An intelligent text analysis method and system for government service systems

Abstract

The application provides an intelligent text analysis method and system for a government service system, and belongs to the technical field of data processing, and specifically comprises the following steps: determining a user portrait updating method according to a file pushing user under a user label corresponding to an interest portrait of the user; updating the interest portrait of the user based on the portrait updating method; determining different user updating delay types based on the portrait updating method; determining an abnormal analysis and identification scheme under different user labels by using an updating delay type determination model; determining the identification deviation of a natural language processing model under different user labels by using the abnormal analysis and identification scheme; and determining the user for real-time updating of the interest portrait according to the identification deviation of the natural language processing model under different user labels, so that the exposure and matching of the file pushing are improved.

Description

Technical Field

[0001] This invention belongs to the field of data processing technology, and in particular relates to an intelligent text analysis method and system for government service systems. Background Technology

[0002] With the deepening of the construction of "digital government," government departments at all levels have generated massive amounts of text data, including policy documents, notices and announcements, service guides, and work summaries. Currently, the distribution of government documents mainly relies on hierarchical forwarding, portal website publication, or unified mass distribution, which has significant pain points: 1) Information overload and inefficient acquisition: Government personnel need to manually filter content related to their responsibilities from massive amounts of information, which is inefficient and prone to missing key information; 2) Insufficient push accuracy: Important policies are difficult to accurately reach all relevant implementers, supervisors, or consultants; 3) Lack of personalization: Existing systems lack a deep understanding of user roles, responsibilities, and interests, and cannot achieve intelligent push notifications tailored to each user.

[0003] Therefore, there is an urgent need for an intelligent text analysis method and system for government service systems. Summary of the Invention

[0004] To achieve the objectives of this invention, the following technical solution is adopted:

[0005] Specifically, this application provides an intelligent text analysis method for government service systems, which includes:

[0006] S1 determines the recommendation data under different user tags based on the text recommendation data of the government service system, and determines the update requirement type of the user's interest profile based on the recommendation data under different user tags. When the update requirement type does not belong to the target requirement type, the method for updating the user's profile is determined based on the file push user under the user tag corresponding to the user's interest profile.

[0007] S2 updates the user's interest profile based on the profile update method, determines the update delay type for different users based on the profile update method, and uses the update delay type to determine the anomaly parsing and identification scheme of the model under different user tags;

[0008] S3 uses the aforementioned anomaly analysis and identification scheme to determine the identification deviation of the natural language processing model under different user tags, and determines the users whose interest profiles are updated in real time based on the identification deviation of the natural language processing model under different user tags.

[0009] The beneficial effects of this invention are as follows:

[0010] Based on recommendation data under different user tags, the update needs of user interest profiles are determined, thereby enabling the number of recommendations for files under different user tags, determining the file exposure rate, and updating user interest profiles according to the differences in exposure rate. This improves the timeliness of interest profile updates when the exposure rate is low, thereby increasing the file's exposure rate and reach.

[0011] Based on the recognition deviations of natural language processing models under different user tags, users are identified and their interest profiles are updated in real time. This enables timely adjustment of the interest profile update time for users with delayed updates, especially when there are user tags with a high probability of labeling errors in multiple natural language processing models. This improves the matching degree between recommended documents under user tags and the user's true user tags. In addition to anomaly identification of user tags based on click data, it can effectively eliminate the impact of untimely updates of user tags on click data, further improving the reliability of user tag recognition and processing.

[0012] Furthermore, the recommendation data under the user tag includes the number of recommended texts under the user tag.

[0013] Furthermore, the method for determining the update request type of the user's interest profile is as follows:

[0014] Based on recommendation data under different user tags, determine the recommended text under the user tags;

[0015] Based on the user's user tags and the recommended text under the user tags, determine the number of users recommended for different recommended texts;

[0016] Based on the number of users recommending different recommended texts, the update requirement type of the user's interest profile is determined.

[0017] Furthermore, the method for determining the user profile update method is as follows:

[0018] The number of users who push files to a user under the user's user tag corresponding to the user's interest profile is determined.

[0019] Based on the number of users to whom the file was pushed, determine the valid user tags among the user tags of the users;

[0020] Based on the valid user tags in the user's user tags and the composition data of the valid user tags of all users, the user's profile update method is determined.

[0021] Furthermore, the effective user tags are those whose number of file-pushing users exceeds a preset threshold for the number of recommended users.

[0022] Furthermore, the method for determining users in the real-time update process of the interest profile is as follows:

[0023] Based on the recognition deviation of the natural language model under different user tags, the number of recognition deviations of the recommended text under different user tags is determined. Based on the proportion of the number of recognition deviations of the recommended text under the user tag in the total number of recommended texts under the user tag, the recognition deviation factor of the user tag is determined.

[0024] Based on the identification bias factor of the user tag, identify the identification bias risk tag in the user tag;

[0025] Using the identification bias risk label data and combined with user data containing the identification bias risk label, it is determined whether the user is a user subject to real-time update processing of the interest profile.

[0026] In a second aspect, the present invention provides a computer system comprising: a memory and a processor connected in communication, and a computer program stored in the memory and capable of running on the processor, wherein the processor executes the above-described intelligent text analysis method for a government service system when running the computer program.

[0027] Other features and advantages will be set forth in the following description, and the objects and other advantages of the invention are realized and obtained through the structures particularly pointed out in the description and the drawings.

[0028] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0029] The above and other features and advantages of the present invention will become more apparent from a detailed description of exemplary embodiments thereof with reference to the accompanying drawings;

[0030] Figure 1 This is a flowchart of an intelligent text analysis method for government service systems;

[0031] Figure 2 This is a flowchart illustrating the method for determining the types of user interest profile update requests;

[0032] Figure 3 This is a flowchart illustrating the method for determining the user profile update process.

[0033] Figure 4 This is a flowchart illustrating the method for determining the anomaly parsing and identification scheme under user tags. Detailed Implementation

[0034] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.

[0035] Example 1

[0036] like Figure 1 As shown, this application provides an intelligent text analysis method for government service systems, specifically including:

[0037] S1 determines the recommendation data under different user tags based on the text recommendation data of the government service system, and determines the update requirement type of the user's interest profile based on the recommendation data under different user tags. When the update requirement type does not belong to the target requirement type, the method for updating the user's profile is determined based on the file push user under the user tag corresponding to the user's interest profile.

[0038] The core decision-making objective of this method is to intelligently determine whether high-priority real-time updates of knowledge profiles for relevant staff groups are needed by analyzing the breadth and coverage balance of internal government documents under different staff tags. Its core logic is a multi-layered, progressive decision-making process: first, identifying the existence of widely distributed documents; second, assessing the prevalence of such documents; third, analyzing the concentration of these documents' reception; and finally, judging whether there is a "distribution solidification" phenomenon based on the concentration, thereby deciding whether to optimize document distribution strategies through real-time updates of knowledge profiles, improve the coverage and fairness of government information dissemination, and ensure that internal policy documents reach all relevant personnel in a balanced and effective manner.

[0039] Furthermore, the recommendation data under the user tag includes the number of recommended texts under the user tag.

[0040] Specifically, such as Figure 2 As shown, the method for determining the update request type of the user's interest profile is as follows:

[0041] S11 uses recommendation data under different user tags to determine the recommended text under the user tag;

[0042] The key terms in this step are "staff tags" and "document distribution." "Staff tags" are system-identified classifications based on the job responsibilities, management authority, business areas, historical document types processed, and browsing interests of staff within the government system. Examples include "Financial Approval Specialist," "Project Construction Administrator," and "Policy and Regulations Researcher." "Document distribution" refers to the collection of government documents, policy notices, work guidelines, and other documents proactively distributed by the system to all internal staff with a specific staff tag within a statistical period, regardless of whether the staff actually review them. This step is designed this way because ensuring relevant personnel receive documents related to their responsibilities in a timely manner is a fundamental requirement in government systems, and distribution is a direct manifestation of the system fulfilling this responsibility. Its significance lies in the fact that aggregating distribution behavior data through tags allows for structured analysis of the system's document distribution strategy across different responsibility groups. It focuses on assessing the coverage and completeness of the key administrative link of "document delivery," providing an accurate "system-side" distribution data foundation for subsequent analysis and ensuring an objective evaluation of the government information dissemination mechanism.

[0043] Example: Staff member Section Chief Zhang is tagged with "Safety Production Supervision". The system extracts 50 relevant documents (including safety inspection notices, accident reports, and regulatory updates) that were distributed to all staff members tagged with "Safety Production Supervision" in the past week. These 50 documents are the "distributed documents" determined in this step.

[0044] S12 determines the number of recommended users for different recommended texts based on the user's user tags and the recommended text under the user tags;

[0045] The key term in this step is "number of distribution personnel." This refers to counting how many internal personnel with the staff member tag are distributed to each document identified in step S11. This is an indicator of the "breadth of distribution coverage" of a single document, focusing on the scope of the system's delivery of the document to the potential audience, rather than the actual number of times it is viewed. The reason for this step is that in the dissemination of government documents, "delivery" is a prerequisite for "awareness" and "execution." The number of distribution personnel directly reflects the system's effort and initial scope in trying to reach the relevant staff. Its significance lies in quantifying the system's distribution actions into analyzable data, allowing us to assess from the source whether the document has achieved the necessary delivery coverage and how the system's distribution resources are allocated across different documents. This is a crucial basis for diagnosing whether the document dissemination mechanism has "coverage blind spots" or "insufficient distribution of key documents."

[0046] Example: For the 50 safety production documents mentioned above, the system counted the number of people who received each document. It was found that an emergency notice on flood season safety production was distributed to 190 out of 200 staff members under that tag, while a technical guideline for a specific inspection was only distributed to 15 people.

[0047] S13 determines the type of user interest profile update requirement based on the number of users recommending different recommended texts.

[0048] It is understandable that, based on the number of users recommending different recommended texts, the type of update requirement for the user's interest profile is determined, specifically including:

[0049] S131 uses the number of recommended users of the recommended text to determine whether there is a recommended text with a number of recommended users greater than a preset threshold for the number of recommended users. If yes, proceed to the next step; otherwise, determine the update request type of the user's interest profile as the target request type.

[0050] This step is the overall decision-making step, and its specific logic is implemented through the following sub-steps (S131-S134). The aim is to use a series of threshold judgments to finely determine whether the knowledge profile of the target demand type (i.e., real-time update) needs to be updated from two dimensions: distribution breadth and reception distribution, in order to solve the potential problem of uneven file distribution.

[0051] S131: Determine if a wide-coverage distribution file exists:

[0052] This step introduces the keywords "preset distribution personnel number threshold" and "target demand type". The "preset distribution personnel number threshold" is a threshold value set based on the proportion of the total number of staff under that label, used to determine whether a document has received sufficient basic distribution (e.g., 60% of the total number of staff under that label). "Target demand type" specifically refers to the processing mode that requires triggering real-time, high-frequency knowledge profile updates. The reason for this setting is that if the number of distribution personnel for all documents under a certain label is lower than this threshold, it indicates a serious flaw in the system's document distribution strategy for that group. A large number of staff who should be aware of the documents have not been delivered to a sufficient range, and their knowledge profiles may not receive necessary policy information updates, leading to work disconnect. Its significance lies in serving as an early warning mechanism, immediately identifying the serious problem of "systemic insufficient distribution," directly triggering enhanced real-time updates and adjustments to the distribution strategy, ensuring full coverage of basic government information in a more proactive way, and preventing administrative execution from being affected by information gaps.

[0053] Example: Continuing from the previous example, there are 200 staff members under the "Safety Production Supervision" label, with a preset threshold of 120 people (60%). An inspection revealed that several of the 50 documents were distributed to more than 190 people, thus failing to meet the condition of "all below the threshold," and the process proceeds to S132.

[0054] S132 takes recommended texts with a number of recommended users greater than a preset threshold for the number of recommended users as valid recommended texts, and determines whether the proportion of the number of valid recommended texts in the recommended texts is greater than a preset threshold for the proportion of the number of recommended texts. If so, it is determined that the user's interest profile update request type does not belong to the target request type. If not, proceed to the next step.

[0055] Determine the percentage of validly distributed files:

[0056] The key terms in this step are "effectively distributed files" and "preset quantity percentage threshold." "Effectively distributed files" are those selected in S131 whose number of distributors exceeds a preset threshold. The "preset quantity percentage threshold" is the percentage of effective files out of the total number of distributed files (e.g., 60%). This step is set because if the percentage of widely distributed "effectively distributed files" is too high, it means the system is concentrating most of its distribution resources on a few "universal" or "highly urgent" files. This may reflect the nature of the files or a conservative distribution strategy, potentially leading to insufficient coverage of file types received by staff and inadequate coverage of certain specialized or in-depth files. Its significance lies in assessing the concentration of the system's distribution behavior. If the percentage is too high, it indicates that the current distribution model is stable but may not meet diverse professional information needs, requiring further analysis; if the percentage is moderate or low, it indicates that most files have not been fully distributed, potentially posing a risk of insufficient distribution.

[0057] Example: In the above example, assuming there are 8 valid distribution files distributed to more than 120 people, accounting for 16% (8 / 50), which is lower than the 30% threshold, the condition is not met, and the process proceeds to S133.

[0058] S133 uses the recommended user data of the effective recommended text to determine the proportion of effective recommended texts among different users in all effective recommended texts, and determines whether there are users whose proportion is greater than a preset proportion threshold. If so, proceed to the next step; otherwise, determine that the user's interest profile update requirement type does not belong to the target requirement type.

[0059] The key terms in this step are "personnel reception concentration ratio" and "preset concentration threshold". "Personnel reception concentration ratio" specifically refers to the percentage of validly distributed files (i.e., broad-coverage files) distributed to a single employee, calculated as the proportion of such files distributed to an individual employee relative to the total number of distributions of all valid files. The "preset concentration threshold" is a critical value (e.g., 3%) for determining whether an individual employee has been excessively and centrally distributed with broad-coverage files. This step is set because even if there aren't many broad-coverage files, if the system repeatedly and centrally distributes these files to a small number of identical employees, it indicates a potential bias in the system's distribution logic based on individuals or positions, creating excessive exposure to "high-priority files" for a few individuals, while other employees with the same label receive relatively less. Its significance lies in revealing the uneven distribution of distribution resources at the individual employee level and identifying the existence of "distribution privilege positions," which is an important micro-perspective for assessing the fairness of information transmission within the government system.

[0060] Example: Suppose that out of 50 documents, only the aforementioned 8 are valid distribution documents. Statistics show that Director Wang, the regional supervisor, distributed 4 out of these 8 documents. Therefore, Director Wang's concentration in valid distribution is approximately 4 / 8 times ≈ 50%, exceeding the assumed threshold of 40%. Thus, there is personnel with excessively concentrated distribution, and the process proceeds to S134.

[0061] S134 identifies users whose number proportion exceeds a preset proportion threshold as overlapping users, and determines the type of user interest profile update requirement based on the proportion of overlapping users among the recommended users of valid recommended text.

[0062] It should be noted that when the proportion of overlapping users among the recommended users of the effective recommended text is greater than a preset proportion threshold, the exposure rate of the text is poor, and the user's interest profile update request type is determined to be the target request type.

[0063] The final decision will be made based on the proportion of overlapping personnel.

[0064] The key terms in this step are "overlapping personnel" and "preset proportion threshold." "Overlapping personnel" refers to staff identified in S133 whose concentration of personnel receiving documents exceeds a preset concentration threshold. The "preset proportion threshold" is the critical percentage (e.g., 5%) of these overlapping personnel among all staff who have received valid distribution documents. The reason for setting this step is that even if a small number of "high-frequency recipients" exist, if their proportion in the overall audience is low, it may indicate special responsibilities. Conversely, if the proportion is too high, it means that the audience for widely distributed documents has formed a highly overlapping and rigid group. The system's distribution diversity has failed at the receiving level, leading to a rigid overall document delivery model. Most staff with the same tag receive overly similar combinations of documents, potentially overlooking professional documents more suited to their specific responsibilities. Its significance lies in making a final administrative decision: when the proportion of overlapping personnel is too high, it indicates an unhealthy document distribution ecosystem under that tag, with "systemic distribution rigidity." It is necessary to break this rigid distribution model by updating the knowledge profiles of relevant staff in real time, optimizing the tag system or distribution rules, and improving the targeting and comprehensiveness of document distribution among different staff.

[0065] Example: Continuing the previous example, the total number of staff members who have received valid documents is 180. Among them, there are 12 overlapping staff members, such as Director Wang, who account for 6.67% of the total audience. If the preset threshold is 5%, then 6.67% > 5%. At this point, it is determined that the document distribution model has a risk of becoming rigid, and the knowledge profile update need type for this staff group is the target need type. Real-time updates need to be initiated to optimize the distribution strategy.

[0066] It should be noted that when the user's interest profile update request type is the target request type, the interest profile of all users will be updated in real time to increase the exposure of the text.

[0067] It should be noted that the file push users under the user tags are users who have the user tags.

[0068] The core decision-making objective of this method is to intelligently decide whether to adopt a real-time or delayed update strategy for a user's profile maintenance by analyzing which of a user's interest tags are "effective" (i.e., have a sufficient audience base) and combining this with the prevalence of the effective tags among all users. The core logic is: first, select "effective user tags" with a large audience base; then, conduct a dual evaluation at both the individual user level (number of effective tags) and the group level (prevalence of tag composition); finally, make a final decision based on a comprehensive factor. This logic aims to balance update efficiency and resource consumption—prioritizing real-time updates for users with niche interests or unique tag combinations to quickly explore optimization; and using delayed updates for users with mainstream interests and stable tags to conserve system resources.

[0069] Specifically, such as Figure 3 As shown, the method for determining the user profile update method is as follows:

[0070] S21 Determine the number of users who receive file pushes under the user tags corresponding to the user's interest profile.

[0071] Keyword Explanation: "User Tag" refers to a work attribute identifier constructed based on the legally mandated duties of government personnel and their dynamic behavioral data, such as historical records of browsing and processing documents. Examples include "policy researcher who frequently handles cross-departmental coordination documents" and "frontline law enforcement officer who frequently reviews new environmental regulations." Unlike traditional fixed tags, user tags automatically adjust as personnel behavior changes. "Document Pusher" refers to the set of all staff members currently tagged with this user tag in the government system. "Number of Document Pushers" is the total number of staff members currently covered by each user tag.

[0072] This step forms the basis for assessing the "actual impact" or "real-world prevalence" of user tags. The number of people covered by a user tag reflects not only the legal basis of that job combination or work model, but also how many people actually adopt it in the current organization's operations. It is a key indicator for distinguishing between "core work models" and "peripheral or emerging work models." The significance of this step lies in combining dynamic, personalized user profiles with group analysis, providing an objective basis based on real behavioral data for subsequent judgments on whether tags are "effective." This ensures that the analysis respects the stability of the organizational structure while capturing the dynamic evolution of work practices.

[0073] Example: Section Chief Zhang has the following user tags: "Responsible for infrastructure project approval and frequently reviews green building standards" (covering 180 people), "Participates in fiscal budget supervision and frequently reviews performance reports" (covering 120 people), "Focuses on coastal zone protection and browses relevant scientific research literature" (covering 25 people), and "Occasionally handles consultations on the renovation of historical buildings" (covering 8 people). The system retrieves the number of people to whom files are pushed to each of these four user tags.

[0074] S22 determines the valid user tags among the user tags of the user based on the number of users to whom the file is pushed;

[0075] In the above steps, it is determined whether the user has a valid user tag. If yes, proceed to the next step; otherwise, the user's profile update method is determined to be a real-time update method.

[0076] Keyword Explanation: "Effective User Tags" are defined as user tags whose number of recipients of file pushes exceeds a preset threshold. This threshold is set based on statistical analysis value, aiming to filter out tags with sufficient behavioral sample size to support reliable pattern analysis. "Real-time Update Method" refers to high-frequency, near-real-time incremental updates to the knowledge profiles of staff, enabling user tags to quickly respond to changes in user behavior.

[0077] The concept of "valid user tags" is designed to focus on work patterns that have formed a sizable group in practice. User tags with very limited coverage receive less exposure. This step is significant for critical traffic allocation: if all of a worker's user tags are invalid (i.e., their work pattern is very unique or emerging within the system), the exposure of related user tags is low, thus requiring improved update efficiency.

[0078] Example: Suppose the system sets a preset threshold of 50 users for push notifications. Comparing Section Chief Zhang's user tags: "Responsible for infrastructure project approval and frequently consults green building standards" (180 > 50) and "Participates in fiscal budget supervision and frequently reviews performance reports" (120 > 50) are valid user tags; "Focuses on coastal zone protection and browses relevant scientific literature" (25 < 50) and "Occasionally handles historical building renovation consultations" (8 < 50) are invalid user tags. Since Section Chief Zhang has valid user tags, the process proceeds to S23.

[0079] S23 determines the user profile update method based on the valid user tags in the user tags and the composition data of the valid user tags of all users.

[0080] It should be noted that the effective user tags are those of users whose number of file push users exceeds a preset threshold for the number of recommended users.

[0081] It is understood that the method for updating the user's profile, based on the valid user tags in the user's user tags and the constituent data of valid user tags for all users, specifically includes:

[0082] S231 determines whether the number of valid user tags of the user is greater than the preset threshold for the number of useful tags. If yes, the user's profile update method is determined to be a delayed update method. If no, proceed to the next step.

[0083] Determine whether the number of valid user tags for staff members is greater than the preset threshold for the number of valid tags. Keyword explanation: "Preset threshold for the number of valid tags" is a critical value used to measure the breadth of staff members' mainstream work modes, indicating whether there are user tags with sufficient exposure, such as 1 or 2.

[0084] This step assesses performance at the individual staff level. Staff with a large number of valid user tags indicate higher document exposure under those tags. The significance of this step is to identify "standardized workers" with high user tag exposure, who themselves significantly contribute to document exposure. For them, delayed update methods (such as daily or weekly batch updates) are sufficient to meet their knowledge synchronization needs, while conserving substantial resources used for real-time computation and push notifications. This reflects optimized resource allocation for stable work patterns.

[0085] Example: Section Chief Zhang has 2 valid user tags. Assuming the preset threshold for the number of valid tags is 1, then 2 > 1, the condition is met. According to the rules, this can be directly determined as a delayed update. However, to demonstrate the complete process, let's assume the threshold is set to 3, then 2 < 3, the condition is not met, and the process proceeds to S232.

[0086] S232 determines whether the proportion of the user's valid user tags in the total number of valid user tags of all users is greater than a preset proportion threshold. If yes, the user's profile update method is determined to be a delayed update method. If no, proceed to the next step.

[0087] Keyword Explanation: "Composition Ratio" here specifically refers to the proportion of people covered by a particular valid user tag owned by this staff member out of the total number of people covered by all valid user tags of all staff members. "Preset Composition Ratio Threshold" is used to determine its contribution to high exposure.

[0088] This step involves screening at the group level. If a staff member's valid user tags cover a very high percentage of the population, their impact on increasing the document's exposure is significant, thus reducing the need for timely updates.

[0089] Example: Calculate the composition ratio of Section Chief Zhang's valid tags. Assume the total number of valid user tags in the entire system is 8. "Then his tag ratio is 0.25, which is less than 0.3, so the condition is not met, and the process proceeds to S233."

[0090] S233 determines the user's update requirement factor based on the number of valid user tags and the number of user tags, and determines the user's profile update method based on the update requirement factor.

[0091] It should be noted that the update demand factor is related to the number of valid user tags and the number of user tags. The more valid user tags and the more user tags a user has, the smaller the update demand factor will be.

[0092] The "Update Demand Factor" is a quantitative indicator that comprehensively assesses the urgency of updating staff dynamic profiles. Its design principle is: the more valid user tags a staff member has, and the greater the total number of user tags (valid + invalid), the smaller the update demand factor. The logic is that a higher number of valid tags and a larger total number of tags significantly increase exposure, thus resulting in a lower update demand factor.

[0093] The first two steps (S231, S232) involve rapid decision-making from two extreme perspectives: "whether there are enough mainstream patterns" and "whether it is in the most prevalent pattern." S233 introduces a smoother, more comprehensive indicator to handle a wider range of intermediate situations. The core design principle of the update requirement factor is that the urgency of profile updates is inversely proportional to the "certainty" and "diversity" of the work patterns they reflect. The significance of this step lies in achieving refined resource grading, enabling the scientific allocation of different strategies—from millisecond-level real-time updates to day-level delayed updates—based on the complexity and maturity of staff behavior patterns. This maximizes the overall operational efficiency of the government system while ensuring timely synchronization of knowledge among key personnel.

[0094] Example: Assume the update demand factor U = K / (α * N_valid + β * N_total), where N_valid is the number of valid user tags (2), N_total is the total number of user tags (4), α and β are weighting coefficients, and K is a constant. The calculated U is a numerical value. Assume the preset demand factor threshold is T. If U > T, it indicates a high update demand, and a real-time update method is used; if U ≤ T, a delayed update method is used. Assuming that U > T after calculation, the final method for updating the profile of Section Chief Zhang is the real-time update method.

[0095] Specifically, when the update demand factor is greater than a preset demand factor threshold, the user profile update method is determined to be a real-time update method; otherwise, the user profile update method is determined to be a delayed update method.

[0096] S2 updates the user's interest profile based on the profile update method, determines the update delay type for different users based on the profile update method, and uses the update delay type to determine the anomaly parsing and identification scheme of the model under different user tags;

[0097] Furthermore, the method for determining the user's update delay type is as follows:

[0098] When the user's profile update method is a real-time update method, the user's update delay type is determined to be a real-time update user;

[0099] When the user's profile update method is a delayed update method, the user's update delay type is determined to be a delayed update user.

[0100] The core decision-making objective of this method is to intelligently select the anomaly parsing and identification scheme that best matches the tag by comprehensively analyzing the profile update delay of users matched by user tags and fully integrating the logic of previous embodiments such as user tag construction and profile update requirement determination. The core logic is a tightly connected multi-level decision-making process: first, it determines the "tag-matching user" based on the dynamic tags constructed from the user profile; then, it combines... Figure 3 The implementation example identifies "update delay users" by determining the update delay type. Finally, through a four-step conditional judgment, based on indicators such as system-wide risk, tag popularity, and the degree of delay pollution, a choice is made between two anomaly analysis schemes. This logic aims to establish a complete governance chain—from tag construction to update decision-making to effect monitoring—ensuring the coordinated operation and optimized resource allocation of the recommendation system at each stage.

[0101] Specifically, such as Figure 4 As shown, the method for determining the anomaly parsing and identification scheme under the user tag is as follows:

[0102] S31 uses users whose user tags exist in the interest profile as tag matching users;

[0103] "User Tags" refer to dynamic tags built based on user profiles, such as "personnel handling high-frequency approvals of new energy subsidies" (built based on responsibilities and viewed documents). "Interest Profiles" are dynamic profiles built by the system based on user behavior data, containing a set of tags that are adjusted in real time. "Tag-Matched Users" refers to the set of all users in the current system whose interest profiles are marked with that specific dynamic tag.

[0104] Why this setup is used and its significance: This step is a direct application of the logic used in the previous embodiments to build tags based on user responsibilities and browsing behavior. By transforming abstract user profile tags into specific user groups, it provides clear target objects for subsequent anomaly analysis. Its significance lies in ensuring that the objects of anomaly analysis monitoring are real, dynamic user groups within the business scenario, rather than static, pre-defined categories. This allows monitoring to align with actual business changes, reflecting the dynamism and accuracy of user profile construction. Simultaneously, this step also completes the transition from "personalized tag definition" to "group effect evaluation."

[0105] Example: Based on the implementation of the government system, after the user tag "frontline law enforcement personnel who frequently consult the latest environmental regulations" is constructed, the system determines that there are currently 800 staff members who match this tag, and these 800 staff members become the "tag matching users" of this tag.

[0106] S32 determines the users with update delays based on the different update delay types of each user;

[0107] "Update Delay Type" refers to... Figure 3 The user profile update method is used to classify users into different types based on the implementation examples. When the user profile update method is a real-time update method, the update delay type is real-time update user; when the user profile update method is a delayed update method, the update delay type is delayed update user. "Delayed update user" specifically refers to those who update through... Figure 3 In the implementation example, the decision process determines that the user adopts the delayed update method.

[0108] This step enables a seamless transition between previous and subsequent implementations, directly transforming the results of profile update decisions into risk inputs for anomaly analysis and monitoring. Figure 3The example, by analyzing metrics such as the number and proportion of valid user tags, has scientifically determined which users need real-time updates and which can have delayed updates. This decision is continued here, treating "users with delayed updates" as potential sources of recommendation risk because their profiles may be somewhat outdated. The significance lies in constructing a complete governance chain: first deciding "how to update user profiles," then deciding "how to monitor recommendation effectiveness" based on the update status, forming a closed-loop decision-making process. This avoids redundant analysis and improves the overall efficiency of the system.

[0109] Example: In Figure 3 In this embodiment, User A is determined to require a real-time update method due to the small number of valid tags and their unique composition. User B is determined to require a delayed update method due to the large number and stability of valid tags. In this step, User B is categorized as an "update-delayed user" and will be considered for risk assessment in subsequent anomaly analysis.

[0110] S33 determines the anomaly parsing and identification scheme under the user tag based on the updated delayed user data and the tag matching user of the user tag.

[0111] It is understood that, based on the updated delayed user data and the user tags matching the users, an anomaly parsing and identification scheme under the user tags is determined, specifically including:

[0112] S331 Based on the updated delayed user data, determine the proportion of updated delayed users among all users and use it as the delayed user proportion. Determine whether the delayed user proportion is greater than the preset delayed user proportion threshold. If so, the number of updated delayed users is large. In order to improve the accuracy of recommendation processing, determine the anomaly parsing and identification scheme under all user tags as the preset parsing scheme. If not, proceed to the next step.

[0113] This step is the overall decision-making step, and its specific logic is implemented through the following sub-steps (S331-S334), which aim to select between two anomaly analysis schemes based on the data provided in the preceding steps. The preset analysis scheme uses a lower trigger threshold (the second preset threshold for the percentage of browsing users), while the second preset analysis scheme uses a higher trigger threshold (the preset threshold for the percentage of browsing users).

[0114] Determining whether the percentage of users experiencing global latency is too high is a step in assessing the overall health of the system. A high percentage of users experiencing latency means that the profiles of a large number of users are not being updated, which directly affects the reliability of the entire recommendation system. In this case, a more sensitive preset parsing scheme (low threshold) is needed for comprehensive monitoring. This reflects the linkage mechanism between anomaly parsing and system operation and maintenance status (update latency), automatically increasing the monitoring intensity when a global problem occurs in the system.

[0115] Example: Suppose the system is based on Figure 3 The implementation example determined that 450 out of 3000 users on the entire site experienced update delays (15%). If the preset threshold for delayed user percentage is 10%, then 15% > 10%. At this point, the system is determined to have a serious delay risk, and the anomaly parsing scheme under all user tags is determined to be the preset parsing scheme (using a lower trigger threshold), entering a state of comprehensive high-sensitivity monitoring.

[0116] S332 determines whether the number of users matching the user tag is less than a preset threshold for the number of matching users. If so, the number of users matching the user tag is small, and therefore the anomaly parsing and identification scheme under the user tag is determined to be the preset parsing scheme. If not, proceed to the next step.

[0117] Determining if the number of users matching a tag is too low involves combining user tag construction logic to identify long-tail or emerging business scenarios. If the number of users matching a certain tag is very small (e.g., ... Figure 1 The "niche business" tag in this example indicates that the business area has few participants or is just emerging, with sparse data and poor model stability. A pre-defined parsing scheme (low threshold) is needed to enhance monitoring, which is consistent with the principle of focusing on niche businesses in previous examples.

[0118] Example: In a government system, the tag "responsible for cross-border data security review" might only have 50 matching users (far fewer than the preset threshold of 200 matching users). In this case, regardless of other conditions, the anomaly analysis scheme under this tag will be directly determined as the preset analysis scheme, ensuring close monitoring of this emerging, important, but niche business area.

[0119] S333 determines whether there are users with delayed updates among the users whose tags match the user tags. If yes, proceed to the next step. If no, the update timeliness of the user profile whose tags match the user tags is high. Therefore, through the anomaly analysis under the user tags, the reliability of the model's recommendation can be accurately evaluated. Thus, the anomaly analysis identification scheme under the user tags is determined to be the preset analysis scheme.

[0120] Determining whether there are users with delayed updates among those matching tags involves checking whether specific business tags are "polluted" by profile delays. If all matching users for a certain tag are real-time updated users, it indicates that the profiles of relevant personnel in that business area are up-to-date, and the recommendation environment is good. A pre-defined parsing scheme can then be used for routine monitoring. This reflects the idea of ​​precise policy implementation: applying appropriate monitoring intensity only to user groups that truly pose a risk.

[0121] Example: For the tag "daily document processing personnel", there may be 5,000 matching users, all of whom are real-time updated users. In this case, the preset parsing scheme is directly adopted. However, note that although the "preset parsing scheme" is real-time parsing, the actual conditions for triggering parsing may be relatively lenient (low threshold) because the user profiles are fresh. This is more of a protective rather than error-correcting monitoring.

[0122] S334 determines whether the proportion of users with update delays among the users whose tags match the user tag is greater than a preset delay user proportion threshold. If yes, the abnormal parsing and identification scheme under the user tag is determined to be the second preset parsing scheme. If no, the abnormal parsing and identification scheme under the user tag is determined to be the preset parsing scheme.

[0123] This step is crucial for refined management. When the proportion of delayed users within a tag exceeds a threshold, it means that the profiles of some key personnel in that business area are severely outdated. In this case, the reliability of anomaly analysis based on browsing data is low, so a second preset analysis scheme is adopted. Conversely, if the proportion is controllable, the preset analysis scheme can be used, and in-depth analysis is only performed when the recommendation effect is significantly poor (extremely low browsing rate). This is consistent with the logic of allocating resources according to risk level in the previous embodiment.

[0124] Example: For the tag "emergency management personnel", 1000 users are matched, of which 150 are users with delayed updates (accounting for 15%). If the preset threshold for delayed user proportion is 10%, then 15% > 10%, so the second preset parsing scheme (high threshold) is selected.

[0125] It should be noted that the preset parsing scheme only parses the recommended text when the ratio of the number of users browsing the recommended text under the user tag to the number of recommended users is less than the second preset threshold for the proportion of browsing users (e.g., less than 10%), in order to determine whether there is a parsing anomaly in the model, that is, whether the recommended text is linked to the user tag.

[0126] Additionally, it can be understood that the second preset parsing scheme determines whether parsing is needed based on the number of users who viewed the recommended text under the user tag. Specifically, the recommended text is only parsed when the ratio of the number of users who viewed the recommended text under the user tag to the number of recommended users is less than a preset threshold for the percentage of users who viewed the text (e.g., less than 20%), in order to determine whether there is a parsing anomaly in the model, that is, whether the recommended text is linked to the user tag.

[0127] It should be noted that the second preset threshold for the percentage of browsing users is less than the preset threshold for the percentage of browsing users.

[0128] S3 uses the aforementioned anomaly analysis and identification scheme to determine the identification deviation of the natural language processing model under different user tags, and determines the users whose interest profiles are updated in real time based on the identification deviation of the natural language processing model under different user tags.

[0129] The core decision-making objective of this method is to identify high-risk business tags and corresponding staff by analyzing the recognition bias of natural language models under different business tags in the government system, thereby accurately determining which staff profiles require real-time updates. The core logic is a three-tiered decision-making process based on risk transmission: first, quantifying the degree of model recognition bias for each business tag; second, filtering out "recognition bias risk tags" that exceed the risk threshold; and finally, based on these risk tags and the personnel involved, determining the scope of staff requiring real-time updates through multi-condition judgment. This logic aims to establish a linkage mechanism between model performance monitoring and profile update decisions, ensuring that when document recommendation models in specific business areas encounter problems, the profile update strategy for relevant personnel can be adjusted promptly to improve the efficiency of user profile updates and the reliability of identifying bias tags.

[0130] Specifically, the method for determining users in the real-time update process of the interest profile is as follows:

[0131] S41 determines the number of recognition deviations of recommended texts under different user tags based on the recognition deviation of natural language models under different user tags, and determines the recognition deviation factor of the user tag based on the proportion of the number of recognition deviations of recommended texts under the user tag in the total number of recommended texts under the user tag.

[0132] "Natural Language Model Recognition Bias" refers to errors or biases that occur when the natural language processing model used in the government system to understand document content and match business tags, when processing documents in a specific business domain. For example, the model might fail to correctly identify a document about "renovation of old residential areas" as belonging to the "urban construction" tag, and incorrectly classify it into another tag. "Number of Recognition Bias in Recommended Texts" refers to the number of documents incorrectly identified or recommended by the model under this business tag within a certain statistical period. "Number Percentage" is the proportion of the number of recognition biases to the total number of recommended texts under this tag. "Recognition Bias Factor" is a quantitative indicator used to measure the severity of model recognition bias under this business tag, usually calculated based on the number percentage.

[0133] This step is fundamental to risk quantification. In government systems, the accuracy of natural language models directly impacts the precise delivery of policy documents to relevant staff. By conducting fine-grained monitoring and quantitative evaluation of model performance under each business tag, we can identify which business areas exhibit significant problems with the model. The significance of calculating the "identification bias factor" lies in transforming the qualitative notion that "the model may be biased" into a quantitative, comparable risk value. This provides an objective, data-driven decision-making basis for subsequent risk tag selection, ensuring that resources are prioritized for the business areas with the most prominent problems.

[0134] Example: In a government affairs system, 100 relevant documents were recommended in the past week for the "Flood Control and Drought Relief" business tag. Through manual sampling or system verification, it was found that 8 of these documents had recommendation biases (e.g., flood control plans were incorrectly recommended to unrelated personnel). Therefore, the number of biased recommendations is 8, representing 8% (8 / 100). Assuming this percentage is directly used as the bias factor, the bias factor for the "Flood Control and Drought Relief" tag is 0.08.

[0135] S42 determines the identification bias risk label in the user label based on the identification bias factor of the user label;

[0136] "Preset deviation factor threshold" is a system-defined threshold used to determine whether the model deviation of a business label has reached a level requiring special attention. "Identify deviation risk labels" refers to business labels whose identified deviation factors exceed the preset deviation factor threshold.

[0137] This step is crucial for risk focusing. Not all labels with some degree of deviation require immediate action; only labels with deviations exceeding acceptable limits are considered high-risk. Setting a "preset deviation factor threshold" and filtering "identify deviation risk labels" enables tiered risk management, avoiding overreactions to minor deviations. It helps the system quickly pinpoint the few high-risk areas requiring immediate intervention from numerous business labels, ensuring that subsequent profile updates are targeted and resources are concentrated on resolving the most critical issues.

[0138] Example: The system's preset deviation factor threshold is 0.05. According to the calculation in step S41, the identification deviation factor of the "Flood Control and Drought Relief" label is 0.08 > 0.05, therefore it is identified as a risk label for identification deviation. However, the identification deviation factor of the "Routine Document Circulation" label is 0.02 < 0.05, therefore it is not a risk label.

[0139] S43 uses the identification bias risk label data and combines it with user data containing the identification bias risk label to determine whether the user is a user subject to real-time update processing of the interest profile.

[0140] This step is the final decision-making step, and its specific logic is implemented through the following sub-steps (S431-S433). The aim is to comprehensively determine which staff members need to be included in the real-time update processing based on factors such as the scale of the risk labels and the proportion of delayed personnel involved. Special note: The "users" focused on in this step specifically refer to the original users whose updates were delayed (based on...). Figure 3 / 4 The embodiment has been identified as the person using the delayed update method, because the real-time update user is already updating in real time.

[0141] Furthermore, the identification deviation risk label is a user label whose identification deviation factor is greater than a preset deviation factor threshold.

[0142] Furthermore, the user mentioned is the original user with the update delay.

[0143] Specifically, if the number of identified deviation risk tags is greater than a preset risk tag number threshold, then all users are determined to be users whose interest profiles are updated in real time.

[0144] Furthermore, in step S431, if the number of identification deviation risk tags is not greater than the preset risk tag number threshold, the user data with identification deviation risk tags is used to determine whether the proportion of users with identification deviation risk tags among all users with update delays is less than the preset deviation risk user proportion. If so, it is determined that all users belong to the users of the real-time update processing of interest profiles. If not, proceed to the next step.

[0145] "Preset risk tag quantity threshold" is a critical value used to determine whether there are too many high-risk business areas in the system. "Users with identification bias risk tags" refers to staff whose business profiles contain at least one identification bias risk tag. "Preset bias risk user percentage" refers to the percentage threshold of these high-risk personnel out of all personnel with update delays.

[0146] This step assesses the overall breadth of risks faced by the system. If there are too many risk labels, it indicates widespread model bias, no longer a localized problem, requiring a systemic approach—including all staff in real-time updates. If there are few risk labels, but the proportion of high-risk personnel involved is low, it indicates the risk is concentrated in a small area; a conservative strategy of "full real-time updates" can also be adopted to quickly curb the spread of risk. Both scenarios employ the most thorough intervention, reflecting the management principle of "taking decisive measures when risks spread."

[0147] Example: Suppose the system analyzes 100 business tags and identifies 5 risk tags (such as "flood control and drought relief," "data security," etc.). The preset threshold for the number of risk tags is 3. Since 5 > 3, the condition is met. At this point, it is determined that all staff members belong to the users whose interest profiles are updated in real time. This is scenario one: too many risk tags.

[0148] Another scenario: If the number of risk tags is 2 (≤3), and the check reveals that only 10 users have update delays due to these risk tags, while the total number of users with update delays in the entire system is 100, then the percentage is 10%. If the preset percentage of users with deviation risk is 5%, then 10% is not less than 5%, and the condition is not met.

[0149] S432 determines whether the number of identification deviation risk tags of the user is greater than the risk tag number threshold. If yes, it determines that the user belongs to the user of the real-time update processing of interest profile. If no, it proceeds to the next step.

[0150] "Risk label quantity threshold" (note that this is different from "preset risk label quantity threshold" in S431) is a threshold value set for a single employee to judge the level of their personal risk. It refers to the threshold number of identification deviation risk labels included in the employee's business profile.

[0151] This step involves assessment at the individual staff level. If a staff member's responsibilities involve multiple high-risk business areas (i.e., their profile contains multiple risk tags), it indicates that their work is complex and operates in areas where multiple models are unreliable. The accuracy of their personal profile is crucial for the correct receipt of documents. Therefore, even if the overall risk is manageable, it is necessary to isolate these "high-risk" personnel and update their profiles in real time. This reflects refined management and focused protection for key positions.

[0152] Example: (Following the branch of S431 "No") The system has 5 risk tags, involving high-risk personnel who account for 8% (>5%) of those with delayed updates. Now, we are assessing staff member Director Li. Director Li's business profile includes 2 risk tags: "Flood Control and Drought Relief" and "Data Security". Assuming the threshold for the number of risk tags for an individual is 1, then 2 > 1, the condition is met. Therefore, it is determined that Director Li belongs to the user group for real-time update processing of interest profiles.

[0153] S433 If the number of users with the identification bias risk label is not greater than the preset real-time update user number threshold, then all users with the identification bias risk label will be treated as users for real-time update of interest profiles.

[0154] The "Preset Real-Time Update User Count Threshold" is a percentage or absolute number threshold set for the group of "users with identification bias risk labels" to determine whether this high-risk group has received sufficient real-time update coverage.

[0155] This step serves as a safety net and safeguard. After the S432 screening, some high-risk individuals may still not be included in the real-time updates (because their individual risk tags do not meet the required number). However, if the proportion or number of these high-risk individuals already in the real-time update phase is too low, it means that the high-risk group as a whole lacks timely protection. To prevent the risk from spreading within the group, all remaining high-risk individuals are then included in the real-time updates. The significance of this is to ensure comprehensive protection for high-risk business areas and prevent the group's risk from spiraling out of control due to oversights in individual screening.

[0156] Example: (Following the branch of S432 "No") Staff member Xiao Wang has a risk label, but it does not exceed the personal threshold (assuming the threshold is 2). Checking all 50 staff members with the risk label corresponding to Xiao Wang, we find that only 5 of them are currently real-time updated users. If the preset threshold for real-time updated users is 10 (or 20%), then 5 < 10, and the condition is not met. In this case, all 50 staff members with risk labels (including Xiao Wang) are treated as users for real-time updates of their interest profiles.

[0157] Example 2

[0158] In a second aspect, the present invention provides a computer system comprising: a memory and a processor connected in communication, and a computer program stored in the memory and capable of running on the processor, wherein the processor executes the above-described intelligent text analysis method for a government service system when running the computer program.

[0159] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments of apparatus, devices, and non-volatile computer storage media are basically similar to the method embodiments, so the descriptions are relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0160] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0161] The above description is merely one or more embodiments of this specification and is not intended to limit this specification. Various modifications and variations can be made to the one or more embodiments of this specification by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of one or more embodiments of this specification should be included within the scope of the claims of this specification.

Claims

1. An intelligent text analysis method for government service systems, characterized in that, Specifically, it includes: Based on the text recommendation data of the government service system, recommendation data under different user tags is determined. Based on the recommendation data under different user tags, the update requirement type of the user's interest profile is determined. The update requirement type includes target requirement type and non-target requirement type. When the update requirement type of the user's interest profile is target requirement type, all interest profiles of the user are updated in real time. When the update request type does not belong to the target request type, the method for updating the user's profile is determined based on the file push user under the user tag corresponding to the user's interest profile. The method for determining the user's profile update method is as follows: Based on the number of file push users under the user's user tags, the effective user tags in the user's user tags are determined. The effective user tags are those whose number of file push users is greater than a preset recommended user number threshold. Based on the effective user tags in the user's user tags and the effective user tags of all users, the user's profile update method is determined. When the user's profile update method is a real-time update method, the user's update delay type is determined to be a real-time update user. When the user's profile update method is a delayed update method, the user's update delay type is determined to be a delayed update user. The user's interest profile is updated based on the aforementioned profile update method. An anomaly parsing and identification scheme for the natural language processing model under different user tags is determined using the update delay type. The anomaly parsing and identification scheme for the natural language processing model under different user tags is determined using the update delay type. This includes triggering the parsing of recommended text based on the update delay type and the percentage of users browsing recommended text under the user tags, and determining whether there are parsing anomalies in the natural language processing model, i.e., whether the recommended text is linked to the user tags. Based on the aforementioned anomaly analysis and identification scheme, the identification deviation of the natural language processing model under different user tags is determined, and the user's interest profile is updated in real time according to the identification deviation of the natural language processing model under different user tags.

2. The intelligent text analysis method for government service systems as described in claim 1, characterized in that, The method for determining the update request type of the user's interest profile is as follows: Based on recommendation data under different user tags, determine the recommended text under the user tags; Based on the user's user tags and the recommended text under the user tags, determine the number of users recommended for different recommended texts; Based on the number of users recommending different recommended texts, the update requirement type of the user's interest profile is determined.

3. The intelligent text analysis method for government service systems as described in claim 1, characterized in that, The file push users under the user tags are those whose user tags exist.

4. The intelligent text analysis method for government service systems as described in claim 1, characterized in that, If it is determined that the user does not have a valid user tag, then the user's profile update method is determined to be a real-time update method.

5. The intelligent text analysis method for government service systems as described in claim 1, characterized in that, The method for determining users in the real-time update process of the interest profile is as follows: Based on the recognition deviation of the natural language model under different user tags, the number of recognition deviations of the recommended text under different user tags is determined. Based on the proportion of the number of recognition deviations of the recommended text under the user tag in the total number of recommended texts under the user tag, the recognition deviation factor of the user tag is determined. Based on the identification bias factor of the user tag, identify the identification bias risk tag in the user tag; Using the identification bias risk label data and combined with user data containing the identification bias risk label, it is determined whether the user is a user subject to real-time update processing of the interest profile.

6. A computer system, comprising: A memory and processor connected by communication, and a computer program stored in the memory and capable of running on the processor, characterized in that, when the processor runs the computer program, it executes an intelligent text analysis method for a government service system as described in any one of claims 1-5.

Images