AI-based science and technology reward review system

Abstract

An AI-based science and technology reward review system is characterized by comprising a material pre-review module, an expert matching module and a man-machine cooperation final review module, according to the invention, the project is reviewed in a man-machine cooperation mode, abnormal labeling is carried out at a specific position according to scoring requirements, and the corresponding scoring materials are automatically connected to facilitate comparison and discrimination by experts, so that the scoring speed of the experts is improved, and problems existing in the materials are visually displayed; the POLAR strategy model is used for special training, and the NLP technology, the OCR technology and the GNN technology are used for cooperative processing, so that the AI model can accurately read the content of the declaration material for auxiliary scoring.

Description

Technical Field

[0001] This invention relates to the field of science and technology award evaluation, and in particular to an AI-based science and technology award evaluation system. Background Technology

[0002] The science and technology award review is an important part of the implementation of science and technology policies and rewards. It involves a thorough review of the basic qualifications of candidate projects and candidates, the validity of supporting materials, and the relevance of supporting materials to the candidate projects / candidates. Candidate projects, candidates and supporting materials that do not meet the application requirements are screened out and eliminated, and the winning projects and companies are finally decided.

[0003] Currently, the review of science and technology awards is generally conducted manually. Due to the fragmented nature of science and technology award data, the diverse range of materials involved, low levels of data sharing, and the lack of effective deduplication methods, reviewers find it difficult to conduct a comprehensive review of the supporting materials for award applications manually. For example, a particular award application may involve numerous supporting materials and a large number of past award recipients. A comprehensive review would require checking the usage of each supporting material in previous award projects, a massive undertaking prone to omissions. This review of material reuse is only one aspect of the science and technology award review process. Therefore, the current manual review method for science and technology awards is not only inefficient but also risks compromising the fairness of the award review process due to the inability to conduct a comprehensive review.

[0004] Currently, AI (Artificial Intelligence) technology is maturing, and more and more industries are using AI as an auxiliary tool to improve productivity and efficiency. AI has a natural advantage in text review because it can quickly search databases for comparison and analysis. Therefore, an applicant designed an AI-based science and technology award review system to address the shortcomings of the existing review methods. Summary of the Invention

[0005] The main objective of this invention is to provide an AI-based science and technology award review system to address the shortcomings of existing technologies.

[0006] To achieve the above objectives, the specific plan is as follows:

[0007] An AI-based science and technology award review system includes: a materials pre-review module, an expert matching module, and a human-machine collaborative final review module;

[0008] The material pre-review module employs NLP (Natural Language Processing), OCR (Optical Character Recognition), and GNN (Graph Neural Network) technologies in a collaborative manner. OCR image recognition scans material images, extracting key information such as text, tables, and seals, and supports handwriting recognition. NLP semantic parsing uses the BERT model to understand text content and identify entities and logical relationships. The POLAR strategy model is used to determine the completeness of the application materials and whether they meet the required standards. If they do, the application proceeds to the expert matching module; otherwise, it is returned for correction.

[0009] The expert matching module selects a corresponding number of experts to participate in the review based on the matching degree of experts in the expert database. The matching degree = domain relevance × activity level × avoidance weight.

[0010] The human-machine collaborative final review module uses NLP (Natural Language Processing), OCR (Optical Character Recognition), and GNN (Graph Neural Network) technologies to work together. It uses the POLAR strategy model to generate corresponding knowledge graphs and innovation level evaluations, and marks logical contradictions, data outliers, and key points on the corresponding material display page to assist in the review. The scoring module connects to the relevant scoring application materials for split-screen display.

[0011] Furthermore, the logical contradictions are verified using NLP technology to validate the material logic.

[0012] Furthermore, the formula for determining key points is Sim = α·Jaccard(Keywords) + β·Cos(BERT_Embedding), where (α + β = 1) if the repetition rate exceeds the threshold, an automatic red warning is issued.

[0013] Furthermore, the data outlier detection is based on the POLAR strategy model to retrieve and map relevant data values. When the deviation of a data value from the currently recorded value exceeds a preset threshold, it is highlighted in red as a warning.

[0014] Furthermore, the training samples used in the POLAR strategy model include review criteria, scoring criteria, samples of historical award-winning projects, feasibility study reports of historical award-winning projects, template samples of material formats, samples of qualification certificates, and historical expert scoring sheets.

[0015] Compared with the prior art, the present invention has the following beneficial effects:

[0016] 1. This invention adopts a human-machine collaborative approach to review projects, marks anomalies at specific locations according to scoring requirements, and automatically connects to corresponding scoring materials to facilitate expert comparison and judgment, thereby improving the speed of expert scoring and intuitively showing the problems existing in the materials.

[0017] 2. This invention uses the POLAR strategy model for specialized training and leverages NLP, OCR, and GNN technologies in a collaborative process to enable the AI ​​model to accurately read the contents of the application materials for auxiliary scoring. Attached Figure Description

[0018] Figure 1 This is a system structure diagram of the present invention; Detailed Implementation

[0019] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings and preferred embodiments.

[0020] Combination Figure 1 The aforementioned AI-based science and technology award review system includes: a materials pre-review module, an expert matching module, and a human-machine collaborative final review module;

[0021] The material pre-review module employs NLP (Natural Language Processing), OCR (Optical Character Recognition), and GNN (Graph Neural Network) technologies in a collaborative manner. OCR image recognition scans material images, extracting key information such as text, tables, and seals, and supports handwriting recognition. NLP semantic parsing uses the BERT model to understand text content and identify entities and logical relationships. The POLAR strategy model is used to determine the completeness of the application materials and whether they meet the required standards. If they do, the application proceeds to the expert matching module; otherwise, it is returned for correction.

[0022] The expert matching module selects a corresponding number of experts to participate in the review based on the matching degree of experts in the expert database. The matching degree = domain relevance × activity level × avoidance weight.

[0023] The human-machine collaborative final review module uses NLP (Natural Language Processing), OCR (Optical Character Recognition), and GNN (Graph Neural Network) technologies to work together. It uses the POLAR strategy model to generate corresponding knowledge graphs and innovation level evaluations, and marks logical contradictions, data outliers, and key points on the corresponding material display page to assist in the review. The scoring module connects to the relevant scoring application materials for split-screen display.

[0024] In a preferred embodiment, the logical contradictions are verified using NLP technology to validate the material logic.

[0025] In a preferred embodiment, the formula for determining the key points is Sim = α·Jaccard(Keywords) + β·Cos(BERT_Embedding), where (α + β = 1) if the repetition rate exceeds the threshold, an automatic red warning is issued.

[0026] In a preferred embodiment, the data anomaly detection is based on the POLAR strategy model to retrieve and map relevant data values. When the deviation of a data value from the currently recorded value exceeds a preset threshold, it is highlighted in red as a warning.

[0027] In a preferred embodiment, the training samples used in the POLAR strategy model include review criteria, scoring criteria, samples of historical award-winning projects, feasibility study reports of historical award-winning projects, template samples of material formats, samples of qualification certificates, and historical expert scoring sheets.

[0028] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. An AI-based science and technology award review system, characterized in that, include: The module includes a material pre-review module, an expert matching module, and a human-machine collaborative final review module. The material pre-review module employs NLP (Natural Language Processing), OCR (Optical Character Recognition), and GNN (Graph Neural Network) technologies in a collaborative manner. OCR image recognition scans material images, extracting key information such as text, tables, and seals, and supports handwriting recognition. NLP semantic parsing uses the BERT model to understand text content and identify entities and logical relationships. The POLAR strategy model is used to determine the completeness of the application materials and whether they meet the required standards. If they do, the application proceeds to the expert matching module; otherwise, it is returned for correction. The expert matching module selects a corresponding number of experts to participate in the review based on the matching degree of experts in the expert database. The matching degree = domain relevance × activity level × avoidance weight. The human-machine collaborative final review module uses NLP (Natural Language Processing), OCR (Optical Character Recognition), and GNN (Graph Neural Network) technologies to work together. It uses the POLAR strategy model to generate corresponding knowledge graphs and innovation level evaluations, and marks logical contradictions, data outliers, and key points on the corresponding material display page to assist in the review. The scoring module connects to the relevant scoring application materials for split-screen display.

2. The AI-based science and technology award review system according to claim 1, characterized in that, include: The logical contradictions were verified using NLP technology to validate the material logic.

3. The AI-based science and technology award review system according to claim 1, characterized in that, include: The formula for determining key words is Sim = α·Jaccard(Keywords) + β·Cos(BERT_Embedding), where (α + β = 1) if the repetition rate exceeds the threshold, an automatic red warning will be issued.

4. The AI-based science and technology award review system according to claim 1, characterized in that, include: The data anomaly detection is based on the POLAR strategy model to retrieve and map relevant data values. When the deviation of a data value from the currently recorded value exceeds a preset threshold, it is highlighted in red as a warning.

5. The AI-based science and technology award review system according to claim 1, characterized in that, include: The training samples used in the POLAR strategy model include review criteria, scoring criteria, samples of historical award-winning projects, feasibility study reports of historical award-winning projects, template samples of material formats, samples of qualification certificates, and historical expert scoring sheets.

Images