Project review method and device, electronic equipment and computer program product

Abstract

The present disclosure relates to a project review method and device, electronic equipment and computer program product, and belongs to the technical field of computers. The method comprises: obtaining a project file of a target project to be reviewed, and obtaining a full expert list from a review system; obtaining project information from the project file, and inputting the project information and expert information in the full expert list into a neural network to obtain a matching score between the target project and each expert; determining a constraint condition corresponding to the target project, and screening a candidate expert list from the full expert list according to the matching score and the constraint condition; determining a review expert from the candidate expert list, and sending a review task of the target project to the review expert, and obtaining an initial review result of the target project according to a review result of the review expert. The present disclosure automatically matches project and expert information through a neural network and screens candidate experts in combination with a constraint condition, which can improve review efficiency and accuracy and reduce artificial intervention bias.

Description

Technical Field

[0001] This disclosure relates to the field of computer technology, and more specifically, to a project review method, a project review device, an electronic device, and a computer program product. Background Technology

[0002] With the deepening of digital transformation, various associations and enterprises generally use online systems to manage the application and review of various projects. The current project application and review systems generally adopt a basic information architecture. The process design of this architecture is too rigid and lacks dynamic adaptability, which makes it impossible to flexibly adjust the review process to meet the changes in actual scenarios.

[0003] In the expert allocation process, managers must rely on manual methods to screen and match experts from the expert database. This process is not only time-consuming and labor-intensive, but also prone to subjective preferences and human biases, which seriously affect the fairness of the review and the reliability of the results.

[0004] Therefore, there is an urgent need in this field for a project review method that can improve review efficiency and accuracy and reduce human intervention bias.

[0005] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this disclosure is to provide a project review method, a project review device, an electronic device, and a computer program product, which can at least to some extent improve review efficiency and accuracy and reduce human intervention bias.

[0007] According to the first aspect of this disclosure, a method for reviewing a project is provided, comprising: Obtain the project files of the target project to be reviewed, and retrieve the full list of experts from the review system; The project information is obtained from the project file, and the project information and the expert information of each expert in the full expert list are input into the neural network to obtain the matching score between the target project and each expert. Determine the constraints corresponding to the target project, and based on the matching score and the constraints, filter from the full list of experts to obtain a list of candidate experts; Review experts are selected from the list of candidate experts, and the review task of the target project is sent to the review experts. The initial review result of the target project is obtained based on the review results of the review experts.

[0008] In one exemplary embodiment of this disclosure, the constraints include load balancing constraints and domain avoidance constraints. The load balancing constraints are used to limit the maximum number of projects that the expert can review within a specified time period, and the domain avoidance constraints are used to exclude experts who are related to the target project.

[0009] In one exemplary embodiment of this disclosure, after determining the review experts from the list of candidate experts, the method further includes: The network parameters in the neural network are updated based on the expert information of the review experts determined from the list of candidate experts.

[0010] In one exemplary embodiment of this disclosure, the method further includes: The review system provides a project application interface; Based on the basic information of the target project, determine the required and optional fields in the project application interface, as well as the maximum number of project personnel that can be entered.

[0011] In one exemplary embodiment of this disclosure, the method further includes: The multimodal application data in the project file uploaded to the project application interface is identified using optical character recognition technology, and corresponding prompts are displayed on the project application interface based on the recognition results.

[0012] In one exemplary embodiment of this disclosure, the method further includes: The project files uploaded in the project application interface are cross-validated with the application information filled in in the project application interface, and the corresponding prompt information is displayed on the project application interface according to the verification results.

[0013] In one exemplary embodiment of this disclosure, after obtaining the initial review results of the target project, the method further includes: During the meeting discussion phase, a visual list of projects and voting controls are provided in the expert review system.

[0014] In one exemplary embodiment of this disclosure, the step of inputting the project information and the expert information of each expert in the full expert list into a neural network to obtain a matching score between the target project and each expert includes: The project information and the expert information of each expert in the full expert list are input into a graph neural network to obtain the embedding features corresponding to the project information and the embedding features corresponding to the expert information. Based on the embedding features corresponding to the project information and the embedding features corresponding to the expert information, a matching score is obtained between the target project and the expert.

[0015] In one exemplary embodiment of this disclosure, the graph neural network is constructed based on a heterogeneous graph, which includes project nodes, expert nodes, technical field nodes, keyword nodes, and organizational nodes. The project nodes and the expert nodes are connected through the technical field nodes, the keyword nodes, or the organizational nodes.

[0016] In one exemplary embodiment of this disclosure, the method further includes: For the same organizational node, the edge between the project node and the organizational node, and the edge between the expert node and the organizational node are designated as conflict edges. During the forward propagation of the graph neural network, collision detection is performed using collision masks set on the collision edges.

[0017] In one exemplary embodiment of this disclosure, the method further includes: Obtain the project information of the expert's historical projects from the expert information, including the distribution of the projects participated in and the corresponding project time; The technical field nodes connected to the expert nodes are determined based on the distribution of participating projects, and the edge weights corresponding to the edges between the expert nodes and the technical field nodes are determined based on the project time; wherein, the edge weights are negatively correlated with the project time.

[0018] In one exemplary embodiment of this disclosure, the method further includes: Obtain project keywords from the project information of the expert's historical projects, and determine the keyword node connected to the expert node based on the project keywords; Based on the word frequency of the project keywords and the project time, the edge weights corresponding to the edges between the expert nodes and the keyword nodes are determined; wherein, the edge weights are negatively correlated with the project time and positively correlated with the word frequency.

[0019] In one exemplary embodiment of this disclosure, obtaining the matching score between the target project and the expert based on the embedding features corresponding to the project information and the embedding features corresponding to the expert information includes: Natural language processing technology is used to identify the semantic features of the project file, including the potential topics and potential domains of the target project; The review capability characteristics of the experts are obtained from the expert information, including the distribution fields of the projects the experts have participated in and the importance coefficient of the projects they have participated in; The project file's fusion feature is obtained based on its semantic and embedding features, and the expert information's fusion feature is obtained based on the expert's review capability features and the expert information's embedding features. Based on the fusion characteristics of the project documents and the fusion characteristics of the expert information, a matching score is obtained between the target project and the expert.

[0020] According to a second aspect of this disclosure, a project review apparatus is provided, comprising: The project file acquisition module is used to acquire the project files of the target project to be reviewed and to obtain a full list of experts from the review system. The matching score determination module is used to obtain project information from the project file and input the project information and the expert information of each expert in the full expert list into the neural network to obtain the matching score between the target project and each expert; The candidate expert determination module is used to determine the constraints corresponding to the target project, and to filter the candidate expert list from the full expert list based on the matching score and the constraints. The review result determination module is used to determine review experts from the candidate expert list, send the review task of the target project to the review experts, and obtain the initial review result of the target project based on the review results of the review experts.

[0021] According to a third aspect of this disclosure, an electronic device is provided, comprising: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to perform a review method for any of the preceding items by executing the executable instructions.

[0022] According to a fourth aspect of this disclosure, a computer program product is provided, comprising a computer program that, when executed by a processor, implements the review method for the project described in any one of the preceding claims.

[0023] The exemplary embodiments disclosed herein can have the following beneficial effects: In the project review method of this exemplary implementation, a neural network automatically matches project and expert information and selects candidate experts in combination with constraints. On the one hand, this effectively solves the problems of low efficiency in expert allocation and obvious subjective bias, reduces human intervention bias, and improves the fairness and reliability of the review process. On the other hand, by providing a safe and intelligent application review process, it improves the process adaptability of the application plan, as well as the review efficiency and the accuracy of the results.

[0024] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0025] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure. It is obvious that the drawings described below are merely some embodiments of this disclosure, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0026] Figure 1 A flowchart illustrating the project review method according to an exemplary embodiment of this disclosure is shown; Figure 2 A schematic diagram of the process for obtaining matching scores between a target project and various experts through a neural network in an exemplary embodiment of this disclosure is shown. Figure 3 A schematic diagram of the process for obtaining matching scores between a target item and various experts based on embedded features is shown in an exemplary embodiment of this disclosure. Figure 4 A block diagram of an item review apparatus according to an exemplary embodiment of this disclosure is shown; Figure 5 A schematic diagram of the structure of a computer system suitable for implementing the embodiments of the present disclosure is shown. Detailed Implementation

[0027] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided to make this disclosure more comprehensive and complete, and to fully convey the concept of the example embodiments to those skilled in the art. The described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to give a full understanding of embodiments of this disclosure. However, those skilled in the art will recognize that the technical solutions of this disclosure can be practiced with one or more of the specific details omitted, or other methods, components, apparatus, steps, etc., can be employed. In other instances, well-known technical solutions are not shown or described in detail to avoid obscuring various aspects of this disclosure.

[0028] Furthermore, the accompanying drawings are merely illustrative of this disclosure and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0029] This example implementation first provides a method for project review. (See references) Figure 1 As shown, the review method for the above projects may include the following steps: Step S110. Obtain the project files of the target project to be reviewed, and obtain the full list of experts from the review system.

[0030] Step S120. Obtain project information from the project file, and input the project information and the expert information of each expert in the full expert list into the neural network to obtain the matching score between the target project and each expert.

[0031] Step S130. Determine the constraints corresponding to the target project, and based on the matching score and constraints, filter from the full list of experts to obtain a list of candidate experts.

[0032] Step S140. Select review experts from the list of candidate experts, send the review task of the target project to the review experts, and obtain the initial review result of the target project based on the review results of the review experts.

[0033] In the project review method of this exemplary implementation, a neural network automatically matches project and expert information and selects candidate experts in combination with constraints. On the one hand, this effectively solves the problems of low efficiency in expert allocation and obvious subjective bias, reduces human intervention bias, and improves the fairness and reliability of the review process. On the other hand, by providing a safe and intelligent application review process, it improves the process adaptability of the application plan, as well as the review efficiency and the accuracy of the results.

[0034] The steps described above in this example implementation will now be explained in more detail.

[0035] In step S110, the project file of the target project to be reviewed is obtained, and a full list of experts is obtained from the review system.

[0036] In this example implementation, project documents refer to all relevant documents submitted by the target project to be reviewed. These may include project applications, technical solutions, budget reports, team member information, and supporting materials, describing the project's background, objectives, content, methods, expected outcomes, and implementation plan. The review system is a software platform used to manage project applications, expert matching, review processes, and result summarization. It integrates functions such as project information management, expert database management, review task allocation, and review opinion collection. It can also provide functions such as assigned tasks, completed tasks, process circulation, and process monitoring, aiming to support and optimize the project review lifecycle. The full expert list refers to the collection of all expert information stored in the review system that can be used for project review. This list may include each expert's basic information, professional field, academic background, review experience, and historical review records.

[0037] Project documents can be obtained in several ways. For example, the project applicant can send the project documents to the review agency via email, and the staff can manually upload them to the review system; alternatively, the project applicant can upload the project documents to the review system themselves. The review system provides a complete list of experts, set up and maintained by the administrator.

[0038] In this example implementation, a project application interface can be provided in the review system; the required and optional fields in the project application interface, as well as the maximum number of project personnel that can be entered, are determined based on the basic information of the target project.

[0039] The project application interface refers to a user-friendly interface built into the review system for project applicants to submit project-related information and documents. This interface aims to standardize the entry of project information, ensuring its completeness and accuracy, and providing a high-quality data foundation for subsequent expert matching and review. The system can dynamically adjust the information items required in the application interface based on the actual situation of the project application. For example, the system can preset multiple project type templates, each with a set of required and optional field rules. When the user selects a project type, the interface automatically loads the corresponding fields. Alternatively, the system can dynamically adjust field requirements according to the stage of the project application, ensuring that necessary information is collected at different stages. Furthermore, it can dynamically generate field requirements based on basic information such as the project's field and funding scale through a configured rule engine. After the project application is completed, approved projects will be synchronized to the project adjustment module, allowing for adjustments and modifications to individual projects.

[0040] The maximum number of personnel that can be entered for a project refers to the system's restriction on the number of project participants that a project applicant can submit. For example, the system can preset the maximum number of personnel based on the project type or level; for instance, a research project may limit the number of main participants to no more than 10. Alternatively, the system can dynamically calculate the maximum number of personnel allowed based on the project budget. Or, administrators can manually configure the maximum number of personnel limits for different projects according to actual needs.

[0041] Through the above technical solutions, the review system can effectively guide project applicants to submit high-quality and standardized project information, solving the problem of incomplete or non-standard project information from the source, and making the entire project review process more efficient, transparent and scientific.

[0042] In this example implementation, optical character recognition technology can also be used to identify the multimodal application data in the project file uploaded in the project application interface, and the corresponding prompt information can be displayed on the project application interface based on the recognition results.

[0043] Optical Character Recognition (OCR) is a technology that converts text content in images into editable and searchable digital text. Its function is to automatically extract text information from unstructured or semi-structured image data. Multimodal application data refers to project documents that contain not only plain text information but also various other data formats such as tables, charts, images, and handwritten signatures. This data may exist in different formats and structures within the project documents, requiring comprehensive processing. OCR technology can automatically process multimodal application data in uploaded project documents to extract the structured and unstructured information contained within. After completing the recognition of multimodal data in the project documents, the system provides feedback to the user with key information, potential problems, or suggestions. For example, if a key field in the project document is found to be missing, the system will prompt the user to supplement it; if the recognized data is inconsistent with the data already entered on the application interface, the system will prompt the user to verify it.

[0044] For example, OCR (Optical Character Recognition) can be used to identify the location of the seal and the area of ​​the signature, and to detect whether the official seal or the signature of the person in charge is missing. This avoids problems such as missing seals, missing signatures, and incorrect formats in the attachments of the application. It eliminates the need for manual inspection of each document, making it highly efficient and less prone to omissions.

[0045] In this example implementation, the project files uploaded in the project application interface can also be cross-validated with the application information filled in in the project application interface, and the corresponding prompt information can be displayed on the project application interface according to the verification results.

[0046] Cross-validation refers to comparing and verifying the same project data—that is, the content of the file uploaded by the user and the structured information manually entered by the user in the interface—to ensure the consistency and accuracy of the data. This validation aims to identify and correct problems such as data entry errors and discrepancies between file content and application information, thereby improving the overall quality and reliability of the application data. Cross-validation can be implemented in various ways. For example, it can utilize machine learning models to train the model to identify specific entities in the project file (such as project number, applicant institution, research direction, etc.) and perform semantic comparison with the structured data entered in the application interface.

[0047] Through the above technical solutions, during the project application stage, the system can automatically extract multimodal application data from user-uploaded project files and provide real-time prompts based on the recognition results, greatly improving the efficiency and accuracy of project application information entry and avoiding incomplete or inconsistent application information caused by manual verification and entry errors. Simultaneously, cross-validation effectively solves the problem of data inconsistencies caused by manual errors or discrepancies between file content and application information, significantly improving the efficiency and quality of the entire project review process.

[0048] In step S120, project information is obtained from the project file, and the project information and the expert information of each expert in the full expert list are input into the neural network to obtain the matching score between the target project and each expert.

[0049] In this example implementation, project information refers to structured data extracted from project documents that describes the core content of the target project. This data may include the project name, applicant organization, research direction, technical keywords, field of study, and project leader. Expert information refers to detailed information for each expert in the full expert list. This information may include the expert's name, title, affiliation, research direction, professional expertise, past review projects, and academic achievements, used to assess the match between the expert and the project. The matching score is a numerical value calculated using a neural network to quantify the degree of correlation between the target project and a specific expert. Generally, the higher the score, the higher the match between the expert and the project, and the more suitable the expert is for reviewing the project.

[0050] Project information can be obtained using text parsing techniques, such as keyword extraction and topic modeling, to identify structured information like project names, research areas, and technical keywords from unstructured project documents. Expert information can be directly retrieved from a full list of experts. Subsequently, this project and expert information can be encoded into numerical vectors and fed into a pre-trained neural network model. This neural network model learns from a large amount of historical project and expert matching data, outputting a matching score; generally, a higher score indicates a stronger relevance between the project and the expert. This neural network model can be, for example, a graph neural network, a graph matching neural network, or a large language model.

[0051] In this example implementation, as Figure 2 As shown, the project information and the expert information of each expert in the full expert list are input into the neural network to obtain the matching score between the target project and each expert. This can specifically include the following steps: Step S210. Input the project information and the expert information of each expert in the full expert list into the graph neural network to obtain the embedding features corresponding to the project information and the embedding features corresponding to the expert information.

[0052] In this example implementation, a graph neural network can be used to learn the embeddings of project nodes and expert nodes. A graph neural network is a deep learning model specifically designed for graph-structured data. Its core principle is to learn node representations by aggregating information from neighboring nodes. Through the processing of the graph neural network, embedding features corresponding to project information and expert information can be obtained. Embedding features are representations that map high-dimensional, sparse, or unstructured data into a low-dimensional, dense vector space. These vectors can effectively capture the semantic and structural information of the original data.

[0053] In this example implementation, the graph neural network is built based on a heterogeneous graph, which includes project nodes, expert nodes, technical field nodes, keyword nodes, and organizational nodes. Project nodes and expert nodes are connected through technical field nodes, keyword nodes, or organizational nodes.

[0054] Compared to homogeneous graphs where all nodes and edges are of the same type, heterogeneous graphs can more naturally represent the diverse relationships between entities in the real world. For example, in a project review scenario, projects, experts, technical fields, keywords, and organizations are all different types of entities, and the connections between them are also different. Constructing heterogeneous graphs allows for more refined modeling of these different types of entities and their interactions. When graph neural networks propagate and aggregate information on heterogeneous graphs, they can distinguish different types of nodes and edges and perform weighted processing based on their semantic information, thereby learning more expressive project embedding features and expert embedding features. These embedding features not only contain the attribute information of the projects and experts themselves, but also incorporate their structural information and semantic associations in complex relational networks.

[0055] Heterogeneous graphs can include project nodes, expert nodes, technology field nodes, keyword nodes, and organizational nodes. Project nodes represent the various project entities awaiting review; each node can be associated with attributes such as project name, project description, and applicant organization. Expert nodes represent the various expert entities participating in the review; each node can be associated with attributes such as expert name, research direction, affiliated institution, and past review experience. Technology field nodes represent the professional technical fields involved in the projects and experts, such as artificial intelligence, biomedicine, and new energy. These nodes act as a bridge between projects and experts, reflecting the degree of matching in professional knowledge. Keyword nodes represent specific keywords or themes in the research content of projects and experts, such as deep learning and photovoltaic power generation. These nodes can more granularly characterize the research focus of projects and experts. Organizational nodes represent the applicant organization or the institution to which the expert belongs, such as universities, research institutes, and enterprises. These nodes can be used to identify potential conflicts of interest or collaborative relationships.

[0056] Project nodes and expert nodes can be connected through technology nodes, keyword nodes, or organizational nodes. For example, if a project belongs to a certain technology field, there is an edge between the project node and the corresponding technology field node; similarly, if an expert is proficient in a certain technology field, there is an edge between the expert node and the corresponding technology field node. Similarly, projects and experts can also be connected through keywords and organizational structures. A graph convolutional network (GCNN) can then be used as the graph neural network. This GCNN learns the embedding features of each project node and expert node by aggregating the neighbor information of each node through multiple convolutional operations. For example, each node can be represented as a 128-dimensional embedding vector. After obtaining the project embedding features and expert embedding features, the matching score between the target project and the expert can be obtained by calculating the cosine similarity between these two vectors.

[0057] The above technical solution can fully utilize the multi-dimensional and heterogeneous correlation information between projects and experts, such as technical fields, keywords, and organizational structures. This refined modeling approach enables graph neural networks to learn more expressive and discriminative project and expert embedding features, thereby significantly improving the accuracy of matching scores between target projects and various experts. This helps to more accurately identify review experts who are highly compatible with the project content and have no potential conflicts of interest, thus improving the quality and efficiency of review expert selection.

[0058] In this example implementation, for the same organizational node, the edges between the project node and the organizational node, as well as the edges between the expert node and the organizational node, are designated as conflict edges; during the forward propagation of the graph neural network, conflict detection is performed using conflict masks set on the conflict edges.

[0059] When both a project node and an expert node are connected to nodes from the same organization, these connections are defined as conflict edges. During information transfer and feature learning in graph neural networks, conflict masks can be used to block or weaken information flow along conflict edges, thus preventing conflicting information from negatively impacting matching score calculations. Conflict detection ensures that direct or indirect connections between projects and experts from the same organization are not incorrectly reinforced when calculating matching scores. For each pair of nodes, if an edge is identified as a conflict edge when calculating attention weights, its attention weight can be forced to a very small value, or it can be excluded from the attention calculation altogether. In this way, the contribution of conflicting neighbors is significantly reduced when aggregating neighbor node information, thus achieving effective conflict detection and avoidance.

[0060] The above technical solution effectively identifies and avoids potential conflicts of interest or review biases that may arise during project review when the target project and the review expert belong to the same organization. By defining the edges between project nodes and organization nodes, as well as the edges between expert nodes and organization nodes, as conflict edges, and performing conflict detection using conflict masks set on these edges during the forward propagation of the graph neural network, interference from organizational affiliation is eliminated when calculating the matching score between the project and the expert. This allows the graph neural network to more fairly and objectively evaluate the professional matching degree between the project and the expert, avoiding distortion of the matching score due to non-professional factors, thereby significantly improving the accuracy and fairness of the expert selection process and the overall impartiality of the project review process.

[0061] In this example implementation, project information of the expert's historical participation in projects can be obtained from the expert information. The project information includes the distribution fields of the participating projects and the corresponding project time. Based on the distribution fields of the participating projects, the technical field nodes connected to the expert nodes are determined, and the edge weights corresponding to the edges between the expert nodes and the technical field nodes are determined based on the project time. The edge weights are negatively correlated with the project time.

[0062] Specifically, retrieving project information from expert profiles aims to deeply analyze an expert's past performance in actual projects, serving as the data foundation for the current matching process. For example, when registering or updating their personal information, experts can proactively fill in or upload a list of projects they have participated in, including detailed information such as project name, field, and start and end dates. Furthermore, data mining and correlation analysis can be performed on publicly available academic paper databases, patent databases, and research project application systems to automatically extract historical project information associated with the expert's identity, and parse out the project's field and completion date.

[0063] The goal is to identify technology nodes connected to expert nodes based on the distribution of their project involvement. This involves mapping an expert's historical project experience to technology nodes in a graph neural network, thereby establishing a direct link between expert nodes and technology nodes. The distribution of their project involvement reflects the expert's accumulated practical experience in different technical areas. For example, the domain labels of an expert's historical project involvement can be matched with predefined technology nodes. If an expert has participated in projects in the field of artificial intelligence / machine learning, then that expert node is connected to the "machine learning" technology node in the graph neural network.

[0064] Project time is a crucial indicator of the timeliness of an expert's expertise. Determining edge weights based on project time can reflect an expert's activity and most recent contributions within a specific technical field, ensuring that recently participated projects contribute more to an expert's weight within that field, while earlier projects contribute less. For example, a decay function, such as an exponential or linear decay function, can be designed, taking project time as input and outputting a weight value; projects closer to the current time have a higher weight, while projects further back have a lower weight. Alternatively, project time can be divided into different time periods, each assigned a different weight factor, and then edge weights can be calculated based on the corresponding time period for each historical project.

[0065] The above technical solution enables a more accurate assessment of the depth and timeliness of experts' expertise in different technical fields. By acquiring information on experts' historical project participation and dynamically adjusting the edge weights between expert nodes and technical field nodes based on project timelines, the graph neural network can fully consider the timeliness and depth of experts' knowledge when calculating expert-project matching scores. This avoids inaccurate matching caused by outdated expert experience, significantly improving the accuracy and effectiveness of expert matching and ensuring that the most suitable and professional review experts are selected for the target project.

[0066] In this example implementation, project keywords can also be obtained from the project information of the expert's historical participation in projects, and keyword nodes connected to the expert nodes can be determined based on the project keywords; the edge weights corresponding to the edges between the expert nodes and the keyword nodes can be determined based on the word frequency of the project keywords and the project time; wherein, the edge weights are negatively correlated with the project time and positively correlated with the word frequency.

[0067] Extracting project keywords from an expert's past project information involves analyzing relevant documents, reports, or descriptions of past projects to extract keywords that represent their specific technical expertise and research direction. These keywords provide more granular professional information than just the technical field. This process can utilize natural language processing technology to analyze the text of project documents from the expert's past projects, identifying and extracting high-frequency or representative keywords. Alternatively, a pre-defined keyword dictionary can be used to match project descriptions and identify relevant project keywords.

[0068] In graph neural networks, a corresponding keyword node can be created for each identified project keyword. If a project in which an expert participated contains a specific keyword, a connection is established between the corresponding expert node and the keyword node, thus explicitly representing the association between the expert and the specific technical keyword in the graph structure. The term frequency of the project keyword can be calculated by counting the total number of times a keyword appears in an expert's historical projects, or the number of times it appears within a specific time window. A higher term frequency generally indicates more experience the expert has with that keyword. Edge weights are negatively correlated with project time and positively correlated with term frequency, meaning that the more recent the keyword appears, the greater the edge weight; the older the keyword appears, the smaller the edge weight. Furthermore, the higher the term frequency of the keyword, the greater the edge weight. For example, the edge weights can be designed as a function that comprehensively considers term frequency and project time, such as using a weighted average, product, or a more complex nonlinear function.

[0069] The above technical solutions enable more precise identification of experts' specific technical expertise and current activity levels, allowing graph neural networks to generate more accurate expert embedding features. This significantly improves the accuracy of the matching scores between target projects and experts, thereby helping to more effectively select review experts who are highly compatible with project needs and improving the quality and efficiency of project reviews.

[0070] Step S220. Based on the embedding features corresponding to the project information and the embedding features corresponding to the expert information, obtain the matching score between the target project and the expert.

[0071] Subsequently, based on the embedded features corresponding to the project information and the expert information, a matching score between the target project and the expert can be obtained. For example, the matching score can be obtained by calculating the cosine similarity of the two embedded feature vectors. After obtaining the matching score, experts can be automatically recommended in order of matching degree on the expert settings page. This not only supports one-click batch adoption but also allows administrators to manually adjust the settings.

[0072] In this example implementation, as Figure 3 As shown, based on the embedding features corresponding to the project information and the embedding features corresponding to the expert information, the matching score between the target project and the expert is obtained, which may include the following steps: Step S310. Use natural language processing technology to identify the semantic features of the project files, including the potential topics and potential domains of the target project.

[0073] Natural Language Processing (NLP) is an artificial intelligence technology designed to enable computers to understand, interpret, manipulate, and generate human language. This technology can be used to extract structured semantic information from unstructured project documents. The semantic features of project documents refer to the deeper meanings and thematic information contained within the text content. Latent themes and domains represent the core content and professional scope of the project document. These features provide a deeper understanding of the project content and reflect its actual research direction and application areas.

[0074] Step S320. Obtain the review capability characteristics of experts from the expert information. The review capability characteristics include the distribution fields of the projects the experts have participated in and the importance coefficient of the projects they have participated in.

[0075] Expert review competence characteristics are indicators that measure an expert's professional level, experience, and influence in a specific field. Among these, the "distribution of projects participated in" refers to the professional scope covered by projects the expert has previously reviewed or researched, while the "importance coefficient of projects participated in" is a quantitative indicator measuring the expert's influence or contribution to those projects. These characteristics are used to comprehensively evaluate an expert's professional background and practical review experience, ensuring that the matched experts are not only relevant to the field but also possess sufficient authority and experience. The distribution of projects participated in can be constructed by analyzing data such as the types of projects reviewed in the expert's history, the fields of published papers, and the fields of research projects undertaken; the importance coefficient of projects participated in can be derived by comprehensively considering factors such as the expert's role in historical projects, project level, and project outcomes.

[0076] Step S330. Obtain the fusion features of the project documents based on the semantic features and embedding features of the project documents, and obtain the fusion features of the expert information based on the expert's review ability features and the embedding features of the expert information.

[0077] The fusion feature of project documents is formed by fusing the semantic features and embedded features of project documents. This feature combines the interpretability of semantics with the abstract representational power of embedded features, making the project features more accurate and representative. Similarly, the fusion feature of expert information is formed by fusing the expert's review ability features with the embedded features of their expert information. This feature combines the expert's actual review experience and specific information from their professional field with the abstract representation learned in a graph neural network, making the expert features more accurate and representative.

[0078] Step S340. Based on the fusion characteristics of the project documents and the fusion characteristics of the expert information, obtain the matching score between the target project and the expert.

[0079] By further integrating the semantic features of project documents and the review capabilities of experts, the matching process can gain a more comprehensive and in-depth understanding of the relationship between project content and expert professional background. Therefore, the resulting matching score can more accurately reflect the actual fit between the project and the expert, significantly improving the accuracy of expert selection.

[0080] Continue to refer to Figure 1 As shown, in step S130, the constraints corresponding to the target project are determined, and a list of candidate experts is obtained from the full list of experts based on the matching score and the constraints.

[0081] In this example implementation, constraints refer to the restrictive conditions that need to be met when screening candidate experts, and the candidate expert list refers to the set of experts who are qualified and eligible for review, which are initially screened from the full expert list based on matching scores and constraints.

[0082] Specifically, you can first filter out all experts who meet the constraints, then sort the filtered experts from high to low according to their matching scores, and select a specified number of experts to add to the candidate expert list according to the sorting results, or set a minimum matching score threshold and add experts with scores greater than the threshold to the candidate expert list.

[0083] In this example implementation, the constraints may include load balancing constraints and domain avoidance constraints. The load balancing constraints are used to limit the maximum number of projects that an expert can review within a specified time period, while the domain avoidance constraints are used to exclude experts who are related to the target project.

[0084] Load balancing constraints aim to rationally allocate review tasks, prevent a few experts from bearing an excessive workload, and ensure that review tasks are evenly distributed among the expert group. This can be achieved by setting a maximum number of projects each expert can review within a specific time period (e.g., a week, a month, or a review cycle), or by dynamically evaluating factors such as the expert's current free time and historical review efficiency.

[0085] Domain avoidance constraints are used to identify and exclude experts who have potential conflicts of interest or connections with the project under review, in order to maintain the impartiality and objectivity of the review. For example, the system can check whether the expert has any affiliation, cooperation, or kinship with the project applicant, project leader, or project team members, or whether the expert has participated in any competing projects directly related to this project.

[0086] Through the above technical solutions, the selection of review experts not only considers the professional matching degree between experts and projects, but also fully considers the reasonable allocation of expert review tasks and the fairness of the review process. The introduction of load balancing constraints effectively avoids a few experts from undertaking too many review tasks due to high matching degree, thereby ensuring the sustainability of the review work and the overall efficiency of the expert team. The setting of domain avoidance constraints can exclude experts with potential ties to the project from the source, greatly reducing the risk of bias, unfairness, or conflicts of interest in the review process, thereby significantly improving the objectivity, fairness, and credibility of the review results.

[0087] In step S140, review experts are selected from the list of candidate experts, and the review task of the target project is sent to the review experts. The initial review result of the target project is obtained based on the review results of the review experts.

[0088] In this example implementation, review experts refer to the experts ultimately selected from the list of candidate experts and assigned review tasks. They will review the target project and provide review comments and results. The initial review result refers to the preliminary review conclusion obtained by the system based on the review comments and scores submitted by the experts after they have completed their review tasks.

[0089] The selection of review experts can employ various strategies. For example, a specified number of experts can be randomly selected from a list of candidate experts, or selected based on matching scores from highest to lowest. Alternatively, experts can be manually selected by administrators. Administrators can add or remove experts individually or in batches, and removed experts can be restored to the expert addition module. Upon receiving the task, review experts log into the review system to review the project documents and evaluate them according to preset review criteria, submitting their review comments and scores. The system then aggregates the review results from all experts, such as calculating the average score or a weighted average score, to obtain the initial review results for the target project.

[0090] Expert opinions can be adjusted via a control. After logging into the system, experts can fill in their initial evaluation opinions by clicking on the initial evaluation management. After filling in the opinions, they can click the save button and return to the main list interface to see that the initial evaluation status has been completed.

[0091] In this example implementation, after the review experts are determined from the list of candidate experts, the network parameters in the neural network can be updated based on the expert information of the review experts determined from the list of candidate experts.

[0092] The system records the adjustment behaviors of review experts in the candidate expert list and feeds this information back to the model for incremental learning. This further refines the parameters within the neural network, enabling it to better learn and identify the matching relationships between projects and experts. Alternatively, offline retraining can be performed periodically or after accumulating a certain amount of review expert information to obtain even better network parameters.

[0093] The above technical solution significantly improves the accuracy and adaptability of neural networks in subsequent expert matching. Because neural networks can continuously adjust their internal parameters based on actual feedback, they can more accurately identify the matching relationship between projects and experts, thereby selecting more suitable review experts and improving the overall efficiency of the review task and the quality of the review results. This adaptive learning mechanism ensures the long-term effectiveness and robustness of the expert selection system.

[0094] In this example implementation, after obtaining the initial review results of the target project, a visual project list and voting controls can be provided in the review system of the meeting experts during the meeting discussion phase.

[0095] The discussion phase refers to the stage in the project review process where, after the preliminary review results are formed, a group of experts jointly discuss, evaluate, and make decisions on the project. Its purpose is to enhance the fairness, authority, and accuracy of the review results. The visualized project list refers to the list of projects to be discussed displayed in the expert review system. The voting control is an interactive component in the expert review system that allows experts to vote. This control allows experts to vote on specific projects or issues, such as supporting, opposing, abstaining, or scoring the project. Experts participating in the discussion phase can also be added or removed using single or multiple selections. Projects participating in the meeting can also be adjusted and maintained by triggering the adjustment menu.

[0096] In addition, the review system can also provide award ledger management function. Specifically, it can obtain the ledger and generate a preview version, which can then be confirmed by the user before the release process is initiated. Once the process is approved, the release version is generated.

[0097] The review system in this example implementation can also be applied to the review and management of various projects, such as award application platforms, advanced quality projects, and EPC (Engineering, Procurement, and Construction) general contracting projects.

[0098] It should be noted that although the steps of the method in this disclosure are described in a specific order in the accompanying drawings, this does not require or imply that the steps must be performed in that specific order, or that all the steps shown must be performed to achieve the desired result. Additional or alternative steps may be omitted, multiple steps may be combined into one step, and / or a step may be broken down into multiple steps.

[0099] Furthermore, this disclosure also provides a project review apparatus. (See reference) Figure 4 As shown, the review apparatus for this project may include a project document acquisition module 410, a matching score determination module 420, a candidate expert determination module 430, and a review result determination module 440. Wherein: The project file acquisition module 410 can be used to acquire the project files of the target project to be reviewed and to obtain a full list of experts from the review system. The matching score determination module 420 can be used to obtain project information from the project file and input the project information and the expert information of each expert in the full expert list into the neural network to obtain the matching score between the target project and each expert. The candidate expert determination module 430 can be used to determine the constraints corresponding to the target project, and to filter the candidate expert list from the full expert list based on the matching score and constraints. The review result determination module 440 can be used to determine review experts from the list of candidate experts, send the review task of the target project to the review experts, and obtain the initial review result of the target project based on the review results of the review experts.

[0100] In some exemplary embodiments of this disclosure, a project review apparatus provided by this disclosure may further include a network parameter update module, which can be used to update the network parameters in the neural network based on the expert information of the review experts determined from the candidate expert list.

[0101] In some exemplary embodiments of this disclosure, a project review apparatus provided by this disclosure may further include a project application module, which may include an application interface providing unit and an application field determining unit. Wherein: The application interface provider can be used to provide a project application interface in the review system; The application field determination unit can be used to determine the required and optional fields in the project application interface, as well as the maximum number of project personnel that can be entered, based on the basic information of the target project.

[0102] In some exemplary embodiments of this disclosure, the project application module may further include a multimodal data recognition unit, which can be used to recognize the multimodal application data in the project file uploaded in the project application interface through optical character recognition technology, and display corresponding prompt information on the project application interface according to the recognition result.

[0103] In some exemplary embodiments of this disclosure, the project application module may further include a cross-validation unit, which can be used to cross-validate the project files uploaded in the project application interface with the application information filled in in the project application interface, and display corresponding prompt information on the project application interface according to the verification results.

[0104] In some exemplary embodiments of this disclosure, a project review apparatus provided by this disclosure may further include a meeting visualization module, which can be used to provide a visualized list of projects and voting controls in the review system of meeting experts during the meeting discussion phase.

[0105] In some exemplary embodiments of this disclosure, the matching score determination module 420 may include an embedded feature determination unit and a matching score determination unit. Wherein: The embedding feature determination unit can be used to input project information and expert information of each expert in the full expert list into the graph neural network to obtain the embedding features corresponding to the project information and the embedding features corresponding to the expert information. The matching score determination unit can be used to obtain the matching score between the target project and the expert based on the embedding features corresponding to the project information and the embedding features corresponding to the expert information.

[0106] In some exemplary embodiments of this disclosure, the graph neural network is constructed based on a heterogeneous graph, which includes project nodes, expert nodes, technical field nodes, keyword nodes, and organizational nodes. Project nodes and expert nodes are connected through technical field nodes, keyword nodes, or organizational nodes.

[0107] In some exemplary embodiments of this disclosure, the matching score determination module 420 may further include a conflict edge determination unit and a conflict detection unit. Wherein: The conflict edge determination unit can be used to identify conflict edges between project nodes and organizational nodes, as well as between expert nodes and organizational nodes, for nodes of the same organization. The collision detection unit can be used to detect collisions during the forward propagation of a graph neural network by using collision masks set on collision edges.

[0108] In some exemplary embodiments of this disclosure, the matching score determination module 420 may further include a project information acquisition unit and a technical field node determination unit. Wherein: The project information acquisition unit can be used to obtain project information about the historical projects that experts have participated in from expert information. The project information includes the distribution of the projects participated in and the corresponding project time. The technical field node determination unit can be used to determine the technical field nodes connected to the expert nodes based on the distribution of participating projects, and to determine the edge weights corresponding to the edges between the expert nodes and the technical field nodes based on the project time; wherein, the edge weights are negatively correlated with the project time.

[0109] In some exemplary embodiments of this disclosure, the matching score determination module 420 may further include a keyword node determination unit and a keyword weight determination unit. Wherein: The keyword node determination unit can be used to obtain project keywords from the project information of the expert's historical participation in projects, and determine the keyword nodes connected to the expert nodes based on the project keywords; The keyword weight determination unit can be used to determine the edge weights between expert nodes and keyword nodes based on the keyword frequency and project time. The edge weights are negatively correlated with project time and positively correlated with keyword frequency.

[0110] In some exemplary embodiments of this disclosure, the matching score determination unit may include a semantic feature determination unit, a review capability feature determination unit, a fusion feature determination unit, and a matching score calculation unit. Wherein: The semantic feature determination unit can be used to identify the semantic features of project files from project files using natural language processing technology. The semantic features include the potential topics and potential domains of the target project. The review capability characteristic determination unit can be used to obtain the review capability characteristics of experts from expert information. The review capability characteristics include the distribution fields of the expert's projects and the importance coefficient of the projects participated in. The fusion feature determination unit can be used to obtain the fusion features of the project file based on the semantic features and embedding features of the project file, and to obtain the fusion features of the expert information based on the expert's review ability features and the embedding features of the expert information. The matching score calculation unit can be used to obtain the matching score between the target project and the expert based on the fusion characteristics of the project file and the fusion characteristics of the expert information.

[0111] The specific details of each module / unit in the review device for the above-mentioned project have been described in detail in the corresponding method implementation section, and will not be repeated here.

[0112] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to exemplary embodiments of this disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0113] Figure 5 A schematic diagram of the structure of a computer system suitable for implementing the embodiments of the present disclosure is shown.

[0114] It should be noted that, Figure 5 The computer system 500 of the electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments disclosed herein.

[0115] like Figure 5 As shown, the computer system 500 includes a central processing unit (CPU) 501, which can perform various appropriate actions and processes based on programs stored in read-only memory (ROM) 502 or programs loaded from storage section 508 into random access memory (RAM) 503. The RAM 503 also stores various programs and data required for system operation. The CPU 501, ROM 502, and RAM 503 are interconnected via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0116] The following components are connected to I / O interface 505: an input section 506 including a keyboard, mouse, etc.; an output section 507 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 508 including a hard disk, etc.; and a communication section 509 including a network interface card such as a LAN card, modem, etc. The communication section 509 performs communication processing via a network such as the Internet. A drive 510 is also connected to I / O interface 505 as needed. A removable medium 511, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on drive 510 as needed so that computer programs read from it can be installed into storage section 508 as needed.

[0117] In particular, according to embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 509, and / or installed from removable medium 511. When the computer program is executed by central processing unit (CPU) 501, it performs various functions defined in the system of this disclosure.

[0118] An exemplary embodiment of this disclosure also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, implements the review method described above.

[0119] In one implementation, the computer program product can be a tangible product containing a computer program, such as a computer-readable storage medium storing the computer program. The readable storage medium can be a storage medium based on electrical, magnetic, optical, electromagnetic, infrared, or other signals, including but not limited to: random access memory (RAM), read-only memory (ROM), magnetic tape, floppy disk, flash memory, hard disk drive (HDD), solid-state drive (SSD), etc. For example, the computer program product can be implemented as a non-volatile storage medium storing a computer program, such as read-only memory, NAND flash memory, etc.

[0120] In one implementation, the computer program product can be an intangible product containing a computer program. For example, the computer program product can be implemented as a virtual digital product, such as an executable file, installation package, or other digital file storing the computer program.

[0121] Computer program code can be written in one or more programming languages. Examples of programming languages ​​include C, Java, and C++. Program code can execute entirely on the user's computing device, partially on the user's computing device, or as a standalone software package. It can also execute partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, such as a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via an internet connection provided by a mobile network operator).

[0122] Computer programs can be carried or transmitted via signals such as electrical, magnetic, optical, electromagnetic, and infrared rays. Electronic devices can convert the signals carrying computer programs into digital signals, thereby running the computer programs. When a computer program runs on an electronic device, its code is used to cause the electronic device to execute (more specifically, to execute) the method steps of various exemplary embodiments of this disclosure, such as the review method described above.

[0123] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0124] It should be noted that although several modules for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to embodiments of this disclosure, the features and functions of two or more modules described above can be embodied in one module. Conversely, the features and functions of one module described above can be further divided and embodied by multiple modules.

[0125] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This disclosure is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein.

[0126] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A method for reviewing projects, characterized in that, include: Obtain the project files of the target project to be reviewed, and retrieve the full list of experts from the review system; The project information is obtained from the project file, and the project information and the expert information of each expert in the full expert list are input into the neural network to obtain the matching score between the target project and each expert. Determine the constraints corresponding to the target project, and based on the matching score and the constraints, filter from the full list of experts to obtain a list of candidate experts; Review experts are selected from the list of candidate experts, and the review task of the target project is sent to the review experts. The initial review result of the target project is obtained based on the review results of the review experts.

2. The project review method according to claim 1, characterized in that, The constraints include load balancing constraints and domain avoidance constraints. The load balancing constraints limit the maximum number of projects that the expert can review within a specified time period, and the domain avoidance constraints exclude experts who are related to the target project.

3. The project review method according to claim 1, characterized in that, After determining the review experts from the list of candidate experts, the method further includes: The network parameters in the neural network are updated based on the expert information of the review experts determined from the list of candidate experts.

4. The project review method according to claim 1, characterized in that, The method further includes: The review system provides a project application interface; Based on the basic information of the target project, determine the required and optional fields in the project application interface, as well as the maximum number of project personnel that can be entered.

5. The project review method according to claim 4, characterized in that, The method further includes: The multimodal application data in the project file uploaded to the project application interface is identified using optical character recognition technology, and corresponding prompts are displayed on the project application interface based on the recognition results.

6. The project review method according to claim 5, characterized in that, The method further includes: The project files uploaded in the project application interface are cross-validated with the application information filled in in the project application interface, and the corresponding prompt information is displayed on the project application interface according to the verification results.

7. The project review method according to claim 1, characterized in that, After obtaining the initial review results of the target project, the method further includes: During the meeting discussion phase, a visual list of projects and voting controls are provided in the expert review system.

8. The project review method according to claim 1, characterized in that, The step of inputting the project information and the expert information of each expert in the full expert list into the neural network to obtain the matching score between the target project and each expert includes: The project information and the expert information of each expert in the full expert list are input into a graph neural network to obtain the embedding features corresponding to the project information and the embedding features corresponding to the expert information. Based on the embedding features corresponding to the project information and the embedding features corresponding to the expert information, a matching score is obtained between the target project and the expert.

9. The project review method according to claim 8, characterized in that, The graph neural network is built on a heterogeneous graph, which includes project nodes, expert nodes, technical field nodes, keyword nodes, and organizational nodes. The project nodes and expert nodes are connected through the technical field nodes, keyword nodes, or organizational nodes.

10. The project review method according to claim 9, characterized in that, The method further includes: For the same organizational node, the edge between the project node and the organizational node, and the edge between the expert node and the organizational node are designated as conflict edges. During the forward propagation of the graph neural network, collision detection is performed using collision masks set on the collision edges.

11. The project review method according to claim 9, characterized in that, The method further includes: Obtain the project information of the expert's historical projects from the expert information, including the distribution of the projects participated in and the corresponding project time; The technical field nodes connected to the expert nodes are determined based on the distribution of participating projects, and the edge weights corresponding to the edges between the expert nodes and the technical field nodes are determined based on the project time; wherein, the edge weights are negatively correlated with the project time.

12. The project review method according to claim 11, characterized in that, The method further includes: Obtain project keywords from the project information of the expert's historical projects, and determine the keyword node connected to the expert node based on the project keywords; Based on the word frequency of the project keywords and the project time, the edge weights corresponding to the edges between the expert nodes and the keyword nodes are determined; wherein, the edge weights are negatively correlated with the project time and positively correlated with the word frequency.

13. The project review method according to claim 8, characterized in that, The step of obtaining the matching score between the target project and the expert based on the embedding features corresponding to the project information and the expert information includes: Natural language processing technology is used to identify the semantic features of the project file, including the potential topics and potential domains of the target project; The review capability characteristics of the experts are obtained from the expert information, including the distribution fields of the projects the experts have participated in and the importance coefficient of the projects they have participated in; The project file's fusion feature is obtained based on its semantic and embedding features, and the expert information's fusion feature is obtained based on the expert's review capability features and the expert information's embedding features. Based on the fusion characteristics of the project documents and the fusion characteristics of the expert information, a matching score is obtained between the target project and the expert.

14. A project review device, characterized in that, include: The project file acquisition module is used to acquire the project files of the target project to be reviewed and to obtain a full list of experts from the review system. The matching score determination module is used to obtain project information from the project file and input the project information and the expert information of each expert in the full expert list into the neural network to obtain the matching score between the target project and each expert; The candidate expert determination module is used to determine the constraints corresponding to the target project, and to filter the candidate expert list from the full expert list based on the matching score and the constraints. The review result determination module is used to determine review experts from the candidate expert list, send the review task of the target project to the review experts, and obtain the initial review result of the target project based on the review results of the review experts.

15. An electronic device, characterized in that, include: processor; as well as A memory for storing one or more programs, which, when executed by the processor, cause the processor to implement the review method for the project as described in any one of claims 1 to 13.

16. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the project review method as described in any one of claims 1 to 13.

Images