Method and device for dynamic evaluation of patent value and technical information analysis

Abstract

The embodiment of the specification provides a patent value dynamic evaluation and technical information analysis method and device, wherein the method comprises the following steps: acquiring multi-source patent data, adopting a conditional separation strategy to construct a double-layer topological graph, calculating a time sequence evolution feature set of each node of the double-layer topological graph, outputting a technical life cycle stage label and an identification confidence, positioning a main associated node and a corresponding technical life cycle stage label, generating a probabilistic valuation interval through Monte Carlo simulation, outputting a current value score and a confidence interval, outputting a future basic value trend curve and an annual renewal probability curve, generating a future patent value prediction curve by adopting a stage self-adaptive weighting formula based on the future basic value trend curve, the value decay factor and the impact coefficient, and integrating the stage identification evidence chain, the current patent value valuation interval, the technical life cycle stage label and the technical information chart into an analysis result and outputting.

Description

Technical Field

[0001] This document relates to the field of patent valuation and technical intelligence analysis, and in particular to a method and apparatus for dynamic patent valuation and technical intelligence analysis. Background Technology

[0002] Patent valuation plays a crucial role in technology transfer, patent pledge financing, mergers and acquisitions, and patent portfolio management. Traditional patent valuation methods primarily rely on intrinsic patent metrics, such as citation count, number of claims, and family size, obtaining a comprehensive score through weighted summation. However, existing methods have the following shortcomings: 1. Static nature: Using fixed indicators and weights fails to reflect the dynamic changes in patent value as technology develops. The value formation mechanisms of nascent patents and mature patents are completely different, and a uniform model leads to evaluation bias.

[0003] 2. Lag: Indicators such as the number of citations require many years of accumulation to reflect value, making it impossible to effectively evaluate newly filed patents or patents in emerging technology fields.

[0004] 3. Coarse granularity: Technical fields are usually divided by IPC classification numbers, but IPC classification is coarse-grained, updates are slow, and semantic information is lost, making it impossible to accurately represent the semantic nodes and relationships of patents in the technological evolution.

[0005] 4. Lack of predictive ability: Existing methods only output the valuation at the current point in time and cannot predict the future value trend of patents, making it difficult to support decisions such as patent abandonment, maintenance, and licensing.

[0006] 5. Limited analytical dimensions: Existing technologies only focus on the valuation of individual components or combinations, failing to utilize data from segmented technology fields for in-depth technical intelligence analysis (such as technology trends, geographical distribution, competitive landscape, life cycle tracking, etc.), thus failing to meet higher-level needs such as corporate strategic planning, competitive intelligence, and investment due diligence.

[0007] Therefore, there is an urgent need for a method and system that can dynamically and precisely assess patent value, predict future trends, and conduct in-depth technical intelligence analysis based on fine-grained technical field data, in order to overcome the shortcomings of the aforementioned existing technologies. Summary of the Invention

[0008] The purpose of this invention is to provide a method and apparatus for dynamic evaluation of patent value and analysis of technical information, aiming to solve the above-mentioned problems in the prior art.

[0009] This invention provides a method for dynamic evaluation of patent value and analysis of technical intelligence, including: Acquire multi-source patent data, preprocess the multi-source patent data using a distributed computing framework to obtain standardized patent data, and construct a two-layer topology map based on the standardized patent data using a conditional separation strategy. Calculate the temporal evolution feature set of each node in the two-layer topology map, construct a historical patent dataset, train the temporal evolution feature set, generate a patent stage identification model, input the feature vector of the target node into the patent stage identification model, and output the technology life cycle stage label and identification confidence. The entity set extracted from the target patent is semantically matched with the graph nodes of the two-layer topology graph to locate the main associated nodes and the corresponding technology life cycle stage labels. A dynamic mapping table of "patent technology stage - evaluation dimension weight" is established. The weights are configured differently according to the time separation logic. The index values ​​of each evaluation dimension are calculated and normalized. The basic score is obtained by stage adaptive weight weighting and summing. With historical transaction data as the distribution prior, a probabilistic valuation interval is generated through Monte Carlo simulation. The current value score and confidence interval are output. An LSTM model is employed, trained via a sliding window based on the node's historical degree sequence and neighborhood size, to output a future basic value trend curve. Based on the existing years of existence, family size, renewal records, and historical value scores, an annual renewal probability curve is output, mapped to a value decay factor. The topological evolution of the technology cluster to which the main associated nodes of the two-layer topological graph belong is monitored. When highly similar replacement nodes or technology fusion events are identified, a technology replacement risk index is calculated and converted into an impact coefficient. Based on the future basic value trend curve, the value decay factor, and the impact coefficient, a stage-adaptive weighted formula is used to generate a future patent value prediction curve. Based on the structure of the dual-layer topology map and the future patent value prediction curve, multi-dimensional drill-down aggregation is performed on the technology theme clusters, integrating the stage identification evidence chain, the current patent value valuation range, the technology life cycle stage labels, and the technology intelligence charts into analysis results and outputting them.

[0010] This invention provides a device for dynamic patent value assessment and technical intelligence analysis, comprising: A construction module is used to acquire multi-source patent data and preprocess the multi-source patent data using a distributed computing framework to acquire standardized patent data. Based on the standardized patent data, a two-layer topology map is constructed using a conditional separation strategy. The technology life cycle stage identification module is used to calculate the temporal evolution feature set of each node of the two-layer topology map, construct a historical patent dataset, train the temporal evolution feature set, generate a patent stage identification model, input the feature vector of the target node into the patent stage identification model, and output the technology life cycle stage label and identification confidence. The value assessment module is used to semantically match the entity set extracted from the target patent with the graph nodes of the two-layer topology graph, locate the main associated nodes and the corresponding technology life cycle stage labels, establish a dynamic mapping table of "patent technology stage - assessment dimension weight", configure weights differently according to time separation logic, calculate and normalize the index values ​​of each assessment dimension, obtain the basic score by stage adaptive weight weighting and summing, use historical transaction data as the distribution prior, generate probabilistic valuation intervals through Monte Carlo simulation, and output the current value score and confidence interval. The patent value prediction module is used to perform sliding window training based on the node's historical degree sequence and neighborhood size, outputting a future basic value trend curve. Based on the existing years of existence, family size, renewal records, and historical value scores, it outputs an annual renewal probability curve, which is mapped to a value decay factor. It monitors the topological evolution of the technology cluster to which the main associated nodes of the two-layer topological graph belong. When a highly similar replacement node or technology fusion event is identified, it calculates a technology replacement risk index and converts it into an impact coefficient. Based on the future basic value trend curve, the value decay factor, and the impact coefficient, it uses a stage-adaptive weighted formula to generate a future patent value prediction curve. The results output module is used to perform multi-dimensional drill-down aggregation on the technology topic cluster based on the structure of the two-layer topology map and the future patent value prediction curve, and integrate the stage identification evidence chain, the current patent value valuation range, the technology life cycle stage label and the technology intelligence chart into the analysis results and output them.

[0011] This invention also provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, it implements the steps of the above-described dynamic patent value assessment and technical intelligence analysis method.

[0012] This invention also provides a computer-readable storage medium storing an information transmission implementation program, which, when executed by a processor, implements the steps of the above-described dynamic patent value assessment and technical intelligence analysis method.

[0013] This invention provides a method and system for dynamic patent value assessment and technical intelligence analysis based on a technology development map. It constructs a fine-grained technology development map using a large language model, identifies the technology development stage by utilizing the temporal evolution characteristics of map nodes, and outputs current valuation and future value prediction curves based on a stage-adaptive model. Simultaneously, it aggregates patent data based on the divided technology nodes to achieve multi-dimensional technical intelligence analysis, providing data support for enterprise strategic planning, competitive intelligence, and investment decisions. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in one or more embodiments of this specification or in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a flowchart of the patent value dynamic evaluation and technical intelligence analysis method according to an embodiment of the present invention; Figure 2 This is a schematic diagram illustrating the detailed processing of the patent value dynamic evaluation and technical intelligence analysis method according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the development map construction process according to an embodiment of the present invention; Figure 4 This is a schematic diagram illustrating the lifecycle of the node temporal evolution feature recognition technology according to an embodiment of the present invention; Figure 5 This is a schematic diagram of the weight configuration table of evaluation indicators for different technical stages in an embodiment of the present invention; Figure 6 This is a visual schematic diagram of the future value prediction curve according to an embodiment of the present invention; Figure 7 This is a schematic diagram of the patent value dynamic assessment and technical intelligence analysis device according to an embodiment of the present invention; Figure 8 This is a schematic diagram of a preferred example of the patent value dynamic evaluation and technical intelligence analysis device according to an embodiment of the present invention; Figure 9 This is a schematic diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0016] To enable those skilled in the art to better understand the technical solutions in one or more embodiments of this specification, the technical solutions in one or more embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this specification, and not all of the embodiments. Based on one or more embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of this document.

[0017] Method Implementation Examples According to embodiments of the present invention, a method for dynamic evaluation of patent value and analysis of technical intelligence is provided. Figure 1 This is a flowchart of the patent value dynamic assessment and technical intelligence analysis method according to an embodiment of the present invention, such as... Figure 1As shown, the patent value dynamic assessment and technical intelligence analysis method according to an embodiment of the present invention specifically includes: Step S101 involves acquiring multi-source patent data, preprocessing the multi-source patent data using a distributed computing framework to obtain standardized patent data, and constructing a two-layer topology map based on the standardized patent data using a conditional separation strategy; specifically including: By connecting to a multi-source patent database through an API interface, the system collects full-text patents, bibliographic data, legal status, and citation network data in the target technology field. A distributed computing framework is used to perform deduplication, noise filtering, text segmentation, and structured parsing on the multi-source patent data. The unstructured multi-source patent data and structured metadata are stored in a hybrid storage architecture of graph database and relational database. Based on pre-set structured constraint prompt word templates, the large language model is guided to output technical terms, methods, materials, processes and application domain entities according to the preset JSON Schema, and to identify four types of semantic relations: reference, improvement, substitution and collaboration, and generate standardized triples; All triples are aggregated into a heterogeneous graph, with nodes bearing a timestamp of their first appearance and edges bearing relation type and time attributes. An improved temporal Louvain community discovery algorithm is used to dynamically aggregate and generate technical topic clusters according to a sliding window of predetermined time length while preserving the underlying fine-grained nodes.

[0018] Step S102 involves calculating the temporal evolution feature set of each node in the dual-layer topology map, constructing a historical patent dataset, training the temporal evolution feature set to generate a patent stage identification model, inputting the feature vector of the target node into the patent stage identification model, and outputting the technology lifecycle stage label and identification confidence score; specifically including: Calculate the temporal evolution feature set of each node in the two-layer topology graph, wherein the temporal evolution feature set specifically includes: node degree sequence, annual growth rate, neighboring node change rate, and community split / fusion index; A historical patent dataset is constructed, with lifecycle stages pre-annotated by intellectual property experts as supervision labels. The XGBoost classifier is used to train the temporal evolution feature set. K-fold cross-validation and grid search are used to determine the decision boundary function and confidence output mapping for each stage, thereby generating a patent stage identification model. The feature vector of the target node is input into the trained patent stage identification model, and the patent stage identification model outputs the technology life cycle stage label and identification confidence. The technology life cycle stage label specifically includes: budding stage, growth stage, maturity stage and decline stage.

[0019] Step S103 involves semantically matching the entity set extracted from the target patent with the graph nodes of the two-layer topology graph, locating the main associated nodes and their corresponding technology lifecycle stage labels, establishing a dynamic mapping table of "patent technology stage - evaluation dimension weights," configuring weights differently according to time separation logic, calculating and normalizing the index values ​​of each evaluation dimension, obtaining a basic score through stage adaptive weighted summation, using historical transaction data as a distribution prior, generating a probabilistic valuation interval through Monte Carlo simulation, and outputting the current value score and confidence interval; specifically including: The entity set extracted from the target patent is semantically matched with the graph nodes of the two-layer topology graph to locate the main associated node and the corresponding technology life cycle stage label. Establish a dynamic mapping table of "patent technology stage - evaluation dimension weight", and configure weights differently according to the time separation logic: increase the weight of technological foresight and market potential in the budding / growth stage, and dynamically increase the weight of legal stability, renewal willingness and synergistic value in the maturity / decline stage. The values ​​of each evaluation dimension are calculated and normalized, and a basic score is obtained by weighted summation using stage-adaptive weights. Using historical transaction data from the same field, stage, and legal status as the prior, a probabilistic valuation range is generated through Monte Carlo simulation, and the current value score and confidence interval are output.

[0020] Step S104: Using an LSTM model, a sliding window training is performed based on the node's historical degree sequence and neighborhood size to output the future basic value trend curve. Based on the existing years of existence, family size, renewal records, and historical value scores, an annual renewal probability curve is output and mapped to a value decay factor. The topological evolution of the technology cluster to which the main associated nodes of the two-layer topological graph belong is monitored. When a highly similar replacement node or technology fusion event is identified, a technology replacement risk index is calculated and converted into an impact coefficient. Based on the future basic value trend curve, the value decay factor, and the impact coefficient, a stage-adaptive weighted formula is used to generate a future patent value prediction curve. Specifically, this includes: The LSTM model is used, with the node's historical degree sequence and neighborhood size as input, and is trained with a sliding window to output the future basic value trend curve. Cox survival analysis is used to output an annual renewal probability curve based on the number of years the product has been in existence, the size of the family, the renewal history, and the historical value score, which is then mapped to a value decay factor. Monitor the topological evolution of the technology cluster to which the main associated node belongs. When a high-similarity replacement node or technology fusion event is identified, calculate the technology replacement risk index and convert it into an impact coefficient. The high-similarity replacement node is defined as two nodes whose similarity is greater than or equal to a predetermined similarity threshold.

[0021] Based on the future fundamental value trend curve, the value decay factor, and the impact coefficient, a future patent value prediction curve is generated using a stage-adaptive weighted formula as shown in Formula 1. Formula 1; in, For stage-dependent weights, This represents the future fundamental value trend curve. This represents the value decay component obtained by mapping renewal probability. This represents the impact coefficient.

[0022] Step S105: Based on the structure of the dual-layer topology map and the future patent value prediction curve, perform multi-dimensional drill-down aggregation on the technology theme cluster, and integrate the stage identification evidence chain, the current patent value valuation range, the technology life cycle stage label and the technology intelligence chart into the analysis results and output them.

[0023] Specifically, it includes: Based on the structure of the aforementioned two-layer topology map and the future patent value prediction curve, multi-dimensional drill-down aggregation is performed on technology theme clusters to conduct technology field heat analysis, regional distribution and flow analysis, competitive landscape quantification, and life cycle dynamic tracking. Specifically, the technology field heat analysis includes: statistically analyzing the number of patents, total value, and development stage percentage for each theme over the years to identify emerging technology inflection points; the regional distribution and flow analysis includes: generating heat maps by theme-country / region matrix and tracking cross-border technology transfer paths by combining applicant change records; the competitive landscape quantification includes: calculating the patent concentration index of applicants under each theme to identify core rights holders and potential cooperation / merger targets; and the life cycle dynamic tracking includes: mapping the key nodes of a theme from its initial appearance to its decline, and outputting technology substitution early warning signals based on evolution trends. The system integrates the evidence chain for stage identification, the current valuation range, the future value prediction curve, and technical intelligence charts into an interactive dashboard. The interactive dashboard provides multi-dimensional filtering, data drill-down, scenario simulation, and standardized report export. All analysis results are connected to enterprise patent management systems, technology trading platforms, or financial institution risk control systems through standardized API interfaces.

[0024] The technical solutions of the embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0025] Figure 2 This is a schematic diagram illustrating the detailed processing steps of the patent value dynamic assessment and technical intelligence analysis method according to an embodiment of the present invention, as shown below. Figure 2 As shown, the specific processing includes the following: Step S1: Patent Data Acquisition and Distributed Preprocessing By connecting to multi-source patent databases through API interfaces, the system collects full-text patents, bibliographic data, legal status, and citation network data in the target technology field. Using a distributed computing framework, the system performs deduplication, noise filtering, text segmentation, and structured parsing on the original data. Unstructured text and structured metadata are stored in a hybrid storage architecture of graph database and relational database, providing a standardized data flow for subsequent graph construction.

[0026] Step S2: Construction of a development map of fine-grained technologies based on the principle of conditional separation. Figure 3 This is a schematic diagram of the development map construction process according to an embodiment of the present invention, such as... Figure 3 The diagram illustrates the transformation process from patent text to technical entities, relation extraction, and graph clustering. Specifically, it includes the following processing steps: To address the physical contradiction between "technical semantic precision and graph computation efficiency," a conditional separation strategy is employed to construct a two-layer topological graph: Fine-grained entity extraction at the lower level: Design structured constraint prompt word templates to guide the large language model to output technical terms, methods, materials, processes and application domain entities according to the preset JSONSchema, and identify four types of semantic relations: reference, improvement, substitution and collaboration, and generate standardized triples; Top-level temporal topic aggregation: Aggregate all triples into a heterogeneous graph, with nodes bearing a timestamp of their first appearance and edges bearing relation type and time attributes; Employ an improved temporal Louvain community discovery algorithm, while preserving the fine-grained nodes at the bottom level, dynamically aggregate technical topic clusters by annual sliding window, achieving conditional separation of "fine representation during construction and topic aggregation during analysis", reducing the complexity of graph maintenance and ensuring the integrity of semantic associations.

[0027] Step S3: Identification of data-driven technology lifecycle stages. Figure 4 This is a schematic diagram of the lifecycle of the node temporal evolution feature recognition technology according to an embodiment of the present invention, illustrating the correspondence between changes in node degree sequence and stage determination, such as... Figure 4 As shown, the specific processing includes the following: To address the technical contradiction between "assessing static aspects and the dynamic evolution of technology," the principles of feedback and dynamism are introduced, abandoning the setting of empirical thresholds. Calculate the temporal evolution feature set of each node in the graph: including node degree sequence, annual growth rate, neighboring node change rate, and community split / merge index; Threshold-adaptive training: Construct a historical patent dataset, use the life cycle stages labeled by intellectual property experts as supervision labels, and train the time-series evolution feature set using an XGBoost classifier; determine the decision boundary function and confidence output mapping for each stage through K-fold cross-validation and grid search, and generate a stage identification decision tree; The feature vector of the target node is input into the trained decision tree, and the output is the technology life cycle stage label (emergence stage, growth stage, maturity stage, decline stage) and the recognition confidence level, realizing data-driven and non-human intervention in stage determination.

[0028] Step S4: Stage-adaptive current valuation based on the time separation principle To address the contradiction between "fixed weights and technological heterogeneity," a time-separation strategy is adopted to achieve dynamic weighting. Semantic matching is performed between the entity set extracted from the target patent and the graph nodes to locate the main associated nodes and corresponding stage labels; Establish a dynamic mapping table of "technology stage - evaluation dimension weight" and configure weights differently according to the time separation logic: in the budding / growth stage, focus on the weight of technological foresight and market potential, and in the maturity / decline stage, dynamically increase the weight of legal stability, renewal willingness and synergistic value. Figure 5 This is a schematic diagram of the weight configuration table of evaluation indicators for different technical stages in an embodiment of the present invention, as shown below. Figure 5 As shown, the weighting of the six evaluation dimensions across the four lifecycle stages is illustrated.

[0029] The system calculates and normalizes the values ​​of each dimension's indicators, and uses stage-adaptive weighted summation to obtain the basic score. Using historical transaction data from the same field, stage, and legal status as the prior, it generates a probabilistic valuation range through Monte Carlo simulation, and outputs the current value score and confidence interval.

[0030] Step S5: Dynamic Prediction of Future Value through Multi-Model Collaboration To address the contradiction between "improved prediction accuracy and enhanced data dependence," a dynamic fusion mechanism for stage-aware perception is constructed using a combination principle: LSTM time series prediction module: It takes the node's historical degree sequence and neighborhood size as input, trains with a sliding window, and outputs the basic value trend curve for the next 5 years. Cox Survival Analysis Module: Input the number of years of existence, family size, renewal history, and historical score, and output the annual renewal probability curve, which is mapped to the value decay factor; The graph evolution simulation module monitors the topological evolution of the technology cluster to which the main associated node belongs. When a highly similar substitution node or technology fusion event is identified, it calculates the technology substitution risk index and converts it into an impact coefficient. ; Dynamic fusion output: Prediction curves are generated using a stage-adaptive weighted formula.

[0031] in Stage-dependent weights (e.g., growth stage) Focusing on trend extrapolation, recession period Focusing on survival decay, the fusion parameters are optimized through historical backtesting. Experimental results show that this fusion mechanism reduces the average absolute error of 3-year value prediction to 8.2%, demonstrating a significant synergistic effect compared to the single model.

[0032] Figure 6 This is a visualization diagram of the future value prediction curve according to an embodiment of the present invention, such as... Figure 6 As shown, it includes the horizontal axis (year), the vertical axis (value score), the prediction curve, the confidence interval (optimistic / pessimistic scenario), the peak year label, and the abandonment risk warning mark.

[0033] Step S6: Multi-dimensional technical intelligence analysis based on graph node aggregation Using the aforementioned two-layer map structure and stage identification results, multi-dimensional drill-down aggregation is performed on the technology topic clusters: Technological field popularity analysis: Statistics on the number of patents, total value, and development stage of each topic over the years to identify inflection points of emerging technologies; Geographical distribution and flow analysis: Heat maps are generated by theme-country / region matrix, combined with cross-border transfer paths using applicant change record tracking technology; Competitive landscape quantification: Calculate the patent concentration index (CRn, HHI) of applicants under each subject to identify core rights holders and potential cooperation / merger targets; Lifecycle dynamic tracking: Mapping the trajectory of a theme from its first appearance to its decline, and outputting early warning signals for technological alternatives based on evolutionary trends.

[0034] Step S7: Output and Visualization Interaction of Results.

[0035] The system integrates the evidence chain for stage identification, the current valuation range, the future forecast curve, and technical intelligence charts into an interactive dashboard; it supports multi-dimensional filtering, data drill-down, scenario simulation (optimistic / neutral / pessimistic), and standardized report export; all analysis results are connected to enterprise patent management systems, technology trading platforms, or financial institution risk control systems through standardized API interfaces.

[0036] The technical solutions described above in the embodiments of the present invention will be illustrated below.

[0037] Example 1: This embodiment uses the field of artificial intelligence chips as an example to illustrate the method of the present invention.

[0038] Step 1: Patent data collection and preprocessing.

[0039] By connecting to patent databases (such as the China National Intellectual Property Administration (CNIPA), the United States Patent and Trademark Office (USPTO), and the European Patent Office (EPO), patent data in the target technical field is obtained. The data collection scope includes: full text of the patent (title, abstract, claims, specification), IPC classification number (main classification number, subclassification number), citation relationships (list of cited patents, list of cited patents), legal status data (application date, publication date, grant date, annual fee payment record, rights transfer record), and bibliographic information (applicant, inventor, applicant's country / region).

[0040] Data preprocessing includes: deduplication (removing duplicate records based on patent number and application date), word segmentation and text chunking (segmenting the patent text into Chinese words and dividing it into paragraph chunks), and structured parsing (extracting structured fields and unstructured text). The preprocessed data is stored in two databases: a graph database (Neo4j) for storing the patent citation network and entity relationships; and a relational database (MySQL) for storing structured bibliographic data. This hybrid storage architecture provides a standardized data flow for subsequent graph construction.

[0041] Step 2: Construct a development map of the technology field.

[0042] (1) Fine-grained entity extraction: Information extraction is performed using large language models (such as GPT-4, Claude, etc.). Design a structured constraint prompt template: You are a patent analysis expert. Please extract technical entities and relationships from the following patent text. Technical entity types: technical terms, methods, materials, processes, application fields. Relationship types: reference, improvement, substitution, collaboration. Output format: JSON list, containing a list of entities and relation triples. Process the abstract and claims of each patent to obtain a list of technical entities and relation triples. For example, for a patent about "in-memory computing architecture", extract the entities: in-memory computing, GPU architecture, memory wall, near-memory computing; relationships: in-memory computing improves GPU architecture, near-memory computing replaces memory wall.

[0043] (2) Two-layer graph construction: All triples are aggregated into a heterogeneous graph: nodes represent technical entities, edges represent semantic relationships, each node is accompanied by the patent application date of its first appearance as a timestamp, and edges are accompanied by relationship type and time attributes. The temporal Louvain algorithm is used for clustering: Louvain community discovery is performed in each time window (e.g., 2015-2025, one window per year) to aggregate closely related nodes into technical theme clusters. The evolution of the same community in different time windows is tracked to form the temporal evolution trajectory of technical themes. Finally, a two-layer topological graph is formed: the bottom layer is the technical entity graph (fine-grained, 15,632 nodes, 38,921 edges); the top layer is the technical theme graph (aggregation granularity, 325 theme clusters). This conditional separation design of "fine representation during construction and theme aggregation during analysis" not only preserves the integrity of technical semantics but also reduces the computational complexity of graph analysis.

[0044] Step 3: Identify the technology lifecycle stage.

[0045] (1) Temporal feature extraction: For each technology theme node in the graph, calculate the following temporal evolution features: node degree growth rate g_t = (d_t - d_{t-1}) / d_{t-1}, reflecting the change in technology activity; neighborhood node change rate: the proportion of newly added neighborhood nodes to the total number of neighborhood nodes each year, reflecting the speed of technology branch expansion; community split index: track the splitting of the community to which the node belongs over time. If a community splits into multiple sub-communities, it is marked as technology fission; community integration index: if multiple communities merge into one, it is marked as technology integration.

[0046] (2) XGBoost Stage Recognition Model: A historical patent dataset (50,000 patents, 2015-2024) was constructed, with lifecycle stages (emergence / growth / maturity / decline) labeled by intellectual property experts as supervisory labels. The XGBoost classifier was used to train the temporal evolution feature set: the input was a feature vector [node degree growth rate, neighborhood change rate, community split index, community fusion index], and the output was stage label + confidence score. Training method: 5-fold cross-validation + grid search to optimize hyperparameters. Model performance: test set accuracy 91.3%, F1 score 0.89.

[0047] (3) Stage identification output: Input the feature vector of the target node into the trained XGBoost model, and output the technology life cycle stage label and identification confidence. For example: for the topic of in-memory computing technology, the output is the nascent stage (confidence 0.82); for the topic of GPU architecture technology, the output is the mature stage (confidence 0.91).

[0048] Step 4: Current value assessment.

[0049] (1) Patent positioning and technology stage determination: Semantically match the set of technical entities extracted from the target patent with the graph nodes to locate the technical theme node to which the patent belongs. Obtain the technology development stage label corresponding to the node. For example: a certain patent "An AI chip architecture based on in-memory computing" is matched with the technology theme "in-memory computing", and the stage label is growth stage.

[0050] (2) Dynamic weight configuration: Based on the identified technology stage, the corresponding evaluation index weight is selected from the preset "stage-weight" mapping table. This dynamic weighting mechanism based on the time separation principle solves the contradiction between "fixed weight and technological heterogeneity" in traditional methods.

[0051] (3) Indicator Calculation and Valuation Output: Calculate and normalize the indicator values ​​for each evaluation dimension (0-100 points), and obtain the basic score by weighted summation using stage-adaptive weights. Using historical transaction data from the same field, stage, and legal status as the prior distribution, generate a probabilistic valuation range through Monte Carlo simulation (10,000 times). Output: Current value score (e.g., 85 points) and confidence interval (e.g., 800,000-1,200,000 yuan, P25-P75).

[0052] Step 5: Future value prediction.

[0053] (1) LSTM time series prediction module: Construct a Long Short-Term Memory (LSTM) network model: The input is the historical degree sequence of technology topic nodes (window length 5 years), and the output is the predicted trend of the value for the next 5 years. This module mainly captures the time dependence and long-term trend of technology development.

[0054] (2) Cox Survival Analysis Module: Constructs a Cox proportional hazards regression model: Input features include years of existence, IPC classification, family size, renewal history, historical value score, etc. The output is the renewal probability curve for each year in the future, which is mapped to the value decay factor.

[0055] (3) Graph Evolution Simulation Module: Monitor the topological evolution of the graph to which the technology theme node belongs: When a highly similar alternative technology node or technology fusion event is identified, calculate the technology substitution risk index and convert it into the impact coefficient gamma (value range 0.5-0.8).

[0056] (4) Multi-model fusion output: The final prediction curve is generated using a stage-adaptive weighted formula: V_pred(t) = alpha_s V_LSTM(t) + beta_s V_Cox(t) gamma(t). Here, alpha_s and beta_s are stage-dependent weights: during the growth phase, alpha_s > beta_s (emphasizing trend extrapolation), and during the decline phase, beta_s > alpha_s (emphasizing survival decay). A value prediction curve for the next 5 years is generated, for example: 1.1 million in year 1, 1.5 million in year 3 (peak), and 900,000 in year 5 (due to decreased replacement risk). Simultaneously, a risk warning for abandoning subscriptions is output (if the renewal probability in a certain year is less than 30%).

[0057] Step 6: Technical intelligence analysis.

[0058] Based on the aggregation of graph nodes, the following multi-dimensional analysis is performed: (1) Technology heat analysis: Aggregate by technology theme nodes, count the number of patents, total value, average value, and development stage of each theme over the years, generate a heat trend chart (line graph) of the technology field, and identify the inflection point of emerging technologies. (2) Competitive landscape analysis: Aggregate by technology theme nodes and applicants, count the ranking of applicants, number of patents, and value contribution under each technology theme. Calculate market concentration indicators: CR4 (the total market share of the top 4 applicants) and HHI index (the squared market share of each applicant, reflecting market concentration). Identify core rights holders and potential cooperation / merger targets, and generate a bubble chart of the competitive landscape. (3) Geographical distribution analysis: Aggregate by technology theme nodes and applicant countries / regions, count the number and value distribution of patents for each technology theme in different countries, generate a technology theme-country heat map, and analyze the technological advantage areas of each country. (4) Dynamic tracking of technology life cycle: Track the development trajectory of technology from its first appearance to the present according to the technology theme node, record key time nodes (first appearance, entering the growth stage, entering the maturity stage, peak, entering the decline stage), generate technology life cycle trajectory map, and output technology substitution early warning signal based on evolution trend.

[0059] Step 7: Output the results.

[0060] The visualization system outputs the following results: technology development stage identification (including confidence level), current value assessment (score + valuation range), future value prediction curve (including optimistic / neutral / pessimistic scenarios), and technology intelligence analysis report (popularity trends, competitive landscape, geographical distribution, and lifecycle tracking). All analysis results are integrated with enterprise patent management systems, technology trading platforms, or financial institution risk control systems via standardized API interfaces.

[0061] Example 2: Technical intelligence analysis in the biomedical field: This example demonstrates the application of the present invention in the biomedical field, verifies the versatility of the method, and focuses on demonstrating the technical intelligence analysis function.

[0062] Step 1: Data collection.

[0063] Using the publicly available patent database of the United States Patent and Trademark Office (USPTO) as the data source, we obtained patent application data related to antibody drugs in the biopharmaceutical field from 2015 to 2025, totaling 12,500 applications.

[0064] Step 2: Map construction.

[0065] Using a large language model, technical entities were extracted, resulting in technical nodes including: antibody types (monoclonal antibodies, bispecific antibodies, antibody-drug conjugates); target proteins (PD-1, HER2, CD20); and technical methods (immune checkpoint inhibition, CAR-T). A technical atlas was constructed with 3,247 nodes, 7,892 edges, and 56 technical subject clusters.

[0066] Step 3: Stage identification.

[0067] Taking the node "bispecific antibody" as an example: the degree sequence is 2018:8, 2019:15, 2020:32, 2021:68, 2022:105, 2023:142, 2024:168, 2025:185; the average growth rate over the past three years is approximately 30.3%. XGBoost classifier output: the stage label is growth stage, with a confidence level of 0.94.

[0068] Steps 4-5: Assessment and prediction.

[0069] Evaluation results of the target patent "A bispecific antibody drug targeting HER2": Current value score 88 points, estimated value range [8.5 million, 12 million] RMB, peak year 2028, probability of abandonment <10% (before 2030).

[0070] Step 6: Technical intelligence analysis.

[0071] Technological Trends Analysis: Bispecific antibodies CAGR 55% (Growth Stage), Antibody-Drug Conjugates CAGR 30% (Growth Stage), Monoclonal Antibodies CAGR 5% (Maturity Stage), CAR-T CAGR 42% (Growth Stage). Geographic Distribution Analysis: In the bispecific antibody field, the US accounts for 52% and China for 22%; in the antibody-drug conjugate field, the US accounts for 45% and China for 30%. Competitive Landscape Analysis: In the bispecific antibody field, the CR4 (top four companies) is 48% (Roche, AbbVie, Pfizer, BeiGene); in the antibody-drug conjugate field, the CR4 is 35% (Daiichi Sankyo, AstraZeneca, Rongchang Bio).

[0072] Step 7: Output the results.

[0073] Compared with the prior art, the present invention has the following beneficial effects: 1. Dynamic Adaptive Evaluation: By using a technological development map, patents are located at fine-grained technical nodes, and evaluation indicators and weights are dynamically adjusted according to the technology lifecycle stage of each node. This solves the drawbacks of the "one-size-fits-all" evaluation method in existing approaches and improves evaluation accuracy. Small-scale experimental verification (500 AI-related patents) shows that the correlation coefficient between the evaluation results of this invention and the expert evaluation results reaches 89%, which is 17 percentage points higher than the fixed-weight model (76%).

[0074] 2. Fine-grained stage identification: This method utilizes a large language model to extract technical entities, constructs a semantically driven technology graph, and identifies stages through the temporal evolution features of nodes. Compared to traditional methods based on global application volume curves, it can reflect the differentiated evolution paths of different branches within the technical field, resulting in more accurate stage identification. In the field of artificial intelligence chips, this method successfully identified the nascent stage (2023) of the "in-memory computing" technology theme and the mature stage of the "GPU" technology theme, while the global method misjudged the entire AI chip field as being in its growth stage.

[0075] 3. Future Value Prediction Capability: Combining time-series forecasting, survival analysis, and graph evolution simulation, this method outputs a patent future value prediction curve and a risk warning for abandonment, providing a quantitative basis for decisions regarding patent maintenance, licensing, and abandonment. Experimental backtesting shows that the average absolute error of this method in predicting patent value changes over the next three years is 8.2%, which is superior to the comparative method using only survival analysis (12.5%).

[0076] 4. Multi-dimensional technology intelligence analysis capabilities: Based on data from segmented technology fields, in-depth analysis of technology popularity, geographical distribution, competitive landscape, and lifecycle tracking is achieved, upgrading patent data from an "assessment tool" to a "strategic decision-making platform." Through technology-country heat maps, enterprises can intuitively understand the shortcomings in the geographical layout of core technologies; through competitive landscape analysis, potential cooperation / merger targets can be identified; and through lifecycle tracking, the risk of technology substitution can be predicted, assisting in the adjustment of R&D resources.

[0077] 5. High interpretability: The output results include evidence of stage identification (node ​​degree growth rate, neighborhood changes, etc.) and sub-indicator scores, allowing users to trace the evaluation basis and enhancing the credibility of the evaluation results.

[0078] 6. High practicality: It can be widely used in patent transaction pricing, pledge financing evaluation, enterprise patent portfolio management, technology transfer decision-making, technology strategy planning, competitive intelligence analysis, investment due diligence and other scenarios, with significant economic and social benefits.

[0079] Device Example 1 According to embodiments of the present invention, a device for dynamic evaluation of patent value and analysis of technical information is provided. Figure 7 This is a schematic diagram of the patent value dynamic assessment and technical intelligence analysis device according to an embodiment of the present invention, such as... Figure 7 As shown, the patent value dynamic assessment and technical information analysis device according to an embodiment of the present invention specifically includes: The acquisition module 70 is used to acquire multi-source patent data, and to preprocess the multi-source patent data using a distributed computing framework to acquire standardized patent data. Based on the standardized patent data, a two-layer topology map is constructed using a conditional separation strategy. The technology life cycle stage identification module 72 is used to calculate the temporal evolution feature set of each node of the two-layer topology map, construct a historical patent dataset, train the temporal evolution feature set, generate a patent stage identification model, input the feature vector of the target node into the patent stage identification model, and output the technology life cycle stage label and identification confidence. The value assessment module 74 is used to semantically match the entity set extracted from the target patent with the graph nodes of the two-layer topology graph, locate the main associated nodes and the corresponding technology life cycle stage labels, establish a dynamic mapping table of "patent technology stage - assessment dimension weight", configure weights differently according to time separation logic, calculate and normalize the index values ​​of each assessment dimension, obtain the basic score by stage adaptive weight weighting and summing, use historical transaction data as the distribution prior, generate a probabilistic valuation interval through Monte Carlo simulation, and output the current value score and confidence interval. The patent value prediction module 76 is used to perform sliding window training based on the node's historical degree sequence and neighborhood size, outputting a future basic value trend curve, and outputting an annual renewal probability curve based on the existing years, family size, renewal records, and historical value scores, which is mapped to a value decay factor; it monitors the topological evolution of the technology cluster to which the main associated node of the two-layer topological graph belongs, and when a highly similar replacement node or technology fusion event is identified, it calculates a technology replacement risk index and converts it into an impact coefficient. Based on the future basic value trend curve, the value decay factor, and the impact coefficient, it uses a stage-adaptive weighted formula to generate a future patent value prediction curve. The result output module 78 is used to perform multi-dimensional drill-down aggregation on the technology theme cluster based on the structure of the two-layer topology map and the future patent value prediction curve, and to integrate the stage identification evidence chain, the current patent value valuation range, the technology life cycle stage label and the technology intelligence chart into the analysis results and output them.

[0080] Figure 8 This is a schematic diagram of the system structure of the device according to an embodiment of the present invention, as shown below. Figure 8As shown, the data flow between modules including data acquisition, graph construction, stage identification, current valuation, future prediction, technical intelligence analysis, and output is illustrated. ① represents the data acquisition layer, ② the graph construction layer, ③ the analysis engine layer, ④ the application service layer, and ⑤ the user system layer. The vertical arrows indicate the data flow between layers. The modules within each layer process data sequentially from left to right. Each layer has four modules, which correspond one-to-one, forming a clear data processing pipeline.

[0081] The embodiments of the present invention are device embodiments corresponding to the above method embodiments. The specific operation of each module can be understood with reference to the description of the method embodiments, and will not be repeated here.

[0082] Device Example 2 This invention provides an electronic device, such as... Figure 9 As shown, it includes: a memory 90, a processor 92, and a computer program stored on the memory 90 and executable on the processor 92, wherein the computer program, when executed by the processor 92, performs the steps as described in the method embodiment.

[0083] Device Example 3 This invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor 92, performs the steps described in the method embodiment.

[0084] The computer-readable storage media described in this embodiment include, but are not limited to, ROM, RAM, disk, or optical disk.

[0085] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for dynamic evaluation of patent value and analysis of technical intelligence, characterized in that, include: Acquire multi-source patent data, preprocess the multi-source patent data using a distributed computing framework to obtain standardized patent data, and construct a two-layer topology map based on the standardized patent data using a conditional separation strategy. Calculate the temporal evolution feature set of each node in the two-layer topology map, construct a historical patent dataset, train the temporal evolution feature set, generate a patent stage identification model, input the feature vector of the target node into the patent stage identification model, and output the technology life cycle stage label and identification confidence. The entity set extracted from the target patent is semantically matched with the graph nodes of the two-layer topology graph to locate the main associated nodes and the corresponding technology life cycle stage labels. A dynamic mapping table of "patent technology stage - evaluation dimension weight" is established. The weights are configured differently according to the time separation logic. The index values ​​of each evaluation dimension are calculated and normalized. The basic score is obtained by stage adaptive weight weighting and summing. With historical transaction data as the distribution prior, a probabilistic valuation interval is generated through Monte Carlo simulation. The current value score and confidence interval are output. An LSTM model is employed, trained via a sliding window based on the node's historical degree sequence and neighborhood size, to output a future basic value trend curve. Based on the existing years of existence, family size, renewal records, and historical value scores, an annual renewal probability curve is output, mapped to a value decay factor. The topological evolution of the technology cluster to which the main associated nodes of the two-layer topological graph belong is monitored. When highly similar replacement nodes or technology fusion events are identified, a technology replacement risk index is calculated and converted into an impact coefficient. Based on the future basic value trend curve, the value decay factor, and the impact coefficient, a stage-adaptive weighted formula is used to generate a future patent value prediction curve. Based on the structure of the dual-layer topology map and the future patent value prediction curve, multi-dimensional drill-down aggregation is performed on the technology theme clusters, integrating the stage identification evidence chain, the current patent value valuation range, the technology life cycle stage labels, and the technology intelligence charts into analysis results and outputting them.

2. The method according to claim 1, characterized in that, Acquiring multi-source patent data and preprocessing it using a distributed computing framework to obtain standardized patent data, and then constructing a two-layer topology map based on the standardized patent data using a conditional separation strategy, specifically includes: By connecting to a multi-source patent database through an API interface, the system collects full-text patents, bibliographic data, legal status, and citation network data in the target technology field. A distributed computing framework is used to perform deduplication, noise filtering, text segmentation, and structured parsing on the multi-source patent data. The unstructured multi-source patent data and structured metadata are stored in a hybrid storage architecture of graph database and relational database. Based on pre-set structured constraint prompt word templates, the large language model is guided to output technical terms, methods, materials, processes and application domain entities according to the preset JSON Schema, and to identify four types of semantic relations: reference, improvement, substitution and collaboration, and generate standardized triples; All triples are aggregated into a heterogeneous graph, with nodes bearing a timestamp of their first appearance and edges bearing relation type and time attributes. An improved temporal Louvain community discovery algorithm is used to dynamically aggregate and generate technical topic clusters according to a sliding window of predetermined time length while preserving the underlying fine-grained nodes.

3. The method according to claim 1, characterized in that, Calculate the temporal evolution feature set of each node in the two-layer topology graph, construct a historical patent dataset, train the temporal evolution feature set to generate a patent stage identification model, input the feature vector of the target node into the patent stage identification model, and output the technology life cycle stage label and identification confidence score, specifically including: Calculate the temporal evolution feature set of each node in the two-layer topology graph, wherein the temporal evolution feature set specifically includes: node degree sequence, annual growth rate, neighboring node change rate, and community split / fusion index; A historical patent dataset is constructed, with lifecycle stages pre-annotated by intellectual property experts as supervision labels. The XGBoost classifier is used to train the temporal evolution feature set. K-fold cross-validation and grid search are used to determine the decision boundary function and confidence output mapping for each stage, thereby generating a patent stage identification model. The feature vector of the target node is input into the trained patent stage identification model, and the patent stage identification model outputs the technology life cycle stage label and identification confidence. The technology life cycle stage label specifically includes: budding stage, growth stage, maturity stage and decline stage.

4. The method according to claim 1, characterized in that, The entity set extracted from the target patent is semantically matched with the graph nodes of the two-layer topology graph to locate the main associated nodes and their corresponding technology life cycle stage labels. A dynamic mapping table of "patent technology stage - evaluation dimension weight" is established. Weights are configured differently according to the time separation logic. The index values ​​of each evaluation dimension are calculated and normalized. A basic score is obtained by weighted summation using stage adaptive weights. Using historical transaction data as a distribution prior, a probabilistic valuation interval is generated through Monte Carlo simulation. The current value score and confidence interval are output, including: The entity set extracted from the target patent is semantically matched with the graph nodes of the two-layer topology graph to locate the main associated node and the corresponding technology life cycle stage label. Establish a dynamic mapping table of "patent technology stage - evaluation dimension weight", and configure weights differently according to the time separation logic: increase the weight of technological foresight and market potential in the budding / growth stage, and dynamically increase the weight of legal stability, renewal willingness and synergistic value in the maturity / decline stage. The values ​​of each evaluation dimension are calculated and normalized, and a basic score is obtained by weighted summation using stage-adaptive weights. Using historical transaction data from the same field, stage, and legal status as the prior, a probabilistic valuation range is generated through Monte Carlo simulation, and the current value score and confidence interval are output.

5. The method according to claim 1, characterized in that, An LSTM model is employed, trained via a sliding window based on the node's historical degree sequence and neighborhood size, to output a future basic value trend curve. Based on the existing years of existence, family size, renewal records, and historical value scores, an annual renewal probability curve is output, mapped to a value decay factor. The topological evolution of the technology cluster to which the main associated nodes of the two-layer topological graph belong is monitored. When highly similar replacement nodes or technology fusion events are identified, a technology replacement risk index is calculated and converted into an impact coefficient. Specifically, this includes: The LSTM model is used, with the node's historical degree sequence and neighborhood size as input, and is trained with a sliding window to output the future basic value trend curve. Cox survival analysis is used to output an annual renewal probability curve based on the number of years the product has been sold, the size of the family, the renewal history, and the historical value score, which is then mapped to a value decay factor. Monitor the topological evolution of the technology cluster to which the main associated node belongs. When a high-similarity replacement node or technology fusion event is identified, calculate the technology replacement risk index and convert it into an impact coefficient. The high-similarity replacement node is defined as two nodes whose similarity is greater than or equal to a predetermined similarity threshold.

6. The method according to claim 5, characterized in that, Based on the aforementioned future fundamental value trend curve, the aforementioned value decay factor, and the aforementioned impact coefficient, the generation of the future patent value prediction curve using a stage-adaptive weighted formula specifically includes: Based on the future fundamental value trend curve, the value decay factor, and the impact coefficient, a future patent value prediction curve is generated using a stage-adaptive weighted formula as shown in Formula 1. Official 1; in, For stage-dependent weights, This represents the future fundamental value trend curve. This represents the value decay component obtained by mapping renewal probability. This represents the impact coefficient.

7. The method according to claim 1, characterized in that, Based on the structure of the aforementioned two-layer topology map and the future patent value prediction curve, multi-dimensional drill-down aggregation is performed on the technology theme clusters. The stage identification evidence chain, current patent value estimation range, technology lifecycle stage labels, and technology intelligence charts are integrated into the analysis results, which are specifically output as follows: Based on the structure of the aforementioned two-layer topology map and the aforementioned future patent value prediction curve, multi-dimensional drill-down aggregation is performed on technology theme clusters to conduct technology field heat analysis, regional distribution and flow analysis, competitive landscape quantification, and life cycle dynamic tracking. Specifically, the technology field heat analysis includes: statistically analyzing the number of patents, total value, and development stage percentage for each theme over the years to identify emerging technology inflection points; the regional distribution and flow analysis includes: generating heat maps by theme-country / region matrix and tracking cross-border technology transfer paths by combining applicant change records; the competitive landscape quantification includes: calculating the patent concentration index of applicants under each theme to identify core rights holders and potential cooperation / merger targets; and the life cycle dynamic tracking includes: mapping the key nodes of a theme from its initial appearance to its decline, and outputting technology substitution early warning signals based on evolution trends. The system integrates the evidence chain for stage identification, the current valuation range, the future value prediction curve, and technical intelligence charts into an interactive dashboard. The interactive dashboard provides multi-dimensional filtering, data drill-down, scenario simulation, and standardized report export. All analysis results are connected to enterprise patent management systems, technology trading platforms, or financial institution risk control systems through standardized API interfaces.

8. A device for dynamic evaluation of patent value and analysis of technical information, characterized in that, include: A construction module is used to acquire multi-source patent data and preprocess the multi-source patent data using a distributed computing framework to acquire standardized patent data. Based on the standardized patent data, a two-layer topology map is constructed using a conditional separation strategy. The technology life cycle stage identification module is used to calculate the temporal evolution feature set of each node of the two-layer topology map, construct a historical patent dataset, train the temporal evolution feature set, generate a patent stage identification model, input the feature vector of the target node into the patent stage identification model, and output the technology life cycle stage label and identification confidence. The value assessment module is used to semantically match the entity set extracted from the target patent with the graph nodes of the two-layer topology graph, locate the main associated nodes and the corresponding technology life cycle stage labels, establish a dynamic mapping table of "patent technology stage - assessment dimension weight", configure weights differently according to time separation logic, calculate and normalize the index values ​​of each assessment dimension, obtain the basic score by stage adaptive weight weighting and summing, use historical transaction data as the distribution prior, generate probabilistic valuation intervals through Monte Carlo simulation, and output the current value score and confidence interval. The patent value prediction module is used to perform sliding window training based on the node's historical degree sequence and neighborhood size, outputting a future basic value trend curve. Based on the existing years of existence, family size, renewal records, and historical value scores, it outputs an annual renewal probability curve, which is mapped to a value decay factor. It monitors the topological evolution of the technology cluster to which the main associated nodes of the two-layer topological graph belong. When a highly similar replacement node or technology fusion event is identified, it calculates a technology replacement risk index and converts it into an impact coefficient. Based on the future basic value trend curve, the value decay factor, and the impact coefficient, it uses a stage-adaptive weighted formula to generate a future patent value prediction curve. The results output module is used to perform multi-dimensional drill-down aggregation on the technology topic cluster based on the structure of the two-layer topology map and the future patent value prediction curve, and integrate the stage identification evidence chain, the current patent value valuation range, the technology life cycle stage label and the technology intelligence chart into the analysis results and output them.

9. An electronic device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the method for dynamic evaluation of patent value and analysis of technical information as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the patent value dynamic assessment and technical intelligence analysis method as described in any one of claims 1 to 7.

Images