Knowledge item updating and tracing method based on change association analysis and user feedback

Abstract

The application discloses a knowledge item updating and tracing method based on change correlation analysis and user feedback, and relates to the technical field of computer information processing, comprising the following steps: S1: receiving multiple user feedbacks for the same knowledge item, and starting a conflict coordination process when it is determined that there is a content conflict among the user feedbacks. The knowledge item updating and tracing method based on change correlation analysis and user feedback realizes the root cause analysis of user feedback conflicts in the knowledge item updating process by constructing a change influence network and locating a conflict root anchor point, and overcomes the decision-making stalemate caused by simple aggregation rules. By using a simulated consensus deduction and multi-round verification mechanism, the system can pre-evaluate the acceptability of the updating scheme and perform closed-loop optimization based on real feedback, thereby effectively improving the accuracy and stability of the knowledge item updating decision.

Description

Technical Field

[0001] This invention relates to the field of computer information processing technology, specifically to a method for updating and tracing knowledge entries based on change correlation analysis and user feedback. Background Technology

[0002] In existing knowledge entry update and tracing systems, a direct aggregation mechanism based on user feedback is commonly used to drive content evolution. When multiple users submit conflicting modification suggestions for the same knowledge entry, the system typically makes decisions based on pre-set simple rules, such as adopting the majority vote or prioritizing the most recently submitted feedback. This approach exposes a fundamental technical flaw: it cannot intelligently analyze and reconcile the underlying causes of feedback conflicts, leading to decision-making deadlocks or frequent repetitions in the knowledge entry update process. The root cause lies in the fact that existing systems treat user feedback as isolated voting events, failing to dynamically correlate these feedbacks with the complete historical change chain of the knowledge entry, and ignoring the credibility and contextual differences of different feedback sources. Therefore, the update decisions made by the system often lack robust tracing logic support, not only prolonging the consensus-reaching cycle but also potentially causing the final version of the knowledge entry to contain unverified errors, severely damaging the accuracy and reliability of the knowledge base. This invention aims to solve the practical problem of inaccurate update decisions and incomplete tracing logic caused by the lack of dynamic correlation analysis and mediation of conflicting feedback in existing technologies. Summary of the Invention

[0003] The purpose of this invention is to provide a method for updating and tracing knowledge entries based on change correlation analysis and user feedback, so as to solve the problems mentioned in the background art.

[0004] To address the aforementioned technical problems, this invention provides the following technical solution: a method for updating and tracing knowledge entries based on change correlation analysis and user feedback, comprising the following steps:

[0005] S1: Receive multiple user feedbacks for the same knowledge item. When it is determined that there is a content conflict among the user feedbacks, initiate a conflict coordination process. Construct a change impact network for the knowledge item, wherein the change impact network consists of nodes and edges connecting the nodes. Nodes represent the historical versions of the knowledge item and the historical user feedback associated with each historical version, and edges represent the change association strength between different historical versions.

[0006] S2: Based on the changes affecting the network and the currently received conflicting user feedback, locate at least one historical node that caused the current conflict as the root cause anchor point of the conflict; based on the root cause anchor point of the conflict and the content of the current conflicting user feedback, synthesize at least one knowledge item update candidate scheme.

[0007] S3: Perform a simulation consensus deduction on the candidate solutions for updating the knowledge items. The simulation consensus deduction uses a pre-set behavior prediction model to simulate the expected feedback behavior of a specific user group after the candidate solutions are released, and selects the candidate solutions that meet the expected consensus requirements as the solutions to be verified based on the simulation results.

[0008] S4: Publish the verification scheme to be verified within a limited user scope and collect the verification feedback data generated within that limited user scope; make adaptive adjustments to the verification scheme based on the verification feedback data, and generate the final knowledge item update instruction based on the adjustment results;

[0009] S5: Execute the version update of the knowledge entry according to the final update instruction, and generate the corresponding traceability record. The traceability record includes at least the conflict root cause anchor information, key process data of the simulated consensus deduction, and a summary of the verification feedback data.

[0010] Furthermore, constructing the network of changes affecting the knowledge entries specifically includes:

[0011] Extract all historical versions of the knowledge entry, arranged in chronological order;

[0012] Each historical version of the content and its corresponding set of user feedback are abstracted into a network node;

[0013] Calculate the semantic similarity and logical dependency between any two historical versions of content using natural language processing techniques;

[0014] Based on the weighted calculation results of semantic similarity and logical dependency, the weight value of the edge connecting two corresponding historical version nodes is determined, and the weight value is used to characterize the strength of the change association.

[0015] Furthermore, the location of at least one historical node that triggered the current conflict as the root cause anchor point specifically includes:

[0016] Match the content topics involved in the current conflicting user feedback with the content topics carried by each node in the network affected by the change;

[0017] Calculate the user feedback on the current conflict and the distance of each node in the opinion vector space;

[0018] Nodes whose viewpoint vector distance is less than a set threshold and whose edge weight value is higher than a specific threshold are selected and marked as the conflict root source anchor points.

[0019] Furthermore, the simulation consensus deduction of the candidate schemes for updating the knowledge entries specifically includes:

[0020] The candidate solutions are input into a pre-trained user behavior prediction model;

[0021] The user behavior prediction model predicts the possible set of operations and the corresponding probability distribution of users under different user attribute tags for the candidate scheme based on historical interaction data in the change impact network.

[0022] Based on the set of possible operations and probability distribution, calculate the expected conflict index for each candidate solution, and eliminate candidate solutions whose expected conflict index exceeds the acceptable range.

[0023] Furthermore, the training data for the pre-trained user behavior prediction model includes:

[0024] The changes affect the user's actual operational behavior data regarding historical versions recorded in the network;

[0025] The user's historical feedback content and its corresponding reputation rating;

[0026] Semantic feature vectors of changes in the content of knowledge entries.

[0027] Furthermore, the limited user scope for publishing the verification scheme within a limited user range is dynamically determined based on the user's accumulated reputation score in the network affected by the change.

[0028] Furthermore, the adaptive adjustment of the verification scheme based on the verification feedback data specifically includes:

[0029] Analyze the semantics of modification suggestions contained in the verification feedback data;

[0030] The proposed modifications are then integrated with the scheme to be verified to generate at least one adjusted scheme variant.

[0031] The behavioral prediction model is used to simulate and extrapolate the variant scheme again, and the variant scheme with the highest expected consensus is selected as the basis for generating the final update instruction.

[0032] Furthermore, the method also includes:

[0033] The new user feedback data, solution decision data, and final update results generated during this conflict coordination process are added as new nodes and edges to the change impact network to update the training dataset of the behavior prediction model.

[0034] Furthermore, the source tracing records are stored in a structured data format and support reverse querying and visualization through the node identifier of the conflict root cause anchor point.

[0035] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for updating and tracing knowledge entries based on change association analysis and user feedback.

[0036] This invention provides a method for updating and tracing knowledge entries based on change correlation analysis and user feedback. It has the following beneficial effects:

[0037] This knowledge entry update and tracing method, based on change correlation analysis and user feedback, constructs a change impact network and locates the root cause anchor points of conflicts. This enables root cause analysis of user feedback conflicts during the knowledge entry update process, overcoming the decision-making deadlock caused by simple aggregation rules. Utilizing simulated consensus deduction and multi-round verification mechanisms, the system can pre-assess the acceptability of update schemes and perform closed-loop optimization based on real feedback, thereby effectively improving the accuracy and stability of knowledge entry update decisions.

[0038] This knowledge entry update and tracing method, based on change correlation analysis and user feedback, records the complete decision logic in a structured manner as tracing information and establishes a data feedback mechanism to continuously optimize the analysis model. This method not only enhances the traceability and transparency of knowledge change history, but also enables the system to have the ability to continuously evolve and learn, ultimately ensuring the quality reliability and long-term maintenance efficiency of the knowledge base content. Attached Figure Description

[0039] Figure 1 This is a flowchart illustrating the knowledge entry update and tracing method based on change correlation analysis and user feedback according to the present invention.

[0040] Figure 2 This is a system data flow diagram of the knowledge item update and tracing method based on change association analysis and user feedback according to the present invention;

[0041] Figure 3 This is a flowchart illustrating the adaptive adjustment process of the knowledge entry update and tracing method based on change correlation analysis and user feedback according to the present invention. Detailed Implementation

[0042] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0043] Please see Figures 1 to 3 This invention provides a technical solution: a method for updating and tracing knowledge entries based on change correlation analysis and user feedback, comprising the following steps:

[0044] S1: Receive multiple user feedbacks for the same knowledge item. When it is determined that there is a content conflict between the user feedbacks, initiate the conflict coordination process. Construct a change impact network for the knowledge item. The change impact network consists of nodes and edges connecting the nodes. Nodes represent the historical versions of the knowledge item and the historical user feedback associated with each historical version. Edges represent the change association strength between different historical versions.

[0045] S2: Based on the changes affecting the network and the currently received conflicting user feedback, locate at least one historical node that caused the current conflict as the root cause anchor point; based on the root cause anchor point and the content of the current conflicting user feedback, synthesize at least one knowledge entry update candidate scheme.

[0046] S3: Simulate consensus deduction for candidate solutions to knowledge item updates. The simulation consensus deduction uses a pre-set behavior prediction model to simulate the expected feedback behavior of a specific user group after the candidate solution is released, and selects the candidate solutions that meet the expected consensus requirements as the solutions to be verified based on the simulation results.

[0047] S4: Publish the solution to be verified within a limited user scope and collect the verification feedback data generated within that user scope; make adaptive adjustments to the solution to be verified based on the verification feedback data, and generate the final knowledge item update instruction based on the adjustment results;

[0048] S5: Execute the version update of the knowledge entry according to the final update instruction and generate the corresponding source record. The source record shall at least contain the root cause anchor information of the conflict, the key process data of the simulated consensus deduction, and the summary of the verification feedback data.

[0049] It's important to further clarify that when the system detects contradictory user feedback regarding the same knowledge item, it doesn't directly apply simple rules for decision-making. Instead, it automatically triggers a multi-stage intelligent analysis process. First, the system dynamically constructs a change impact network for the knowledge item. This network is a graph structure model based on historical data, where nodes are abstracted from each historical version of the knowledge item and all corresponding user feedback. Edges represent the evolutionary relationships between different historical versions, and their weights are quantified by analyzing the semantic similarity and logical dependencies between versions using natural language processing techniques. This precisely characterizes the strength of the association between changes.

[0050] Based on this, the system maps the feedback content of the current conflict into the network. By calculating its distance to each historical node in the viewpoint vector space and combining it with the weights of the connecting edges, it locates one or more key historical nodes that triggered the current conflict, i.e., the root source anchors of the conflict. Subsequently, based on the located root source anchors and the specific content of the current conflict, the system intelligently synthesizes one or more knowledge entry update candidate solutions. These solutions may include content fusion, version rollback, or other logically derived corrective versions.

[0051] Next, the system introduces a key simulation consensus deduction step, which uses a pre-trained user behavior prediction model. This model is trained based on historical interaction data, user reputation information and content features in the change impact network. It simulates the expected feedback behavior of the user group after the release of each candidate solution, calculates the expected conflict index, and selects the solution with higher consensus as the solution to be verified.

[0052] Subsequently, the system will publish the selected solutions within a limited scope comprised of high-reputation users, and meticulously track and collect verification feedback data generated by users within that scope, including operational behaviors and textual suggestions.

[0053] The system then uses this verification data to adaptively adjust and optimize the verification plan at the content level, ultimately generating and executing knowledge entry update instructions. The entire decision-making process, including the identification of conflict root cause anchors, key parameters and results of simulations, and summaries of verification feedback, is fully recorded as structured traceability information and attached to the updated knowledge entries, ensuring the transparency and traceability of the update decision-making process. This series of steps constitutes a closed-loop processing mechanism that deeply integrates historical correlation analysis and dynamic feedback verification.

[0054] The impact of changes to knowledge entries on the network specifically includes:

[0055] Extract all historical versions of the knowledge entry, arranged in chronological order;

[0056] Each historical version of the content and its corresponding set of user feedback are abstracted into a network node;

[0057] Calculate the semantic similarity and logical dependency between any two historical versions of content using natural language processing techniques;

[0058] The semantic similarity is calculated using the BERT-based cosine similarity algorithm, and the specific steps and formulas are as follows:

[0059] The semantic similarity is calculated using the BERT-based cosine similarity algorithm. The specific steps are as follows:

[0060] Using a pre-trained BERT-base-uncased model, the text content of two historical versions, denoted as Text, is processed. A Text B Encode the text to obtain the vector representation of the [CLS] token, denoted as Vec. A Vec B All dimensions are 768;

[0061] Calculate the cosine similarity S using the following formula. sim :

[0062] ;

[0063] Where “·” represents the vector dot product operation, and “||·||” represents the L2 norm of the vector;

[0064] For the calculated S sim Normalization is performed to limit the value range of S to [0,1]. sim The closer the value is to 1, the higher the semantic similarity between the two historical versions.

[0065] Based on the weighted calculation results of semantic similarity and logical dependency, the weight value of the edge connecting two corresponding historical version nodes is determined. The weight value is used to characterize the strength of the change association.

[0066] Logical dependencies are calculated using a weighted summation algorithm based on dependency parsing. The specific steps and formulas are as follows:

[0067] Logical dependencies are calculated using a weighted accumulation algorithm based on dependency parsing. The specific steps are as follows:

[0068] The Stanford Parser dependency parsing tool was used to analyze the Text. A Text B Perform syntactic analysis to extract core predicates from the text, such as "definition," "modification," and "supplement," as well as their associated objects, modifiers, and other components, and construct two versions of logical dependency trees;

[0069] Define a logical dependency weight table: If the core predicates of the two versions are the same and the overlap of related components is ≥80%, the base weight is set to 0.8; if the core predicates are different but the overlap of related components is ≥60%, the base weight is set to 0.5; if the core predicates are different and the overlap of related components is <60%, the base weight is set to 0.2.

[0070] Calculate the logical dependency L using the following formula. dep :

[0071] ;

[0072] Among them, the "number of core arguments" is identified from the text using keyword extraction tools, such as TF-IDF combined with TextRank. dep The value range is normalized to [0,1], and the closer it is to 1, the stronger the logical dependency between the two versions.

[0073] In this embodiment, the weighted calculation uses a fixed weight coefficient algorithm, and the specific formula and weight determination rules are as follows:

[0074] The edge weight W is calculated using a weighted algorithm with fixed weight coefficients. edge The specific formula is as follows:

[0075] ;

[0076] Where α is the weighting coefficient of semantic similarity, which was determined to be 0.6 through historical data verification, meaning that semantic similarity accounts for 60% and logical dependency accounts for 40%. W edge The value range is [0,1]. When the weight value is ≥0.7, it is determined to be a strongly related edge, and when the weight value is <0.3, it is determined to be a weakly related edge.

[0077] It should be further explained that when constructing the change impact network, the system first extracts all historical versions of the specified knowledge item from the knowledge base in chronological order to ensure the integrity of the historical evolution sequence. Then, it abstracts each independent historical version and all user feedback texts received after its release as a whole information unit into an independent node in the change impact network.

[0078] The system then calls the built-in natural language processing engine to perform deep semantic analysis on the text content of any two different historical versions. It quantifies the semantic similarity by calculating the cosine similarity in the word vector space or the semantic encoding similarity based on the pre-trained language model. At the same time, it evaluates the degree of logical dependence between versions by analyzing logical connectors, factual statement structures and argument dependencies in the text.

[0079] Finally, the system performs a linear weighted fusion of the calculated semantic similarity and logical dependency values, and the resulting comprehensive score is set as the weight value of the edge connecting the two historical version nodes. This weight value directly and quantitatively reflects the strength of the intrinsic correlation between different historical version changes.

[0080] Identifying at least one historical point that triggered the current conflict as the root cause anchor point specifically includes:

[0081] Match the content topics involved in the current conflicting user feedback with the content topics carried by each node in the network affected by the change;

[0082] Calculate the user feedback on the current conflict and the distance of each node in the opinion vector space;

[0083] The distance between viewpoint vectors is calculated using the Euclidean distance algorithm combined with Sentence-BERT encoding. The specific steps and formulas are as follows:

[0084] The Euclidean distance algorithm is used to calculate the distance between viewpoint vectors. The specific steps are as follows:

[0085] The Sentence-BERT model was used, with the pre-trained model being all-MiniLM-L6-v2. The feedback text for the current conflict was denoted as Feedback. C Changes that affect the text content of historical nodes in the network are denoted as Node. H Encode the viewpoint vectors (Vec) to generate the viewpoint vectors. C Vec H All dimensions are 384;

[0086] Calculate the Euclidean distance D using the following formula. dist :

[0087] ;

[0088] Among them, Vec C,i Vec H,i They represent Vec C Vec H The i-th element;

[0089] For D dist Normalization is performed, converting the values ​​to the [0,1] interval using a linear mapping. The distance threshold after normalization is set to 0.3. When the distance is <0.3, the viewpoints are considered highly similar.

[0090] Select nodes whose viewpoint vector distance is less than a set threshold and whose edge weight value is higher than a specific threshold, and mark them as conflict root source anchor points.

[0091] It should be further explained that when locating the root cause anchor point of the conflict, the system first uses natural language processing technology to extract the topic and semantically analyze the user feedback of the current conflict, identify its core content topic, and accurately match the topic with the historical content topics carried by all nodes in the change impact network. Subsequently, the system maps the text content of the current conflict feedback and the historical content of each node to the same high-dimensional vector space through a pre-trained text embedding model to generate corresponding opinion vectors, and calculates the Euclidean distance or cosine distance between the current feedback vector and the vectors of each historical node as a quantitative indicator of opinion difference.

[0092] Based on this, the system comprehensively considers the network topology and selects only those historical nodes whose vector distance to the current feedback viewpoint is less than the dynamic threshold set through historical data analysis, and whose weight values ​​of the connecting edges on the path directly related to the current conflict are higher than a specific threshold determined according to the overall network connection strength distribution. These nodes, which simultaneously satisfy high content relevance and strong structural correlation, are officially marked as the root anchor points of this conflict, thereby accurately identifying the historical evolution source that triggered the current controversy.

[0093] The simulation and consensus deduction of candidate solutions for updating knowledge items specifically includes:

[0094] The candidate solutions are input into the pre-trained user behavior prediction model;

[0095] The user behavior prediction model is based on historical interaction data in the change impact network to predict the possible set of user actions and the corresponding probability distribution for candidate solutions under different user attribute labels.

[0096] Furthermore, the specific structure and training parameter settings of the pre-trained user behavior prediction model in this embodiment are as follows:

[0097] The pre-trained user behavior prediction model adopts a Transformer-based multi-label classification prediction model, and its model structure specifically includes:

[0098] Input layer: Receives three types of feature vectors and concatenates them: semantic feature vector of knowledge item update candidate schemes, user attribute label vector, and feature vector of network topology impact from changes. The concatenation forms an input feature matrix with a dimension of 816. The semantic feature vector is generated by encoding through the BERT-base model and has a dimension of 768. The user attribute label vector includes user reputation level, historical feedback domain matching degree, etc., and has a dimension of 32. The feature vector of network topology impact from changes extracts node connection density, mean edge weight of target node, etc., and has a dimension of 16.

[0099] Encoding layer: A 3-layer Transformer encoder is set up, with 8 attention heads in each layer. The GELU activation function is used to extract deep semantic and related features from the input feature matrix.

[0100] Output layer: A multi-label classification layer using the Sigmoid activation function, outputting the probability distribution of three core operations for the user on the candidate solutions, namely "accept P" and "reject P". accept "Propose to modify P" modify "Oppose P" oppose ", and satisfy P accept +P modify +P oppose =1.

[0101] The specific parameters for the model training process are set as follows: the number of training iterations is 50-100 rounds, the Adam optimizer is used, the initial learning rate is set to 1e-5, and the learning rate is dynamically adjusted through a cosine annealing strategy; the loss function is the cross-entropy loss function, and the loss value is weighted in combination with the user's credit score. The weight of user behavior samples with a credit score ≥80 is set to 1.2, and the weight of samples with a credit score <60 is set to 0.6, in order to improve the impact of high-credibility samples on the model.

[0102] Based on the set of possible operations and probability distribution, calculate the expected conflict index for each candidate solution, and eliminate candidate solutions whose expected conflict index exceeds the acceptable range.

[0103] It should be further explained that during the consensus simulation phase, the system inputs one or more synthetic knowledge item update candidate schemes into the pre-trained user behavior prediction model. This model is a machine learning model obtained by supervised learning based on historical user interaction data accumulated in the change impact network. It establishes its predictive ability by learning the actual operation behavior patterns and feedback content of users with different attributes after the release of historical versions.

[0104] The model receives the content feature vector of the candidate solution, the attribute features of the target user group, and the topology of the network affected by the current change as input. Through its internal neural network calculation, it outputs the set of operation types that users under different user tags may perform on the candidate solution and their corresponding probability distribution. These operation types include, but are not limited to, directly accepting, proposing modification suggestions, or strongly opposing.

[0105] Based on these predictions, the system then calculates a quantitative expected conflict index for each candidate solution by weighting and summarizing the probability and intensity of negative actions by different user groups. Based on a preset acceptable threshold, the system automatically eliminates candidate solutions whose expected conflict index exceeds the threshold, thereby selecting solutions with higher consensus in the simulation environment to enter the subsequent verification stage.

[0106] The training data for the pre-trained user behavior prediction model includes:

[0107] The changes affect user behavior data related to historical versions, which is recorded in the network.

[0108] The user's historical feedback content and its corresponding reputation rating;

[0109] Semantic feature vectors of changes in the content of knowledge entries.

[0110] It should be further explained that the historical user interaction data used for model training is directly extracted from the change impact network. This data contains records of users' actual operational behaviors on historical knowledge item versions. These records are classified and encoded into discrete operation type labels. At the same time, the training data integrates the historical feedback text content of users associated with each operation behavior, as well as the reputation level identifier that the system dynamically calculates and maintains based on the quality of users' past contributions, the adoption of feedback, and community evaluation. This reputation level is a quantitative comprehensive score.

[0111] In addition, the content of each knowledge entry's historical version is transformed into a fixed-dimensional semantic feature vector through a pre-trained semantic model. This vector can capture the deep semantic information of the text. Finally, the model training set is composed of the above-mentioned cleaned and labeled multi-dimensional data, enabling the model to learn the complex mapping relationship between user attributes, content features, and historical behavior.

[0112] The limited user scope for publishing the verification plan within a specified user range is dynamically determined based on the user's accumulated reputation score in the network affected by the change.

[0113] It should be further explained that the core of the dynamic determination mechanism for limiting the scope of users lies in the system maintaining and updating in real time a reputation score system calculated based on the user's long-term behavioral contribution in the change-affected network. This reputation score is a comprehensive value, and its calculation is based on, but is not limited to, the final adoption rate of the user's historical feedback, the frequency with which the feedback provided is verified as correct by other users, and the number of times the feedback content is identified as a key node in subsequent change correlation analysis.

[0114] When a small-scale verification is required, the system does not select a fixed user group. Instead, it dynamically sets a reputation score threshold based on the content characteristics and conflict nature of the current verification plan. Only users with a reputation score higher than the threshold among the currently active users are automatically included in the scope of this verification. This ensures that the user group participating in the verification has the corresponding identification ability and credibility, providing high-quality data input for subsequent decision-making.

[0115] The adaptive adjustments to the verification plan based on the verification feedback data specifically include:

[0116] Analyze the semantic meaning of modification suggestions contained in the verification feedback data;

[0117] The proposed modifications are integrated with the scheme to be verified to generate at least one revised scheme variant.

[0118] The behavioral prediction model is used to simulate and extrapolate the variant schemes again, and the variant scheme with the highest expected consensus is selected as the basis for generating the final update instruction.

[0119] It should be further explained that when the system makes adaptive adjustments to the verification plan based on the verification feedback data, it first conducts in-depth semantic analysis on the collected verification feedback text to identify and extract the specific modification intentions and content addition / deletion suggestions contained therein. Subsequently, the system intelligently integrates these structured modification suggestions with the existing content of the verification plan. This integration process is not a simple replacement, but rather generates multiple plan variants that incorporate reasonable modifications while retaining the core framework of the original plan through semantic alignment and logical consistency verification.

[0120] Next, the system inputs these variant schemes back into the user behavior prediction model to conduct a new round of simulated consensus deduction, predicting the changes in user behavior and consensus that may be triggered after the release of each variant. Finally, the system compares the deduction results of each variant and selects the variant scheme with the highest expected consensus as the basis for determining the final knowledge item update instruction, thereby completing the closed-loop optimization based on real verification feedback.

[0121] The method also includes:

[0122] The new user feedback data, solution decision data, and final update results generated during this conflict coordination process are added as new nodes and edges to the change impact network to update the training dataset of the behavior prediction model.

[0123] It should be further explained that after the knowledge entry version is updated and the corresponding traceability record is generated, the system automatically starts a closed-loop process of data feedback and knowledge enhancement. All key data generated in the entire coordination process from conflict triggering to final decision, including newly received user feedback data, intermediate solutions and simulation data generated in the simulation consensus simulation stage, final adopted update instructions and verification feedback data, are structured and integrated into the change impact network as new knowledge.

[0124] Specifically, the system creates a new node by taking the final version of the knowledge entry and its associated feedback set from this update. Based on the semantic and logical connections between the node and existing nodes in the network, especially the identified conflict root anchors, it calculates and establishes new edge connections and weights. At the same time, all the interaction data accumulated in this process is added to the dataset used to train the user behavior prediction model, thereby enabling incremental updates and optimization of the model parameters. This allows the entire system to continuously learn from each conflict resolution practice, thereby continuously improving its accuracy and efficiency in handling similar conflicts in the future.

[0125] The source tracing records are stored in a structured data format and support reverse queries and visualization via node identifiers of conflict root cause anchors. Further explanation is needed: the system persistently stores the source tracing records generated by each update operation in a machine-readable structured data format. This format explicitly defines multiple fields, including conflict root cause anchor node identifiers, key input parameters and output results during the simulated consensus deduction process, a summary of verification feedback data features, and the final decision logic. The stored source tracing records, through the establishment of a strong correlation index with the corresponding knowledge entry version, support efficient reverse queries by users or external systems by inputting the node identifier of a specific conflict root cause anchor. The system can quickly retrieve all subsequent update records that are causally related to that historical node.

[0126] Meanwhile, the system provides a dedicated visual interactive interface. When it receives a node identifier query request, it can dynamically render a subgraph of the change impact network with the anchor point as the core in a graphical way, and clearly display the subsequent decision chain triggered by it, thereby transforming the abstract tracing logic into an intuitive visual analysis path.

[0127] A computer-readable storage medium having a computer program thereon that, when executed by a processor, implements a method for updating and tracing knowledge entries based on change correlation analysis and user feedback.

[0128] It should be further explained that when the computer program contained in the computer-readable storage medium is executed by the processor, the specific implementation includes all the method steps including constructing the change impact network, locating the root cause anchor of the conflict, synthesizing candidate solutions, conducting simulated consensus deduction, organizing small-scale verification, adjusting the solution based on the verification feedback, and generating traceability records.

[0129] The program performs text semantic analysis by calling a pre-trained natural language processing model, maintains and queries the change impact network using graph computing algorithms, runs a user behavior prediction model trained on historical data to perform simulations, and establishes a dynamic reputation evaluation system to screen and verify users. The execution of the program transforms general-purpose computing devices into dedicated devices for intelligent knowledge item conflict resolution and tracing, fully replicating the defined closed-loop processing mechanism based on historical correlation analysis and multi-stage verification.

[0130] This method constructs a change impact network and locates the root cause anchor points of conflicts, enabling root cause analysis of user feedback conflicts during knowledge item updates and overcoming decision-making deadlocks caused by simple aggregation rules. Utilizing simulated consensus deduction and multi-round verification mechanisms, the system can pre-assess the acceptability of update schemes and perform closed-loop optimization based on real feedback, thereby effectively improving the accuracy and stability of knowledge item update decisions.

[0131] By structurally recording the complete decision-making logic as traceability information and establishing a data feedback mechanism to continuously optimize the analysis model, this method not only enhances the traceability and transparency of knowledge change history, but also enables the system to have continuous learning capabilities, ultimately ensuring the quality, reliability, and long-term maintenance efficiency of the knowledge base content.

[0132] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0133] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for updating and tracing knowledge entries based on change correlation analysis and user feedback, characterized in that: Includes the following steps: S1: Receive multiple user feedbacks for the same knowledge item. When it is determined that there is a content conflict among the user feedbacks, initiate a conflict coordination process. Construct a change impact network for the knowledge item, wherein the change impact network consists of nodes and edges connecting the nodes. Nodes represent the historical versions of the knowledge item and the historical user feedback associated with each historical version, and edges represent the change association strength between different historical versions. S2: Based on the changes affecting the network and the currently received conflicting user feedback, locate at least one historical node that caused the current conflict as the root cause anchor point of the conflict; based on the root cause anchor point of the conflict and the content of the current conflicting user feedback, synthesize at least one knowledge item update candidate scheme. S3: Perform a simulation consensus deduction on the candidate solutions for updating the knowledge items. The simulation consensus deduction uses a pre-set behavior prediction model to simulate the expected feedback behavior of a specific user group after the candidate solutions are released, and selects the candidate solutions that meet the expected consensus requirements as the solutions to be verified based on the simulation results. S4: Publish the verification scheme to be verified within a limited user scope and collect the verification feedback data generated within that limited user scope; make adaptive adjustments to the verification scheme based on the verification feedback data, and generate the final knowledge item update instruction based on the adjustment results; S5: Execute the version update of the knowledge entry according to the final knowledge entry update instruction generated based on the adjustment results in step S4, and generate the corresponding traceability record. The traceability record includes at least the conflict root cause anchor information, key process data of the simulated consensus deduction, and a summary of the verification feedback data.

2. The knowledge entry updating and tracing method based on change correlation analysis and user feedback as described in claim 1, characterized in that: Constructing the network of impacts of changes to the knowledge entries specifically includes: Extract all historical versions of the knowledge entry, arranged in chronological order; Each historical version of the content and its corresponding set of user feedback are abstracted into a network node; Calculate the semantic similarity and logical dependency between any two historical versions of content using natural language processing techniques; Based on the weighted calculation results of semantic similarity and logical dependency, the weight value of the edge connecting two corresponding historical version nodes is determined, and the weight value is used to characterize the strength of the change association.

3. The knowledge item updating and tracing method based on change correlation analysis and user feedback as described in claim 2, characterized in that: The location of at least one historical node that triggered the current conflict as the root cause anchor point of the conflict specifically includes: Match the content topics involved in the current conflicting user feedback with the content topics carried by each node in the network affected by the change; Calculate the user feedback on the current conflict and the distance of each node in the opinion vector space; Nodes whose viewpoint vector distance is less than a set threshold and whose edge weight value is higher than a specific threshold are selected and marked as the conflict root source anchor points.

4. The knowledge item updating and tracing method based on change correlation analysis and user feedback as described in claim 1, characterized in that: The simulation consensus deduction of the candidate schemes for updating the knowledge items specifically includes: The candidate solutions are input into a pre-trained user behavior prediction model; The user behavior prediction model predicts the possible set of operations and the corresponding probability distribution of users under different user attribute tags for the candidate scheme based on historical interaction data in the change impact network. Based on the set of possible operations and probability distribution, calculate the expected conflict index for each candidate solution, and eliminate candidate solutions whose expected conflict index exceeds the acceptable range.

5. The knowledge entry updating and tracing method based on change correlation analysis and user feedback as described in claim 4, characterized in that: The training data for the pre-trained user behavior prediction model includes: The changes affect the user's actual operational behavior data regarding historical versions recorded in the network; The user's historical feedback content and its corresponding reputation rating; Semantic feature vectors of changes in the content of knowledge entries.

6. The knowledge entry updating and tracing method based on change correlation analysis and user feedback according to claim 1, characterized in that: The limited user scope for publishing the proposed verification scheme within a limited user range is dynamically determined based on the user's accumulated reputation score in the network affected by the change.

7. The knowledge entry updating and tracing method based on change correlation analysis and user feedback according to claim 1, characterized in that: The adaptive adjustment of the verification scheme based on the verification feedback data specifically includes: Analyze the semantics of modification suggestions contained in the verification feedback data; The proposed modifications are then integrated with the scheme to be verified to generate at least one adjusted scheme variant. The behavioral prediction model is used to simulate and extrapolate the variant scheme again, and the variant scheme with the highest expected consensus is selected as the basis for generating the final update instruction.

8. The knowledge entry updating and tracing method based on change correlation analysis and user feedback according to claim 1, characterized in that: The method further includes: The new user feedback data, solution decision data, and final update results generated during this conflict coordination process are added as new nodes and edges to the change impact network to update the training dataset of the behavior prediction model.

9. The knowledge entry updating and tracing method based on change correlation analysis and user feedback according to claim 1, characterized in that: The source tracing records are stored in a structured data format and support reverse querying and visualization through the node identifier of the conflict root source anchor point.

10. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the knowledge entry update and tracing method based on change association analysis and user feedback as described in any one of claims 1 to 9.

Images