Deep learning-based intelligent screening and evaluation method and system for innovation and entrepreneurship projects

Abstract

The invention relates to an innovation and entrepreneurship project intelligent screening evaluation method and system based on deep learning. The method comprises the following steps: acquiring multi-source data through a government policy platform, a news aggregation service, a social media stream and a technical monitoring station; performing data enhancement on the multi-source data through a B LSTM-CRF model to obtain enhanced structured data; extracting emergency data from the enhanced structured data to obtain a classified event set; based on a project knowledge base and the classification event set, calculating association between the emergencies and innovation and entrepreneurship projects to obtain a project association network graph; based on the project association network diagram, the influence of the emergency on the innovation and entrepreneurship project is quantified, and a dynamic influence coefficient matrix is obtained; and screening the project based on the dynamic influence coefficient matrix to obtain an intelligent screening report of the innovation and entrepreneurship project. By adopting the method, external information can be monitored in real time, emergencies can be accurately identified, and project value evaluation results can be dynamically adjusted.

Description

Technical Field

[0001] The present invention belongs to the field of financial information technology, and in particular relates to a method and system for intelligent screening and evaluation of innovative and entrepreneurial projects based on deep learning. Background Art

[0002] With the increasing application of information technology in financial venture capital, innovative and entrepreneurial project evaluation methods have emerged within the industry. However, existing evaluation methods mostly rely on a combination of static models and manual intervention to screen projects: first, fixed evaluation indicators are set based on historical financial data and industry benchmarks, followed by qualitative risk analysis through expert meetings, and then supplemented with external information through regularly updated industry reports and risk event libraries. The core problem with existing technologies lies in their rigid architecture, which makes them difficult to adapt to dynamic changes in the external environment. This is particularly true in the face of emergencies, as existing technologies are unable to respond promptly to emergencies, resulting in lags and low accuracy in model-based evaluation projects. Summary of the Invention

[0003] Based on this, it is necessary to address the above technical issues and provide an intelligent screening and evaluation method and system for innovative and entrepreneurial projects based on deep learning that can monitor external information in real time, accurately identify emergencies, and dynamically adjust project value assessment results.

[0004] First, this application provides a method for intelligent screening and evaluation of innovative and entrepreneurial projects based on deep learning, including:

[0005] Obtain multi-source data through government policy platforms, news aggregation services, social media streams, and technology monitoring sites; this multi-source data is used to characterize policy signals and social media sentiment regarding innovation and entrepreneurship projects;

[0006] The BiLSTM-CRF model is used to enhance multi-source data to obtain enhanced structured data.

[0007] Extracting emergency event data from enhanced structured data to obtain a classified event set; the emergency event data is used to characterize the content of the emergency event;

[0008] Based on the project knowledge base and the classified event set, the correlation between emergencies and innovation and entrepreneurship projects is calculated to obtain the project correlation network diagram;

[0009] Based on the project association network diagram, the impact of emergencies on innovation and entrepreneurship projects is quantified to obtain a dynamic impact coefficient matrix;

[0010] Screen projects based on the dynamic impact coefficient matrix and obtain an intelligent screening report on innovative and entrepreneurial projects.

[0011] Furthermore, the BiLSTM-CRF model is used to enhance the multi-source data to obtain enhanced structured data, including:

[0012] Unify the data structure of multi-source data to obtain standardized data flow;

[0013] Based on the bidirectional GRU model, the text in the standardized data stream is detected to determine whether it is advertising content. If so, the advertising content text is removed from the standardized data stream to obtain the purified data.

[0014] Identify domain entities in the purified data based on the BiLSTM-CRF model;

[0015] Link domain entities to standard nodes in the domain knowledge graph to obtain an enhanced entity set;

[0016] Transform the fuzzy spatiotemporal description in the enhanced entity set into a precise spatiotemporal description to obtain spatiotemporal enhanced data;

[0017] Perform cross-modal correlation on spatiotemporally enhanced data to obtain enhanced structured data.

[0018] Furthermore, based on the project knowledge base and the classified event set, the association between emergencies and innovation and entrepreneurship projects is calculated to obtain a project association network diagram, including:

[0019] Encode the project knowledge base and the classified event set separately to obtain the event vector and project document vector;

[0020] Calculate the similarity between the event vector and the project document vector, and remove event-document pairs with similarity less than a threshold to obtain a preliminary set of related events;

[0021] Identify industry chain nodes affected by emergencies based on a preliminary set of related events and the industry chain knowledge graph, which includes upstream and downstream relationships and supplier dependency data.

[0022] Calculate the influence weight of each industrial chain node to obtain the weighted industrial chain influence path;

[0023] Based on the preliminary set of related events and the project knowledge base, the rule matching engine analyzes the policy requirements involved in the emergency and scores them by comparing them with the current project parameters to obtain a policy matching score matrix;

[0024] The preliminary correlation event set, industrial chain impact path and policy matching score matrix are weighted and summed to obtain the project correlation network diagram; among them, the nodes of the project correlation network diagram are project entities and emergency event entities, and the edges are correlation types and strengths.

[0025] Furthermore, based on the project impact association network diagram, the impact of emergencies on the project is quantified, and a dynamic impact coefficient matrix is ​​obtained, including:

[0026] Based on the project association network diagram, the cost change rate is calculated using the following formula to obtain the economic impact vector:

[0027]

[0028] Among them, ΔC is the cost change rate, the set of ΔC of each submodule is the economic impact vector, α i is the direct cost impact coefficient, β j is the industry chain transmission coefficient, ΔP i is the price change of the i-th category caused by the emergency;

[0029] Based on the project association network diagram, the technical failure probability is calculated using the following formula to obtain the technical risk vector:

[0030] R=1-exp(-λ·TRL gap t)

[0031] Among them, R is the probability of technical failure, the set of R of each module is the technical risk vector, λ is the technical replacement difficulty coefficient, TRL gap is the gap in technology readiness level, t is the duration of the impact of the emergency;

[0032] Based on the project association network diagram, the violation risk value is calculated using the following formula to obtain the compliance risk vector:

[0033]

[0034] Among them, CRI is the violation risk value, and the set of CRIs of each policy clause is the compliance risk vector, S policy is the policy matching score, I regulatory is the regulatory intensity indicator, D buffer The remaining days for policy buffer;

[0035] Predict market share loss trends based on project association network diagrams and competitive product intelligence;

[0036] Based on the market share loss trend chart, quantify the impact of unexpected events on the project and obtain the market fluctuation vector;

[0037] Based on the project type, dynamic weights are assigned to the economic impact vector, technical risk vector, compliance risk vector and market volatility vector to obtain the dynamic impact coefficient matrix.

[0038] Furthermore, emergency event data is extracted from the enhanced structured data to obtain a classified event set, including:

[0039] Based on the enhanced structured data, the Z-score of the technical terms is calculated using the following formula, and technical terms with a Z-score less than the threshold are eliminated to obtain the differential technical terms:

[0040]

[0041] Where Z is the Z-score, which is used to quantify the degree of frequency mutation of technical terms, a is the current frequency, μ is the historical mean, and σ is the historical standard deviation;

[0042] Analyze innovative expressions in enhanced structured data;

[0043] The innovative expression and the differential technology term in the same semantic unit are marked as a technological breakthrough event, and a candidate set of technological breakthroughs is obtained; the differential technology is represented by the differential technology term;

[0044] Based on enhanced structured data, we quantify the differences in policy terms and obtain a set of policy change events.

[0045] Based on enhanced structured data, market volatility is quantified to obtain a set of market volatility events;

[0046] Extract core entity features of events from the technology breakthrough candidate set, policy change event set, and market volatility event set; core entity features include execution subject data, key action data, and execution time data;

[0047] The Jaccard similarity between each core entity feature is calculated, and the events represented by the core entity feature pairs with Jaccard similarity greater than a threshold are defined as repeated events; repeated events are merged to obtain a classified event set.

[0048] Furthermore, innovative and entrepreneurial projects are screened based on the dynamic impact coefficient matrix to obtain an intelligent screening report of innovative and entrepreneurial projects, including:

[0049] Normalize the four-dimensional coefficients in the dynamic influence coefficient matrix to obtain a standardized influence matrix;

[0050] Based on the dynamic impact coefficient matrix and the investor risk preference profile, the preset weights are loaded to obtain the weight vector;

[0051] Generate a preliminary project list based on the weight vector and the standardized impact matrix; the preliminary project list includes the four-dimensional risk values ​​of innovative and entrepreneurial projects;

[0052] Based on real-time market volatility, the risk threshold is dynamically adjusted and compared with the four-dimensional risk values ​​of the projects included in the preliminary project list. Projects that do not meet the threshold are eliminated to obtain a filtered project list;

[0053] Perform attribution analysis on the filtered project list and standardized impact matrix to obtain an attribution report;

[0054] Generate an intelligent screening report for innovative and entrepreneurial projects based on the filtered project list and attribution report.

[0055] Furthermore, after screening projects based on the dynamic impact coefficient matrix and obtaining an intelligent screening report of innovative and entrepreneurial projects, it also includes:

[0056] Compare actual investment results data with forecast data and calculate the average error rate;

[0057] Mark the models whose average error rate is greater than the threshold and obtain the error analysis report;

[0058] Based on the error analysis report and recent investment data sets, adjust the model to obtain an updated model version;

[0059] Based on the regulatory inquiry records of multi-source datasets, the knowledge graph nodes are updated to obtain the updated knowledge graph version;

[0060] Deploy the updated model version and the updated knowledge graph version, record the operation data, and obtain the system operation log.

[0061] Secondly, this application also provides an intelligent screening and evaluation system for innovative and entrepreneurial projects based on deep learning, including:

[0062] The source module is used to obtain multi-source data through government policy platforms, news aggregation services, social media streams, and technology monitoring sites. Multi-source data is used to represent policy signals and social media sentiment about innovation and entrepreneurship projects;

[0063] The enhancement module is used to enhance multi-source data through the BiLSTM-CRF model to obtain enhanced structured data;

[0064] The event module is used to extract emergency event data from the enhanced structured data to obtain a classified event set; the emergency event data is used to characterize the content of the emergency event;

[0065] The association module is used to calculate the association between emergencies and projects based on the project knowledge base and the classified event set, and obtain the project association network diagram;

[0066] The quantification module is used to quantify the impact of emergencies on innovation and entrepreneurship projects based on the project association network diagram and obtain a dynamic impact coefficient matrix;

[0067] The screening module is used to screen innovative and entrepreneurial projects based on the dynamic impact coefficient matrix and obtain an intelligent screening report on innovative and entrepreneurial projects.

[0068] In a third aspect, the present application further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the method provided in the first aspect of the present application when executing the computer program.

[0069] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method provided in the first aspect of the present application.

[0070] The above-mentioned deep learning-based intelligent screening and evaluation method and system for innovative and entrepreneurial projects obtains multi-source data through government policy platforms, news aggregation services, social media streams and technology monitoring sites; the multi-source data is used to characterize policy signals and social media sentiments about innovative and entrepreneurial projects; the multi-source data is enhanced through the BiLSTM-CRF model to obtain enhanced structured data; emergency data is extracted from the enhanced structured data to obtain a classified event set; the emergency data is used to characterize the content of the emergency; based on the project knowledge base and the classified event set, the association between the emergency and the innovative and entrepreneurial projects is calculated to obtain a project association network diagram; based on the project association network diagram, the impact of the emergency on the innovative and entrepreneurial projects is quantified to obtain a dynamic impact coefficient matrix; based on the dynamic impact coefficient matrix, projects are screened to obtain a technical means of an intelligent screening report on innovative and entrepreneurial projects, which realizes the effect of real-time monitoring of external information, accurate identification of emergencies and dynamic adjustment of project value assessment results. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments or related technical descriptions. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0072] Figure 1 A flowchart of a method for intelligent screening and evaluation of innovative and entrepreneurial projects based on deep learning provided by the present invention;

[0073] Figure 2 This is a structural diagram of an intelligent screening and evaluation system for innovative and entrepreneurial projects based on deep learning provided by the present invention. DETAILED DESCRIPTION

[0074] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0075] In one embodiment, Figure 1 As shown, a method for intelligent screening and evaluation of innovative and entrepreneurial projects based on deep learning is provided. This embodiment uses the method applied to a terminal as an example. It is understandable that the method can also be applied to a server, and can also be applied to a system including a terminal and a server, and implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0076] Step 101: Acquire multi-source data through government policy platforms, news aggregation services, social media streams, and technology monitoring sites; the multi-source data is used to represent policy signals and social media sentiment regarding innovation and entrepreneurship projects.

[0077] Specifically, the government policy platform is a repository of officially released policy documents, containing structured and / or unstructured text on topics such as industry support and regulatory revisions. News aggregation services are platforms that automatically capture and integrate news media, covering industry dynamics, financing events, and more. Social media streams are a source of real-time public discussion data, reflecting market sentiment and emerging trends. Technology monitoring sites can include technology patent libraries and academic paper platforms, providing information on cutting-edge technological breakthroughs. Policy signals are implicit in policy texts expressing support or restrictions for specific industries. Social media sentiment represents public opinion on technologies and projects on social platforms. Terminals continuously collect raw text, image metadata, and timestamp information from these four data sources through corresponding API interfaces or web crawlers, generating multi-source data.

[0078] Step 102: Perform data enhancement on the multi-source data through the BiLSTM-CRF model to obtain enhanced structured data.

[0079] Specifically, the Bidirectional Long Short-Term Memory-Conditional Random Field (Bidirectional Long Short-Term Memory-Conditional Random Field) model captures contextual semantics through a bidirectional long short-term memory network, and the conditional random field optimizes entity label sequences for named entity recognition. Enhanced structured data is cleaned, linked, and standardized data that can contain precise entities, times, locations, and relationships. The terminal converts multi-source data into a standardized stream in a unified time-entity-event format. A bidirectional GRU model is used to identify and remove promotional text. The BiLSTM-CRF identifies technical terms, company names, policy names, and other field entities in the text, links the entities to the domain knowledge graph, and associates the time and technical descriptions in the text with corresponding patent diagrams to obtain structured data.

[0080] Step 103 : extracting emergency event data from the enhanced structured data to obtain a classified event set; the emergency event data is used to characterize the content of the emergency event.

[0081] Specifically, the breaking event is a policy, technology and market change that deviates significantly from the historical norm; for example, a breaking event can include a sudden policy ban, a technology patent disclosure. The classified event set is the standardized event library after deduplication and merging, which can be classified by technology breakthrough, policy change and market fluctuation. The Z-score of the terminal computing technology term is retained as the difference technology term if the Z-score is greater than the threshold value. The Sci BERT model is used to analyze the innovative expressions similar to "breakthrough" and "global first" in the text, and mark the sentences that appear with the difference technology term as technology breakthrough candidate events. After removing duplicate events, the classified event set is obtained.

[0082] Step 104, based on the project knowledge base and the classified event set, the association between the breaking event and the innovation and entrepreneurship project is calculated, and the project association network graph is obtained.

[0083] Among them, the project knowledge base is a database that stores project technical parameters, team background and supply chain relationship; the project association network graph is a graph structure with projects and events as nodes and association types and strengths as edges. The terminal encodes the event text description and the project document into a vector, and analyzes the impact of the breaking event on the project, locates the nodes affected by the event in the industry chain knowledge graph, and obtains the project association network graph.

[0084] Step 105, based on the project association network graph, the influence of the breaking event on the innovation and entrepreneurship project is quantified, and a dynamic influence coefficient matrix is obtained.

[0085] Specifically, the dynamic influence coefficient matrix is a four-dimensional risk value matrix, where each row represents a project and each column represents an economic, technical, compliance and market risk coefficient. The terminal calculates the influence of the breaking event on multiple dimensions including economy, technology, policy and market based on the association network graph, and fuses the influence of each dimension on the project based on the preset weight to obtain the dynamic influence coefficient matrix.

[0086] Step 106, based on the dynamic influence coefficient matrix, the project is screened, and an intelligent screening report of the innovation and entrepreneurship project is obtained.

[0087] Specifically, the intelligent screening report is a visual decision document containing project ranking, risk dimension decomposition and attribution analysis. The terminal loads the preset investor configuration and adjusts the threshold of each dimension of risk based on the investor configuration, eliminates the innovation and entrepreneurship projects that do not meet the threshold requirements, explains the source of project risk, and outputs a recommended list of innovation and entrepreneurship projects, which can include a four-dimensional risk radar chart and key event attribution explanation.

[0088] The intelligent screening and evaluation method for innovative and entrepreneurial projects based on deep learning provided in this embodiment obtains multi-source data through government policy platforms, news aggregation services, social media streams and technology monitoring sites; the multi-source data is used to characterize policy signals and social media sentiments about innovative and entrepreneurial projects; the multi-source data is enhanced through the BiLSTM-CRF model to obtain enhanced structured data; emergency data is extracted from the enhanced structured data to obtain a classified event set; the emergency data is used to characterize the content of the emergency; based on the project knowledge base and the classified event set, the association between the emergency and the innovative and entrepreneurial projects is calculated to obtain a project association network diagram; based on the project association network diagram, the impact of the emergency on the innovative and entrepreneurial projects is quantified to obtain a dynamic influence coefficient matrix; based on the dynamic influence coefficient matrix, the projects are screened to obtain an intelligent screening report for innovative and entrepreneurial projects. Through the above technical means, the effects of real-time monitoring of external information, accurate identification of emergencies and dynamic adjustment of project value assessment results are achieved.

[0089] In one embodiment, based on a project knowledge base and a set of classified events, the association between emergencies and innovation and entrepreneurship projects is calculated to obtain a project association network diagram, including:

[0090] In one embodiment, data enhancement is performed on multi-source data using a BiLSTM-CRF model to obtain enhanced structured data, including:

[0091] Step 201: unify the data structure of multi-source data to obtain a standardized data stream.

[0092] Specifically, a standardized data stream is a data sequence with aligned fields, consistent encoding, and standardized timestamps. Standardized data is predefined as text containing core fields, including source, timestamp, original content, and media type. The terminal extracts text data from web pages, images, and videos and generates descriptive tags while preserving timestamps.

[0093] Step 202 : Detect text in the standardized data stream based on the bidirectional GRU model to determine whether it is advertising content. If so, remove the advertising content text from the standardized data stream to obtain purified data.

[0094] Specifically, advertising content refers to text containing promotional information or commercial advertising. The terminal inputs the standardized text into a bidirectional GRU (Gated Recurrent Unit) model and receives the advertising probability value output by the model. If the probability value exceeds a threshold, the standardized text is discarded, retaining the remaining text data.

[0095] Step 203: Identify domain entities of the cleansed data based on the BiLSTM-CRF model.

[0096] Among them, domain entities are predefined entity types related to innovation and entrepreneurship, which can include technology, policy, enterprise, person, and location. The terminal extracts key information units from unstructured text. For example, the predefined entity types are: technology, policy, enterprise, person, and location. The bidirectional LSTM captures the bidirectional contextual features of the text, with the number of layers set to 1, the hidden layer dimension set to 128, and the dropout rate set to 0.3; the CRF layer is responsible for learning and ensuring the constraints of the label sequence. The loss function is set to the negative log-likelihood loss of CRF, optimized using the Adam optimizer with a learning rate of 3e-4.

[0097] Step 204: Link the domain entities to standard nodes in the domain knowledge graph to obtain an enhanced entity set.

[0098] Specifically, the domain knowledge graph is a graph database containing standardized entity nodes and attribute relationships. An enhanced entity set is a collection of entities linked to the graph with authoritative information. The terminal performs entity recognition, connecting polysemous words to corresponding entities based on contextual information. It also supplements the industry chain relationships related to emergencies based on the domain knowledge graph.

[0099] Step 205 : convert the fuzzy spatiotemporal description in the enhanced entity set into a precise spatiotemporal description to obtain spatiotemporal enhanced data.

[0100] Among them, the terminal resolves the ambiguous time and space expressions based on the time when the emergency occurs, such as converting "next month" and "Q3" into specific time ranges, and converting "Yangtze River Delta" and "South China" into specific geographic fences and administrative division codes.

[0101] Step 206: Perform cross-modal association on the spatiotemporal enhanced data to obtain enhanced structured data.

[0102] Specifically, enhanced structured data is a multimodal associated data system. The terminal calculates the semantic similarity between news text and images through the CLIP (Constrastive Language-Image Pre-training) model, analyzes descriptive indicators (such as product form, scene characteristics, personnel size, etc.), and associates them with database fields to form a multi-dimensional data verification system. Exemplarily, enhanced structured data may include: structured entities, attributes, and relationships extracted from text; visual indicators and associated fields analyzed from images; and semantic consistency assessment results between text and images.

[0103] This embodiment uses deep structural transformation and information filtering to monitor external information in real time and accurately identify emergencies, thereby improving the credibility of the acquired data and enhancing the accuracy of screening and evaluation.

[0104] Step 301: Encode the project knowledge base and the classified event set respectively to obtain an event vector and a project document vector.

[0105] Specifically, an event vector is a numerical vector of the encoded text description of each emergency event in the classified event set; a project document vector is a numerical vector of the encoded project detail documents in the project knowledge base. The terminal inputs the emergency event description text and project document text into the Sentence-BERT (Sentence Bidirectional Encoder Representations from Transformers) model. Two weighted BERT models encode the input emergency event description text and project document text, generating fixed-length vectors. This captures the core semantics of the text and can transform unstructured text into a computable mathematical representation.

[0106] Step 302 : Calculate the similarity between the event vector and the project document vector, and remove event-document pairs with similarity less than a threshold to obtain a preliminary set of associated events.

[0107] Specifically, the preliminary set of associated events is the set of retained event and project pairs whose similarity exceeds a threshold. The terminal calculates the cosine similarity between each pair of event vectors and project document vectors, discarding pairs with similarity below the threshold and retaining highly correlated pairs. Optionally, the threshold is preset to 0.65.

[0108] Step 303: Identify the nodes in the industrial chain affected by the emergency based on the preliminary set of associated events and the industrial chain knowledge graph; the industrial chain knowledge graph includes upstream and downstream relationships in the industry and supplier dependency data.

[0109] Specifically, the industry chain knowledge graph is a graph structure containing industry entities and relationships; industry chain nodes are entity nodes in the knowledge graph. By mapping emergencies from the initial set of associated events to the industry chain knowledge graph and performing a diffusion search along upstream and downstream relationships, we can identify directly or indirectly affected nodes.

[0110] Step 304: Calculate the influence weight of each industrial chain node to obtain a weighted industrial chain influence path.

[0111] The influence weight is the global importance score of a node in the industry chain, and the weighted industry chain influence path is the transmission path annotated with the node weight. The terminal uses the industry chain knowledge graph as input, runs the PageRank algorithm to calculate the influence weight of each node on innovation and entrepreneurship projects, and annotates each weight value for the identified affected node path.

[0112] Step 305 , based on the preliminary associated event set and the project knowledge base, the policy requirements involved in the emergency are parsed by the rule matching engine, and the current parameters of the project are compared and scored to obtain a policy matching score matrix.

[0113] Specifically, the rule matching engine parses policy clauses based on predefined rules, while the policy matching score matrix is ​​a compliance scoring matrix across the event-project dimension. The terminal uses the rule matching engine to parse policy clauses in an emergency, extract corresponding parameters from the project knowledge base, and calculate the matching score based on the rules.

[0114] In step 306, the preliminary associated event set, the industrial chain impact path, and the policy matching score matrix are weighted and summed to obtain a project association network diagram; wherein the nodes of the project association network diagram are project entities and emergency event entities, and the edges are association types and strengths.

[0115] Specifically, the terminal calculates the comprehensive correlation strength for each event-item pair based on the preset weight ratio, uses entities as nodes, and labels the type and strength of the edges according to the calculation results to construct a network diagram.

[0116] This embodiment breaks through the limitations of a single association model and improves the accuracy of screening and evaluation through the triple integration of semantic analysis, industrial chain topology calculation, and policy rule matching.

[0117] In one embodiment, based on the project impact association network diagram, the impact of the emergency event on the project is quantified to obtain a dynamic impact coefficient matrix, including:

[0118] Step 401: Based on the project association network diagram, the cost change rate is calculated using the following formula to obtain the economic impact vector:

[0119]

[0120] Among them, ΔC is the cost change rate, the set of ΔC of each submodule is the economic impact vector, α i is the direct cost impact coefficient, β j is the industry chain transmission coefficient, ΔP i is the price change of the i-th category caused by the emergency.

[0121] Specifically, the cost change rate is the percentage change in total project costs caused by the incident; the direct cost impact coefficient is the sensitivity of the i-th cost category to the incident; the industry chain transmission coefficient is the amplification factor when the incident is transmitted through the j-th level of the industry chain; the price change is the price change of the i-th cost category directly caused by the incident; and the economic impact vector is the sum of the cost change rates of each submodule, representing the cost changes in each part of the project. The cost parameters related to the incident are extracted from the project association network diagram, and the economic impact vector is derived using a formula.

[0122] Step 402: Based on the project association network diagram, the technical failure probability is calculated using the following formula to obtain a technical risk vector:

[0123] R=1-exp(-λ·TRL gap t)

[0124] Among them, R is the probability of technical failure, the set of R of each module is the technical risk vector, λ is the technical replacement difficulty coefficient, TRL gap is the gap in technology readiness level, and t is the duration of the impact of the emergency.

[0125] Specifically, the probability of technical failure is the probability that a sudden event will cause the project's technical route to fail; the difficulty coefficient of technology substitution is the ease with which technology can be replaced; the technology readiness gap is the difference between the project's current technical maturity and the required technical maturity after the event; the impact duration is the duration of the event's impact on the technical route; and the technology risk vector is the set of technical failure probabilities for each technical module. The terminal extracts the project's technical barrier value from the project's associated network diagram and quantifies it as the difficulty of technology substitution, the technology readiness gap, and the predicted event impact period. The technology risk vector is then calculated using a formula.

[0126] Step 403: Based on the project association network diagram, the violation risk value is calculated using the following formula to obtain the compliance risk vector:

[0127]

[0128] Among them, CRI is the violation risk value, and the set of CRIs of each policy clause is the compliance risk vector, S policy is the policy matching score, I regulatory is the regulatory intensity indicator, D buffer The remaining days for policy buffer.

[0129] Specifically, the violation risk value is the risk intensity of the project facing penalties for non-compliance with the new policy; the policy matching score is the degree of matching between the current status of the project and the policy requirements; the supervision intensity index is the strictness of policy implementation; the policy buffer remaining days is the remaining rectification time before the policy takes effect; and the compliance risk vector is the set of CRI values ​​of each policy clause.

[0130] Step 404: predict the market share loss trend graph based on the project association network graph and competitor intelligence.

[0131] Competitive intelligence can be a dynamic database of competitors' production capacity, pricing, and alternative technologies. A market share churn trend chart is a forecast of project market share changes over time due to events. The terminal analyzes the impact of events on projects based on the associated network diagram and generates a time-share curve. For example, if an event enhances a competitor's advantage, such as a competitor receiving subsidies, the customer churn rate is predicted. If an event causes industry contraction, such as the cancellation of a demand policy, the total market share is predicted to shrink.

[0132] Step 405: Based on the market share loss trend graph, quantify the impact of the emergency on the project and obtain a market fluctuation vector.

[0133] Specifically, the market volatility vector quantifies the impact of market share loss on the project's core metrics. The terminal extracts key indicators from the market share loss trend chart, which can include maximum churn rate, stable period, and probability of recovery. Based on these key indicators, the impact value is calculated, which can include revenue loss and customer churn cost.

[0134] Step 406 , based on the project type, dynamic weights are assigned to the economic impact vector, the technical risk vector, the compliance risk vector, and the market volatility vector to obtain a dynamic impact coefficient matrix.

[0135] Specifically, the rows of the dynamic impact coefficient matrix represent projects, and the columns represent four-dimensional risk values. The terminal assigns weights to the four-dimensional risk values ​​based on the type of innovation and entrepreneurship project, thereby obtaining the dynamic impact coefficient matrix.

[0136] This embodiment forms a closed-loop assessment chain by independently quantifying risks in multiple dimensions, thereby improving the reliability of screening assessments and making various risks traceable.

[0137] In one embodiment, emergency event data is extracted from the enhanced structured data to obtain a classified event set, including:

[0138] Step 501: Based on the enhanced structured data, the Z-score of the technical terms is calculated using the following formula, and technical terms with a Z-score less than a threshold are eliminated to obtain differential technical terms:

[0139]

[0140] Among them, Z is the Z-score, which is used to quantify the degree of frequency mutation of technical terms, a is the current frequency, μ is the historical mean, and σ is the historical standard deviation.

[0141] Specifically, the Z-score is used to measure the number of standard deviations by which the current frequency deviates from the historical average level; the current frequency is the number of times the target technical term appears in the recent window period; the historical mean can be the average daily number of occurrences of the term in the historical period; the historical standard deviation represents the statistical dispersion of historical frequency fluctuations; and the difference technical term is a term whose Z-score exceeds a threshold, optionally, the threshold is set to 3.

[0142] Step 502: Analyze innovative expressions in the enhanced structured data.

[0143] Specifically, innovative expressions are text snippets that contain semantics such as breakthrough, subversive, and original. The terminal inputs the enhanced structured data into the SciBERT (A Pretrained Language Model for Scientific Text) model to obtain the probability of innovation of the text snippet.

[0144] Step 503 , marking the differential technologies in which the innovative expression and the differential technology terms are in the same semantic unit as technology breakthrough events, and obtaining a technology breakthrough candidate set; the differential technologies are represented by the differential technology terms.

[0145] Specifically, the technology breakthrough candidate set is a collection of confirmed major technology breakthrough events. The terminal marks text paragraphs containing both innovative expressions and differentiated technical terms as technology breakthrough events.

[0146] Step 504: quantify the differences in policy terms based on the enhanced structured data to obtain a set of policy change events.

[0147] Specifically, the policy change event set is a collection of policy clauses where significant differences were detected. A pre-trained BERT classifier is fed into the classifier, which then outputs a difference level. The policy text in the augmented data is then scanned to detect clauses with a difference level greater than a threshold. Optionally, the difference level is divided into three levels, with a threshold of 2.

[0148] Step 505: quantify market fluctuations based on the enhanced structured data to obtain a market fluctuation event set.

[0149] Specifically, market volatility events are defined as those in which sentiment fluctuations exceed a threshold. Using the VADER (ValenceAware Dictionary and Sentiment Reasoner) algorithm, a rule-based sentiment analyzer, we calculate a composite sentiment score for market-related texts, combining word strength, modifiers, and negation. If the standard deviation of the sentiment score increases suddenly, for example, by more than 1.5 times the historical mean, the event is flagged as a market volatility event.

[0150] Step 506, extract the core entity features of the events in the technology breakthrough candidate set, the policy change event set and the market fluctuation event set; wherein the core entity features include execution subject data, key action data and execution time data.

[0151] Specifically, the execution subject is the event initiator, the key action is the event core behavior, and the execution time is the time effective time. The terminal extracts structured triples from the three types of event sets to obtain the core entity features.

[0152] Step 507, calculate the Jaccard similarity between each core entity feature, and define the events represented by the core entity features with Jaccard similarity greater than a threshold as repeated events; merge the repeated events to obtain a classified event set.

[0153] Specifically, the classified event set is the final event library after deduplication and classification by technology breakthrough, policy change and market fluctuation. The Jaccard similarity of each pair of events is calculated, and if the similarity is greater than a threshold, it is considered that the two burst events in the event pair are the same event, and the event pair is merged. Optionally, the threshold is set to 0.7.

[0154] The present embodiment processes structured data into a high-purity, traceable burst event set through a multi-layer filtering mechanism, providing high-quality input for project association network construction and improving the accuracy of screening and evaluation.

[0155] In one of the embodiments, the innovative entrepreneurship projects are screened based on the dynamic influence coefficient matrix to obtain an intelligent screening report of the innovative entrepreneurship projects, including:

[0156] Step 601, normalize the four-dimensional coefficients in the dynamic influence coefficient matrix to obtain a standardized influence matrix.

[0157] Specifically, the coefficients in the dynamic influence coefficient matrix are normalized by the Z-score algorithm.

[0158] Step 602, based on the dynamic influence coefficient matrix and the investor risk preference profile, load a preset weight to obtain a weight vector.

[0159] Specifically, the investor risk preference profile is the user's preset risk tolerance data; the weight vector is the final determined four-dimensional weight combination. The terminal loads the user configuration, and if the user sets the upper limit of the market risk weight to 0.4, the weight of this dimension is limited to less than 0.4, and the dimensions not configured by the user use the system default value, and the final output weight vector is 1.

[0160] Step 603, generate a preliminary project list based on the weight vector and the standardized influence matrix; the preliminary project list contains four-dimensional risk values of the innovative entrepreneurship projects.

[0161] Specifically, the preliminary project list is an initial list of recommendations, including project IDs, names, and four-dimensional risk scores. The four-dimensional risk score is a normalized risk score for each project across four dimensions. The terminal calculates the weighted total risk score, generates a list sorted by total score, and outputs the four-dimensional risk score.

[0162] Step 604 , dynamically adjust the risk threshold based on the real-time market volatility, and compare it with the four-dimensional risk values ​​of the projects included in the preliminary project list, eliminate the projects that do not meet the threshold, and obtain a filtered project list.

[0163] Specifically, real-time market volatility is the current financial market volatility index. Optionally, the VIX (CBOE Volatility Index) can be used as the real-time market volatility index. The risk threshold is the maximum allowable risk score, which is dynamically adjusted with market fluctuations. Pre-set rules are used to map volatility to risk thresholds, and projects with total risk scores exceeding the threshold are eliminated.

[0164] Step 605 : Perform attribution analysis on the filtered project list and the standardized impact matrix to obtain an attribution report.

[0165] Specifically, the SHAP (SHapley Additive exPlanations) value attribution algorithm is a game-theory-based model interpretation method used to quantify the contribution of each dimension to total risk. An attribution report is an analytical document that explains the sources of risk for each project. The terminal calculates SHAP values ​​for each of the four dimensions for each retained project. A positive value indicates that the dimension increases the total risk, while a negative value indicates that the dimension decreases the risk. An attribution report is generated based on the SHAP values.

[0166] Step 606: Generate an intelligent screening report of innovative and entrepreneurial projects based on the filtered project list and attribution report.

[0167] The intelligent screening report is the final output of the innovation and entrepreneurship project analysis document. The terminal integrates the filtered project list and SHAP attribution analysis of each project, and adds decision-making recommendations.

[0168] This embodiment improves the efficiency of screening innovative and entrepreneurial projects through dynamic adaptability and risk transparency, and provides recommendations for high-quality projects with controllable risks, traceable sources, and rapid response to market changes.

[0169] In one embodiment, after screening projects based on the dynamic impact coefficient matrix to obtain an intelligent screening report of innovative and entrepreneurial projects, the process further includes:

[0170] Step 701: Compare the actual investment result data with the forecast data and calculate the average error rate.

[0171] Specifically, actual investment results represent the risk loss value of invested projects; forecast data represents the project risk value predicted during previous recommendations; and the average error rate is the arithmetic mean of the difference between the forecast and the actual value. The terminal calculates the error rate for each project and takes the arithmetic mean of all error rates.

[0172] Step 702: Mark the models whose average error rate is greater than a threshold, and obtain an error analysis report.

[0173] Specifically, the error analysis report is a diagnostic document that identifies high-error models and their error patterns. The terminal analyzes the failure scenarios of high-error models, locates the model failure points, and identifies the specific modules that need optimization.

[0174] Step 703: Based on the error analysis report and the recent investment data set, adjust the model to obtain an updated model version.

[0175] The recent investment dataset represents newly generated project investment data, while the updated model version represents a new forecasting model with adjusted parameters and structure. The terminal optimizes the model for high-error models and retrains the model using the new dataset to generate the updated model version.

[0176] Step 704: Based on the multi-source data set supervision inquiry records, the knowledge graph nodes are updated to obtain an updated knowledge graph version.

[0177] Specifically, regulatory inquiry records from multi-source datasets can include inquiries and responses from institutions such as the China Securities Regulatory Commission and exchanges. Knowledge graph nodes are semantic network units representing emergency event entities and their relationships. The terminal parses the key information in the inquiry records and then modifies the graph node attributes.

[0178] Step 705: deploy the updated model version and the updated knowledge graph version, and record the operation data to obtain the system operation log.

[0179] Specifically, the system operation log is time series data that records the operation status such as model inference results and the number of knowledge graph calls.

[0180] This embodiment provides a self-updating means for the model through a self-evolution mechanism and multi-source knowledge fusion, thereby improving the applicability of the screening and evaluation model and the accuracy of the screening and evaluation.

[0181] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above can include multiple steps or multiple stages, and these steps or stages are not necessarily executed and completed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0182] Based on the same inventive concept, the embodiments of the present application also provide a system for intelligent screening and evaluation of innovative and entrepreneurial projects based on deep learning, which is used to implement the aforementioned method for intelligent screening and evaluation of innovative and entrepreneurial projects based on deep learning. The implementation solution provided by this system is similar to the implementation solution described in the aforementioned method. Therefore, the specific limitations of one or more embodiments of the system for intelligent screening and evaluation of innovative and entrepreneurial projects based on deep learning provided below can be found in the limitations of the method for intelligent screening and evaluation of innovative and entrepreneurial projects based on deep learning above, and will not be repeated here.

[0183] In an exemplary embodiment, Figure 2 As shown, a deep learning-based intelligent screening and evaluation system 800 for innovative and entrepreneurial projects is provided, comprising:

[0184] Source module 801 is used to obtain multi-source data through government policy platforms, news aggregation services, social media streams, and technology monitoring sites. The multi-source data is used to represent policy signals and social media sentiment regarding innovation and entrepreneurship projects.

[0185] An enhancement module 802 is used to perform data enhancement on multi-source data using a BiLSTM-CRF model to obtain enhanced structured data;

[0186] The event module 803 is used to extract emergency event data from the enhanced structured data to obtain a classified event set; the emergency event data is used to characterize the content of the emergency event;

[0187] The association module 804 is used to calculate the association between the emergency events and the projects based on the project knowledge base and the classified event set to obtain a project association network diagram;

[0188] Quantification module 805 is used to quantify the impact of emergencies on innovation and entrepreneurship projects based on the project association network diagram to obtain a dynamic impact coefficient matrix;

[0189] The screening module 806 is configured to screen the innovative entrepreneurship project based on the dynamic influence coefficient matrix, and obtain an intelligent screening report of the innovative entrepreneurship project.

[0190] Further, the association module 804 is further configured to:

[0191] encode the project knowledge base and the classified event set respectively to obtain an event vector and a project document vector;

[0192] calculate the similarity of the event vector and the project document vector, and eliminate event document pairs with a similarity less than a threshold value to obtain a preliminary associated event set;

[0193] identify an industrial chain node affected by the sudden event based on the preliminary associated event set and an industrial chain knowledge graph; the industrial chain knowledge graph comprises industry upstream and downstream relationships and supplier dependency data;

[0194] calculate the influence weight of each industrial chain node to obtain an industrial chain influence path with a weight;

[0195] based on the preliminary associated event set and the project knowledge base, analyze the policy requirements involved in the sudden event through a rule matching engine, and compare the current parameters of the project to obtain a policy matching score matrix;

[0196] weight and sum the preliminary associated event set, the industrial chain influence path and the policy matching score matrix to obtain a project association network graph; wherein the nodes of the project association network graph are project entities and sudden event entities, and the edges are association types and strengths.

[0197] Further, the quantification module 805 is further configured to:

[0198] based on the project association network graph, calculate the cost change rate through the following formula to obtain an economic influence vector:

[0199]

[0200] wherein, ΔC is the cost change rate, the set of ΔC of each sub-module is the economic influence vector, α i is a direct cost influence coefficient, β j is an industrial chain transmission coefficient, ΔP i is the i-th type of price change amount caused by the sudden event;

[0201] based on the project association network graph, calculate the technology failure probability through the following formula to obtain a technology risk vector:

[0202] R = 1 - exp (- λ · TRL gap · t)

[0203] Among them, R is the probability of technical failure, the set of R of each module is the technical risk vector, λ is the technical replacement difficulty coefficient, TRL gap is the gap in technology readiness level, t is the duration of the impact of the emergency;

[0204] Based on the project association network diagram, the violation risk value is calculated using the following formula to obtain the compliance risk vector:

[0205]

[0206] Among them, CRI is the violation risk value, and the set of CRIs of each policy clause is the compliance risk vector, S policy is the policy matching score, I regulatory is the regulatory intensity indicator, D buffer Buffer remaining days for policy;

[0207] Predict market share loss trends based on project association network diagrams and competitive product intelligence;

[0208] Based on the market share loss trend chart, quantify the impact of unexpected events on the project and obtain the market fluctuation vector;

[0209] Based on the project type, dynamic weights are assigned to the economic impact vector, technical risk vector, compliance risk vector and market volatility vector to obtain the dynamic impact coefficient matrix.

[0210] Furthermore, the event module 803 is further configured to:

[0211] Based on the enhanced structured data, the Z-score of the technical terms is calculated using the following formula, and technical terms with a Z-score less than the threshold are eliminated to obtain the differential technical terms:

[0212]

[0213] Where Z is the Z-score, which is used to quantify the degree of frequency mutation of technical terms, a is the current frequency, μ is the historical mean, and σ is the historical standard deviation;

[0214] Analyze innovative expressions in enhanced structured data;

[0215] The innovative expression and the differential technology term in the same semantic unit are marked as a technological breakthrough event, and a candidate set of technological breakthroughs is obtained; the differential technology is represented by the differential technology term;

[0216] Based on enhanced structured data, we quantify the differences in policy terms and obtain a set of policy change events.

[0217] Based on enhanced structured data, market volatility is quantified to obtain a set of market volatility events;

[0218] Extract core entity features of events from the technology breakthrough candidate set, policy change event set, and market volatility event set; core entity features include execution subject data, key action data, and execution time data;

[0219] The Jaccard similarity between each core entity feature is calculated, and the events represented by the core entity feature pairs with Jaccard similarity greater than a threshold are defined as repeated events; repeated events are merged to obtain a classified event set.

[0220] Furthermore, the enhancement module 802 is further configured to:

[0221] Unify the data structure of multi-source data to obtain standardized data flow;

[0222] Based on the bidirectional GRU model, the text in the standardized data stream is detected to determine whether it is advertising content. If so, the advertising content text is removed from the standardized data stream to obtain the purified data.

[0223] Identify domain entities in the purified data based on the BiLSTM-CRF model;

[0224] Link domain entities to standard nodes in the domain knowledge graph to obtain an enhanced entity set;

[0225] Transform the fuzzy spatiotemporal description in the enhanced entity set into a precise spatiotemporal description to obtain spatiotemporal enhanced data;

[0226] Perform cross-modal correlation on spatiotemporally enhanced data to obtain enhanced structured data.

[0227] Furthermore, the screening module 806 is further configured to:

[0228] Normalize the four-dimensional coefficients in the dynamic influence coefficient matrix to obtain a standardized influence matrix;

[0229] Based on the dynamic impact coefficient matrix and the investor risk preference profile, the preset weights are loaded to obtain the weight vector;

[0230] Generate a preliminary project list based on the weight vector and the standardized impact matrix; the preliminary project list includes the four-dimensional risk values ​​of innovative and entrepreneurial projects;

[0231] Based on real-time market volatility, the risk threshold is dynamically adjusted and compared with the four-dimensional risk values ​​of the projects included in the preliminary project list. Projects that do not meet the threshold are eliminated to obtain a filtered project list;

[0232] Perform attribution analysis on the filtered project list and standardized impact matrix to obtain an attribution report;

[0233] Generate an intelligent screening report for innovative and entrepreneurial projects based on the filtered project list and attribution report.

[0234] Furthermore, the system further includes an update module for:

[0235] Compare actual investment results data with forecast data and calculate the average error rate;

[0236] Mark the models whose average error rate is greater than the threshold and obtain the error analysis report;

[0237] Based on the error analysis report and recent investment data sets, adjust the model to obtain an updated model version;

[0238] Based on the regulatory inquiry records of multi-source datasets, the knowledge graph nodes are updated to obtain the updated knowledge graph version;

[0239] Deploy the updated model version and the updated knowledge graph version, record the operation data, and obtain the system operation log.

[0240] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the intelligent screening and evaluation method for innovative and entrepreneurial projects based on deep learning as described above are implemented.

[0241] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0242] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to the partial description of the method embodiments. The device embodiments described above are merely illustrative, wherein the components described as separate parts may or may not be physically separated, and the parts displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the disclosed solution. A person of ordinary skill in the art can understand and implement it without expending creative work.

[0243] The above-described embodiments merely represent several implementation methods of the embodiments of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that a person skilled in the art may make various modifications and improvements without departing from the concept of the embodiments of the present application, and these modifications and improvements fall within the scope of protection of the embodiments of the present application.

Claims

1. A method for intelligent screening and evaluation of innovative and entrepreneurial projects based on deep learning, characterized by: The method comprises: Obtaining multi-source data through government policy platforms, news aggregation services, social media streams, and technology monitoring sites; said multi-source data is used to characterize policy signals and social media sentiment regarding innovation and entrepreneurship projects; Performing data enhancement on the multi-source data through a BiLSTM-CRF model to obtain enhanced structured data; Extracting emergency event data from the enhanced structured data to obtain a classified event set; the emergency event data is used to characterize the content of the emergency event; Based on the project knowledge base and the classified event set, calculating the association between the emergency event and the innovation and entrepreneurship project to obtain a project association network diagram; Based on the project association network diagram, quantify the impact of the emergency on the innovation and entrepreneurship project to obtain a dynamic impact coefficient matrix; The projects are screened based on the dynamic impact coefficient matrix to obtain an intelligent screening report on innovative and entrepreneurial projects.

2. The method for intelligent screening and evaluation of innovative and entrepreneurial projects based on deep learning according to claim 1 is characterized in that: The multi-source data is enhanced by the BiLSTM-CRF model to obtain enhanced structured data, including: Unifying the data structure of the multi-source data to obtain a standardized data stream; Detecting text in the standardized data stream based on a bidirectional GRU model to determine whether it is advertising content; if so, removing the advertising content text from the standardized data stream to obtain purified data; Identify domain entities of the cleansed data based on the BiLSTM-CRF model; Linking the domain entities to standard nodes in the domain knowledge graph to obtain an enhanced entity set; Converting the fuzzy spatiotemporal description in the enhanced entity set into a precise spatiotemporal description to obtain spatiotemporal enhanced data; Cross-modal association is performed on the spatiotemporally enhanced data to obtain the enhanced structured data.

3. The method for intelligent screening and evaluation of innovative and entrepreneurial projects based on deep learning according to claim 1 is characterized in that: The calculation of the association between the emergency event and the innovation and entrepreneurship project based on the project knowledge base and the classified event set to obtain a project association network diagram includes: Encode the project knowledge base and the classified event set respectively to obtain an event vector and a project document vector; Calculating the similarity between the event vector and the project document vector, and eliminating event-document pairs whose similarity is less than a threshold, to obtain a preliminary set of associated events; Based on the preliminary set of associated events and the industry chain knowledge graph, identifying the industry chain nodes affected by the emergency; the industry chain knowledge graph includes industry upstream and downstream relationships and supplier dependency data; Calculate the influence weight of each industrial chain node to obtain the weighted industrial chain influence path; Based on the preliminary associated event set and the project knowledge base, the policy requirements involved in the emergency event are parsed by a rule matching engine, and the policy requirements are scored by comparing them with the current parameters of the project to obtain a policy matching score matrix; The project association network diagram is obtained by weighted summing of the preliminary association event set, the industrial chain impact path and the policy matching score matrix; wherein the nodes of the project association network diagram are project entities and emergency event entities, and the edges are association types and strengths.

4. The method for intelligent screening and evaluation of innovative and entrepreneurial projects based on deep learning according to claim 1 is characterized in that: The method of quantifying the impact of the emergency on the project based on the project impact association network diagram to obtain a dynamic impact coefficient matrix includes: Based on the project association network diagram, the cost change rate is calculated using the following formula to obtain the economic impact vector: Among them, ΔC is the cost change rate, the set of ΔC of each submodule is the economic impact vector, α i is the direct cost impact coefficient, β j is the industry chain transmission coefficient, ΔP i is the price change of the i-th category caused by the emergency; Based on the project association network diagram, the technical failure probability is calculated using the following formula to obtain the technical risk vector: R=1-exp(-λ·TRL gap ·t) Among them, R is the probability of technical failure, the set of R of each module is the technical risk vector, λ is the technical replacement difficulty coefficient, TRL gap is the gap in technology readiness level, t is the duration of the impact of the emergency; Based on the project association network diagram, the violation risk value is calculated using the following formula to obtain the compliance risk vector: Among them, CRI is the violation risk value, and the set of CRIs of each policy clause is the compliance risk vector, S policy is the policy matching score, I regulatory is the regulatory intensity indicator, D buffer Buffer remaining days for policy; Based on the project association network diagram and competitive product intelligence, predict the market share loss trend chart; Based on the market share loss trend graph, quantify the impact of the emergency on the project to obtain a market fluctuation vector; Based on the project type, dynamic weights are assigned to the economic impact vector, the technical risk vector, the compliance risk vector, and the market volatility vector to obtain the dynamic impact coefficient matrix.

5. The method for intelligent screening and evaluation of innovative and entrepreneurial projects based on deep learning according to claim 1 is characterized in that: Extracting emergency event data from the enhanced structured data to obtain a classified event set includes: Based on the enhanced structured data, the Z-score of the technical terms is calculated using the following formula, and the technical terms with a Z-score less than a threshold are eliminated to obtain differential technical terms: Where Z is the Z-score, which is used to quantify the degree of frequency mutation of technical terms, a is the current frequency, μ is the historical mean, and σ is the historical standard deviation; Analyzing innovative expressions in the enhanced structured data; Marking the difference technology in the same semantic unit as the innovative expression and the difference technology term as a technology breakthrough event to obtain a technology breakthrough candidate set; the difference technology is represented by the difference technology term; Based on the enhanced structured data, the differences in policy terms are quantified to obtain a set of policy change events; quantifying market volatility based on the enhanced structured data to obtain a market volatility event set; Extracting core entity features of events in the technology breakthrough candidate set, the policy change event set, and the market volatility event set; wherein the core entity features include execution subject data, key action data, and execution time data; The Jaccard similarity between each of the core entity features is calculated, and the events represented by the core entity feature pairs whose Jaccard similarity is greater than a threshold are defined as repeated events; and the repeated events are merged to obtain the classified event set.

6. The method for intelligent screening and evaluation of innovative and entrepreneurial projects based on deep learning according to any one of claims 1 to 5, characterized in that: The step of screening the innovation and entrepreneurship projects based on the dynamic influence coefficient matrix to obtain an intelligent screening report of the innovation and entrepreneurship projects includes: Normalizing the four-dimensional coefficients in the dynamic influence coefficient matrix to obtain a standardized influence matrix; Based on the dynamic impact coefficient matrix and the investor risk preference profile, loading the preset weights to obtain a weight vector; Based on the weight vector and the standardized impact matrix, a preliminary project list is generated; the preliminary project list includes the four-dimensional risk value of the innovation and entrepreneurship project; Dynamically adjust the risk threshold based on real-time market volatility, compare it with the four-dimensional risk values ​​of the projects included in the preliminary project list, and eliminate the projects that do not meet the threshold to obtain a filtered project list; Performing attribution analysis on the filtered project list and the standardized impact matrix to obtain an attribution report; An intelligent screening report for the innovation and entrepreneurship projects is generated based on the filtered project list and the attribution report.

7. The method for intelligent screening and evaluation of innovative and entrepreneurial projects based on deep learning according to claim 1 is characterized in that: After screening the projects based on the dynamic impact coefficient matrix to obtain an intelligent screening report of innovative and entrepreneurial projects, the method further includes: Compare actual investment results data with forecast data and calculate the average error rate; Mark the model whose average error rate is greater than the threshold, and obtain an error analysis report; Adjusting the model based on the error analysis report and the recent investment data set to obtain an updated model version; Based on the regulatory inquiry records of the multi-source dataset, updating the knowledge graph nodes to obtain an updated knowledge graph version; Deploy the updated model version and the updated knowledge graph version, and record the operation data to obtain the system operation log.

8. An intelligent screening and evaluation system for innovative and entrepreneurial projects based on deep learning, characterized by: The system comprises: A source module is used to obtain multi-source data through government policy platforms, news aggregation services, social media streams, and technology monitoring sites; the multi-source data is used to represent policy signals and social media sentiment regarding innovation and entrepreneurship projects; An enhancement module, configured to perform data enhancement on the multi-source data through a BiLSTM-CRF model to obtain enhanced structured data; An event module is used to extract emergency event data from the enhanced structured data to obtain a classified event set; the emergency event data is used to characterize the content of the emergency event; An association module, configured to calculate the association between the emergency events and the projects based on the project knowledge base and the classified event set, and obtain a project association network diagram; A quantification module, configured to quantify the impact of the emergency on the innovation and entrepreneurship project based on the project association network diagram, and obtain a dynamic impact coefficient matrix; A screening module is used to screen the innovation and entrepreneurship projects based on the dynamic influence coefficient matrix to obtain an intelligent screening report on the innovation and entrepreneurship projects.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images