A multi-dimensional scientific and technological achievement and innovation resource matching method based on big data

Abstract

The application discloses a kind of multi-dimension scientific and technological achievements and innovation resource matching method based on big data, it is related to scientific and technological financial technical field, including obtaining enterprise working condition data and insurance financial clause, input dynamic classification model, output enterprise type, associated to achievement transformation stage;In the insurance interface rendering insurance scheme priority, corresponding insurance scheme is topped according to enterprise type, and embedded format contract clause to insurance declaration, generates insurance decision data;Insurance decision data is fed back to entropy risk assessment mechanism, and the weight coefficient of difference factor and probation period parameter is updated.The application generates migration risk value dynamically by entropy risk assessment mechanism fusion credit agency historical performance data, real-time probation period parameter and matching decision, realizes the risk quantization of multi-source heterogeneous data, and accurately captures enterprise risk migration track.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of science and technology finance, and particularly relates to a multi-dimensional science and technology achievement and innovation resource matching method based on big data. BACKGROUND

[0002] In recent years, insurance financial tools have been gradually introduced into the field of science and technology achievement transformation to reduce the risk of enterprises, and the dynamic matching technology based on big data has become a research hotspot. The prior art has realized the preliminary association of enterprise working condition data and insurance clauses, and has constructed an entropy risk assessment mechanism relying on historical performance data provided by credit institutions. In the aspect of achievement transformation stage division, the double-path mechanism of free trial and paid transformation has gradually matured, and some have supported trial period parameter monitoring and achievement pool matching decision. Interface interaction technology has also made progress, and dynamic rendering of the insurance interface and embedded delivery of contract clauses have become industry standards.

[0003] At present, there are deficiencies in the real-time cooperation of insurance schemes and dynamic risks. The insurance financial clauses are fixed and cannot be dynamically adjusted in priority according to the risk state of enterprises. After the migration risk value is generated, only a simple alarm is triggered, and the depth adaptation of insurance liability scope and achievement buffer options is not driven. This leads to the fact that high-risk enterprises cannot obtain high-protection schemes in real time, and the overload of interface information aggravates the decision-making burden. SUMMARY

[0004] In view of the above-mentioned existing problems, the present application is proposed.

[0005] Therefore, the present application provides a multi-dimensional science and technology achievement and innovation resource matching method based on big data, which solves the problem of insufficient adaptability of insurance financial clauses to dynamic risks of science and technology achievement transformation enterprises.

[0006] To solve the above technical problems, the present application provides the following technical solutions:

[0007] The present application provides a multi-dimensional science and technology achievement and innovation resource matching method based on big data, which comprises: obtaining enterprise working condition data and insurance financial clauses, inputting a dynamic classification model, outputting enterprise types, and associating to the achievement transformation stage;

[0008] If it is a free trial stage, a free license agreement is generated by calling a preset agreement template, and trial period parameters are monitored. If it is a paid transformation stage, an achievement identifier is extracted from an achievement pool, a difference degree with an enterprise identifier is calculated, a difference factor is generated, and a matching decision is generated in combination with a demand list;

[0009] A migration risk value is generated by calling an entropy risk assessment mechanism to integrate historical performance data provided by a credit institution, in combination with a matching decision and trial period parameters;

[0010] When the migration risk value exceeds the preset risk threshold, the non-core clauses of the insurance financial clause are folded and the risk buffer result option is highlighted, the interface rendering instruction is triggered and bound to the insurance interface;

[0011] In the insurance interface, the insurance scheme priority is rendered, the corresponding insurance scheme is topped according to the enterprise type, and the format contract clause is embedded into the insurance instruction to generate the insurance decision data;

[0012] The insurance decision data is fed back to the entropy risk assessment mechanism, and the weight coefficient of the difference factor and the probation period parameter are updated.

[0013] As a preferred scheme of the multi-dimensional technology achievement and innovation resource matching method based on big data, wherein: the enterprise working condition data includes equipment operation log, revenue scale interval, demand list and enterprise identification;

[0014] The enterprise type includes emergency failure type, sub-health type and normal operation type.

[0015] As a preferred scheme of the multi-dimensional technology achievement and innovation resource matching method based on big data, wherein: the association to the achievement transformation stage is as follows:

[0016] Real-time acquisition of equipment operation log, revenue scale interval, demand list, enterprise identification and insurance financial clause;

[0017] According to the equipment operation log, the equipment failure rate is obtained, the revenue scale interval is converted into a grade label, the demand list is parsed into a demand density, and the insurance financial clause is parsed to generate a responsibility strength index;

[0018] Fusion of equipment failure rate, grade label, demand density and responsibility strength index, construction of multi-dimensional feature vector input dynamic classification model, output of enterprise type and insurance adaptation identification, association to achievement transformation stage.

[0019] As a preferred scheme of the multi-dimensional technology achievement and innovation resource matching method based on big data, wherein: the monitoring probation period parameter is as follows:

[0020] According to the enterprise type and the insurance adaptation identification, a preset agreement template is matched, the insurance financial clause constraint is injected, the equipment failure rate and the demand density are combined, the probation period parameter and the monitoring index set are generated, and the free license agreement is output;

[0021] Through the distributed nodes, the equipment operation log and the achievement calling behavior are captured in real time, the abnormal fluctuation is identified by comparing the monitoring index set, and the probation period compression and insurance responsibility upgrade instruction is triggered.

[0022] As a preferred scheme of the multi-dimensional technology achievement and innovation resource matching method based on big data, wherein: the matching decision is generated in combination with the demand list, and the specific steps are as follows:

[0023] Extract achievement identifiers that have not been transformed in recent years from the achievement pool, and filter achievement identifiers with a matching degree lower than a preset matching degree threshold from the demand list;

[0024] Encode the pre-screened achievement identifiers and enterprise identifiers into achievement identifier vectors and enterprise identifier vectors through a multi-modal spatio-temporal encoder;

[0025] Calculate the difference degree of the achievement identifier vectors and the enterprise identifier vectors through a Transformer cross-attention mechanism to generate a difference factor, and generate a matching decision in combination with the demand list.

[0026] As a preferred scheme of the multi-dimensional technology achievement and innovation resource matching method based on big data, wherein: the difference factor refers to a dynamic quantitative index for quantifying the matching degree of the technology achievement and the enterprise demand.

[0027] As a preferred scheme of the multi-dimensional technology achievement and innovation resource matching method based on big data, wherein: the matching decision is generated in combination with the demand list, and the specific steps are as follows:

[0028] Call the interface of the credit institution to obtain historical performance data, synchronize the trial period parameters and the difference factor, clean up abnormal data through a dynamic fluctuation filtering algorithm, and output a standardized data matrix;

[0029] Input the standardized data matrix into a feature fusion layer, dynamically adjust the influence weight of the historical performance data, the trial period parameters and the difference factor, and generate a multi-dimensional fusion feature tensor;

[0030] Based on the multi-dimensional fusion feature tensor, calculate the migration risk value through an entropy risk assessment mechanism.

[0031] As a preferred scheme of the multi-dimensional technology achievement and innovation resource matching method based on big data, wherein: the interface rendering instruction is triggered and bound to the insurance interface, and the specific steps are as follows:

[0032] Parse the insurance financial clauses, identify the core clauses and non-core clauses through natural language recognition, and when the migration risk value exceeds a preset risk threshold, automatically fold the non-core clauses into an expandable gray-scale panel to generate a folded clause set;

[0033] Extract the matched buffer achievements from the achievement pool, and generate a highlighted achievement set according to the enterprise working condition data and the matched buffer achievements;

[0034] inputting the folded clause set and the highlighted result set into a neural rendering engine, dynamically modulating rendering priorities by migrating risk values, and generating a visual instruction set;

[0035] pushing the visual instruction set to the insurance interface node through a low-delay communication protocol to update a visual component.

[0036] As a preferred scheme of the multi-dimensional technology achievement and innovation resource matching method based on big data, the specific steps of generating the insurance decision data are as follows:

[0037] According to the enterprise type and the migration risk value, the temporal trend of the equipment failure rate and the topology relationship of the equipment cluster are fused, and a risk portrait vector is generated through a spatio-temporal graph convolution network.

[0038] The risk portrait vector is input into a neural symbol joint engine, the enterprise type and the migration risk value are combined, the risk mode is analyzed through a neural network component, the symbol rule component is superimposed, and an insurance scheme priority score is output.

[0039] According to the insurance scheme priority score, the display order of the insurance financial clauses is dynamically adjusted, and an interactive clause component is output.

[0040] According to the insurance scheme priority score, the insurance scheme is rendered in descending order, the interactive clause component is embedded, the user selection behavior is captured in real time, and the insurance decision data is generated.

[0041] As a preferred scheme of the multi-dimensional technology achievement and innovation resource matching method based on big data, the specific steps of updating the weight coefficient of the difference factor and the trial period parameter are as follows:

[0042] The user behavior characteristics in the insurance decision data are analyzed, including risk sensitivity, clause attention, and decision hesitation index.

[0043] Based on the risk sensitivity and the clause attention, the weight coefficient of the difference factor is adjusted, the decision hesitation index and the risk sensitivity are combined, and the trial period parameter is adjusted.

[0044] The updated weight coefficient of the difference factor and the trial period parameter are fed back to an entropy risk assessment mechanism, the feature fusion layer is reconstructed, and the risk assessment accuracy is optimized.

[0045] The application has the beneficial effects that: through the entropy risk assessment mechanism, historical performance data of a credit institution, real-time probation period parameters and matching decisions are fused, a migration risk value is dynamically generated, risk quantification of multi-source heterogeneous data is realized, and an enterprise risk migration track is accurately captured; when the risk value exceeds a threshold value, insurance clause folding and achievement highlighting are automatically triggered, and second-level feedback is generated to a policy interface to generate visual decision support; meanwhile, through policy decision data closed-loop feedback, difference factor weight and probation period parameters are dynamically calibrated, the entropy risk assessment mechanism is continuously self-optimized, and risk prediction accuracy is improved. BRIEF DESCRIPTION OF DRAWINGS

[0046] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed to be used in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0047] Figure 1 The flowchart of the multi-dimensional technology achievement and innovation resource matching method based on big data.

[0048] Figure 2 The flowchart of the enterprise type classification and achievement transformation stage association.

[0049] Figure 3 The flowchart of the entropy risk assessment mechanism.

[0050] Figure 4 The flowchart of the policy decision data generation. DETAILED DESCRIPTION

[0051] In order to make the above-mentioned purposes, features and advantages of the present application more apparent and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings of the specification.

[0052] In the following description, many specific details are set forth in order to provide a thorough understanding of the present application, but the present application can also be implemented in other ways different from those described herein, and those skilled in the art can make similar generalizations without departing from the connotation of the present application, therefore the present application is not limited to the specific embodiments disclosed below.

[0053] Secondly, the "one embodiment" or "embodiment" referred to herein means that the specific features, structures or characteristics can be included in at least one implementation of the present application. "In one embodiment" appearing in different places in the specification does not mean the same embodiment, nor is it an independent or alternative embodiment that excludes other embodiments.

[0054] REFERENCE Figure 1For an embodiment of the present application, the embodiment provides a big data-based multi-dimensional technology achievement and innovation resource matching method, comprising the following steps:

[0055] S1: Obtain enterprise working condition data and insurance financial clauses, input dynamic classification model, output enterprise type, and associate to achievement transformation stage. For details, please refer to Figure 2 .

[0056] S1.1: The enterprise working condition data includes equipment operation log, revenue scale interval, demand list, and enterprise identification.

[0057] It should be noted that the equipment operation log records the equipment running state, fault event and timestamp.

[0058] The revenue scale interval represents the enterprise annual revenue range, which is stored in the form of a numerical interval.

[0059] The demand list lists the current technical demand items of the enterprise, each of which contains demand description and priority.

[0060] The enterprise identification is a unique code used to identify the enterprise entity.

[0061] S1.2: The enterprise type includes emergency failure type, sub-health type and normal operation type.

[0062] It should be noted that the enterprise type classification standard is as follows: the emergency failure type indicates high equipment failure rate and high demand density, the sub-health type indicates medium equipment failure rate or unstable revenue, and the normal operation type indicates low equipment failure rate and stable revenue.

[0063] S1.3: Real-time collection of equipment operation log, revenue scale interval, demand list, enterprise identification and insurance financial clauses.

[0064] Deploy a data collection agent program to read equipment operation logs from enterprise device sensors in real time through an API interface. The equipment operation log includes real-time equipment state data.

[0065] Subscribing to revenue scale interval updates in real time through enterprise financial data flow.

[0066] Call the demand management platform API to grab the demand list in real time.

[0067] Call the enterprise registration database to query the enterprise identification in real time. The enterprise identification is used as a unique key value.

[0068] Call the insurance database interface to obtain insurance financial clauses in real time. The insurance financial clauses are text data describing the scope of insurance liability.

[0069] S1.4: Obtain the device failure rate from the device operation log, convert the revenue scale interval into a level label, parse the demand list into a demand density, and analyze the insurance financial clauses to generate a liability intensity index.

[0070] Specifically, the ratio of the number of failure events to the total running time in the device operation log is calculated as the device failure rate.

[0071] The revenue scale interval value is mapped to a level label. For example, the revenue scale interval is 0-1 million yuan for a low level label, 1-5 million yuan for a medium level label, and more than 5 million yuan for a high level label.

[0072] The number of demand items and the urgency level in the demand list are analyzed to output the demand density.

[0073] The insurance financial clause text is analyzed using natural language to extract liability clause keywords (such as "compensation range" and "exemption clause"), and a liability intensity index is obtained.

[0074] S1.5: Fuse device failure rate, level label, demand density, and liability intensity index to construct a multi-dimensional feature vector input dynamic classification model, output enterprise type and insurance adaptation identification, and associate it to the achievement transformation stage.

[0075] It should be noted that the pre-training process of the dynamic classification model is as follows: collect historical enterprise working condition data and insurance financial clauses, and form historical multi-dimensional feature vectors composed of historical device failure rate, historical level label, historical demand density, and historical liability intensity index. At the same time, the corresponding historical enterprise type and historical insurance adaptation identification are obtained; the historical multi-dimensional feature vector is used as the input feature, and the historical enterprise type and historical insurance adaptation identification are used as the supervision label to train the dynamic classification model; the enterprise type classification error rate and the insurance adaptation identification prediction error are minimized during training; after training, the dynamic classification model parameters are saved to obtain the dynamic classification model.

[0076] Specifically, the device failure rate, level label, demand density, and liability intensity index are combined into a multi-dimensional feature vector, and the multi-dimensional feature vector has a fixed dimension of four, with each dimension corresponding to an index value.

[0077] The multi-dimensional feature vector is input into the dynamic classification model, which classifies the multi-dimensional feature vector based on historical training data and pre-set classification rules, and outputs the enterprise type and insurance adaptation identification. The insurance adaptation identification is a binary value (0 or 1).

[0078] According to the enterprise type and insurance adaptation identification output by the dynamic classification model, the achievement transformation stage is associated through a pre-set mapping rule:

[0079] Free trial stage applicable conditions:

[0080] When the enterprise type is an emergency failure type (high device failure rate and high demand density), the free trial stage is forcibly associated.

[0081] Conditions for applying the paid conversion stage:

[0082] When the enterprise type is a normal operation type (device failure rate less than 5% and stable revenue), the paid conversion stage is directly associated.

[0083] When the enterprise type is sub-healthy, the insurance adaptation identifier needs to be combined:

[0084] If the insurance adaptation identifier is 1 (adapt insurance), the paid conversion stage is associated;

[0085] If the insurance adaptation identifier is 0 (no insurance), return to adjust the demand list.

[0086] It should be noted that the preset classification rule refers to the decision rule formed by the dynamic classification model in the training process through the supervised learning algorithm, which is used to output the enterprise type and insurance adaptation identifier according to the multi-dimensional feature vector.

[0087] The preset mapping rule refers to the pre-defined key-value pair logic, with the enterprise type and insurance adaptation identifier as the "key" and the "value" of the outcome conversion stage or operation, which is used to associate the enterprise type and insurance adaptation identifier output by the dynamic classification model to the outcome conversion stage.

[0088] S2: If it is a free trial stage, call the preset agreement template to generate a free license agreement, monitor the trial period parameters, and if it is a paid conversion stage, extract the outcome identifier from the outcome pool, calculate the difference degree with the enterprise identifier, generate a difference factor, and generate a matching decision combined with the demand list.

[0089] S2.1: According to the enterprise type and insurance adaptation identifier, match the preset agreement template, inject insurance financial clause constraints, combine device failure rate and demand density, generate trial period parameters and monitoring index set, and output free license agreement.

[0090] It should be noted that the preset agreement template refers to a document pre-stored in the template library containing a basic clause framework. After selecting a specific template according to the enterprise type and insurance adaptation identifier, dynamically injecting insurance financial clause constraints, and filling in the trial period parameters combined with the device failure rate and demand density, a free license agreement is generated.

[0091] Specifically, according to the enterprise type and insurance adaptation identifier, a corresponding free license agreement template is selected from the preset agreement template.

[0092] Parse the insurance financial terms, extract the scope of liability, exemption clauses, compensation limits, premium calculation rules, claim time limits and dispute resolution methods, and inject them into the selected agreement template.

[0093] According to the device failure rate and the demand density, the device failure rate is converted into a risk level, the risk level includes a high risk level, a medium risk level and a low risk level, the demand density is converted into an emergency level, the emergency level includes a high emergency level, a medium emergency level and a low emergency level, the risk level and the emergency level are combined to generate a trial period parameter, the trial period parameter includes a trial period length and a result call upper limit.

[0094] Based on the device failure rate, a monitoring index set is generated, the monitoring index set includes a device failure threshold and a result call frequency threshold, the agreement template, the insurance constraint, the trial period parameter and the monitoring index set are integrated to generate a free license agreement.

[0095] It should be noted that the device failure threshold is set based on the risk level (high / medium / low) converted from the device failure rate, the high risk level corresponds to the high failure threshold, and the example value is: the high risk level device failure threshold is set to the upper limit of the single-day failure number 10 times; the result call frequency threshold is set based on the emergency level (high / medium / low) converted from the demand density, the high emergency level corresponds to the high call frequency threshold; the example value is: the high emergency level result call frequency threshold is set to the lower limit of the daily call number 5 times.

[0096] S2.2: Real-time capture of device operation logs and result call behavior through distributed nodes, comparison of monitoring index set to identify abnormal fluctuations, triggering of trial period length compression and insurance liability upgrade instructions.

[0097] Specifically, a lightweight data collection node is deployed on the enterprise server to capture device operation logs and result call behavior in real time, and the result call behavior includes API call records and function usage times.

[0098] Compare the real-time captured device operation logs and result call behavior with the monitoring index set:

[0099] When the single-day device failure number exceeds the high failure threshold, it is marked as an abnormal fluctuation;

[0100] When the result call frequency is lower than the result call high call frequency threshold, it is marked as an abnormal fluctuation;

[0101] According to the abnormal fluctuation type, if the device failure rate is abnormal for consecutive days, the trial period length compression instruction is triggered, and if the result call frequency is lower than the result call frequency threshold and the device failure rate is high, the insurance liability upgrade instruction is triggered.

[0102] S2.3: Extract the achievement identifiers that have not been converted in recent years from the achievement pool, and filter the achievement identifiers with a matching degree lower than the preset matching degree threshold with the demand list.

[0103] Specifically, the achievement identifiers that have not been converted in the past three years are screened from the achievement pool, each achievement identifier is compared with the demand list in terms of semantic similarity, the achievement identifiers with a semantic similarity lower than the preset matching degree threshold are removed, and the pre-screened achievement identifiers are retained.

[0104] It should be noted that the preset matching degree threshold is set based on the demand density generated by analyzing the demand list, and the higher the demand density, the stricter the matching degree threshold; for example, when the demand density is high (more than 8 items / month), the preset matching degree threshold is set to 0.6.

[0105] S2.4: Encode the pre-screened achievement identifiers and enterprise identifiers into achievement identifier vectors and enterprise identifier vectors through a multi-modal spatio-temporal encoder;

[0106] Specifically, the pre-screened achievement identifiers and enterprise identifiers are input into the multi-modal spatio-temporal encoder, the technology field classification code, the R&D time stamp, and the registration region code associated with the achievement identifier are analyzed, the technology field classification code is mapped into a field embedding vector, the R&D time stamp is converted into a time feature vector through a time encoder, and the registration region code is generated into a spatial feature vector through a geographic encoder, and the field embedding vector, the time feature vector, and the spatial feature vector are fused to generate the achievement identifier vector.

[0107] According to the three features of device failure rate, level label, and demand density, the device failure rate is converted into a failure risk vector, the level label is mapped into a level vector, and the demand density is converted into a demand intensity vector, and the failure risk vector, the level vector, and the demand intensity vector are fused to generate the enterprise identifier vector.

[0108] S2.5: Calculate the difference degree of the achievement identifier vector and the enterprise identifier vector through the Transformer cross-attention mechanism, generate a difference factor, and generate a matching decision in combination with the demand list.

[0109] Specifically, the achievement identifiers are arranged in descending order of the difference factor in combination with the priority order of the demand list, and a matching decision is generated.

[0110] The difference degree of the achievement identifier vector and the enterprise identifier vector is calculated through the Transformer cross-attention mechanism, and a difference factor is generated, and the expression is:

[0111] ;

[0112] In the formula, the difference factor is represented by d, the normalization function is represented by n, the achievement identifier vector is represented by a, represents an enterprise identification vector, represents a vector dimension, represents a demand list vector.

[0113] It should be noted that, is dimensionless, is dimensionless, is dimensionless, is dimensionless, is dimensionless, and finally maintains dimensional consistency.

[0114] It should be noted that the difference factor refers to a dynamic quantitative index for quantifying the matching degree of scientific and technological achievements and enterprise demand, and the value lies in resolving information barriers, improving conversion efficiency, and dynamically adapting to demand.

[0115] S3: Call the entropy risk assessment mechanism to integrate historical performance data provided by the credit agency, combine the matching decision and the probation period parameter to generate a migration risk value, please refer to Figure 3 .

[0116] S3.1: Call the credit agency interface to obtain historical performance data, synchronize the probation period parameter and the difference factor, and clean up abnormal data through a dynamic fluctuation filtering algorithm to output a standardized data matrix.

[0117] Specifically, historical performance data is obtained through the credit agency API, which includes the number of historical insurance claims, the amount of claims, and the default record of the enterprise, and is synchronized with the probation period parameter and the difference factor.

[0118] Calculate the average value and fluctuation range of the historical performance data sliding window, and remove data points deviating from the average value by more than the preset fluctuation range.

[0119] Standardize the retained historical performance data, probation period parameter and difference factor, calculate the overall average value and dispersion degree of the historical performance data, probation period parameter and difference factor, adjust the data distribution according to the unified dimension, generate a standardized data matrix with the number of rows as the effective data amount and the number of columns as three, and the number of columns as three respectively corresponding to the historical performance data, probation period parameter and difference factor.

[0120] It should be noted that the preset fluctuation range refers to a three-standard-deviation interval formed by the historical performance data in the sliding window calculation, which is used to remove abnormal data points deviating from the average value by more than the interval by the dynamic fluctuation filtering algorithm.

[0121] S3.2: Input the standardized data matrix into the feature fusion layer, dynamically adjust the influence weight of the historical performance data, probation period parameter and difference factor, and generate a multi-dimensional fusion feature tensor.

[0122] Specifically, the weight proportion is allocated according to the numerical size of the difference factor: the higher the difference factor, the greater the weight proportion (for example, when the difference factor is 90, the weight proportion reaches 60%).

[0123] The weight of historical performance data and probation period parameters is inversely adjusted (for example, when the weight of the difference factor is increased, the weight of historical performance data and probation period parameters is correspondingly reduced).

[0124] The three values of each row of the standardized data matrix are weighted and summed according to the weight proportion (for example, historical performance data accounts for 20%, probation period parameters account for 20%, and the difference factor accounts for 60%), and a multi-dimensional fusion feature tensor is output.

[0125] S3.3: Specifically, based on the multi-dimensional fusion feature tensor, the migration risk value is calculated through an entropy risk evaluation mechanism, and the expression is:

[0126] ;

[0127] ;

[0128] ;

[0129] In the formula, represents the migration risk value, represents the Sigmoid function, represents the second weight matrix, represents the rectified linear unit function, represents the first weight matrix, represents the tensor vectorization operator, represents the multi-dimensional fusion feature tensor, represents the multi-dimensional fusion feature tensor vector, represents the first bias vector, represents the second bias scalar.

[0130] It should be noted that, is dimensionless, is dimensionless, is dimensionless, is dimensionless, is dimensionless, is dimensionless, and finally the dimension is unified.

[0131] It should be noted that the first weight matrix refers to a pre-trained parameter matrix, which is used to linearly transform the vectorized multi-dimensional fusion feature tensor to the hidden layer space, and its dimension is dynamically determined according to the number of hidden layer nodes (for example, when the input vector is 128-dimensional, the first weight matrix is 256x128).

[0132] The second weight matrix refers to a pre-training parameter matrix for linearly mapping the hidden layer output to the risk probability space, which has a fixed dimension of 1x the number of hidden layer nodes (for example, the second weight matrix has a dimension of 1x256 when the hidden layer has 256 nodes).

[0133] The first bias vector refers to a pre-training bias parameter vector of the hidden layer, which has a dimension consistent with the number of hidden layer nodes (for example, the first bias vector is a 256-dimensional vector when there are 256 nodes), and is used to adjust the output offset of the hidden layer activation function.

[0134] The second bias scalar refers to a pre-training bias constant scalar (single value) of the output layer, which is used to adjust the baseline offset of the final migration risk value (for example, the baseline value adjustment constant is -0.3).

[0135] Preferably, by dynamically fusing three types of heterogeneous data sources, i.e., historical performance data of credit institutions, real-time generated matching decisions, and probation period parameters, the entropy risk assessment mechanism realizes real-time quantification of multi-dimensional risk migration trajectories. Compared with the limitations of traditional credit scoring card models that rely only on static historical data, the use of probation period parameters enables the migration risk value to respond immediately to changes in enterprise operating status; at the same time, by injecting the technology achievement transformation adaptation degree index through the difference factor in the matching decision, innovation risk is included in the evaluation framework. The real-time fusion mechanism of multi-source heterogeneous data improves the accuracy of the migration risk value and effectively solves the problem of insurance plan lag caused by data fragmentation.

[0136] S4: When the migration risk value exceeds the preset risk threshold, fold the non-core clauses of the insurance financial clauses and highlight the risk buffer result options, trigger the interface rendering instruction and bind to the insurance interface.

[0137] S4.1: Analyze the insurance financial clauses, identify the core clauses and non-core clauses through natural language recognition, and when the migration risk value exceeds the preset risk threshold, automatically fold the non-core clauses into an expandable gray-scale panel to generate a set of folded clauses.

[0138] It should be noted that the preset risk threshold is set based on historical risk assessment experience values to divide the high-risk and low-risk decision boundaries. For example, the migration risk value threshold is set to 60.

[0139] Specifically, the dependency syntax analysis algorithm is used to identify the core clauses and non-core clauses of the insurance financial clauses, the core clauses include the scope of liability and the compensation limit, and the non-core clauses include the dispute resolution method and the premium calculation details, and the core clause determination rule is an independent sentence containing the keywords "compensation", "liability" and "limit".

[0140] When the migration risk value exceeds the preset risk threshold: the non-core clauses are converted into expandable gray-scale panels (default folded state, background color set to light gray), the core clauses remain in the original display style, structured as a JSON array, and the folded clause set is output.

[0141] S4.2: Extract the matched buffer results from the result pool, and generate the highlighted result set according to the enterprise working condition data and the matched buffer results.

[0142] Specifically, access the result pool to filter the result identifiers marked as "optional transformation" in the matching decision, for each result identifier, extract the associated result fault compatibility coefficient and result demand coverage coefficient, combine the equipment failure rate and demand density in the enterprise working condition data to generate the buffer result adaptation degree, filter the result identifiers with buffer result adaptation degrees reaching the preset highlight threshold, and generate the highlighted result set.

[0143] It should be noted that the preset highlight threshold is set based on the dynamic distribution of the buffer result adaptation degree, for example: taking 80% of the highest buffer result adaptation degree as the reference value, when the highest buffer result adaptation degree is 5.0, the preset highlight threshold is set to 4.0.

[0144] S4.3: Input the folded clause set and the highlighted result set into the neural rendering engine, dynamically modulate the rendering priority by the migration risk value, and generate the visualization instruction set.

[0145] Specifically, read the structured data of the folded clause set and the structured data of the highlighted result set, the structured data of the folded clause set includes clause ID, text content, panel state and style attribute, and the structured data of the highlighted result set includes result ID and highlight reason.

[0146] Generate the rendering weight factor dynamically by the migration risk value and the preset rendering weight rule, and adjust the rendering weight factor proportionally based on the migration risk value.

[0147] Input the rendering weight factor into the neural rendering engine, analyze the folded state distribution of the folded clause set and the spatial relationship of the highlighted result set, output the layer layout parameters, and integrate the layer layout parameters into the visualization instruction set.

[0148] It should be noted that the preset rendering weight rule refers to the basic rendering weight constant and additional rendering weight constant preconfigured by the neural rendering engine, and the parameter combination of the rendering weight factor is dynamically calculated by the proportion of the migration risk value in the upper limit of the risk scale.

[0149] S4.4: Push the visualization instruction set to the insurance interface node through the low-latency communication protocol to update the visualization component.

[0150] Specifically, the WebSocket protocol is used to push the visualization instruction set to the insurance interface in real time.

[0151] The insurance interface parses the visualization instruction set and dynamically updates three components:

[0152] Insurance clause display area: fold non-core clause panel;

[0153] Achievement recommendation area: display the highlighted achievement set at the top;

[0154] Risk prompt bar: embed the migration risk value label.

[0155] Preferably, a risk-driven dynamic interface reconstruction mechanism is established. Compared with the traditional method, which can only statically display all clauses, the high-risk enterprise faces information overload and key risk prompts are submerged. Through natural language processing intelligent identification and folding of non-core clauses, the information redundancy problem is solved. Based on enterprise working condition data, real-time matching of high adaptability buffer achievements is carried out, and risk response guidance is strengthened by visual highlighting. Through the neural rendering engine, the rendering priority is dynamically modulated according to the migration risk value to ensure that the key information is placed at the top within seconds. The problem of disconnection between insurance solutions and dynamic risks caused by interface solidification in conventional methods is effectively solved.

[0156] S5: Render the insurance solution priority in the insurance interface, place the corresponding insurance solution at the top according to the enterprise type, and embed the formatted contract clauses into the insurance explanation to generate insurance decision data. For details, please refer to Figure 4 .

[0157] S5.1: According to the enterprise type and migration risk value, the time trend of equipment failure rate and the relationship between equipment cluster topology are fused to generate a risk portrait vector through a spatio-temporal graph convolution network.

[0158] It should be noted that the pre-training process of the spatio-temporal graph convolution network is as follows: collect the historical enterprise type, historical migration risk value, historical time trend of equipment failure rate, and historical relationship between equipment cluster topology to form a historical training data set. The historical relationship between equipment cluster topology is input into the graph convolution network branch to extract the spatial feature vector, and the historical time trend of equipment failure rate is input into the time series convolution network branch to extract the time feature vector. The spatial feature vector and the time feature vector are spliced to generate a spatio-temporal fusion feature, and the historical enterprise type is encoded into a classification vector and the historical migration risk value is normalized into a scalar. The complete input feature vector is formed by splicing the spatio-temporal fusion feature. The complete input feature vector is input into the fully connected layer to output the predicted risk portrait vector. The difference between the predicted risk portrait vector and the true risk portrait vector is calculated by the mean square error loss function, and the parameters of the graph convolution network, the time series convolution network, and the fully connected layer are updated by back propagation until convergence to obtain the spatio-temporal graph convolution network.

[0159] Specifically, the enterprise type is encoded into a three-dimensional classification vector, the migration risk value is normalized into a scalar between zero and one, the failure rate fluctuation curve of the equipment failure rate in the near month is extracted, the time trend of the equipment failure rate is generated, and the equipment connection matrix in the equipment operation log is extracted to generate the equipment cluster topology relationship.

[0160] The equipment cluster topology relationship is input into a graph convolution network to generate a spatial topology association, and the time trend of the equipment failure rate is input into a time series convolution network to generate a time fluctuation feature.

[0161] The spatial and temporal fusion features generated by the graph convolution network output and the time series convolution network output are fused to generate a spatial and temporal fusion feature, and the three-dimensional classification vector, the normalized migration risk value, and the spatial and temporal fusion feature are spliced to be compressed into a risk portrait vector through a fully connected layer.

[0162] S5.2: Input the risk portrait vector into the neural-symbol joint engine, combine the enterprise type and the migration risk value, analyze the risk pattern through the neural network component, superimpose the symbol rule component, and output the insurance plan priority score.

[0163] Specifically, the risk portrait vector is input into a three-layer fully connected layer of the neural network component, each layer is activated by a rectified linear unit function, and a risk pattern probability distribution value is output.

[0164] At the same time, the enterprise type and the migration risk value are input into the symbol rule component, and the symbol rule is executed to obtain a basic score correction value.

[0165] The risk pattern probability distribution value is converted into a benchmark score, and the basic score correction value output by the symbol rule component is superimposed to generate an insurance plan priority score.

[0166] S5.3: Dynamically adjust the display order of the insurance financial clauses according to the insurance plan priority score, and output the interactive clause component.

[0167] Specifically, read the insurance financial clauses, and establish a mapping relationship between the unique identifier of each clause and the insurance plan priority score.

[0168] Arrange the clause set in descending order according to the insurance plan priority score from high to low, generate a sorted clause sequence, and configure an interactive event handler for each clause in the sorted clause sequence, including event binding to expand the clause details when the mouse hovers and event binding to record the user operation timestamp when clicked.

[0169] Encapsulate the clause identifier, sorting number, and interactive event configuration into an interactive clause component.

[0170] S5.4: Render the insurance plan in descending order according to the insurance plan priority score, embed the interactive clause component, capture user selection behavior in real time, and generate insurance decision data.

[0171] Specifically, an insurance plan display container is created in the insurance interface area, an insurance plan list is sorted from high to low according to the insurance plan priority score, the plan with the highest score is displayed in full details, and the remaining plans are displayed in a folded manner.

[0172] An interactive clause component is embedded for each insurance plan, and a mouse hover event handler and a click event handler are bound.

[0173] The user's mouse hover duration, click selection action, and check confirmation operation are captured in real time through an event listener. When the user submits an insurance application, the current enterprise type, the user's final selected insurance plan identifier, the risk confirmation check status, and the operation timestamp are integrated to generate insurance decision data.

[0174] S6: Feedback the insurance decision data to the entropy risk assessment mechanism to update the weight coefficient of the difference factor and the probation period parameter.

[0175] S6.1: Analyze the user behavior characteristics in the insurance decision data, including risk sensitivity, clause attention, and decision hesitation index.

[0176] Specifically, the risk confirmation flag, the clause expansion timestamp sequence, the first operation timestamp, and the final decision timestamp are extracted from the insurance decision data.

[0177] The risk sensitivity is calculated according to the cumulative number of times the risk confirmation flag is true, the total clause attention duration is generated according to the sum of the adjacent time point differences in the clause expansion timestamp sequence, and the actual decision duration is generated according to the difference between the first operation timestamp and the final decision timestamp.

[0178] The risk sensitivity is equal to the proportion of the number of times the risk confirmation flag is true in the total insurance event, the clause attention is equal to the ratio of the total clause attention duration to the actual decision duration, and the decision hesitation index is equal to the ratio of the actual decision duration to the benchmark decision duration.

[0179] S6.2: Adjust the weight coefficient of the difference factor based on the risk sensitivity and the clause attention, and adjust the probation period parameter in combination with the decision hesitation index and the risk sensitivity.

[0180] Specifically, the weight coefficient of the difference factor is adjusted according to the risk sensitivity percentage value by a proportional scaling rule, and the higher the risk sensitivity, the greater the weight coefficient increase.

[0181] At the same time, the weight coefficient increase is added according to the clause attention percentage value by an auxiliary proportion rule, and the higher the clause attention, the greater the additional amount.

[0182] The decision hesitation index percentage value and the risk sensitivity percentage value are combined, and a trial duration compression ratio is obtained through a linear adjustment rule, and the higher the decision hesitation index and the higher the risk sensitivity, the greater the compression ratio.

[0183] The achievement call frequency upper limit promotion ratio is generated according to a step adjustment rule based on the risk sensitivity percentage value, and the higher the risk sensitivity, the greater the call frequency upper limit promotion range.

[0184] S6.3: The updated weight coefficient of the difference factor and the probation period parameter are fed back to the entropy risk assessment mechanism, the feature fusion layer is reconstructed, and the risk assessment accuracy is optimized.

[0185] Specifically, the updated weight coefficient of the difference factor is written into the feature fusion layer of the entropy risk assessment mechanism, replacing the original weight coefficient, and the updated probation period parameter is synchronized to the entropy risk assessment mechanism, replacing the original probation period parameter.

[0186] In the feature fusion layer, the weighted fusion proportion rule of the historical performance data, the probation period parameter and the difference factor is reconfigured according to the new weight coefficient.

[0187] The error rate of the predicted migration risk value and the actual migration risk value is recalculated using the updated feature fusion layer parameters, and when the error rate decreases by more than a preset optimization threshold, it is confirmed that the risk assessment accuracy optimization is completed.

[0188] It should be noted that the preset optimization threshold is set based on the historical risk assessment error fluctuation range, and is used to determine the effectiveness of parameter updating. An example value is that the error rate decrease threshold is set to 5%.

[0189] In summary, the present application dynamically generates a migration risk value by fusing historical performance data of a credit institution, real-time probation period parameters and matching decisions through an entropy risk assessment mechanism, realizes risk quantification of multi-source heterogeneous data, and accurately captures the enterprise risk migration trajectory; when the risk value exceeds the threshold, the insurance clause folding and achievement highlighting are automatically triggered, and the second-level feedback is fed back to the insurance interface to generate visual decision support; at the same time, through the closed-loop feedback of the insurance decision data, the weight of the difference factor and the probation period parameter are dynamically calibrated, so that the entropy risk assessment mechanism is continuously self-optimized, and the risk prediction accuracy is improved.

[0190] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and not to limit it, although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced equivalently without departing from the spirit and scope of the technical solutions of the present application, and they should be covered in the scope of the claims of the present application.

Claims

1. A method for matching multi-dimensional scientific and technological achievements and innovation resources based on big data, characterized in that: The method comprises the following steps: Obtain enterprise working condition data and insurance financial terms, input dynamic classification model, output enterprise type, and associate with the achievement transformation stage. The enterprise working condition data includes equipment operation log, revenue scale interval, demand list, and enterprise identification. The enterprise type includes emergency failure type, sub-health type, and normal operation type. The association with the achievement transformation stage includes the following steps: Real-time collection of equipment operation log, revenue scale interval, demand list, enterprise identification, and insurance financial terms; According to the equipment operation log, obtain the equipment failure rate. Convert the revenue scale interval into a grade label. Analyze the demand list to obtain the demand density. Analyze the insurance financial terms to generate the responsibility intensity index; Fusion of equipment failure rate, grade label, demand density, and responsibility intensity index to construct a multi-dimensional feature vector input dynamic classification model, output enterprise type and insurance adaptation identification, and associate with the achievement transformation stage; If it is a free trial stage, call the preset agreement template to generate a free license agreement, monitor the trial period parameters, and if it is a paid transformation stage, extract the achievement identification from the achievement pool, calculate the difference degree with the enterprise identification, generate the difference factor, and generate the matching decision combined with the demand list; The enterprise identification is obtained by converting the equipment failure rate into a failure risk vector, mapping the grade label into a grade vector, and converting the demand density into a demand intensity vector, and then fusing the failure risk vector, the grade vector, and the demand intensity vector to generate the enterprise identification vector; Call the entropy risk assessment mechanism to integrate the historical performance data provided by the credit institution, and generate the migration risk value combined with the matching decision and the trial period parameters; When the migration risk value exceeds the preset risk threshold, fold the non-core terms of the insurance financial terms and highlight the risk buffer achievement options, trigger the interface rendering instruction and bind to the insurance interface; Render the insurance scheme priority in the insurance interface, top the corresponding insurance scheme according to the enterprise type, and embed the format contract terms into the insurance explanation to generate the insurance decision data; Feedback the insurance decision data to the entropy risk assessment mechanism to update the weight coefficient of the difference factor and the trial period parameters.

2. The big data-based multi-dimensional technology achievement and innovation resource matching method according to claim 1, characterized in that: The specific steps of monitoring the trial period parameters are as follows: According to the enterprise type and the insurance adaptation identification, match the preset agreement template, inject the insurance financial term constraints, combine the equipment failure rate and the demand density, generate the trial period parameters and the monitoring index set, and output the free license agreement; Through the distributed nodes, real-time capture of equipment operation log and achievement calling behavior, comparison of monitoring index set to identify abnormal fluctuations, triggering of trial period time compression and insurance liability upgrade instructions. 3.The big data based multi-dimension technology achievement and innovation resource matching method according to claim 2, characterized in that: The specific steps of generating the matching decision combined with the demand list are as follows: Extract the achievement identification that has not been transformed in recent years from the achievement pool, and filter the achievement identification with a matching degree lower than the preset matching degree threshold; Encode the pre-screened achievement identification and enterprise identification into achievement identification vector and enterprise identification vector through the multi-modal spatio-temporal encoder; Calculate the difference degree between the achievement identification vector and the enterprise identification vector through the Transformer cross-attention mechanism to generate the difference factor, and generate the matching decision combined with the demand list.

4. The big data-based multi-dimensional technology achievement and innovation resource matching method of claim 3, wherein: The difference factor refers to a dynamic quantitative index quantifying the matching degree of scientific and technological achievements and enterprise needs.

5. The big data-based multi-dimensional technology achievement and innovation resource matching method according to claim 4, characterized in that: The combination of the matching decision and the probation period parameter generates a migration risk value, and the specific steps are as follows: Call the credit agency interface to obtain historical performance data, synchronize the probation period parameter and the difference factor, and clean up abnormal data through a dynamic fluctuation filtering algorithm to output a standardized data matrix; Input the standardized data matrix into the feature fusion layer, dynamically adjust the influence weight of the historical performance data, the probation period parameter and the difference factor, and generate a multi-dimensional fusion feature tensor; Based on the multi-dimensional fusion feature tensor, calculate the migration risk value through the entropy risk assessment mechanism.

6. The big data-based multi-dimensional technology achievement and innovation resource matching method according to claim 5, characterized in that: The trigger interface rendering instruction is bound to the insurance interface, and the specific steps are as follows: Parse the insurance financial clauses, identify the core clauses and non-core clauses through natural language recognition, and when the migration risk value exceeds the preset risk threshold, automatically fold the non-core clauses into an expandable gray-scale panel to generate a folded clause set; Extract the matched buffer achievements from the achievement pool, and generate a highlighted achievement set according to the enterprise working condition data and the matched buffer achievements; Input the folded clause set and the highlighted achievement set into the neural rendering engine, dynamically modulate the rendering priority through the migration risk value, and generate a visual instruction set; Push the visual instruction set to the insurance interface node through a low-latency communication protocol to update the visual components.

7. The big data-based multi-dimensional technology achievement and innovation resource matching method according to claim 6, characterized in that: The generation of the insurance decision data is as follows: According to the enterprise type and the migration risk value, combine the time series trend of the equipment failure rate and the equipment cluster topology relationship, and generate a risk portrait vector through a spatio-temporal graph convolution network; Input the risk portrait vector into the neural symbol joint engine, combine the enterprise type and the migration risk value, analyze the risk mode through the neural network component, superimpose the symbol rule component, and output the insurance scheme priority score; According to the insurance scheme priority score, dynamically adjust the display order of the insurance financial clauses, and output the interactive clause component; Render the insurance scheme in descending order of the insurance scheme priority score, embed the interactive clause component, capture the user selection behavior in real time, and generate the insurance decision data.

8. The big data-based multi-dimensional technology achievement and innovation resource matching method according to claim 7, characterized in that: The weight coefficient of the difference factor and the probation period parameter are updated as follows: Analyze the user behavior characteristics in the insurance decision data, including risk sensitivity, clause attention and decision hesitation index; Based on the risk sensitivity and the clause attention, adjust the weight coefficient of the difference factor, combine the decision hesitation index and the risk sensitivity, and adjust the probation period parameter; Feedback the updated weight coefficient of the difference factor and the probation period parameter to the entropy risk assessment mechanism, reconfigure the feature fusion layer, and optimize the risk assessment accuracy.

Images