Intelligent point distribution system based on user portrait

Abstract

The invention discloses an intelligent point distribution system based on a user portrait, and relates to the technical field of knowledge management, and the system comprises a user portrait collection module, a portrait feature extraction module, a user portrait evaluation module, a user point distribution module, a dynamic early warning module and a visual display module. The user portrait acquisition module acquires behavior and transaction data; the portrait feature extraction module extracts behavior features and transaction features; the user portrait evaluation module constructs an activeness and contribution evaluation model to output points; the user point distribution module distributes user points according to weights; and the visual display module generates an integral thermodynamic diagram and a high-value user clustering tree diagram. According to the method, the resource allocation fairness is improved through dynamic adjustment of the integral weight and real-time anomaly monitoring, knowledge contribution is accurately stimulated in combination with an AI-driven user portrait, an auxiliary management layer is visually displayed to optimize decisions, and the activeness and the service quality of an enterprise knowledge base are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of knowledge management, and in particular to an intelligent points allocation system based on user portraits. Background Art

[0002] The point allocation of the enterprise knowledge base management platform refers to a management mechanism designed by the enterprise to motivate employees to actively participate in the contribution and use of the knowledge base. Specifically, the enterprise gives corresponding point rewards based on the employee's behavior in the knowledge base, such as uploading documents, sharing experiences, answering questions, participating in discussions, etc. These points can be exchanged for physical prizes, virtual rewards and career development opportunities, so as to enhance the employee's contribution to the knowledge base and enthusiasm for use, promote the sharing and circulation of knowledge, optimize the content quality and practicality of the knowledge base, and thus promote the enterprise's knowledge management and innovative development.

[0003] In order to solve the problems of subjective and static resource allocation and insufficient incentives for user participation in the point allocation of enterprise knowledge base management platforms, the existing technology uses fixed rule weighting and manual experience to adjust the points. However, the weight solidification leads to the inability to adapt to dynamic business strategies, low efficiency of manual intervention and susceptibility to subjective bias, which in turn leads to rigid resource allocation, decoupling of user contribution behavior and incentives, and delayed knowledge base content updates and uneven quality. In order to solve the above problems, an intelligent point allocation system based on user portraits is proposed. Summary of the invention

[0004] The purpose of the present invention is to provide an intelligent points allocation system based on user portraits to solve the problems raised in the above background technology.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is: an intelligent points allocation system based on user portrait, including a user portrait acquisition module, a portrait feature extraction module, a user portrait evaluation module, a user points allocation module, a dynamic warning module and a visual display module; The user portrait collection module collects and pre-processes the behavior data and transaction data of internal users of the enterprise; The portrait feature extraction module extracts behavior features and transaction features from the preprocessed behavior data and transaction data; The user portrait evaluation module combines behavioral characteristics and transaction characteristics to build an activity evaluation model and a contribution value evaluation model, and outputs user activity points and contribution points; The user points allocation module calculates the weighted sum of the user activity points and contribution points in proportion, constructs a points allocation model, and allocates user points; The dynamic early warning module divides users into different value levels according to their points, gives corresponding point rewards, detects abnormal point allocation behavior in real time, and establishes a user feedback channel; The visualization display module draws a heat map of the integral distribution and a dendrogram of high-value user clustering.

[0006] A further improvement of the technical solution of the present invention is that in the user portrait collection module, the collection and preprocessing process of the behavior data and transaction data of internal users of the enterprise includes: Lightweight data collection scripts are embedded in key operation nodes where users access the enterprise knowledge base platform, and a distributed log server cluster is built to receive and store user behavior logs in real time. The user behavior includes search behavior and knowledge use behavior. When the user triggers an operation, the data collection script captures the event and generates structured user behavior data. The HTTP protocol is used to transmit the user behavior data to the log server in real time, and the message middleware is used to buffer the high-concurrency log stream. Introduce the transaction system database, establish a two-way communication channel between the enterprise service bus and API gateway and the transaction system database, configure the scheduled task scheduler, deploy the data synchronization tool, and trigger the data extraction task through the task scheduler. Regularly filter the incremental transaction data through the structured query language. The transaction data includes the user's contribution to the knowledge in the enterprise knowledge base, the acquisition and the transaction flow record; Clean up duplicate operations in collected user behavior data, filter out abnormal values, convert unstructured text into structured fields, remove duplicate transactions in collected transaction data, mark abnormal transactions, associate behavior data and transaction data with transaction ID through user ID, and add environment variables to the original collected data, including user department and knowledge classification labels; The preprocessed user behavior data and transaction data are stored in the distributed file system according to user ID segments, integrated into a user-behavior-transaction association table, and transmitted to the portrait feature extraction module via the message queue.

[0007] A further improvement of the technical solution of the present invention is that: in the image feature extraction module, the process of extracting behavior features includes: Behavioral characteristics include search frequency, document opening frequency, and path complexity; Filter the user ID, operation type, and timestamp fields from the user-behavior-transaction association table, group by user ID, count the search frequency and document opening frequency within the specified time window, and identify high-frequency operation periods based on the time interval standard deviation; Extract the single operation details field, which includes the reading time and the number of annotations. Take the difference between the end time and the start time of the user's single document opening as the single reading time. Count the ratio of the number of annotation operations to the number of document openings to get the annotation ratio. According to the number of steps in the user's link from search, screening, preview to download, get the average operation step length and obtain the path complexity.

[0008] A further improvement of the technical solution of the present invention is that in the image feature extraction module, the process of extracting transaction features includes: Transaction characteristics include contribution complexity and knowledge value; The total word count of the document and the number of version updates are extracted from the user's knowledge contribution behavior to quantify the contribution complexity. The knowledge value is evaluated through weighted summation based on the knowledge authority level, number of citations and timeliness decay coefficient. A timeliness decay model is introduced to reduce the weight of obsolete knowledge according to the difference between the knowledge creation time and the current time.

[0009] A further improvement of the technical solution of the present invention is that: in the user portrait evaluation module, the process of constructing an activity evaluation model and outputting user activity scores includes: Set and dynamically update search frequency weights based on corporate knowledge management strategies , Document opening frequency weight and path complexity weight The sum is 1, the maximum allowed operation step is set by the enterprise business rules, and the search frequency is extracted and document opening frequency Daily statistics of path complexity Normalize the path complexity after normalization. Perform weighted summation, build an activity evaluation model, and obtain the user activity score , the calculation process is as follows; ; in, It means that the lower the path complexity, the higher the activity contribution value.

[0010] A further improvement of the technical solution of the present invention is that: in the user portrait evaluation module, a contribution value evaluation model is constructed, and the process of outputting contribution points includes: Dynamically assign weight coefficients for contribution complexity and knowledge value according to the enterprise knowledge management strategy and , and satisfies , the total word count L of the document and the number of version updates V are normalized to the historical maximum values, and the knowledge authority level A is mapped to the numerical value and the number of citations. pass Compression, time-dependent attenuation coefficient , and obtain the standardized contribution complexity and knowledge value ,in, is the difference between the knowledge creation time and the current time, is the decay rate parameter; The contribution value evaluation model is constructed using a multivariate linear regression algorithm. The standardized contribution complexity and knowledge value are input into the contribution value evaluation model to obtain the contribution score. , the calculation process is as follows: ; The total word count of the document, number of version updates, knowledge authority level, number of citations, knowledge creation time and contribution points with manual annotations are used as training data for the contribution value assessment model. The mean square error between the predicted contribution points and the annotated values ​​is minimized through the least squares method. The contribution value assessment model is retrained every month to incorporate new training data.

[0011] A further improvement of the technical solution of the present invention is that: in the user points allocation module, a points allocation model is constructed, and the process of allocating user points includes: Dynamically assign weight coefficients to activity points and contribution points based on the enterprise knowledge management strategy and If the user activity needs to be incentivized, the weight of activity points should be allocated. If the knowledge contribution value needs to be improved, the weight of contribution points should be allocated first. A points allocation model should be constructed to allocate user points. , the calculation process is as follows: ; in, and The intelligent points allocation system is updated in real time by reading corporate rules.

[0012] A further improvement of the technical solution of the present invention is that: in the dynamic warning module, the process of dividing users into different value levels and giving corresponding point rewards includes: Obtaining a user points data set from the points distribution module, clustering the user points by using the K-means algorithm, wherein the clustering operation includes setting the number of clusters k=3, initializing the centroid by using the percentile method of historical points distribution, dividing each user's points into clusters with the closest Euclidean distance based on the Euclidean distance from each user's points to each centroid, updating each cluster's centroid to the mean of the points within the cluster, setting a maximum number of iterations, and stopping the clustering operation when the maximum number of iterations is reached; Sort by centroid in ascending order, use the midpoints of adjacent centroids as the level boundaries, and divide users into high-value users, medium-value users, and low-value users. After the centroids are iteratively updated until convergence, the level boundaries are dynamically adjusted to meet the level ratio set by the enterprise business rules; For users of different value levels, corresponding point reward rules are configured. If they are high-value users, advanced knowledge access rights will be opened, they will be allowed to participate in core knowledge contribution reviews, and computing resources and knowledge recommendation exposure weights will be allocated preferentially. If they are medium-value users, department-level knowledge base extension rights will be unlocked, resource application priority will be increased, computing resources will be allocated on demand, and task processing priority will be higher than that of low-value users. If they are low-value users, basic permissions will be restricted, and standard knowledge access rights will be opened after the first contribution. Basic computing resource quotas must be redeemed with activity points.

[0013] A further improvement of the technical solution of the present invention is that: in the dynamic early warning module, the process of detecting abnormal point allocation behavior in real time and establishing a user feedback channel includes: Calculate the average of daily points changes based on historical points data With standard deviation , set the initial fluctuation threshold , adjust the volatility threshold range according to the enterprise's risk preference, and set volatility thresholds for users of different value levels. Update the mean and standard deviation every morning. If the number of points fluctuations exceeding the threshold exceeds 5% in a single day, adjust the volatility threshold to adapt it to the latest user points distribution and enterprise business rules; The distributed stream computing engine receives user points change event streams in real time, and calculates the points change rate for the day by user ID. If the absolute value of the points change rate for the day is greater than the fluctuation threshold, it is marked as abnormal behavior. The operation log is combined to detect related behaviors, and the points change permissions of abnormal accounts are automatically frozen, isolated to the pending review queue, and an alarm is pushed to the administrator interface to trigger the manual review process. The user submits a complaint form and uploads evidence through the knowledge base platform. The intelligent points allocation system extracts key fields through natural language processing and stores them in the audit library, triggering the system's initial review, manual review and administrator's final review. The key fields include user ID, abnormal time and reason for appeal. The audit result synchronously updates the user status, and normal data that passes the review is added to the training set. The points allocation model is retrained every month, and exemption rules are added to the policy library based on high-frequency false alarm scenarios.

[0014] A further improvement of the technical solution of the present invention is that: in the visualization display module, the process of drawing the integral distribution heat map and the high-value user clustering dendrogram includes: Obtain the user points data set from the points allocation module, associate the user department and knowledge classification label, group the users by user department and knowledge classification label, count the average points of each group of users, divide the intervals according to the points value range, map the low points interval to light color, and map the high points interval to dark color. Fill in the corresponding points mean with user departments as rows and knowledge classification labels as columns to form a two-dimensional matrix. Use the visualization engine ECharts to render the two-dimensional matrix into a points distribution heat map with the horizontal and vertical axes marked with user departments and knowledge classification labels respectively, and the color of the color block reflects the points density; The user value level and the centroid coordinates of the cluster to which they belong are obtained from the dynamic warning module, and the user department, total word count of the document and number of version updates are associated. The AGNES hierarchical clustering algorithm is used to obtain the distance between users based on user points, total word count of the document and number of version updates. The two clusters with the closest distance are iteratively merged to generate a tree structure, and the merging order and height are recorded. Leaf nodes are defined as individual users, and non-leaf nodes are merged clusters. The node height reflects the merged distance. It supports clicking nodes to expand and collapse sub-clusters, and displays the user department distribution, total word count of documents and number of version updates within the cluster.

[0015] Due to the adoption of the above technical solution, the present invention has the following technical advances compared with the prior art: 1. The present invention provides an intelligent points allocation system based on user portraits. Through a dynamic early warning module, fluctuation thresholds are set and user feedback channels are established. Abnormal points behavior is detected in real time and manual review is supported. This effectively avoids resource allocation rigidity and points-brushing loopholes, improves system fairness, protects the rights and interests of high-value users, and drives a virtuous cycle of knowledge contribution and resource allocation.

[0016] 2. The present invention provides an intelligent points allocation system based on user portraits, which adopts a multivariate linear regression algorithm and a dynamic weighted fusion mechanism to quantify user behavior characteristics and transaction characteristics, realize accurate modeling and dynamic adaptation of points allocation, significantly improve the incentive effect of user activity and knowledge contribution, and meet the company's phased operational strategy needs.

[0017] 3. The present invention provides an intelligent points allocation system based on user portraits, which generates a points distribution heat map and a user clustering tree diagram through a visualization display module, intuitively presents the points density of user departments and knowledge classifications and the structure of high-value user groups, provides data-driven decision support for management, optimizes resource allocation strategies and accelerates the quality iteration and value precipitation of knowledge base content. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0019] Figure 1 A block diagram of the present invention. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0021] Examples, such as Figure 1 As shown, the present invention provides an intelligent points allocation system based on user portraits, including a user portrait acquisition module, a portrait feature extraction module, a user portrait evaluation module, a user points allocation module, a dynamic warning module and a visualization display module; The user portrait collection module collects and preprocesses the behavior data and transaction data of internal enterprise users, embeds lightweight data collection scripts at key operation nodes where users access the enterprise knowledge base platform, builds a distributed log server cluster, receives and stores user behavior logs in real time. The user behavior includes search behavior and knowledge use behavior. When the user triggers an operation, the data collection script captures the event and generates structured user behavior data. The HTTP protocol is used to transmit user behavior data to the log server in real time, and the message middleware is used to buffer high-concurrency log streams. The transaction system database is introduced, and the enterprise service bus and API gateway respectively establish two-way communication channels with the transaction system database, configure the scheduled task scheduler, deploy data synchronization tools, and task scheduling. The controller triggers the data extraction task and regularly screens incremental transaction data through structured query language. The transaction data includes the user's contribution to the knowledge in the enterprise knowledge base, acquisition and transaction flow records, cleans the repeated operations in the collected user behavior data, filters out abnormal values, and converts unstructured text into structured fields, removes repeated flows in the collected transaction data, marks abnormal transactions, and associates the behavior data and transaction data with the transaction ID through the user ID. Environment variables are added to the original collected data. The environment variables include user departments and knowledge classification labels. The pre-processed user behavior data and transaction data are stored in the distributed file system according to the user ID segment, integrated into a user-behavior-transaction association table, and transmitted to the portrait feature extraction module through the message queue; The portrait feature extraction module extracts behavioral features and transaction features from the preprocessed behavioral data and transaction data. The behavioral features include search frequency, document opening frequency and path complexity. The user ID, operation type and timestamp fields are filtered from the user-behavior-transaction association table, grouped by user ID, and the search frequency and document opening frequency within the specified time window are counted. The high-frequency operation period is identified by combining the standard deviation of the time interval, and the single operation detail field is extracted. The operation details include reading time and annotation times. The difference between the end time and the start time of a single document opening by the user is taken as the single reading time. The ratio of the number of annotation operations to the number of document openings is counted to obtain the annotation ratio. The average operation step length is obtained according to the number of link steps from search, screening, preview to download, and the path complexity is obtained. The transaction features include contribution complexity and knowledge value. The total number of words in the document and the number of version updates are extracted from the user's knowledge contribution behavior, and the contribution complexity is quantified. The knowledge value is evaluated by weighted summation in combination with the knowledge authority level, the number of citations and the timeliness attenuation coefficient. The timeliness attenuation model is introduced to reduce the weight of obsolete knowledge according to the difference between the knowledge creation time and the current time. The user portrait evaluation module combines behavioral characteristics and transaction characteristics to build an activity evaluation model and a contribution value evaluation model, outputs user activity points and contribution points, and sets and dynamically updates search frequency weights based on the enterprise knowledge management strategy. , Document opening frequency weight and path complexity weight The sum is 1, the maximum allowed operation step is set by the enterprise business rules, and the search frequency is extracted and document opening frequency Daily statistics of path complexity Normalize the path complexity after normalization. Perform weighted summation, build an activity evaluation model, and obtain the user activity score , the calculation process is as follows; ; in, The lower the path complexity, the higher the activity contribution value. According to the enterprise knowledge management strategy, the weight coefficient is dynamically allocated for the contribution complexity and knowledge value. and , and satisfies , the total word count L of the document and the number of version updates V are normalized to the historical maximum values, and the knowledge authority level A is mapped to the numerical value and the number of citations. pass Compression, time-dependent attenuation coefficient , and obtain the standardized contribution complexity and knowledge value ,in, is the difference between the knowledge creation time and the current time, The contribution value evaluation model is constructed by using the multivariate linear regression algorithm as the decay rate parameter. The standardized contribution complexity and knowledge value are input into the contribution value evaluation model to obtain the contribution score. , the calculation process is as follows: ; The total word count, version update times, knowledge authority level, citation times, knowledge creation time, and contribution points with manual annotations in the document are used as training data for the contribution value assessment model. The mean square error between the predicted contribution points and the annotated values ​​is minimized using the least squares method. The contribution value assessment model is retrained every month to incorporate new training data. The user points allocation module calculates the weighted sum of user activity points and contribution points in proportion, builds a points allocation model, allocates user points, and dynamically allocates weight coefficients for activity points and contribution points according to the enterprise knowledge management strategy. and If the user activity needs to be incentivized, the weight of activity points should be allocated. If the knowledge contribution value needs to be improved, the weight of contribution points should be allocated first. A points allocation model should be constructed to allocate user points. , the calculation process is as follows: ; in, and The intelligent points allocation system is updated in real time by reading the enterprise rules; The dynamic early warning module divides users into different value levels according to their points and gives corresponding point rewards, detects abnormal point allocation behavior in real time, establishes a user feedback channel, obtains the user point data set from the point allocation module, and clusters the user points through the K-means algorithm. The clustering operation includes setting the number of clusters k=3, initializing the centroid through the percentile method of historical point distribution, and dividing each user's points into clusters with the closest Euclidean distance based on the Euclidean distance from each user's points to each centroid. The centroid of each cluster is updated to the mean of the points within the cluster, and the maximum number of iterations is set. When the maximum number of iterations is reached, the clustering operation is stopped, and the centroids are sorted in ascending order. The midpoints of adjacent centroids are used as the level boundaries to divide users into high-value users, medium-value users, and low-value users. For users and low-value users, after the centroid is iteratively updated to convergence, the level boundary is dynamically adjusted to meet the level ratio set by the enterprise business rules. For users of different value levels, corresponding point reward rules are configured. If they are high-value users, advanced knowledge access rights are opened, and they are allowed to participate in core knowledge contribution reviews, and computing resources and knowledge recommendation exposure weights are allocated preferentially. If they are medium-value users, department-level knowledge base extension rights are unlocked, resource application priority is increased, computing resources are allocated on demand, and task processing priority is higher than that of low-value users. If they are low-value users, basic permissions are restricted, and standard knowledge access rights are opened after the first contribution. Basic computing resource quotas must be redeemed through activity points, and the average daily point changes are calculated based on historical point data. With standard deviation , set the initial fluctuation threshold , adjust the volatility threshold range according to the enterprise's risk preference, and set volatility thresholds for users of different value levels. Update the mean and standard deviation at dawn every day. If the daily score fluctuation exceeds the threshold for more than 5%, adjust the volatility threshold to adapt the volatility threshold to the latest user score distribution and enterprise business rules. Receive the user score change event stream in real time through the distributed stream computing engine, and count the daily score change rate by user ID. If the absolute value of the daily score change rate is greater than the volatility threshold, it is marked as abnormal behavior. Combined with the operation log to detect related behaviors, the score change permission of the abnormal account is automatically frozen, isolated to the pending review queue, and the alarm is pushed to the administrator interface to trigger the manual review process. The user submits the appeal form through the knowledge base platform and uploads evidence. The intelligent score allocation system extracts key fields through natural language processing and stores them in the review library, triggering the system's initial review, manual review and administrator's final review. The key fields include user ID, abnormal time and reason for appeal. The user status is updated synchronously with the review result. The normal data that has passed the review is added to the training set. The score allocation model is retrained every month. According to the high-frequency false alarm scenario, exemption rules are added to the policy library. The visualization display module draws the points distribution heat map and the high-value user clustering tree diagram, obtains the user points data set from the points allocation module, associates the user department and the knowledge classification label, groups the users by the user department and the knowledge classification label, counts the average points of each group of users, divides the intervals according to the points value range, maps the low points interval to light color, and maps the high points interval to dark color. With the user department as the row and the knowledge classification label as the column, fill in the corresponding points mean value to form a two-dimensional matrix. The visualization engine ECharts renders the two-dimensional matrix with the horizontal and vertical axes marked with the user department and the knowledge classification label respectively, and the color of the color block reflects the points. The integral distribution heat map of density obtains the user value level and the centroid coordinates of the cluster to which it belongs from the dynamic warning module, associates the user department, total number of words in the document and the number of version updates, and uses the AGNES hierarchical clustering algorithm to obtain the distance between users based on user points, total number of words in the document and the number of version updates. It iteratively merges the two clusters with the closest distance to generate a tree structure, and records the merging order and height. It defines leaf nodes as individual users and non-leaf nodes as merged clusters. The node height reflects the merged distance. It supports clicking nodes to expand and collapse sub-clusters, and displays the user department distribution, total number of words in the document and the number of version updates within the cluster.

[0022] First, when users visit the enterprise knowledge base platform, the system captures user behavior data in real time at key operation nodes through lightweight data collection scripts, and accesses the transaction system database through the enterprise service bus and API gateway to collect user transaction data. Then, the collected raw data is cleaned, deduplicated, anomaly filtered, and converted into unstructured text, and then stored in the distributed file system according to user ID shards, integrated into an association table and transmitted to the portrait feature extraction module. Subsequently, the portrait feature extraction module extracts behavioral features and transaction features from the association table, where the path complexity is normalized, and the contribution complexity is dynamically calculated in combination with the number of document words, the number of version updates, and the timeliness attenuation coefficient. The user portrait evaluation module constructs an activity evaluation model through a multivariate linear regression algorithm. The model and contribution value assessment model output activity points and contribution points respectively, and dynamically adjust the weight coefficient according to the enterprise stage strategy. Then, the user point allocation module weights and fuses the activity points and contribution points in a dynamic proportion to generate the user's total points, and automatically allocates resource permissions based on the point value. The dynamic early warning module divides users into levels through the K-means clustering algorithm, sets a single-day point fluctuation threshold, detects abnormal behaviors in real time, freezes abnormal accounts and opens a user complaint channel, and optimizes model parameters based on manual review results. Finally, the visualization display module draws a point distribution heat map and a high-value user clustering tree map to provide resource allocation decision support for the management. At the same time, through the feedback data closed-loop iteration model, the fairness and incentive effect of the system are continuously improved.

[0023] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. An intelligent points allocation system based on user portrait, characterized by: It includes user portrait collection module, portrait feature extraction module, user portrait evaluation module, user points allocation module, dynamic warning module and visual display module; The user portrait collection module collects and pre-processes the behavior data and transaction data of internal users of the enterprise; The portrait feature extraction module extracts behavior features and transaction features from the preprocessed behavior data and transaction data; The user portrait evaluation module combines behavioral characteristics and transaction characteristics to build an activity evaluation model and a contribution value evaluation model, and outputs user activity points and contribution points; The user points allocation module calculates the weighted sum of the user activity points and contribution points in proportion, constructs a points allocation model, and allocates user points; The dynamic early warning module divides users into different value levels according to their points, gives corresponding point rewards, detects abnormal point allocation behavior in real time, and establishes a user feedback channel; The visualization display module draws a heat map of the integral distribution and a dendrogram of high-value user clustering.

2. The intelligent points allocation system based on user portrait according to claim 1, characterized in that: In the user portrait collection module, the collection and preprocessing process of the behavior data and transaction data of internal enterprise users includes: Embed lightweight data collection scripts at key operation nodes where users access the enterprise knowledge base platform, build a distributed log server cluster, receive and store user behavior logs in real time, including search behavior and knowledge use behavior. When users trigger an operation, the data collection script captures the event and generates structured user behavior data. Introduce the transaction system database, establish a two-way communication channel between the enterprise service bus and API gateway and the transaction system database respectively, configure the scheduled task scheduler, deploy the data synchronization tool, and the task scheduler triggers the data extraction task. The incremental transaction data is regularly screened through the structured query language. The transaction data includes the user's contribution to the knowledge in the enterprise knowledge base, the acquisition and the transaction flow record; Clean up duplicate operations in collected user behavior data, filter out abnormal values, convert unstructured text into structured fields, remove duplicate transactions in collected transaction data, mark abnormal transactions, associate behavior data and transaction data with transaction ID through user ID, and add environment variables to the original collected data, including user department and knowledge classification labels; The preprocessed user behavior data and transaction data are stored in the distributed file system according to user ID segments, integrated into a user-behavior-transaction association table, and transmitted to the portrait feature extraction module via the message queue.

3. The intelligent points allocation system based on user portrait according to claim 2 is characterized in that: In the image feature extraction module, the process of extracting behavioral features include: Behavioral characteristics include search frequency, document opening frequency, and path complexity; Filter the user ID, operation type, and timestamp fields from the user-behavior-transaction association table, group by user ID, count the search frequency and document opening frequency within the specified time window, and identify high-frequency operation periods based on the time interval standard deviation; Extract the single operation details field, which includes the reading time and the number of annotations. Take the difference between the end time and the start time of the user's single document opening as the single reading time. Count the ratio of the number of annotation operations to the number of document openings to get the annotation ratio. According to the number of steps in the user's link from search, screening, preview to download, get the average operation step length and obtain the path complexity.

4. The intelligent points allocation system based on user portrait according to claim 3 is characterized in that: In the image feature extraction module, the process of extracting transaction features includes: Transaction characteristics include contribution complexity and knowledge value; The total word count of the document and the number of version updates are extracted from the user's knowledge contribution behavior to quantify the contribution complexity. The knowledge value is evaluated through weighted summation based on the knowledge authority level, number of citations and timeliness decay coefficient. A timeliness decay model is introduced to reduce the weight of obsolete knowledge according to the difference between the knowledge creation time and the current time.

5. The intelligent points allocation system based on user portrait according to claim 4 is characterized in that: In the user portrait evaluation module, the process of constructing an activity evaluation model and outputting user activity scores includes: Set and dynamically update search frequency weights based on corporate knowledge management strategies , Document opening frequency weight and path complexity weight The sum is 1, the maximum allowed operation step is set by the enterprise business rules, and the search frequency is extracted and document opening frequency Daily statistics of path complexity Normalize the path complexity after normalization. Perform weighted summation, build an activity evaluation model, and obtain the user activity score .

6. The intelligent points allocation system based on user portrait according to claim 5, characterized in that: In the user portrait evaluation module, the process of constructing a contribution value evaluation model and outputting contribution points includes: Dynamically assign weight coefficients for contribution complexity and knowledge value according to the enterprise knowledge management strategy and , and satisfies , the total word count L of the document and the number of version updates V are normalized to the historical maximum values, and the knowledge authority level A is mapped to the numerical value and the number of citations. pass Compression, time-dependent attenuation coefficient , and obtain the standardized contribution complexity and knowledge value ,in, is the difference between the knowledge creation time and the current time, is the decay rate parameter; The contribution value evaluation model is constructed using a multivariate linear regression algorithm. The standardized contribution complexity and knowledge value are input into the contribution value evaluation model to obtain the contribution score. ; The total word count of the document, number of version updates, knowledge authority level, number of citations, knowledge creation time and contribution points with manual annotations are used as training data for the contribution value assessment model. The mean square error between the predicted contribution points and the annotated values ​​is minimized through the least squares method. The contribution value assessment model is retrained every month to incorporate new training data.

7. The intelligent points allocation system based on user portrait according to claim 6 is characterized in that: In the user points allocation module, a points allocation model is constructed, and the process of allocating user points includes: Dynamically assign weight coefficients to activity points and contribution points based on the enterprise knowledge management strategy and If the user activity needs to be motivated, the weight of activity points will be focused on; if the knowledge contribution value needs to be improved, the weight of contribution points will be prioritized. The intelligent points allocation system is updated in real time by reading the enterprise rules. and , build a points allocation model to allocate user points .

8. The intelligent points allocation system based on user portrait according to claim 7 is characterized in that: In the dynamic warning module, the process of dividing users into different value levels and giving corresponding points rewards includes: Obtaining a user points data set from the points distribution module, clustering the user points by using the K-means algorithm, wherein the clustering operation includes setting the number of clusters k=3, initializing the centroid by using the percentile method of historical points distribution, dividing each user's points into clusters with the closest Euclidean distance based on the Euclidean distance from each user's points to each centroid, updating each cluster's centroid to the mean of the points within the cluster, setting a maximum number of iterations, and stopping the clustering operation when the maximum number of iterations is reached; Sort by centroid in ascending order, use the midpoints of adjacent centroids as the level boundaries, and divide users into high-value users, medium-value users, and low-value users. After the centroids are iteratively updated until convergence, the level boundaries are dynamically adjusted to meet the level ratio set by the enterprise business rules; For users of different value levels, corresponding point reward rules are configured. If they are high-value users, advanced knowledge access rights will be opened, they will be allowed to participate in core knowledge contribution reviews, and computing resources and knowledge recommendation exposure weights will be allocated preferentially. If they are medium-value users, department-level knowledge base extension rights will be unlocked, resource application priority will be increased, computing resources will be allocated on demand, and task processing priority will be higher than that of low-value users. If they are low-value users, basic permissions will be restricted, and standard knowledge access rights will be opened after the first contribution. Basic computing resource quotas must be redeemed with activity points.

9. The intelligent points allocation system based on user portrait according to claim 8, characterized in that: In the dynamic early warning module, the process of detecting abnormal points allocation behavior in real time and establishing a user feedback channel includes: Calculate the average of daily points changes based on historical points data With standard deviation , set the initial fluctuation threshold , adjust the volatility threshold range according to the enterprise's risk preference, and set volatility thresholds for users of different value levels. Update the mean and standard deviation every morning. If the number of points fluctuations exceeding the threshold exceeds 5% in a single day, adjust the volatility threshold to adapt it to the latest user points distribution and enterprise business rules; The distributed stream computing engine receives user points change event streams in real time, and calculates the points change rate for the day by user ID. If the absolute value of the points change rate for the day is greater than the fluctuation threshold, it is marked as abnormal behavior. The operation log is combined to detect related behaviors, and the points change permissions of abnormal accounts are automatically frozen, isolated to the pending review queue, and an alarm is pushed to the administrator interface to trigger the manual review process. The user submits a complaint form and uploads evidence through the knowledge base platform. The intelligent points allocation system extracts key fields through natural language processing and stores them in the audit library, triggering the system's initial review, manual review and administrator's final review. The key fields include user ID, abnormal time and reason for appeal. The audit result synchronously updates the user status, and normal data that passes the review is added to the training set. The points allocation model is retrained every month, and exemption rules are added to the policy library based on high-frequency false alarm scenarios.

10. The intelligent points allocation system based on user portrait according to claim 9, characterized in that: In the visualization display module, the process of drawing the score distribution heat map and the high-value user clustering dendrogram includes: Obtain the user points data set from the points allocation module, associate the user department and knowledge classification label, group the users by user department and knowledge classification label, count the average points of each group of users, divide the intervals according to the points value range, map the low points interval to light color, and map the high points interval to dark color. Fill in the corresponding points mean with user departments as rows and knowledge classification labels as columns to form a two-dimensional matrix. Use the visualization engine ECharts to render the two-dimensional matrix into a points distribution heat map with the horizontal and vertical axes marked with user departments and knowledge classification labels respectively, and the color of the color block reflects the points density; The user value level and the centroid coordinates of the cluster to which they belong are obtained from the dynamic warning module, and the user department, total word count of the document and number of version updates are associated. The AGNES hierarchical clustering algorithm is used to obtain the distance between users based on user points, total word count of the document and number of version updates. The two clusters with the closest distance are iteratively merged to generate a tree structure, and the merging order and height are recorded. Leaf nodes are defined as individual users, and non-leaf nodes are merged clusters. The node height reflects the merged distance. It supports clicking nodes to expand and collapse sub-clusters, and displays the user department distribution, total word count of documents and number of version updates within the cluster.

Images