiliation edges, and inter-employee collaboration edges, the department node attributes include at least the number of pending posts, the average post processing time, and the professional field code, and the employee node attributes include at least the current status, skill feature vector, and historical satisfaction score; Analyzing the multidimensional feature vector and the graph neural network using a first deep Q-network to obtain a target node in the graph neural network for processing the target post, wherein the state in the first deep Q-network includes: the multidimensional feature vector of the current post, the topological structure and node attributes of the graph neural network, and the actions in the first deep Q-network include: selecting a target department node, selecting a target employee node, and performing multi-hop decision-making along graph edges; The target post is forwarded to the target node, a processing result of the target node on the target post is obtained, and the processing result is forwarded to an information receiving area or a target communication address associated with the target post.

2. The method according to claim 1, characterized in that The reward function of the first deep Q network is determined by the following method: Determine a first sub-reward function based on a matching score, wherein the matching score is used to represent the degree of semantic matching between the post content and the node's professional field. The matching score is obtained by calculating the similarity between the post keywords and the node's preset tags using natural language processing technology; determining a second sub-reward function according to a time delay, wherein the time delay is used to represent a time difference from a post creation timestamp to a successful assignment timestamp; Determining a third sub-reward function according to a load index, wherein the load index is used to represent a ratio of a current task queue length of the node to a maximum processing capacity of the node; determining a fourth sub-reward function based on the satisfaction score, wherein the satisfaction score is used to represent a feedback rating provided by the user after the post is processed; The reward function is determined according to the first sub-reward function, the second sub-reward function, the third sub-reward function, and the fourth sub-reward function.

3. The method according to claim 2, characterized in that Determining the reward function according to the first sub-reward function, the second sub-reward function, the third sub-reward function, and the fourth sub-reward function includes: Get the initial weight coefficient of each sub-reward function; Continuously monitor post assignment success rate, average processing delay, and user satisfaction average. When it is detected that the post distribution success rate indicator decreases, the first initial weight coefficient corresponding to the first sub-reward function and the fourth initial weight coefficient corresponding to the fourth sub-reward function are increased; when it is detected that the average processing delay indicator increases, the second initial weight coefficient corresponding to the second sub-reward function and the third initial weight coefficient corresponding to the third sub-reward function are increased; when it is detected that the user satisfaction mean indicator decreases, the fourth initial weight coefficient is increased; After each weight coefficient is updated, a normalization operation is performed so that the sum of the updated weight coefficients is 1, thereby obtaining a first weight coefficient, a second weight coefficient, a third weight coefficient, and a fourth weight coefficient; Determine the reward function of the first deep Q network based on the first sub-reward function, the first weight coefficient, the second sub-reward function, the second weight coefficient, the third sub-reward function, the third weight coefficient, the fourth sub-reward function, and the fourth weight coefficient.

4. The method according to claim 1, wherein Determining a multidimensional feature vector of the target post includes: Extracting the semantic embedding vector of the target post using a pre-trained language model; calculating different probability distributions corresponding to different topic categories through a topic attention layer; and normalizing the different probability distributions to obtain a first feature component; Calculating a vocabulary strength score for the target post based on a financial sentiment dictionary, wherein the financial sentiment dictionary includes a plurality of different preset words for representing positive sentiment, neutral sentiment, and negative sentiment, and the vocabulary strength score is used to quantify the strength of the sentiment explicitly expressed in the target post; Determining the sentiment probability of the target post through a deep learning model, wherein the sentiment probability is used to represent the probability of the target post belonging to positive sentiment, neutral sentiment, and negative sentiment; Dynamically weighting and fusing the vocabulary strength score and the sentiment probability according to financial term density to obtain a second feature component, wherein the financial term density is the ratio of the number of financial professional terms appearing in the target post to the total number of words in the target post; Removing common stop words and high-frequency financial function words from the target post, and retaining professional terms whose term frequency-inverse document frequency exceeds a preset threshold; Filtering core words in the professional terms using a centrality metric of a term co-occurrence graph, and determining a third feature component based on the core words, wherein the term co-occurrence graph is used to represent the co-occurrence relationship of professional terms retained after screening in the target post, the nodes of the term co-occurrence graph are the screened professional terms, and the edges of the term co-occurrence graph are the semantic associations between professional terms; The first feature component, the second feature component, and the third feature component are concatenated, and a dimension reduction process is performed on the concatenated result through an orthogonal constrained linear transformation layer to obtain the multidimensional feature vector.

5. The method according to claim 1, wherein After obtaining the target post to be processed in the communication platform, the method further includes: Encode the target post using a pre-trained language model to obtain a semantic feature vector; determine a normalized click-through rate value and a weight of the poster's rank for the target post; concatenate the semantic feature vector, the normalized click-through rate value, and the weight of the poster's rank, and determine the concatenation result as a state vector; A second deep Q network is used to analyze the state vector to determine whether the target post is an essence post, wherein the action space of the second deep Q network includes: marking essence posts and not marking essence posts; the reward function of the second deep Q network includes: compliance score, expert review pass rate, and mislabeling penalty, wherein the compliance score is the quantitative result of risk scanning of post content through a preset financial compliance rule library, and the compliance score is the difference between 1 and the target ratio, and the target ratio is the ratio of the number of occurrences of risk keywords to the total number of keywords; the expert review pass rate is the ratio of the number of essence posts marked and confirmed by experts to the total number of marked essence posts; the mislabeling penalty includes a first cost and a second cost, wherein the first cost is the product of the post exposure and the preset unit attention cost, and the second cost is the product of the post knowledge value coefficient and the preset attenuation factor, and the preset unit attention cost is determined based on the working hours lost by employees reading the preset content.

6. The method according to claim 5, characterized in that The method also includes: if the target post is determined to be a featured post using the second deep Q network, after obtaining the processing result of the target node for the target post, sending the processing result to the target object for review, obtaining a review result, and forwarding the review result to an information receiving area associated with the target post or the target communication address.

7. The method according to claim 1, characterized in that The processing result includes at least: target content, processing conclusion type, user privacy level identifier and result sensitivity label; Forwarding the processing result to the information receiving area associated with the target post or the target communication address information receiving area, including: If the processing conclusion type is a first type for indicating a public reply request, forwarding the target content to the information receiving area; if the processing conclusion type is a second type for indicating personal business processing, forwarding the target content to the target communication address; When the user privacy level identifier is a first identifier for indicating public authority, the target content is forwarded to the information receiving area; when the user privacy level identifier is a second identifier for indicating private authority, the target content is forwarded to the target communication address; In a case where the result sensitivity label is a first label for representing private data, the target content is forwarded to the target communication address.

8. An intelligent collaborative management platform driven by artificial intelligence, characterized by: include: An acquisition unit is used to acquire a target post to be processed in the communication platform and determine a multidimensional feature vector of the target post, wherein the multidimensional feature vector includes: topic classification features, sentiment features, and keyword coding features; A construction unit is configured to construct a graph neural network, wherein the nodes in the graph neural network include department nodes and employee nodes, the edges in the graph neural network include inter-department collaboration edges, department-employee affiliation edges, and inter-employee collaboration edges, the department node attributes include at least the number of pending posts, the average post processing time, and the professional field code, and the employee node attributes include at least the current status, skill feature vector, and historical satisfaction score; an analyzing unit, configured to analyze the multidimensional feature vector and the graph neural network using a first deep Q network to obtain a target node in the graph neural network for processing the target post, wherein the state in the first deep Q network includes: the multidimensional feature vector of the current post, the topological structure and node attributes of the graph neural network, and the actions in the first deep Q network include: selecting a target department node, selecting a target employee node, and performing multi-hop decision-making along graph edges; The forwarding unit is configured to forward the target post to the target node, obtain a processing result of the target node on the target post, and forward the processing result to an information receiving area or a target communication address associated with the target post.

9. A non-volatile storage medium, characterized in that: The non-volatile storage medium includes a stored program, wherein when the program is running, the device where the non-volatile storage medium is located is controlled to execute the artificial intelligence-driven intelligent collaborative management method described in any one of claims 1 to 7.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the intelligent collaborative management method based on artificial intelligence driving is implemented as described in any one of claims 1 to 7.

Images