C-end user demand intelligent matching article selection method based on multi-dimensional data

Abstract

The invention relates to a multi-dimensional data-based C-end user demand intelligent matching article selection method, which belongs to the technical field of computer data processing, and specifically comprises the following steps: S1, after user explicit authorization is obtained, obtaining user interaction data, and carrying out anonymization processing to generate decision state parameters; s2, performing normalization processing on the decision state parameters to generate standardized decision state parameters; s3, calculating a user trust degree; a decision space stability index is generated in combination with the user trust degree and the standardized decision state parameters; and S4, comparing the decision space stability index with a preset stability critical threshold to generate an intervention instruction, and adjusting the user interface based on the intervention instruction, thereby laying a solid foundation for subsequent accurate intervention.

Description

Technical Field

[0001] This invention relates to the field of computer data processing, specifically to a method for intelligently matching and selecting products based on the needs of C-end users using multi-dimensional data. Background Technology

[0002] In online shopping platforms targeting highly complex products, existing technologies rely on predicting users' purchase intentions based on their historical behavior, but this approach fails in certain scenarios. When the platform's selection architecture is highly guiding and key opinion leaders hold polarized views, a discrepancy arises between the user's cognitive construction process and decision-making convergence process. This discrepancy is amplified through a cascading effect of information overload, comparison conflict, and trust dissipation, easily triggering a topological collapse of the decision space in the mid-stage of decision-making. This manifests as the counterintuitive phenomenon that the more recommendations a user receives, the lower their decision-making efficiency, leading to the loss of potentially high-value users. Existing technologies have failed to effectively quantify and intervene in this deep-seated contradiction.

[0003] The purpose of this invention is to provide a method for intelligent matching of C-end user needs based on multi-dimensional data to solve the above problems. The core of this method is to construct a dynamic evaluation model for decision space stability and intelligently switch between two intervention modes based on the model output.

[0004] The information disclosed in the background section above is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to provide a method for intelligently matching product selection based on the needs of C-end users using multi-dimensional data, so as to solve the problems mentioned in the background art.

[0006] The technical solution of the present invention includes the following specific steps: S1. After obtaining explicit authorization from the user, acquire user interaction data and anonymize it to generate decision state parameters; S2. Normalize the decision state parameters to generate standardized decision state parameters; S3. Calculate user trust level; and combine user trust level with standardized decision state parameters to generate decision space stability index; S4. Compare the decision space stability index with the preset stability threshold to generate intervention instructions, and adjust the user interface based on the intervention instructions.

[0007] Preferably, in S1, the generated decision state parameters include: Based on the user's interaction behavior sequence, a pre-set hidden Markov model is used to calculate the cognitive script execution entropy; Based on the user's product candidate set, a hypergraph is constructed to calculate the Laplace energy of the candidate set hypergraph; The information cocoon penetration rate is calculated based on user information source data.

[0008] Preferably, in S2, the generation of standardized decision state parameters includes: Divide the cognitive script execution entropy by the preset theoretical maximum entropy value to obtain the normalized cognitive entropy; Divide the Laplace energy of the candidate set of hypergraphs by a preset reference energy value to obtain the normalized hypergraph energy; Standardized decision-making state parameters include normalized cognitive entropy, normalized hypergraph energy, and information cocoon penetration rate.

[0009] Preferably, S3 includes: Collect the number of invalid user interactions and calculate the user's trust level based on the preset initial trust level and trust decay rate; By combining user trust levels, standardized decision state parameters, and preset weighting coefficients, a decision space stability index is generated.

[0010] Preferably, in S4, generating intervention instructions includes: When the decision space stability index is higher than the critical stability threshold, a decision convergence auxiliary instruction is generated. When the stability index of the decision space is lower than or equal to the stability critical threshold, cognitive structure auxiliary instructions are generated.

[0011] Preferably, adjusting the user interface based on decision convergence assist instructions includes: Highlight the key differences between the alternatives; Provides detailed one-on-one parameter comparison; Remove the entry point for exploratory features.

[0012] Preferably, the user interface is adjusted based on cognitive structure-assisted instructions, including: Simplify the filter; Guide users to compare data in groups; Incorporate neutral expert evaluation content; Increase the transparency of the reasons for the recommendation.

[0013] This invention provides an improved method for intelligently matching and selecting products based on multi-dimensional data to meet the needs of end-users. Compared with existing technologies, it has the following improvements and advantages: 1. After obtaining explicit authorization from the user, this method acquires and anonymizes the data, generating a set of decision-making state parameters to create a comprehensive characterization of the user's decision-making process. These parameters include: cognitive script execution entropy, used to quantify the uncertainty of the user's cognition, which measures the clarity or confusion of the user's behavioral patterns; the Laplace energy of the candidate set hypergraph, used to measure the degree of conflict within the user's set of alternative products, which quantifies the intensity of the contradictions the user faces when weighing product attributes; and the information cocoon penetration rate, used to assess the diversity of the user's information acquisition channels. These three parameters, established from the micro-cognitive, meso-comparative, and macro-information levels respectively, enable the system to transcend the limitations of traditional analysis, accurately identifying whether the user is trapped in cognitive disorientation, comparative indecision, or information bias, laying a solid foundation for subsequent precise intervention. 2. This method introduces user trust as a dynamic variable into the evaluation system. By collecting the number of invalid user interactions and calculating user trust based on a pre-set decay model, the evaluation results reflect the degree of user reliance on the platform. Combining user trust, standardized decision state parameters, and pre-set weighting coefficients, a comprehensive decision space stability index is generated. This index uses user trust as the evaluation basis, cognitive confusion and comparative conflict as risk items, and information diversity as a buffer item. It logically integrates multiple core elements affecting decision-making, and its single quantitative output provides a sensitive and reliable basis for subsequent intervention. 3. The core advancement of this method lies in establishing a closed-loop intervention mechanism encompassing state awareness, risk assessment, strategy switching, and interface adaptation. By comparing the decision space stability index with a critical stability threshold determined through historical data analysis, the system can accurately determine whether a user's decision is in a stable convergence state or a confused state on the verge of instability, and automatically generate different intervention instructions. When the system determines that the user's decision-making process is stable, it generates decision convergence assistance instructions and assists the user in focusing and making a final choice by highlighting key differences between alternative solutions, providing one-to-one detailed parameter comparisons, and removing exploratory function entry points. Conversely, when the system determines that the user's decision-making process is at risk of instability, it generates cognitive structure assistance instructions. The purpose of these instructions is not to rush the decision-making process, but rather to help the user rebuild a clear cognitive framework through a series of interface adjustments, such as simplifying the filter to reduce cognitive complexity, guiding users to conduct group comparisons to resolve comparison conflicts, introducing neutral expert evaluation content to break down information barriers, and increasing the transparency of the reasons for recommendation to rebuild user trust. Attached Figure Description

[0014] The present invention will be further explained below with reference to the accompanying drawings and embodiments: Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0016] It should be noted that the acquisition and use of all user interaction data involved in this invention strictly comply with national laws and regulations regarding personal information protection and data security. Before performing any data collection step, this method will provide users with a clear privacy policy, and subsequent data processing steps can only be carried out after obtaining the user's explicit consent and authorization. In addition, all collected raw data will be immediately anonymized or pseudonymous to ensure that the data does not contain any personally identifiable information, thereby technically protecting user privacy.

[0017] Example 1 Please see Figure 1 This invention provides a method for intelligently matching and selecting products based on the needs of C-end users using multi-dimensional data. The specific steps include: S1. After obtaining explicit authorization from the user, acquire user interaction data and anonymize it to generate decision state parameters; S2. Normalize the decision state parameters to generate standardized decision state parameters; S3. Calculate user trust level; and combine user trust level with standardized decision state parameters to generate decision space stability index; S4. Compare the decision space stability index with the preset stability threshold to generate intervention instructions, and adjust the user interface based on the intervention instructions. This invention provides a method for intelligent matching and product selection based on C-end user needs using multi-dimensional data. This method, by constructing a dynamic closed-loop control system, achieves a shift from predicting user intent to maintaining a healthy decision-making process. The method includes the following steps: Step S1, acquiring and processing user interaction data to generate decision state parameters; user interaction data refers to the original sequence of user behavior on the interface. The anonymized original behavior sequence, such as clicks, scrolling, filter usage, and dwell time, has had all identifiable personal information removed; this data is processed to generate a set of decision state parameters; the decision state parameters are a series of... This indicator is used to quantitatively describe a user's current decision-making psychological state from multiple dimensions. Its purpose is to comprehensively characterize the user's cognitive state, degree of conflict, and information environment during the decision-making process. Step S2 involves normalizing the decision-making state parameters to generate standardized decision-making state parameters. The purpose of this step is to eliminate the differences in dimensions and numerical ranges among the decision-making state parameters generated in S1, ensuring that they can be unbiasedly integrated in subsequent mathematical models. The standardized decision-making state parameters obtained after processing are dimensionless and constrained to similar scales. Step S3 calculates the user's trust level and combines it with the standardized decision-making state parameters to generate a decision space stability index. Trust level refers to the degree of trust users place in the current product selection platform, modeled as a dynamic variable that decays with negative interactions. The decision space stability index is the core evaluation output of this method; it is a comprehensive index designed to assess the health and stability of the user's decision-making process in real time, providing a quantitative basis for subsequent interventions. Step S4 compares the decision space stability index with a preset stability threshold to generate intervention instructions, and adjusts the user interface based on these instructions. The preset stability threshold is a pre-defined judgment benchmark, determined based on statistical analysis of historical user successful and failed decision-making session data, used to distinguish whether the decision-making process is in a stable convergence state or on the verge of instability. The user is in a state of confusion. Comparison results generate corresponding intervention commands, which trigger adaptive adjustments to the user interface, thus influencing the user's decision-making process and forming a closed loop of state awareness, risk assessment, strategy switching, and interface adaptation. Through these steps, this method can dynamically assess the stability of the user's decision space and intelligently switch intervention strategies based on the assessment results. It provides cognitive structure assistance when the user is stuck in a decision-making dilemma and decision convergence assistance when the user is close to the decision endpoint. The effect of this technology is that it effectively avoids the reduction in decision-making efficiency and even user churn caused by information overload, comparison conflicts, and trust dissipation, significantly improving the user experience and decision success rate in online purchasing scenarios for highly complex products.

[0018] In S1, the parameters for generating the decision state include: Based on the user's interaction behavior sequence, a pre-set hidden Markov model is used to calculate the cognitive script execution entropy; Based on the user's product candidate set, a hypergraph is constructed to calculate the Laplace energy of the candidate set hypergraph; Calculate the information cocoon penetration rate based on user information source data; In this embodiment, the specific method for generating decision state parameters in step S1 is further defined; for more accurate evaluation, the decision state parameters consist of three key indicators at different levels, namely, cognitive script execution entropy, candidate set hypergraph Laplace energy, and information cocoon penetration rate; cognitive script execution entropy is denoted as... This is a micro-parameter used to quantify cognitive uncertainty when users perform psychological processes such as information gathering and comparison. Its function is to measure whether the user is in a state of exploration with a clear goal or in a state of being lost and confused. In this embodiment, the calculation is based on a preset Hidden Markov Model. The working principle of this model is to predefine a set of implicit cognitive states related to the product selection task, such as exploring, comparing, confirming, and being lost. The model will take the real-time interactive behavior sequence generated by the user on the interface as the observation sequence and infer the most likely implicit cognitive state sequence and its transition probability based on this sequence. To implement this model, the steps are as follows: Define a set of discrete, observable user behaviors as observation states, where each observation state is a quantized mapping of the user's original behaviors; for example, the following observation states can be defined: Quick browsing: After a user clicks to enter the product details page, the time spent on the page is less than 3 seconds; Detailed review: Users spend more than 10 seconds on the product details page, and the page scrolls more than 70%; Generating interest: Users add products to their shopping cart or favorites; Explicit comparison: Users use the platform's sorting, filtering, or product comparison functions; Return to list: The user returns to the product list page from the product details page; By using historical user session data annotated by experts, whose implicit cognitive states at each time point (e.g., exploring, comparing, etc.) are labeled according to behavioral sequences, the model is trained using the standard Baum-Welch algorithm to determine the various probabilistic parameters of the Hidden Markov Model (HMM), including the initial state probability vector, state transition probability matrix, observation probability matrix, and emission matrix. This training process ensures that the model parameters reflect the behavioral patterns of real users during the product selection process, thereby ensuring the accuracy of subsequent entropy calculations. Specifically, the HMM is defined by three sets of parameters: the initial state probability vector... The state transition probability matrix A and the observation probability matrix B are denoted as follows: Assume there is a set of N implicit cognitive states in the model. and a set of M observable user behaviors. The specific definitions of the model parameters are as follows: Initial state probability vector : An N-dimensional vector, where elements This indicates that the system is in state at initial time t. The probability, i.e. ,in ; State transition probability matrix A: a A matrix, where elements This indicates that the system is in state t at time t. Under the conditions, Shifting to a new state The probability, i.e. ,in ; Observation probability matrix B: a A matrix, where elements This indicates that the system is in state t at time t. Under these conditions, user behavior was observed. The probability, i.e. ,in Here, k is the index for a specific user behavior; The Baum-Welch algorithm, as an expectation-maximization algorithm, utilizes expert-annotated observation sequences and corresponding hidden state sequences to solve for the probability of occurrence of events in the training data through iterative re-estimation. Largest model parameters ; This is the Shannon entropy, calculated as the probability distribution of transitions between these implicit cognitive states within a specific time window; a higher entropy value indicates more chaotic user behavior patterns and greater cognitive uncertainty; the Laplace energy of the candidate set hypergraph is denoted as... , is a mesoscopic parameter used to measure the degree of conflict within a user's current product candidate set; its function is to quantify the trade-off dilemma faced by users when comparing products; in this embodiment, the calculation method is to construct a hypergraph based on the user's product candidate set; in this hypergraph, each product item is a node, and each product attribute dimension that the user relies on when making comparisons, such as price, brand, and performance parameters, constitutes a hyperedge connecting related products. User behavior that relies on a certain product attribute dimension is clearly defined as follows: the user has used the attribute to perform at least one explicit filtering or sorting operation in the session, or the mouse pointer hovers over the attribute area for more than 2 seconds in the side-by-side comparison view; if any of the above conditions are met, the attribute is regarded as a hyperedge, connecting all products in the current candidate set that have the attribute. Hypergraph Laplacian Matrix The calculation method is as follows: Constructing the association matrix of the hypergraph ,element At the node Belongs to superedge The value is 1 if it is true, and 0 otherwise. Calculate the node degree matrix and hypermarginality matrix They are all diagonal matrices, with the diagonal elements representing the degree of a node, the number of hyperedges the node belongs to, the degree of the hyperedge, and the number of nodes contained in the hyperedge. Hypergraph Laplacian Matrix Defined as: ; Where L represents the hypergraph Laplacian matrix; : Represents the node degree matrix, whose diagonal elements are the degree of the nodes, i.e. how many hyperedges the node belongs to; H: Represents the hypergraph incidence matrix. When node v belongs to hyperedge e, the element H(v,e) in the matrix is ​​1, otherwise it is 0; : Represents the hyperedge degree matrix, which is a diagonal matrix whose diagonal elements are the degree of the corresponding hyperedge, that is, the number of nodes contained in the hyperedge; : Represents the transpose of the incidence matrix H; Alternative Set: Supergraph Laplace Energy That is, the matrix The sum of the absolute values ​​of all eigenvalues; It is calculated as the sum of the absolute values ​​of all eigenvalues ​​of the Laplacian matrix of the hypergraph; the higher the energy value, the more differences and conflicts there are between products in the candidate set, and the more intense the comparison conflicts that users need to deal with; information cocoon penetration rate, denoted as It is a macro-level parameter used to assess the diversity of information sources received by users; its function is to assess whether users are trapped in an information cocoon state with a single information acquisition channel; in this embodiment, the calculation is based on the analysis of the user's information source data, for example, by statistically analyzing the ratio of the number of information source categories browsed by the user in a unit of time to the preset total number of categories; In this embodiment, the categories of information sources are explicitly preset to the following five types: Platform official product description: Official specifications and graphic descriptions on the product details page; User reviews within the platform: User comments and Q&A sections at the bottom of the product page; Third-party professional reviews: Articles or videos from neutral review organizations embedded in or linked to on the platform; Community / Forum Content: Discussion posts about the product in user communities or forums embedded in or linked to on the platform; Video content: Unboxing and user experience videos embedded or linked to on the platform; Therefore, the preset total number of categories is 5, and the unit of time is defined as the entire time period from when the user enters the current product selection session until the current moment; for example, if the user only browses the official description and user reviews in this session, the number of categories browsed is 2, and the information cocoon penetration rate is... Calculated as ; The value of this indicator is normalized to the [0,1] range, with lower values ​​indicating a stronger information cocoon effect. The advantage of this technique lies in its ability to provide a more comprehensive, multi-dimensional, and profound characterization of users' decision-making states by introducing these three parameters at the micro, meso, and macro levels, surpassing traditional single-dimensional analysis relying solely on historical behavioral data. This multi-dimensional diagnosis lays the foundation for subsequent precise and effective intervention, enabling the central processing module to identify the root cause of users' decision-making difficulties—whether it stems from cognitive dissonance or other factors. I'm quite conflicted, high Or is it a case of information bias, low... ; This method achieves in-depth, multi-dimensional diagnosis of users' decision-making states. Existing technologies typically rely on single-dimensional historical behavioral data for analysis, failing to provide a deep understanding of users' current psychological state. This method acquires user interaction data and generates a set of decision-making state parameters, thus completing a three-dimensional characterization of the user's decision-making process. These parameters include: cognitive script execution entropy, which quantifies the clarity or confusion of user behavior patterns; candidate set hypergraph Laplace energy, which measures the degree of conflict within the user's set of alternative products, quantifying the intensity of contradictions faced by users when weighing product attributes; and information cocoon penetration rate, which assesses the diversity of users' information acquisition channels. These three parameters, established from the micro-cognitive, meso-comparative, and macro-information levels respectively, enable the system to transcend the limitations of traditional analysis, accurately identifying whether users are trapped in cognitive disorientation, comparative indecision, or information bias, laying a solid foundation for subsequent precise intervention.

[0019] In S2, the standardized decision state parameters are generated as follows: Divide the cognitive script execution entropy by the preset theoretical maximum entropy value to obtain the normalized cognitive entropy; Divide the Laplace energy of the candidate set of hypergraphs by a preset reference energy value to obtain the normalized hypergraph energy; Standardized decision-making state parameters include normalized cognitive entropy, normalized hypergraph energy, and information cocoon penetration rate; In this embodiment, the specific method for generating standardized decision state parameters in step S2 is further defined; this step aims to convert the decision state parameters with different dimensions and numerical ranges calculated in the preceding steps into dimensionless indices under a unified scale; the generation of normalized cognitive entropy is to convert the cognitive script execution entropy Divide by the preset theoretical maximum entropy value The processed result is the normalized cognitive entropy. The preset theoretical maximum entropy value It is a reference benchmark, its function is to provide a scale for the original entropy value; in this embodiment, for those with A hidden Markov model that implicitly represents cognitive states. Possible values This represents the greatest uncertainty in theory; ; The generation of normalized hypergraph energy is achieved by converting the Laplace energy of the candidate set of hypergraphs. Divide by the preset reference energy value The processed result is the normalized hypergraph energy. Preset reference energy value This benchmark, obtained through empirical statistics, reflects the relative position of the conflict level of the current candidate set among all historical user behaviors. In this embodiment, it is derived from backtracking calculations on massive amounts of historical user data, taking all calculated values. A specific high quantile of the value, such as the 95th quantile, is used as a reference. ; ; The standardized decision state parameters were determined to include normalized cognitive entropy. Normalized hypergraph energy and the penetration rate of information cocoons ;because It has been normalized to the [0,1] interval in its definition, so no additional processing is required; the effect of this technique is that, through normalization, it ensures that all feature components input into the decision space stability evaluation model are on a similar, dimensionless scale; this avoids model calculation bias caused by huge differences in the original parameter values, making the weight coefficients of each parameter in the subsequent model have a fairer and more robust explanatory power, and improving the accuracy and reliability of the entire evaluation system; A more objective and reliable decision-making risk assessment model was constructed. To ensure unbiased integration of decision-making state parameters from different sources within a unified framework, this method normalized each parameter. The normalized cognitive entropy was obtained by dividing the cognitive script execution entropy by the theoretical maximum entropy value, and the normalized hypergraph energy was obtained by dividing the candidate set hypergraph Laplace energy by the empirically derived reference energy value. This process eliminated the inherent differences in the dimensions and numerical ranges of the parameters, ensuring the accuracy of subsequent model calculations. This method introduced the dynamic variable of user trust into the evaluation system. By collecting the number of invalid user interactions and calculating user trust based on a preset decay model, the evaluation results reflect the degree of user trust in the platform. Combining user trust, standardized decision-making state parameters, and preset weighting coefficients, a comprehensive decision space stability index was generated. This index uses user trust as the evaluation basis, cognitive confusion and comparative conflict as risk items, and information diversity as a buffer item, logically integrating multiple core elements affecting decision-making. Its single quantitative output provides a sensitive and reliable basis for subsequent intervention.

[0020] S3 includes: Collect the number of invalid user interactions and calculate the user's trust level based on the preset initial trust level and trust decay rate; By combining user trust levels, standardized decision state parameters, and preset weighting coefficients, a decision space stability index is generated. In this embodiment, the specific implementation of step S3 is further defined; the core of this step is to construct a mathematical model that can integrate multi-dimensional information and dynamically assess decision-making risks; to perform this assessment, user trust is calculated; in this embodiment, user trust... Estimation is performed using a decay model based on the number of invalid interactions collected. Number of invalid interactions This is a count of user interactions that do not receive positive feedback after being recorded in real time by the user behavior monitoring module, such as clicking invalid links or setting filter conditions that yield no results; the calculation formula is: ; In this formula, : Represents user trust level, This is the preset initial trust level, representing the basic trust level of users when they enter the platform. It is a constant in the range [0,1], such as 1.0. It is the preset trust decay rate, a positive real number, which controls the rate at which trust decreases as the number of invalid interactions increases; and The values ​​can all be optimized and determined through statistical fitting of large-scale historical user session data; This is the real-time accumulated count of invalid interactions; e: represents the base of the natural logarithm; Invalid interaction count The counting range is further explicitly limited to the cumulative number of the following actions: Clicking a link that either does not respond or redirects to an error page; Filtering or searching operations that return zero results upon submission; The action of quickly opening and going back or entering and exiting the same product details page within 3 seconds; Repeatedly clicking the same invalid button without any change in state; parameter and The statistical fitting process is as follows: A large number of historical user sessions are collected, and each session is labeled with whether a purchase was ultimately completed (successful label) or abandoned midway (failure label), and the number of invalid interactions during the session is recorded. Using session abandonment rate as a proxy indicator of decreased trust, a nonlinear least squares method is used to analyze the change in session abandonment rate with respect to... The optimal initial confidence level is obtained by fitting the changing curve. and trust decay rate ; Based on the calculation results of user trust, combined with standardized decision state parameters ( (and a set of preset weighting coefficients) to generate a decision space stability index. The calculation model is as follows: ; In this formula, The current trust level is calculated from the trust decay model. , These are the standardized decision state parameters obtained from the preceding steps; It is a preset non-negative coefficient used to adjust the positive contribution of information cocoon penetration rate to stability; and These are pre-defined dimensionless weight coefficients, which represent the relative importance of normalized cognitive entropy and normalized hypergraph energy in the evaluation model, respectively. They also represent the relative weights of the contributions of normalized cognitive entropy and normalized hypergraph energy to the risk of decision failure. Their values ​​are optimized and determined by training a machine learning model on historical user decision data. The values ​​of these coefficients can be optimized and determined by training a machine learning model on historical datasets. The training process is as follows: Construct a training dataset. From a massive amount of historical user sessions, extract the decision state parameter triples for each session before its end. As a feature; at the same time, based on whether the session ultimately resulted in a successful order, it was assigned a binary classification label: 1 represents a successful decision, and 0 represents a failed decision and user churn; Secondly, logistic regression was chosen as the training model; the goal of this model is to predict the probability of a user making a successful decision given decision state parameters. ; ; By training on a labeled dataset, the optimal weight coefficients of the model are obtained with the goal of minimizing the binary cross-entropy loss function. These learned weighting coefficients constitute the weighting coefficients in the formula of this invention. The quantitative determination of the coefficients ensures the effectiveness and reproducibility of the model; e: represents the base of the natural logarithm; Coefficients obtained during training It reflects the marginal effect of each parameter on the probability of decision success; and provides a more interpretable stability index for constructing a more robust structure. These coefficients can be converted; for example, setting , , And a baseline constant is introduced for scaling to ensure that each weight is within the acceptable range. The model should play a role commensurate with its risk / gain contribution; the specific transformation function and scaling can be optimized on the validation set. The distinguishing performance of the indicators is used to determine this; The advantage of this technology lies in the fact that this implementation provides a rationally structured and computable decision space stability assessment model. This model uses user trust as a foundation, cognitive entropy and comparative energy as risk terms, and information diversity as a buffer term, logically reflecting the interaction of multiple core elements influencing user decisions. Its output is a single index... This provides a reliable, sensitive, and comprehensive quantitative basis for subsequent intervention decisions.

[0021] Example 2 In S4, the generation of intervention instructions includes: When the decision space stability index is higher than the critical stability threshold, a decision convergence auxiliary instruction is generated. When the decision space stability index is lower than or equal to the stability critical threshold, cognitive structure auxiliary instructions are generated. In this embodiment, the specific logic for generating the intervention instruction in step S4 is further defined; this logic is based on the decision space stability index. With stability critical threshold The comparison results are used to decide which intervention mode to adopt; the stability critical threshold These are key preset parameters that distinguish between stable and unstable states in the decision-making process. Their values ​​can be determined based on the analysis of extensive historical user behavior data; for example, they can be set to differentiate between users who successfully place an order and those who abandon the process midway. The optimal split point of the value; The optimal split point was scientifically determined through standard receiver operating characteristic curve analysis, using a labeled validation dataset to calculate the decision space stability index for each session in the dataset. By traversing all possible Using the value as a candidate threshold, the true positive rate, the proportion of sessions with successful decisions correctly identified, and the false positive rate are calculated at each threshold. The proportion of sessions with failed decisions incorrectly identified as successful is then plotted as an ROC curve. The point on the curve that maximizes the Youden index: J = sensitivity + specificity - 1 is selected. The value is then determined as the stability critical threshold. This method ensures the objectivity and optimality of threshold selection. When the decision space stability index is calculated in real time Above this stability critical threshold When the system determines that the user's decision-making process is stable, orderly, and converging towards the goal of completing the selection, it generates a decision convergence assistance instruction. This instruction aims to help the user eliminate final distractions and focus on the final choice to complete the decision. Conversely, when the decision space stability index... Below or equal to the stability critical threshold When the system determines that the user's decision-making process is at high risk of instability, and the user may be in a state of confusion, hesitation, or distrust, the system will generate cognitive structure assistance instructions. The core goal of these instructions is not to urge the user to make a decision, but to help the user reduce cognitive load, rebuild a clear cognitive framework, and get their decision-making process back on track. When the system generates cognitive structure assistive instructions, it does not randomly execute all adjustments, but rather based on the factors that lead to... The specific sources of risk leading to a decline in the indicator are addressed using a rule-based priority decision-making logic: Rule 1: If normalized cognitive entropy It is a major factor leading to instability, for example, The system will prioritize implementing simplified filters and increasing the transparency of the reasons for recommendations in order to directly address users' cognitive confusion and distrust. Rule 2: If the hypergraph energy is normalized It is a major factor, for example, The system will prioritize guiding users to perform group comparisons in order to resolve severe conflicts within the candidate sets; Rule 3: If the information cocoon penetration rate It is a major factor, for example, The system will prioritize the execution of neutral expert evaluation content in order to break down information barriers. If multiple parameters simultaneously exceed their normal range, the system can combine various intervention measures, or select the intervention measure targeting the most serious problem based on the magnitude of the parameter's deviation from its healthy state. This refined decision-making logic ensures the targeted nature and effectiveness of the intervention. The effectiveness of this technology lies in the fact that by setting clear decision-switching thresholds, this method achieves intelligent and contextualized intervention strategies; it avoids using a single, unchanging assistance method, but can accurately identify whether the user needs a push or a helping hand, thereby providing just the right amount of help; this adaptive intervention mode greatly improves the efficiency and humanization of human-computer interaction, and can address the core pain points at different decision-making stages in a targeted manner. This approach achieves intelligent and contextualized adaptive intervention strategies. Its core advancement lies in establishing a closed-loop intervention mechanism encompassing state perception, risk assessment, strategy switching, and interface adaptation. By comparing the decision space stability index with a critical stability threshold determined through historical data analysis, the system can accurately determine whether a user's decision is in a stable convergence state or a confused state on the verge of instability, and automatically generate different intervention instructions. When the system determines that the user's decision-making process is stable, it generates decision convergence assistance instructions and helps the user focus and make a final choice through interface adjustments such as highlighting key differences between alternative solutions, providing one-to-one detailed parameter comparisons, and removing exploratory function entry points. Conversely, when the system determines that the user's decision-making process is at risk of instability, it generates cognitive structure assistance instructions. The purpose is not to hasten decision-making, but rather to help the user rebuild a clear cognitive framework through a series of interface adjustments, including simplifying the filter to reduce cognitive complexity, guiding users to perform group comparisons to resolve comparison conflicts, introducing neutral expert evaluations to break down information barriers, and increasing transparent explanations of recommendation reasons to rebuild user trust.

[0022] Adjust the user interface based on decision convergence assist instructions, including: Highlight the key differences between the alternatives; Provides detailed one-on-one parameter comparison; Remove the entry point for exploratory features; In this embodiment, the method of adjusting the user interface based on decision convergence assistance instructions is refined. When the system determines that the user is in the decision convergence stage, it will perform one or more of the following interface adjustments to assist the user in making the final choice: Highlighting the key differences between alternatives is one interface adjustment method; the system will automatically analyze the last few options in the user's alternative set, identify the most decisive and significantly different product attributes among them, and visually emphasize them on the interface through highlighting, enlarging, or labeling to guide the user to focus on the core trade-offs; Providing a one-to-one detailed parameter comparison is another interface adjustment method; the system will generate a dedicated and concise comparison view, displaying the two alternatives specified by the user side by side, and only listing the detailed parameters. Detailed parameter specifications facilitate precise, item-by-item comparisons by users, eliminating information interference. Removing exploratory function entry points is also an effective convergence aid. To prevent users from getting sidetracked and reverting to divergent information exploration in the final decision-making stage, the system temporarily hides or weakens entry points on the interface that might lead to exploratory behavior, such as "Discover More," "Related Recommendations," or links to other product categories. The effect of this technology is that the aforementioned interface adjustments work synergistically to serve the core goal of decision convergence. By focusing information, simplifying the view, and reducing interference, they effectively reduce cognitive friction and choice anxiety at the final decision-making stage, helping users make final decisions faster and more confidently, thereby directly improving the platform's conversion rate.

[0023] User interface adjustments based on cognitive structure-based assistive instructions include: Simplify the filter; Guide users to compare data in groups; Incorporate neutral expert evaluation content; Increase transparency in explaining the reasons for the recommendation; In this embodiment, the method of adjusting the user interface based on cognitive structure-assisted instructions is refined. When the system determines that the user is in a cognitive dilemma, it will perform one or more of the following interface adjustments, the core of which is to repair the decision-making environment and reduce the user's cognitive load. The simplified filter is a type of filter designed for high cognitive script execution entropy. The system adjusts its settings accordingly. When it detects chaotic user behavior, it temporarily hides some advanced or complex filtering conditions, retaining only the most core and basic filters. This helps users rebuild a clear filtering logic and reduces the complexity of cognitive processing. It should be noted that the "most core and basic filters" mentioned above are determined based on statistical analysis of historical user successful transaction session data. Specifically, the criteria are: calculating and ranking the total frequency of use of each filtering condition in all user sessions that ultimately successfully completed a purchase; defining the filtering conditions with the highest frequency of use (ranked Q) as the "most core and basic filters," where Q is a system preset parameter. Guiding users to perform group comparisons is a method targeting the Laplace energy of high-alternative hypergraphs. The system adjusts its approach; when there is too much conflict within a user's selection set, it avoids displaying all options flatly. Instead, it intelligently divides them into groups and guides the user to compare products within each group first, and then make a final comparison among the winners, thus resolving the complex comparison task; introducing neutral expert reviews is a way to combat information cocoons and reduce... Adjustments are made to mitigate the impact of polarized viewpoints; the system proactively pushes content from neutral professional evaluation organizations onto the interface, providing users with a balanced and authoritative perspective to offset the decision-making risks associated with biased information; increasing the transparency of explanations for recommendations is aimed at rebuilding user trust. The system adjusts its interface to provide clear and understandable explanations for any recommendations or suggestions. It explains which preferences or behaviors the recommendation is based on, for example, recommending this model because the user prioritizes photography performance. This increased transparency helps restore and enhance user trust. The effectiveness of this technology lies in the fact that these interface adjustments constitute a set of interventions aimed at restoring the stability of the user's decision-making process. By addressing multiple dimensions such as reducing cognitive confusion, resolving comparative conflicts, breaking down information barriers, and rebuilding user trust, the system precisely intervenes in the sources of instability identified by the decision-making model. This series of coordinated adjustments effectively pulls users back from a near-collapsed decision-making space, helping them rebuild their cognitive structure and thus solving the problem of decision-making failure due to cognitive overload. This invention solves the problem of user churn caused by the inability to understand and guide users' decision-making dilemmas in complex product selection scenarios when existing technologies quantify user decision-making status, establish a comprehensive risk assessment model, and realize contextualized adaptive intervention. It effectively avoids the reduction in decision-making efficiency caused by information overload, comparison conflict, and trust dissipation, and significantly improves user experience and decision success rate.

[0024] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1.A method for intelligent matching of product selection based on C-end user demand in multi-dimensional data, characterized in that, The specific steps include: S1, after obtaining explicit authorization of the user, acquiring user interaction data and performing anonymization processing to generate decision state parameters; S2, normalizing the decision state parameters to generate standardized decision state parameters; S3, calculating the user trust degree; and combining the user trust degree with the standardized decision state parameters to generate a decision space stability index; S4, comparing the decision space stability index with a preset stability critical threshold to generate an intervention instruction, and adjusting the user interface based on the intervention instruction. 2.The method of claim 1, wherein, In S1, generating the decision state parameters includes: Based on the user's interaction behavior sequence, a preset hidden Markov model is used to calculate the cognitive script execution entropy; Based on the user's product candidate set, a hypergraph is constructed to calculate the candidate set hypergraph Laplace energy; Based on the user's information source data, the information cocoon house penetration rate is calculated. 3.The method of claim 1, wherein, In S2, generating the standardized decision state parameters includes: Divide the cognitive script execution entropy by the preset theoretical maximum entropy value to obtain the normalized cognitive entropy; Divide the candidate set hypergraph Laplace energy by the preset reference energy value to obtain the normalized hypergraph energy; The standardized decision state parameters include the normalized cognitive entropy, the normalized hypergraph energy, and the information cocoon house penetration rate. 4.The method of claim 1, wherein, S3 includes: Collect the number of invalid interactions of the user, and calculate the user trust degree according to the preset initial trust degree and trust decay rate; Combine the user trust degree, the standardized decision state parameters, and the preset weight coefficient to generate the decision space stability index. 5.The method of claim 1, wherein, In S4, generating the intervention instruction includes: When the decision space stability index is higher than the stability critical threshold, generate a decision convergence auxiliary instruction; When the decision space stability index is lower than or equal to the stability critical threshold, generate a cognitive structure auxiliary instruction. 6.The method of claim 5, wherein, Adjusting the user interface based on the decision convergence auxiliary instruction includes: Highlighting the key differences between the alternative solutions; Providing one-to-one detailed parameter comparison; Removing the exploratory function entry. 7.The method of claim 5, wherein, Adjusting the user interface based on the cognitive structure auxiliary instruction includes: Simplify the filter; Guide the user to compare in groups; Introduce neutral expert evaluation content; Increase the transparency of the explanation of the recommended reasons.

Images