Scientific and technological achievement value estimation system

Abstract

The invention discloses a scientific and technological achievement value estimation system, and the system comprises a data collection module which is configured to obtain technical data, market data, legal data and social data through an API interface; the feature fusion module is configured to adopt a heterogeneous data converter to convert the multi-source heterogeneous data into a unified feature space, and the feature fusion module comprises a text data conversion unit, a numerical data conversion unit and a time sequence data conversion unit; the correlation analysis module is used for constructing a cross-domain knowledge graph; the dynamic evaluation module is configured to adopt an attention mechanism weight distribution model; by integrating the multi-source heterogeneous data acquisition module, the system can comprehensively collect information from technologies, markets, laws and society, limitation of a single data source is avoided, the cross-modal alignment unit realizes alignment of multi-source data spaces through comparative learning, information consistency among different data sources is ensured, and the accuracy of the multi-source data space alignment is improved. And the reliability of the evaluation result is further enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and specifically to a scientific and technological achievement value estimation system. Background Art

[0002] In the context of innovation-driven development of science and technology, the transformation efficiency of scientific and technological achievements and the accuracy of value assessment have become key factors affecting the process of technology industrialization. In the existing technology, most scientific and technological achievement value estimation systems are based on static data models or single-dimensional evaluation methods, and they have significant defects in data integration, dynamic adaptability, multi-factor collaborative analysis, etc. The specific problems are as follows:

[0003] Existing systems usually rely on limited expert experience or historical data (such as the number of patents, the number of paper citations), lacking the ability to integrate multi-source heterogeneous data of market dynamics (such as technology supply and demand, competition pattern), legal risks (such as the stability of intellectual property rights), and social benefits (such as environmental impact). The phenomenon of data islands leads to one-sided evaluation results and it is difficult to reflect the potential value of the achievements. Scientific and technological achievement data are scattered in independent modules such as technology, market, and law. For example, the technology module may only analyze patent texts and laboratory data, while the market module relies on industry reports, and the two are not associated through algorithms, resulting in a single evaluation dimension. Summary of the Invention

[0004] The purpose of the present invention is to provide a scientific and technological achievement value estimation system to solve the problems raised in the above background art.

[0005] To achieve the above purpose, the present invention provides the following technical solution: A scientific and technological achievement value estimation system, which includes a data acquisition module configured to obtain technical data, market data, legal data, and social data through an API interface;

[0006] A feature fusion module configured to convert multi-source heterogeneous data into a unified feature space by using a heterogeneous data converter. The feature fusion module includes a text data conversion unit, a numerical data conversion unit, and a time series data conversion unit. The text data conversion unit uses a BERT model to generate text embedding vectors; the numerical data conversion unit performs Z-score normalization processing; the time series data conversion unit extracts time series features through an LSTM network;

[0007] An association analysis module that constructs a cross-domain knowledge graph, whose nodes include technical entities, market entities, and legal entities, and the edge weights reflect the dynamic association relationships between entities;

[0008] A dynamic evaluation module configured to adopt an attention mechanism weight distribution model, and the calculation formula is:

[0009] W t = Softmax(MLP([Tt ; M t ; L t ; S t ))

[0010] Among them, T t , M t , M t , L t , S t respectively represent the technology, market, legal and social feature vectors.

[0011] Preferably, the data acquisition module includes a technology data acquisition unit, a market data acquisition unit, a legal data acquisition unit and a social data acquisition unit;

[0012] Among them, the technology data acquisition unit is connected to the patent database, the SCI paper database and the technology maturity report interface; the market data acquisition unit integrates the supply chain data API and the technology transaction platform interface; the legal data acquisition unit is docked with the intellectual property litigation database and the patent reexamination document database; the social data acquisition unit obtains the environmental assessment report and the policy text data stream.

[0013] Preferably, the feature fusion module further includes a cross-modal alignment unit, which realizes the alignment of multi-source data spaces through contrastive learning, and its loss function is:

[0014] L = max(0, d(p, q+) - d(p, q-) + margin)

[0015] Among them, p is the patent feature vector, q+ is the associated market data vector, q- is the irrelevant data vector, and margin is the preset boundary threshold.

[0016] Preferably, in the association analysis module, the technology entity is encoded using the patent IPC classification number; the market entity is labeled using the industry classification code; the legal entity is vectorized and represented by the lawsuit cause keywords; the edge weight is dynamically updated according to the technology transfer record, and the calculation formula is:

[0017]

[0018] Among them, λ is the attenuation factor, which controls the retention ratio of the historical weight; is the association weight from entity i to j at time t; is the historical weight from entity i to entity j in the previous time period; is the technology transfer frequency from entity i to entity j at time t.

[0019] Preferably, the dynamic evaluation module further includes a spatio-temporal graph convolutional network unit, where the node features include patent citation frequency, litigation risk index, and technology substitution rate; and a dynamic risk assessment subunit configured to perform risk early warning based on the spatio-temporal propagation of node features.

[0020] Preferably, the dynamic evaluation module further includes a social benefit quantification unit that constructs an environment-technology association matrix and calculates the carbon footprint impact factor through input-output analysis:

[0021]

[0022] where α i is the technology diffusion coefficient, and TechCoeff i is the industry carbon emission factor, which is solved by the Leontief inverse matrix method.

[0023] Preferably, the output interface of the dynamic evaluation module includes a three-dimensional radar chart and a time-axis prediction component;

[0024] wherein, the three-dimensional radar chart synchronously displays the evaluation values of four dimensions: technical feasibility, commercial value, legal risk, and social benefit; the time-axis prediction component generates a value trend curve based on the LSTM time series prediction model.

[0025] Preferably, the dynamic evaluation method of the scientific and technological achievement value estimation system includes the following steps:

[0026] S1: Perform feature space unification processing on multi-source data through a heterogeneous data converter;

[0027] S2: Execute the Personalized PageRank algorithm in the cross-domain knowledge graph to mine potential technical associations:

[0028] PPR(u) = (1 - α) × M × PPR(u) + α × v

[0029] where M is the transition matrix, v is the initial distribution vector, and α is the damping factor;

[0030] S3: Dynamically adjust the evaluation weight coefficient. When a sudden policy event is detected, automatically increase the weight proportion of the social benefit dimension.

[0031] Preferably, the data acquisition module further includes a data cleaning and verification unit for denoising, missing value processing, and data consistency verification of the obtained technical data, market data, legal data, and social data to ensure data quality.

[0032] Preferably, the scientific and technological achievement value estimation system further includes a user interaction interface, which provides a customized query function and allows users to filter technical fields, market scopes, legal environments, and social impact parameters according to specific needs.

[0033] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0034] 1. By integrating the multi-source heterogeneous data acquisition module, the system can comprehensively collect information from technology, market, law, and society, avoiding the limitations of a single data source. The cross-modal alignment unit realizes the alignment of multi-source data spaces through contrastive learning, ensuring the information consistency between different data sources and further enhancing the reliability of the evaluation results.

[0035] 2. Through the dynamic evaluation module combined with the attention mechanism and spatio-temporal graph convolutional network, the present invention can dynamically adjust the evaluation weights according to real-time data and timely discover potential risks through the risk warning system, making the evaluation results more accurate and timely. The social benefit quantification unit provides important indicators for decision-makers to measure the social impact of technologies by constructing an environment-technology correlation matrix and calculating the carbon footprint impact factor. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a system composition block diagram of the scientific and technological achievement value estimation system of the present invention.

[0037] Figure 2 It is a flowchart of the processing method of the scientific and technological achievement value estimation system of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0038] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0039] Please refer to Figures 1 to 2 , the present invention provides a technical solution: a scientific and technological achievement value estimation system, which includes a data acquisition module configured to obtain technical data, market data, legal data, and social data through an API interface;

[0040] The data acquisition module includes a technical data acquisition unit, a market data acquisition unit, a legal data acquisition unit, and a social data acquisition unit;

[0041] Among them, the technical data collection unit accesses the Derwent patent database interface to synchronize the national patent publication data hourly; connects to the Web of Science API to obtain the citation network data of SCI papers; integrates a technology maturity assessment report parser to automatically extract technical feasibility parameters;

[0042] The market data collection unit constructs a competitive intelligence monitoring system through the Scrapy framework to monitor the transaction records of the technology trading platform; connects to the cloud market API to obtain the information on the record-filing of technology license contracts;

[0043] The legal data collection unit accesses the open interface of the National Judgment Documents Network to update the intellectual property litigation case data daily; constructs a patent reexamination document parsing engine to extract the key information of the claim amendment records and invalidation decision documents;

[0044] The social data collection unit establishes a policy text stream processing pipeline to capture the policy documents issued by the Ministry of Ecology and Environment in real time; uses OCR technology to analyze the pollutant emission indicators in the environmental assessment reports.

[0045] The data collection module also includes a data cleaning and verification unit, which is used to perform denoising, missing value processing, and data consistency verification on the obtained technical data, market data, legal data, and social data to ensure data quality.

[0046] Use the Z-score analysis and statistical method to identify outliers in the data, apply the moving average or exponential smoothing method to time series data to reduce the impact of random fluctuations. For text data, remove irrelevant characters, stop words, and eliminate format errors through regular expressions; if the missing data is not much, consider directly deleting the records containing missing values. Use the methods of mean, median, and mode to fill in the missing values; check whether the data values are within a reasonable range. For example, the patent application year should not be earlier than the implementation time of the Patent Law, and verify whether the data fields conform to the business logic. For example, the number of patents in a certain technical field should not be greater than the total number of applications in that field.

[0047] The feature fusion module is configured to convert multi-source heterogeneous data into a unified feature space by using a heterogeneous data converter. The feature fusion module includes a text data conversion unit, a numerical data conversion unit, and a time series data conversion unit;

[0048] The text data conversion unit uses the pre-trained SciBERT model to tokenize the patent abstract and claims to generate 768-dimensional text embedding vectors, and uses the TF-IDF weighted algorithm to extract keyword features from the government policy text.

[0049] The numerical data conversion unit performs Z-score standardization on numerical data such as technology transaction amount and patent citation times. The formula is:

[0050]

[0051] Among them, μ is the mean of historical data. σ is the standard deviation.

[0052] The time-series data conversion unit constructs a bidirectional LSTM network. The input layer is set with 128 neurons to extract long-term and short-term dependence features from the time-series data of technology transfer frequency and market volatility index, and outputs a 32-dimensional time-series feature vector.

[0053] The feature fusion module also includes a cross-modal alignment unit, which realizes the spatial alignment of multi-source data through contrastive learning. Its loss function is:

[0054] L = max(0, d(p, q+) - d(p, q-) + margin)

[0055] Among them, p is the patent feature vector, q+ is the associated market data vector, q- is the irrelevant data vector, and margin is the preset boundary threshold.

[0056] Design a contrastive learning loss function. Using the patent feature vector p as the anchor point, the associated market data vector q+ as the positive sample, and randomly sampling the irrelevant data vector q- as the negative sample. The boundary threshold margin is set to 0.5. Through Triplet Loss training, the distance of the positive sample pair is reduced, and the distance of the negative sample pair is increased.

[0057] The association analysis module constructs a cross-domain knowledge graph. Its nodes include technology entities, market entities, and legal entities. The edge weights reflect the dynamic association relationships between entities;

[0058] In the association analysis module, technology entities are encoded using patent IPC classification numbers. For example, "H04L" represents digital information transmission technology; market entities are labeled using industry classification codes, such as "S104022" representing the 5G communication equipment manufacturing industry; legal entities are represented by vectorizing lawsuit case keywords; the edge weights are dynamically updated according to technology transfer records, and the calculation formula is:

[0059]

[0060] Among them, λ is the attenuation factor, which controls the retention ratio of historical weights; is the association weight from entity i to j at time t; is the historical weight from entity i to entity j in the previous time period; is the technology transfer frequency from entity i to entity j at time t.

[0061] The dynamic evaluation module is configured to adopt an attention mechanism weight assignment model, and the calculation formula is:

[0062] W t = Softmax(MLP([T t ; M t ; L t ; S t ))

[0063] where T t , M t , M t , L t , S t represent the technology, market, legal, and social feature vectors respectively; when a sudden policy event is detected, the weight of the social benefit dimension is automatically increased by 20%.

[0064] The dynamic evaluation module also includes a spatio-temporal graph convolutional network unit, and the node features include patent citation frequency, litigation risk index, and technology substitution rate; the litigation risk index is measured by the number of litigation cases or the patent family size, and the technology substitution rate is equivalent to the growth rate of competitor patents;

[0065] The dynamic risk assessment subunit is configured to perform risk early warning based on the spatio-temporal propagation of node features, such as showing red for high litigation risk areas, orange for technology substitution hotspots, and blue for policy-sensitive areas.

[0066] The dynamic evaluation module also includes a social benefit quantification unit, which executes to construct an environment-technology association matrix and calculates the carbon footprint impact factor through the input-output analysis method:

[0067]

[0068] where α i is the technology diffusion coefficient, and TechCoeff i is the industry carbon emission factor, which is solved by the Leontief inverse matrix method.

[0069] The output interface of the dynamic evaluation module includes a three-dimensional radar chart and a time-axis prediction component;

[0070] wherein, the three-dimensional radar chart synchronously displays the evaluation values of the four dimensions of technical feasibility, commercial value, legal risk, and social benefit; the time-axis prediction component generates a value trend curve based on the LSTM time series prediction model, inputs the technical transaction data of the past 12 months, and outputs the value trend curve of the next 6 months, and the confidence interval is set to 95%.

[0071] The dynamic evaluation method of the scientific and technological achievement value estimation system includes the following steps:

[0072] S1: Perform feature space unification processing on multi-source data through a heterogeneous data converter;

[0073] S2: Execute the Personalized PageRank algorithm in the cross - domain knowledge graph to mine potential technical associations:

[0074] PPR(u)=(1 - α)×M×PPR(u)+α×v

[0075] Where M is the transition matrix, v is the initial distribution vector, and α is the damping factor;

[0076] S3: Dynamically adjust the evaluation weight coefficients. When a sudden policy event is detected, automatically increase the weight ratio of the social benefit dimension.

[0077] The scientific and technological achievement value prediction system also includes a user interaction interface. The interface provides a customized query function, allowing users to filter technical fields, market scope, legal environment, and social impact parameters according to specific needs.

[0078] In actual use, assume that we are evaluating the market value and social impact of a new emerging environmental protection technology, such as a new type of high - efficiency solar panel.

[0079] S1: Perform feature space unification processing on multi - source data through a heterogeneous data converter;

[0080] Technical data: Collect indicators such as the efficiency, cost, and durability of the solar panel.

[0081] Market data: Obtain the market demand, competitor situation, and potential market size of the panel.

[0082] Legal data: Analyze the patent protection scope, legal litigation situation, and the impact of environmental protection regulations on the technology.

[0083] Social data: Evaluate the public acceptance of the technology, the evaluation of environmental protection organizations, and the policy support situation.

[0084] These data are processed by the converter, such as converting text data into vectors, normalizing numerical data, and extracting features from time - series data, and finally forming a unified data feature space.

[0085] S2: Execute the Personalized PageRank algorithm in the cross - domain knowledge graph to mine potential technical associations;

[0086] Construct a knowledge graph, where the nodes include technical entities (such as patents, papers), market entities (such as enterprises, products), legal entities (such as legal provisions, litigation cases), and social entities (such as environmental protection organizations, policies).

[0087] Define the transition matrix M, where the element M[i][j] represents the transition probability from node i to node j;

[0088] Set the initial distribution vector v, which reflects the initial importance or attention of each node;

[0089] Set the damping factor α (for example, α = 0.85), execute the Personalized PageRank algorithm, and iteratively calculate the ranking of each node to discover potential technical associations.

[0090] S3: Dynamically adjust the evaluation weight coefficient;

[0091] If during the evaluation process, the government issues a new environmental protection subsidy policy that directly affects the market prospect of this solar panel, and the system detects this sudden policy event, it automatically increases the weight ratio of the social benefit dimension. For example, it raises the weight of the social benefit dimension from the original 20% to 30%. After re-evaluation, the market value and social impact score of this solar panel are significantly improved, indicating that this technology has good development prospects.

[0092] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A scientific and technological achievement value estimation system, characterized by: The forecasting system includes a data acquisition module configured to obtain technical data, market data, legal data, and social data through an API interface; A feature fusion module is configured to use a heterogeneous data converter to convert multi-source heterogeneous data into a unified feature space. The feature fusion module includes a text data conversion unit, a numerical data conversion unit, and a time series data conversion unit. The text data conversion unit uses a BERT model to generate a text embedding vector; the numerical data conversion unit performs Z-score standardization processing; and the time series data conversion unit extracts time series features through an LSTM network; The association analysis module builds a cross-domain knowledge graph whose nodes include technical entities, market entities, and legal entities, and the edge weights reflect the dynamic associations between entities; The dynamic evaluation module is configured to adopt the attention mechanism weight distribution model, and the calculation formula is: W t =Softmax(MLP([T t ;M t ;L t ;S t ])) Where T t , M t , M t , L t , S t Represent the technical, market, legal and social feature vectors respectively.

2. A scientific and technological achievement value estimation system according to claim 1, characterized in that: The data collection module includes a technical data collection unit, a market data collection unit, a legal data collection unit, and a social data collection unit; Among them, the technical data collection unit connects to the patent database, SCI paper library and technology maturity report interface; the market data collection unit integrates the supply chain data API and the technology trading platform interface; the legal data collection unit connects to the intellectual property litigation database and the patent review document database; the social data collection unit obtains the environmental assessment report and policy text data stream.

3. A scientific and technological achievement value estimation system according to claim 1, characterized in that: The feature fusion module also includes a cross-modal alignment unit, which realizes multi-source data spatial alignment through contrastive learning, and its loss function is: L=max(0,d(p,q+)-d(p,q-)+margin) Among them, p is the patent feature vector, q+ is the associated market data vector, q- is the irrelevant data vector, and margin is the preset boundary threshold.

4. A scientific and technological achievement value estimation system according to claim 1, characterized in that: In the association analysis module, the technical entity is encoded using the patent IPC classification number; the market entity is annotated using the industry classification code; the legal entity is vectorized by the litigation case keyword; the edge weight is dynamically updated according to the technology transfer record, and the calculation formula is: Among them, λ is the decay factor, which controls the retention ratio of historical weights; is the association weight from entity i to j in period t; is the historical weight from entity i to entity j in the previous period; is the frequency of technology transfer from entity i to entity j in period t.

5. A scientific and technological achievement value estimation system according to claim 1, characterized in that: The dynamic assessment module also includes a spatiotemporal graph convolutional network unit, whose node features include patent citations, litigation risk index and technology substitution rate; a dynamic risk assessment subunit, which is configured to provide risk warning based on the spatiotemporal propagation of node features.

6. A scientific and technological achievement value estimation system according to claim 1, characterized in that: The dynamic evaluation module also includes a social benefit quantification unit, which constructs an environment-technology association matrix and calculates carbon footprint impact factors through input-output analysis: where α i is the technology diffusion coefficient, TechCoeff i is the industry carbon emission factor, which is solved by the Leontief inverse matrix method.

7. A scientific and technological achievement value estimation system according to claim 1, characterized in that: The output interface of the dynamic evaluation module includes a three-dimensional radar chart and a time axis prediction component; Among them, the three-dimensional radar chart simultaneously displays the evaluation values ​​of four dimensions: technical feasibility, commercial value, legal risks and social benefits; the timeline prediction component generates a value trend curve based on the LSTM time series prediction model.

8. A scientific and technological achievement value estimation system according to claim 1, characterized in that: The dynamic evaluation method of the scientific and technological achievement value estimation system comprises the following steps: S1: Unify the feature space of multi-source data through heterogeneous data converters; S2: Execute the Personalized PageRank algorithm in the cross-domain knowledge graph to mine potential technical associations: PPR(u)=(1-α)×M×PPR(u)+α×v Where M is the transfer matrix, v is the initial distribution vector, and α is the damping factor; S3: Dynamically adjust the evaluation weight coefficient. When a sudden policy event is detected, the weight of the social benefit dimension will be automatically increased.

9. A scientific and technological achievement value estimation system according to claim 1, characterized in that: The data collection module also includes a data cleaning and verification unit for performing denoising, missing value processing and data consistency verification on the acquired technical data, market data, legal data and social data to ensure data quality.

10. The scientific and technological achievement value estimation system according to claim 1, characterized in that: The scientific and technological achievement value estimation system also includes a user interaction interface, which provides a customized query function, allowing users to filter technical fields, market scope, legal environment and social impact parameters according to specific needs.

Images