Intelligent service method and system based on innovation and entrepreneurship platform

Abstract

The invention discloses an intelligent service method and system based on an innovation and entrepreneurship platform, and belongs to the technical field of innovation and entrepreneurship service platform, the intelligent service method and system based on the innovation and entrepreneurship platform comprises the following specific steps: step 1, multi-source data acquisition and preprocessing: collecting structured data and unstructured data, then cleaning, labeling and storing the collected data; and 2, multi-modal data fusion and feature extraction: carrying out feature mapping on the data acquired in the step 1, then carrying out adversarial training fusion, and finally carrying out dimension reduction and compression. Through multi-modal data fusion and integration of market and user behavior data, the system breaks through the single-dimension limitation of a traditional platform, supports real-time updating and self-learning through a dynamic knowledge graph, adapts to a rapidly changing entrepreneurship environment, automatically identifies potential demands of a user through an appeal management module, generates a service task, optimizes a service calling process, and improves the service efficiency. And the response speed is increased to a millisecond level.

Description

Technical Field

[0001] The present invention belongs to the technical field of innovation and entrepreneurship service platforms, and particularly relates to a smart service method and system based on an innovation and entrepreneurship platform. Background Art

[0002] An innovation and entrepreneurship platform is a comprehensive platform that provides support, resources, and services for innovators and entrepreneurs. The goal of these platforms is to help entrepreneurs transform ideas into actual products or services and promote the development of innovation and entrepreneurship activities. Innovation and entrepreneurship platforms generally include the following aspects: Resource support: Provide financial support, technical support, marketing support, etc. For example, some platforms provide startup funds or investor networks to help entrepreneurs obtain financing. Training and coaching: Provide the knowledge training, skill improvement, and professional coaching required by entrepreneurs. Many platforms will organize startup competitions, lectures, seminars, etc. to help entrepreneurs learn how to manage enterprises, conduct market research, and formulate business plans. Cooperation network: Build a wide innovation cooperation network to promote cooperation and interaction among entrepreneurs, other entrepreneurs, enterprises, research institutions, and investors. Technical support and experimental environment: Some platforms provide technical support and R & D environment for innovators to help them conduct product R & D and experimental verification and lower the startup threshold. Market channels: Through the promotion and resource integration of the platform, help entrepreneurs better reach the target market and customers and enhance market competitiveness.

[0003] Existing innovation and entrepreneurship platforms have the following problems: scattered data, the resource library does not integrate multi-modal data, resulting in information silos, the recommendation algorithm depends on a single dimension, lacks dynamic analysis of the demand for skills in the industry, the service call response delay is high, and it is difficult to meet the real-time needs in complex scenarios. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to overcome the above-mentioned disadvantages of the prior art and provide a smart service method and system based on an innovation and entrepreneurship platform.

[0005] The technical solution adopted to solve the above technical problem is: A smart service method based on an innovation and entrepreneurship platform includes the following specific steps:

[0006] Step 1: Multi-source data collection and preprocessing:

[0007] Collect structured data and unstructured data, and then clean, label, and store the collected data;

[0008] Step 2: Multi-modal data fusion and feature extraction:

[0009] Perform feature mapping on the data collected in Step 1, then perform adversarial training fusion, and finally reduce the dimension and compress;

[0010] Step 3: Dynamic Knowledge Graph Construction and Update:

[0011] Graph Initialization: Define the nodes of the graph, define the relationships between the nodes, and use Neo4j to store the initial graph;

[0012] Incremental Update: Detect new data, trigger local graph updates, predict relationship changes based on historical event sequences, and update the embedding vector h of the affected subgraph nodes v , reducing the global computational overhead;

[0013] Step 4: Intelligent Recommendation and Resource Matching:

[0014] Input user registration information and behavior logs, extract temporal behavior features through the LSTM model to generate a dynamic portrait vector, then perform multi-objective recommendation generation, and finally feedback the user's click and ignore behaviors on the recommendation results to the DQN model in real time to update the policy network parameters;

[0015] Step 5: Automated Service Invocation and Execution:

[0016] First, parse the service requirements, then dynamically allocate weights according to the load of the microservice instances, use the M / M / c model to calculate the optimal number of instances c, and finally call the robotic process automation to execute standardized tasks;

[0017] Step 6: User Interaction and Immersive Experience:

[0018] Through user questions, the system extracts associated nodes from the knowledge graph to generate structured answers;

[0019] Step 7: Performance Monitoring and Dynamic Optimization:

[0020] Monitor the recommendation quality and system efficiency, and then perform non-stop updates on the deficiencies of the system.

[0021] Through the above technical solutions, through multi-modal data fusion, integrating market and user behavior data, breaking through the limitations of the single dimension of traditional platforms, optimizing the service call process, and improving the response speed to the millisecond level.

[0022] Furthermore, the structured data in Step 1 includes enterprise registration information and financing records, the unstructured data in Step 1 includes social media comments, startup forum texts, product roadshow videos, and industry reports. The cleaning in Step 1 includes removing duplicate and noisy data, standardizing the format, and the annotation in Step 1 uses NLP tools to classify texts and perform scene annotation on images.

[0023] Through the above technical solution, noise and redundant data are removed by cleaning, the error of subsequent analysis is reduced, the accuracy of model training is improved, and unstructured data is labeled and classified to make the data more structured and facilitate algorithm processing.

[0024] Further, the feature mapping in the second step includes:

[0025] Text data X t : Converted into a vector through word embedding

[0026] Image data X i : Using CNN to extract features

[0027] Structured data X s : Numerical fields are normalized, and categorical fields are one-hot encoded as

[0028] Through the above technical solution, data heterogeneity can be eliminated and feature consistency can be improved.

[0029] Further, the adversarial training fusion includes the following specific formula:

[0030] Input X = {X t , X i , X s} to the generator G, generating the fusion feature F = [f t , f i , f s ;

[0031] The discriminator D distinguishes real data from generated features, and the optimization objective:

[0032]

[0033] After that, the high-dimensional feature F is mapped to the low-dimensional latent space through the VAE model Optimizing the reconstruction loss and KL divergence.

[0034] Through the above technical solution, the computational complexity is reduced, the subsequent model calculation amount is reduced, and the feature expression ability is improved.

[0035] Further, the following formula is adopted for predicting the relationship change based on the historical event sequence in the third step:

[0036]

[0037] Where e i is the event representation, Q is the query vector, indicating the target to be predicted currently; K i is the key vector, indicating the features of each historical event; α iis the attention weight, which assigns a higher weight to recent technological breakthrough events and a lower weight to outdated policy events. is the prediction result.

[0038] Through the above technical solutions, the real-time performance is enhanced, the computational overhead is reduced, and the trend prediction is more accurate.

[0039] Furthermore, the DQN model is specifically defined as follows: Define the state s t (current user demand), action a t (recommended resources), and reward r t = α·CTR + β·conversion rate, and optimize the Q function:

[0040]

[0041] Through the above technical solutions, accurate personalized recommendations are provided, the real-time adaptability is improved, and the diversity of recommendations can be effectively guaranteed.

[0042] Furthermore, it includes a data integration layer, an intelligent analysis layer, and a service execution layer. The data integration layer is responsible for integrating multi-source data, adopting multi-source heterogeneous data access technology, and supporting the fusion of structured and unstructured data; the intelligent analysis layer is responsible for constructing a knowledge graph, based on an unsupervised adaptive algorithm, combining user interests, skill gaps, and market trends, to provide personalized resource recommendations; the service execution layer is responsible for integrating RPA (Robotic Process Automation) to handle routine tasks, optimizing the service call efficiency, achieving millisecond-level response, and supporting intelligent question answering.

[0043] The beneficial effects of the present invention are as follows: Through multi-modal data fusion, the present invention integrates market and user behavior data, breaks through the single-dimensional limitation of traditional platforms, supports real-time update and self-learning through a dynamic knowledge graph, adapts to the rapidly changing entrepreneurial environment, automatically identifies potential user needs and generates service tasks through a demand management module, optimizes the service call process, and improves the response speed to the millisecond level. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 is the flowchart of the method of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0045] In order to make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0046] As Figure 1 shown, the intelligent service method and system based on an innovation and entrepreneurship platform in this embodiment include the following specific steps:

[0047] Step 1: Multi-source data collection and preprocessing:

[0048] Collect structured and unstructured data, and then clean, annotate, and store the collected data;

[0049] Step 2: Multimodal data fusion and feature extraction:

[0050] Perform feature mapping on the data collected in Step 1, then conduct adversarial training fusion, and finally reduce the dimension and compress;

[0051] Step 3: Dynamic knowledge graph construction and update:

[0052] Graph initialization: Define the nodes of the graph, define the relationships between the nodes, and use Neo4j to store the initial graph;

[0053] Incremental update: Detect new data, trigger local graph update, predict relationship changes based on historical event sequences, and update the embedding vector h of the nodes in the affected subgraph v , reducing the global computational overhead;

[0054] Step 4: Intelligent recommendation and resource matching:

[0055] Input user registration information and behavior logs, extract temporal behavior features through the LSTM model to generate dynamic portrait vectors, then perform multi-objective recommendation generation, and finally, based on the user's click and ignore behaviors on the recommendation results, feedback to the DQN model in real time to update the policy network parameters;

[0056] Step 5: Automated service invocation and execution:

[0057] First, parse the service requirements, then dynamically allocate weights according to the load of the microservice instances, use the M / M / c model to calculate the optimal real instance c, and finally call the robotic process automation to execute standardized tasks;

[0058] Step 6: User interaction and immersive experience:

[0059] Through user questions, the system extracts associated nodes from the knowledge graph to generate structured answers;

[0060] Step 7: Performance monitoring and dynamic optimization:

[0061] Monitor the recommendation quality and system efficiency, and then perform non-stop updates on the deficiencies of the system.

[0062] Through multimodal data fusion, integrate market and user behavior data, break through the single-dimensional limitation of traditional platforms, optimize the service invocation process, and improve the response speed to the millisecond level.

[0063] The structured data in step one includes enterprise registration information and financing records. The unstructured data in step one includes social media comments, startup forum texts, product roadshow videos, and industry reports. The cleaning in step one includes removing duplicate and noisy data, and standardizing the format. The annotation in step one uses NLP tools to classify texts and perform scene annotation on images.

[0064] Remove noise and redundant data through cleaning, reduce errors in subsequent analysis, improve the accuracy of model training, annotate and classify unstructured data to make the data more structured and facilitate algorithm processing.

[0065] The feature mapping in step two includes:

[0066] Text data X t : Convert to a vector through word embedding

[0067] Image data X i : Use CNN to extract features

[0068] Structured data X s : Normalize numerical fields, and one-hot encode categorical fields into

[0069] This can eliminate data heterogeneity and improve feature consistency.

[0070] The adversarial training fusion includes the following specific formula:

[0071] Input X = {X t , X i , X s} to the generator G, and generate the fused feature F = [f t , f i , f s ;

[0072] The discriminator D distinguishes real data from generated features, and the optimization objective:

[0073]

[0074] After that, map the high-dimensional feature F to the low-dimensional latent space through the VAE model Optimize the reconstruction loss and KL divergence.

[0075] Reduce the computational complexity, reduce the computational amount of the subsequent model, and improve the feature expression ability.

[0076] The formula for predicting relationship changes based on the historical event sequence in step three is as follows:

[0077]

[0078] where e i is the event representation, Q is the query vector representing the target to be predicted currently; K i is the key vector representing the features of each historical event; α i is the attention weight, which assigns higher weights to recent technological breakthrough events and lower weights to obsolete policy events, and is the prediction result.

[0079] Enhance real-time performance, reduce computational overhead, and make trend prediction more accurate.

[0080] The DQN model is specifically as follows: Define the state s t (current user demand), action a t (recommended resource), reward r t = α·CTR + β·conversion rate, and optimize the Q function:

[0081]

[0082] Precise personalized recommendation, improve real-time adaptability, and can effectively ensure the diversity of recommendations.

[0083] It includes a data integration layer, an intelligent analysis layer, and a service execution layer. The data integration layer is responsible for integrating multi-source data, adopts multi-source heterogeneous data access technology, and supports the fusion of structured and unstructured data; the intelligent analysis layer is responsible for constructing a knowledge graph, based on an unsupervised adaptive algorithm, combines user interests, skill gaps, and market trends to provide personalized resource recommendations; the service execution layer is responsible for integrating RPA (Robotic Process Automation) to handle routine tasks, optimize service call efficiency, achieve millisecond-level response, and support intelligent question answering.

[0084] The above is only a preferred embodiment of the present invention and is not intended to limit the protection scope of the present invention.

Claims

1. The intelligent service method based on the innovation and entrepreneurship platform is characterized by: The specific steps include: Step 1: Multi-source data collection and preprocessing: Collect structured and unstructured data, and then clean, annotate and store the collected data; Step 2: Multimodal data fusion and feature extraction: The data collected in step 1 is feature mapped, then fused through adversarial training, and finally dimensionally reduced and compressed; Step 3: Dynamic knowledge graph construction and update: Graph initialization: define the nodes of the graph, define the relationships between the nodes, and use Neo4j to store the initial graph; Incremental update: detect new data, trigger local graph update, predict relationship changes based on historical event sequences, and update the embedding vector h for the affected subgraph nodes v , reduce global computational overhead; Step 4: Intelligent recommendation and resource matching: Input user registration information and behavior logs, extract temporal behavior features through the LSTM model, generate dynamic portrait vectors, and then generate multi-target recommendations. Finally, based on the user's click or ignore behavior on the recommendation results, the DQN model is fed back in real time to update the policy network parameters. Step 5: Automated service call and execution: First, analyze the service requirements, then dynamically allocate weights according to the load of the microservice instance, use the M / M / c model to calculate the optimal real number instance c, and finally call the robotic process automation to perform standardized tasks; Step 6: User interaction and immersive experience: Through user questions, the system extracts related nodes from the knowledge graph and generates structured answers; Step 7: Performance monitoring and dynamic optimization: Monitor recommendation quality and system efficiency, and perform ongoing updates to address system deficiencies.

2. The intelligent service method based on the innovation and entrepreneurship platform according to claim 1 is characterized in that: The structured data in step one includes enterprise registration information and financing records; the unstructured data in step one includes social media comments, entrepreneurial forum texts, product roadshow videos and industry reports; the cleaning in step one includes removing duplicate and noisy data and standardizing the format; the annotation in step one uses NLP tools to classify text and perform scene annotation on images.

3. The intelligent service method based on the innovation and entrepreneurship platform according to claim 2 is characterized in that: The feature mapping in step 2 includes: Text dataX t : Convert to vector through word embedding Image data X i : Use CNN to extract features Structured DataX s : Numeric fields are normalized, and categorical fields are one-hot encoded as 4. The intelligent service method based on the innovation and entrepreneurship platform according to claim 3 is characterized in that: The adversarial training fusion includes the following specific formula: Input X = {X t ,X i ,X s } to the generator G, generating fusion feature F = [f t ,f i ,f s ]; The discriminator D distinguishes between real data and generated features, optimizing the goal: Then the high-dimensional feature F is mapped to the low-dimensional latent space through the VAE model Optimize reconstruction loss and KL divergence.

5. The intelligent service method based on the innovation and entrepreneurship platform according to claim 4 is characterized in that: In step 3, the following formula is used to predict the relationship change based on the historical event sequence: where e i is the event representation, Q is the query vector, which represents the target that needs to be predicted; K i is the key vector, representing the characteristics of each historical event; α i is the attention weight, giving higher weight to recent technological breakthroughs and lower weight to old policy events. is the prediction result.

6. The intelligent service method based on the innovation and entrepreneurship platform according to claim 5 is characterized in that: The DQN model is specifically as follows: define the state s t (current user demand), action a t (Recommended Resources), Rewards t =α·CTR+β·conversion rate, optimize Q function:

7. The system of the intelligent service method based on the innovation and entrepreneurship platform according to claim 6 is characterized in that: It includes data integration layer, intelligent analysis layer and service execution layer. The data integration layer is responsible for integrating multi-source data, adopts multi-source heterogeneous data access technology, and supports the integration of structured and unstructured data. The intelligent analysis layer is responsible for building a knowledge graph and providing personalized resource recommendations based on unsupervised adaptive algorithms, combined with user interests, skill gaps, and market trends. The service execution layer is responsible for integrating RPA (robotic process automation) to handle routine tasks, optimize service call efficiency, achieve millisecond-level response, and support intelligent question and answer.

Images