A system and method for matching supply and demand of achievement transformation

Abstract

The present invention belongs to the field of scientific and technological achievement transformation, and discloses a system and method for matching supply and demand for achievement transformation, including: a feature extraction unit, which is used to convert the collected technical achievements and demand information into high-dimensional vectors; a supply-demand correlation mining unit, which is used to dynamically adjust the matching priority of achievement feature data in different fields by using an entropy weight algorithm, and mine the supply-demand correlation between technical achievements and demand information; an optimal matching result acquisition unit, which is used to define a multi-objective function and obtain the optimal matching result by a genetic algorithm; a matching result verification and optimization unit, which is used to optimize the matching process by user feedback data; and a matching result display unit, which is used to display the supply-demand matching result by using a visualization tool. The present invention improves the efficiency of information processing by converting technical achievements and demand information into high-dimensional vectors that are easy to process.

Description

Technical Field

[0001] The present invention relates to the field of scientific and technological achievement transformation, and in particular to an achievement transformation supply and demand matching system and method. Background Art

[0002] Transformation of scientific research results refers to the complex process of advancing scientific research results from the laboratory or theoretical stage to practical application, usually involving the development and deployment of new technologies, new products, new processes or new services. This transformation is not only aimed at enabling scientific research results to have a profound and positive impact on economic and social development, but also committed to fully exploring and realizing their potential value through commercialization or direct application in solving social problems.

[0003] However, in the current environment, the process of transforming scientific research results into practical applications is facing a series of challenges, the most notable of which is the mismatch between technical results and demand information. On the one hand, many scientific research results may not have reached the stability and reliability standards required for commercial applications in their early stages; on the other hand, the understanding of demand information is not accurate or not deeply explored, resulting in the inability of developed products or services to accurately meet user needs, which in turn affects market acceptance and application scope. These problems work together to not only hinder the effective transformation of a large number of scientific research results into real productivity, but also cause the waste of valuable scientific research resources and fail to give full play to their due benefits.

[0004] Therefore, how to provide a supply and demand matching system and method for achievement transformation is an urgent problem to be solved. Summary of the invention

[0005] The embodiments of the present invention provide a system and method for matching supply and demand of achievement transformation to solve the above-mentioned technical problems existing in the prior art.

[0006] In order to have a basic understanding of some aspects of the disclosed embodiments, a brief summary is given below. This summary is not a general review, nor is it intended to identify key / important components or to describe the scope of protection of these embodiments. Its only purpose is to present some concepts in a simple form as a prelude to the detailed description that follows.

[0007] According to a first aspect of an embodiment of the present invention, a system for matching supply and demand of achievement transformation is provided.

[0008] In one embodiment, a system for matching supply and demand of achievement transformation includes:

[0009] A feature extraction unit is used to convert the collected technical achievements and demand information into a high-dimensional vector, and extract matching feature data based on the high-dimensional vector to establish a matching feature set;

[0010] The supply-demand correlation mining unit is used to dynamically adjust the matching priority of achievement feature data in different fields based on the established matching feature set using the entropy weight algorithm, and to mine the supply-demand correlation between technical achievements and demand information;

[0011] The optimal matching result acquisition unit is used to define a multi-objective function based on the mined supply-demand correlation and obtain the optimal matching result through a genetic algorithm;

[0012] The matching result verification and optimization unit is used to verify the matching results using game theory algorithms and optimize the matching process through user feedback data in combination with reinforcement learning mechanisms;

[0013] The matching result display unit is used to build a supply and demand matching algorithm framework and use visualization tools to display the supply and demand matching results.

[0014] In one embodiment, the feature extraction unit includes: a data acquisition and preprocessing module, a high-dimensional vector generation module, a feature data extraction module and a feature set establishment module;

[0015] The data acquisition and preprocessing module is used to obtain technical achievements and demand information from data sources using text extraction technology, and to preprocess the obtained technical achievements and demand information;

[0016] A high-dimensional vector generation module is used to input the pre-processed technical achievements and demand information into the pre-built sentence vector embedding model, generate the corresponding high-dimensional vector, and normalize the high-dimensional vector;

[0017] The feature data extraction module is used to extract the achievement feature data and demand feature data from the normalized high-dimensional vector according to the deep learning algorithm;

[0018] The feature set establishment module is used to use a similarity algorithm to calculate the similarity between the result feature data and the demand feature data, and obtain matching feature data based on the similarity to establish a matching feature set.

[0019] In one embodiment, the feature data extraction module includes: a feature extraction model construction submodule, a language information acquisition submodule and a feature data acquisition submodule;

[0020] The feature extraction model construction submodule is used to select a pre-trained bidirectional encoder representation model as the basic architecture according to the deep learning algorithm, and add a shared layer and a multi-task learning layer to the basic architecture to build a feature extraction model;

[0021] The language information acquisition submodule is used to input the normalized high-dimensional vector into the feature extraction model, and capture the semantic information in the high-dimensional vector through the multi-layer encoder of the feature extraction model;

[0022] The feature data acquisition submodule is used to transfer the captured language information to the multi-task learning layer and obtain the result feature data and the requirement feature data from the middle layer of the multi-task learning layer.

[0023] In one embodiment, the supply-demand correlation mining unit includes: a characteristic entropy value calculation module, a priority matching module, a supply-demand matrix construction module, a supply-demand matrix completion module and a supply-demand correlation module;

[0024] A feature entropy value calculation module is used to calculate the feature entropy value of each field according to the feature data of the results in the established matching feature set;

[0025] The priority matching module is used to calculate the weight coefficient of each field based on the entropy value of the achievement characteristics of each field using the entropy weight formula, and dynamically adjust the matching priority of the achievement characteristic data in different fields through the weight coefficient;

[0026] A supply and demand matrix construction module is used to combine the priority-adjusted outcome feature data with the demand feature data to construct a supply and demand matrix;

[0027] The supply and demand matrix completion module is used to analyze the supply and demand matrix by using non-negative matrix decomposition technology to obtain the implicit feature space of the supply and demand matrix, and combine it with the matrix completion algorithm to obtain a new supply and demand matrix;

[0028] The supply-demand correlation module is used to mine the supply-demand correlation between technological achievements and demand information based on the new supply-demand matrix using the graph neural network algorithm.

[0029] In one embodiment, the supply-demand matrix completion module includes: a supply-demand matrix decomposition submodule, an implicit feature space acquisition submodule and a matrix completion submodule;

[0030] A supply-demand matrix decomposition submodule is used to randomly initialize two non-negative matrices based on the supply-demand matrix using a non-negative matrix decomposition technique, wherein the two non-negative matrices correspond to a non-negative matrix representing an implicit feature of supply and a non-negative matrix representing an implicit feature of demand, respectively;

[0031] The implicit feature space acquisition submodule is used to iteratively optimize the two non-negative matrices using an alternating minimization algorithm, and output the final two optimized non-negative matrices to obtain the implicit feature space of the supply and demand matrix;

[0032] The matrix completion submodule is used to combine the matrix completion algorithm to process the missing values ​​in the implicit feature space and obtain a new supply and demand matrix.

[0033] In one embodiment, the combination of the matrix completion algorithm and processing the missing values ​​in the implicit feature space to obtain a new supply and demand matrix includes:

[0034] According to the implicit feature space of the supply and demand matrix, the target matrix is ​​defined, and the low-rank matrix is ​​obtained to fill the missing values ​​by minimizing the nuclear norm of the target matrix;

[0035] Combined with slack variables, an optimization model with low rank and data fitting is constructed, and the alternating direction multiplier algorithm is used for iterative optimization to update the supply and demand matrix;

[0036] Extract the complete supply and demand correlation information from the updated supply and demand matrix to generate a new supply and demand matrix.

[0037] In one embodiment, the formula of the optimization model is:

[0038] ;

[0039] In the formula, X represents a low-rank matrix; E represents the slack variable; represents the nuclear norm of the low-rank matrix; s represents the regularization parameter; represents the square of the norm of the slack variable; subject to Indicates constraints; P μ represents the projection operator; M Represents the target matrix.

[0040] In one embodiment, the matching result verification and optimization unit includes: a matching result verification module, an optimization guidance module and a matching process optimization module;

[0041] The matching result verification module is used to use the game theory algorithm to build a game model involving both supply and demand parties, and to verify the stability of the matching results in combination with the mixed strategy Nash equilibrium algorithm, and to mark unstable matching pairs;

[0042] The optimization guidance module is used to build a reinforcement learning model based on the reinforcement learning mechanism, define the adjustment matching strategy as an action space, and formulate a reward function based on the matching stability to guide the optimization process;

[0043] The matching process optimization module is used to obtain user feedback data and optimize the matching process in combination with the deep Q learning algorithm.

[0044] In one embodiment, the formula of the deep Q learning algorithm is:

[0045] ;

[0046] In the formula, L ( i ) represents the loss function; E Indicates expected value; sIndicates status; a Indicates action; r Indicates immediate reward; s ´ Indicates the next state; c represents the discount factor; a´ Indicates the next action; Q express Q value; i Represents the set of parameters of a neural network.

[0047] According to a second aspect of an embodiment of the present invention, a method for matching supply and demand of achievement transformation is provided.

[0048] In one embodiment, the method for matching supply and demand of achievement transformation comprises:

[0049] Convert the collected technical achievements and demand information into high-dimensional vectors, extract matching feature data based on the high-dimensional vectors, and establish a matching feature set;

[0050] According to the established matching feature set, the entropy weight algorithm is used to dynamically adjust the matching priority of the achievement feature data in different fields, and to explore the supply and demand correlation between technical achievements and demand information;

[0051] Based on the supply-demand correlation mined, a multi-objective function is defined, and the optimal matching result is obtained through a genetic algorithm;

[0052] Use game theory algorithms to verify matching results, and combine reinforcement learning mechanisms to optimize the matching process through user feedback data;

[0053] Build a supply and demand matching algorithm framework and use visualization tools to display the supply and demand matching results.

[0054] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0055] 1. The present invention improves the efficiency of information processing by converting technological achievements and demand information into high-dimensional vectors that are easy to process. At the same time, by extracting matching feature data and deeply exploring the supply-demand correlation between technological achievements and demand information, supply-demand matching is achieved, providing a strong guarantee for achievement transformation.

[0056] 2. The present invention ensures the stability and reliability of matching results by adjusting the matching priorities of technical achievements in different fields; in addition, it defines multi-objective functions and uses genetic algorithms to obtain the optimal matching results, which provides a scientific basis for decision-making and fully supports and serves the transformation of achievements.

[0057] 3. The present invention collects user feedback data through a reinforcement learning mechanism to optimize the matching process, thereby improving the matching accuracy and user satisfaction; at the same time, it uses visualization tools to display the supply and demand matching results, so that users can intuitively understand the matching situation and make more accurate decisions.

[0058] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0060] Figure 1 It is a principle block diagram of a method for matching supply and demand of achievement transformation according to an exemplary embodiment;

[0061] Figure 2 It is a flow chart of a system for matching supply and demand of achievement transformation according to an exemplary embodiment;

[0062] Figure 3 The figure is a schematic diagram showing the structure of a computer device according to an exemplary embodiment. DETAILED DESCRIPTION

[0063] The following description and accompanying drawings fully illustrate the specific embodiments of this article so that those skilled in the art can practice them. Parts and features of some embodiments may be included in or replace parts and features of other embodiments. The scope of the embodiments of this article includes the entire scope of the claims, as well as all available equivalents of the claims. Herein, the terms "first", "second", etc. are only used to distinguish one element from another, without requiring or implying any actual relationship or order between these elements. In fact, the first element can also be called the second element, and vice versa. Moreover, the terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that the structure, device or equipment including a series of elements includes not only those elements, but also other elements that are not explicitly listed, or also include elements inherent to such structure, device or equipment. In the absence of more restrictions, the elements defined by the sentence "including one..." do not exclude the existence of other identical elements in the structure, device or equipment including the elements. Each embodiment is described in a progressive manner herein, and each embodiment focuses on the differences from other embodiments, and the same and similar parts between the embodiments can be referred to each other.

[0064] As used herein, the term "plurality" means two or more than two, unless otherwise specified.

[0065] In this document, the character " / " indicates that the preceding and following objects are in an "or" relationship. For example, A / B means: A or B.

[0066] In this article, the term "and / or" is a description of the association relationship between objects, indicating that three relationships may exist. For example, A and / or B means: A or B, or, A and B.

[0067] It should be understood that, although the various steps in the flow chart are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear description in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the figure may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these sub-steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of the sub-steps or stages of other steps.

[0068] Each module in the device or system of the present application can be implemented in whole or in part by software, hardware, or a combination thereof. The above modules can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to the above modules.

[0069] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0070] Figure 1 An embodiment of a system for matching supply and demand of achievement transformation of the present invention is shown.

[0071] In this optional embodiment, the achievement transformation supply and demand matching system includes:

[0072] The feature extraction unit 101 is used to convert the collected technical achievements and demand information into a high-dimensional vector, and extract matching feature data based on the high-dimensional vector to establish a matching feature set.

[0073] In this optional embodiment, the feature extraction unit 101 includes: a data acquisition and preprocessing module, a high-dimensional vector generation module, a feature data extraction module and a feature set establishment module;

[0074] The data acquisition and preprocessing module is used to obtain technical achievements and demand information from data sources using text extraction technology, and to preprocess the obtained technical achievements and demand information.

[0075] The high-dimensional vector generation module is used to input the pre-processed technical achievements and demand information into the pre-built sentence vector embedding model, generate the corresponding high-dimensional vector, and normalize the high-dimensional vector.

[0076] The feature data extraction module is used to extract achievement feature data and demand feature data from the normalized high-dimensional vector based on the deep learning algorithm.

[0077] In this optional embodiment, the feature data extraction module includes: a feature extraction model construction submodule, a language information acquisition submodule and a feature data acquisition submodule;

[0078] The feature extraction model construction submodule is used to select a pre-trained bidirectional encoder representation model as the basic architecture according to the deep learning algorithm, and add a shared layer and a multi-task learning layer to the basic architecture to build a feature extraction model;

[0079] The language information acquisition submodule is used to input the normalized high-dimensional vector into the feature extraction model, and capture the semantic information in the high-dimensional vector through the multi-layer encoder of the feature extraction model;

[0080] The feature data acquisition submodule is used to transfer the captured language information to the multi-task learning layer and obtain the result feature data and the requirement feature data from the middle layer of the multi-task learning layer.

[0081] The feature set establishment module is used to use a similarity algorithm to calculate the similarity between the result feature data and the demand feature data, and obtain matching feature data based on the similarity to establish a matching feature set.

[0082] It should be noted that the data source is assumed to be an online document library containing technical achievements and demand information, and the document formats are PDF and Word.

[0083] Step 1: Data acquisition and preprocessing module

[0084] (1) Data acquisition:

[0085] Use the PyPDF2 library and python-docx library in Python to read PDF and Word documents respectively. Use regular expressions and natural language processing technology (NLP) to extract technical achievements and demand information in documents. It should be noted that technical achievements are generally technical solutions for products, processes, materials, etc. formed with the help of scientific and technological knowledge, while demand information is the expectations, requirements and application scenario information expressed by market entities when scientific and technological achievements are transferred from the knowledge creation end to the application end. It may be directly proposed by market entities or obtained through industry, technology, and market information analysis and prediction. The demand information in this technical solution includes clear demand information proposed by market entities, as well as implicit demand information that can be deeply extracted and analyzed from massive market transaction data, industry report data, business activity data, news data, patent technology literature data and other multi-source data using advanced data mining algorithms, machine learning models and artificial intelligence analysis technologies. Through in-depth demand mining, the potential market demand is accurately analyzed, so as to match the technical achievements with the potential market opportunities that match them, and effectively build a bridge for the connection between technology and the market.

[0086] (2) Data preprocessing:

[0087] The extracted information is cleaned to remove irrelevant characters and noise data.

[0088] Use word segmentation technology to segment the text for subsequent analysis.

[0089] Store the cleaned and tokenized data in a structured format (such as CSV or JSON) for use by subsequent modules.

[0090] Step 2: High-dimensional vector generation module

[0091] (1) Sentence embedding model:

[0092] Use the pre-trained BERT model or Sentence-BERT (SBERT) model as the sentence vector embedding model, input the pre-processed technical achievements and demand information into the BERT model, and generate the corresponding high-dimensional vector.

[0093] In addition, high-dimensional vectors refer to numerical vector representations of preprocessed technical achievement descriptions and requirement description texts converted through sentence vector embedding models (such as BERT or Sentence-BERT). These vectors usually have high dimensions (for example, hundreds or even thousands of dimensions). Each dimension of the high-dimensional vector represents the projection or representation of the text on a specific feature or semantic space. These features can be lexical, syntactic, or higher-level semantic features. By training the sentence vector embedding model, the model can learn the semantic similarity between texts and map similar texts to similar positions in the vector space. For example, by calculating the similarity between two high-dimensional vectors (such as cosine similarity, Euclidean distance, etc.), the semantic similarity between two texts can be evaluated.

[0094] (2) Normalization processing:

[0095] The generated high-dimensional vectors are normalized so that the vector length is unified to 1, which is convenient for subsequent similarity calculation.

[0096] Step 3: Feature data extraction module

[0097] (1) Deep learning algorithm:

[0098] Use deep learning algorithms such as convolutional neural networks (CNN) or recurrent neural networks (RNN) to extract feature data from high-dimensional vectors.

[0099] (2) Feature data extraction:

[0100] The normalized high-dimensional vector is input into the deep learning model.

[0101] The outcome feature data and requirement feature data are extracted through the convolutional layer or loop layer of the model.

[0102] The extracted feature data is stored in an appropriate format for use by subsequent modules.

[0103] In addition, achievement characteristic data and demand characteristic data refer to numerical representations extracted from technical achievement texts and demand texts, respectively, which can represent their respective core contents and key information. Achievement characteristic data is extracted from information related to technical achievements, such as technical achievement description information and technical achievement related information; among them, the collection method of technical achievement description information includes receiving information filled in by the achievement holder according to the preset form, and also includes information extracted from official filing documents of technical achievements, technical achievement scientific and technological literature reports and other document materials. For example, the description information of technical achievements may include patent descriptions, scientific research results summaries, technical documents, bibliographic project information, etc.; the determination of technical achievement related information is based on the semantic analysis of the technical achievement description information to extract key concepts, terms, technical points, etc.; then based on the semantic analysis results, the related element information such as similar technical principles, upstream and downstream application technologies, cross-domain technology applications, predecessor technologies, iterative technologies, etc. is sorted out and summarized; then according to the characteristics of the related element information and the field to which it belongs, the type of retrieval library that matches it is determined. For example, if the related element information involves similar technical principles, such as the technical achievement is based on the principle of electromagnetic induction, then it can be located in the technical principle retrieval database containing motor technology, transformer technology, etc. that also use the principle of electromagnetic induction, so as to obtain relevant technical solution optimization ideas and application cases; if it involves upstream and downstream application technologies, taking chip manufacturing technology as an example, if it is an upstream chip design technology achievement, it can be located in the chip manufacturing process. , chip packaging and testing and other downstream links, so as to clarify the application needs and adaptation directions of technological achievements in subsequent links of the industrial chain; if the related element information is a cross-domain technology application, such as technology involving artificial intelligence algorithms, then it can correspond to the medical technology retrieval database that uses artificial intelligence algorithms for disease screening in the field of medical imaging diagnosis, and the algorithm technology database that uses artificial intelligence algorithms to optimize route planning in the intelligent traffic dispatching system, etc., to provide a reference for expanding the application scenarios of technology; if it is a predecessor technology, such as the current high-definition video encoding technology, it can be located in the database of its predecessor such as the MPEG series and other early video encoding technologies, from which the technology evolution context and reference points can be explored; if the related element information is an iterative technology, such as the iteration of a smartphone operating system, it can be located in the technology database of past versions of the operating system and the technical data database of similar operating system iterations of competitors, to help analyze the direction of technology iteration and improvement strategies.In the search library that matches the associated element information, the associated element information is used as the search keyword or condition to conduct a comprehensive and in-depth search operation, so as to obtain information related to the technical achievements, including the research progress of related technologies, application cases, technical parameters, etc. When the correlation relevance exceeds the preset threshold, the retrieved information is determined as the technical achievement related information, so as to ensure that the obtained technical achievement related information and the technical achievement description information are highly relevant and targeted, and provide a solid and reliable data foundation for the subsequent extraction of achievement feature data and the precise operation of the entire achievement transformation supply and demand matching system. By inputting these texts into the pre-built sentence vector embedding model, the corresponding high-dimensional vectors can be generated; some dimensions or features in these high-dimensional vectors can reflect the core content and key information of the technical achievements, such as technical methods, application fields, performance indicators, etc.; further, by processing these high-dimensional vectors through deep learning algorithms (such as convolutional neural networks, recurrent neural networks, etc.), more representative feature data, namely achievement feature data, can be extracted.

[0104] Demand feature data comes from the integration and mining of multi-source data, including extraction from clear demand information proposed by market players. This information is usually presented in the form of enterprise technology demand applications, technology project bidding requirements, industry-university-research technology cooperation letters of intent, etc., covering direct demands for specific technical achievements in terms of function, performance, cost, application scenarios, etc.; it also includes in-depth mining of market transaction data, industry report data, business activity data, news data, enterprise patent technology and layout, etc. Advanced means such as text mining algorithms based on deep learning, machine learning models based on cluster analysis, and artificial intelligence analysis systems driven by natural language processing technology are used to extract and analyze transaction information such as product sales trends, customer purchase preferences, technology achievement transfer / licensing / pledge in market transaction data, and industry development trends, technology hot spots and potential demand directions are mined from industry report data. The layout of enterprises is explored in business activity data, and the hot technology demand information and capital flow information behind the investment and financing and bidding data of enterprises are observed. The focus and expected application areas of emerging technologies are captured with the help of news data. The technology layout of enterprise technology gaps and potential competitive needs is analyzed through patent technology literature data. Similar to processing the text of technical achievements, the category of the demand information library that matches it is determined based on the analyzed potential demand characteristics. In the determined demand information library, the demand characteristic information is used as the screening condition or keyword to conduct a comprehensive and in-depth search operation to obtain information that is highly relevant to the demand characteristics, including historical solutions for similar needs, successful application cases, and the technical parameter range that meets the needs. When the matching degree between the obtained information and the demand characteristics exceeds the preset threshold, it is determined as effective demand characteristic data, so as to accurately match the technical achievements with the potential market opportunities that match them, successfully build a bridge for the connection between technology and the market, and provide strong data support for the entire achievement transformation supply and demand matching system, ensuring the accuracy and effectiveness of demand orientation in the process of technical achievement transformation. The demand text is also converted into a high-dimensional vector, and the key features, namely the demand characteristic data, are extracted through a deep learning algorithm; these feature data can reflect the core content and key requirements of the demand, such as functional requirements, performance indicators, application scenarios, etc.

[0105] Step 4: Feature set establishment module

[0106] (1) Similarity algorithm:

[0107] Use cosine similarity or Euclidean distance to calculate the similarity between the outcome feature data and the requirement feature data.

[0108] (2) Establishment of matching feature set:

[0109] The cosine similarity is calculated for the extracted outcome feature data and requirement feature data.

[0110] A similarity threshold is set, and feature data pairs with similarity higher than the threshold are regarded as matching feature data.

[0111] The matching feature data is stored in a matching feature set for subsequent analysis and application.

[0112] In addition, matching feature data refers to feature data that is determined to have a high similarity or matching relationship between technical achievements and requirements after a series of processing flows; these data are usually extracted from high-dimensional vectors and represent the core content or key features of technical achievements and requirements.

[0113] In a specific embodiment, it is assumed that there are the following technical achievements and demand information:

[0114] Technical achievement 1: "An image recognition method based on deep learning".

[0115] Technical achievement 2: "Natural language processing technology based on machine learning".

[0116] Requirement 1: "A technology is needed to identify objects in images."

[0117] Requirement 2: "Hope to develop an algorithm that can process natural language text."

[0118] After being processed by the data acquisition and preprocessing module, the following high-dimensional vector is obtained:

[0119] Technical results 1 vector: [0.1, 0.2, ..., 0.8].

[0120] Technical result 2 vector: [0.2, 0.1, ..., 0.7].

[0121] Required information 1 vector: [0.15, 0.25, ..., 0.85].

[0122] Required information 2 vector: [0.3, 0.05, ..., 0.65].

[0123] After being processed by the high-dimensional vector generation module and the feature data extraction module, the following similarities are calculated:

[0124] The similarity between technical achievement 1 and requirement information 1 is 0.95 (higher than the threshold of 0.8).

[0125] The similarity between technical achievement 2 and requirement information 1 is 0.6 (lower than the threshold of 0.8).

[0126] The similarity between technical achievement 1 and requirement information 2 is 0.55 (lower than the threshold of 0.8).

[0127] The similarity between technical achievement 2 and requirement information 2 is 0.85 (higher than the threshold of 0.8).

[0128] Therefore, the following matching feature set is established:

[0129] Matching feature set 1: {technical achievement 1, demand information 1}.

[0130] Matching feature set 2: {technical achievement 2, demand information 2}.

[0131] The supply-demand correlation mining unit 103 is used to dynamically adjust the matching priority of achievement feature data in different fields according to the established matching feature set by using the entropy weight algorithm, and to mine the supply-demand correlation between technical achievements and demand information.

[0132] In this optional embodiment, the supply-demand correlation mining unit 103 includes: a characteristic entropy value calculation module, a priority matching module, a supply-demand matrix construction module, a supply-demand matrix completion module and a supply-demand correlation module;

[0133] The feature entropy value calculation module is used to calculate the achievement feature entropy value of each field based on the achievement feature data in the established matching feature set.

[0134] It should be noted that the calculation formula of the characteristic entropy value of the outcome is:

[0135] ;

[0136] In the formula, H j Indicates j The characteristic entropy value of the results in each field; n Indicates the number of data items in the field; i Represents the index value; j Indicates the field; p ij Indicates j The field in i The proportion of data in the row.

[0137] The priority matching module is used to calculate the weight coefficient of each field based on the entropy value of the achievement characteristics of each field using the entropy weight formula, and dynamically adjust the matching priority of the achievement characteristic data in different fields through the weight coefficient.

[0138] It should be noted that the entropy weight formula is:

[0139] ;

[0140] In the formula, oh j Indicates jThe weight coefficient of each field; H j Indicates j The characteristic entropy value of the results in each field; m Indicates the field; k Indicates the index value.

[0141] The supply and demand matrix construction module is used to combine the priority-adjusted outcome feature data and demand feature data to construct a supply and demand matrix.

[0142] The supply and demand matrix completion module is used to analyze the supply and demand matrix by using non-negative matrix decomposition technology to obtain the implicit feature space of the supply and demand matrix, and combine it with the matrix completion algorithm to obtain a new supply and demand matrix.

[0143] In this optional embodiment, the supply-demand matrix completion module includes: a supply-demand matrix decomposition submodule, an implicit feature space acquisition submodule and a matrix completion submodule;

[0144] The supply and demand matrix decomposition submodule is used to randomly initialize two non-negative matrices based on the supply and demand matrix using non-negative matrix decomposition technology, wherein the two non-negative matrices correspond to a non-negative matrix representing the implicit characteristics of supply and a non-negative matrix representing the implicit characteristics of demand.

[0145] The implicit feature space acquisition submodule is used to iteratively optimize two non-negative matrices using an alternating minimization algorithm, and output the final two optimized non-negative matrices to obtain the implicit feature space of the supply and demand matrix.

[0146] The matrix completion submodule is used to combine the matrix completion algorithm to process the missing values ​​in the implicit feature space and obtain a new supply and demand matrix.

[0147] In this optional embodiment, the combination of the matrix completion algorithm to process the missing values ​​in the implicit feature space to obtain a new supply and demand matrix includes:

[0148] According to the implicit feature space of the supply and demand matrix, the target matrix is ​​defined, and a low-rank matrix is ​​obtained to fill the missing values ​​by minimizing the nuclear norm of the target matrix.

[0149] Combined with slack variables, an optimization model with low rank and data fitting is constructed, and the alternating direction multiplier algorithm is used for iterative optimization to update the supply and demand matrix.

[0150] In this optional embodiment, the formula of the optimization model is:

[0151] ;

[0152] In the formula, X represents a low-rank matrix; E represents the slack variable; represents the nuclear norm of the low-rank matrix; s represents the regularization parameter; represents the square of the norm of the slack variable; subject to Indicates constraints; P μ represents the projection operator; M Represents the target matrix.

[0153] Extract the complete supply and demand correlation information from the updated supply and demand matrix to generate a new supply and demand matrix.

[0154] The supply-demand correlation module is used to mine the supply-demand correlation between technological achievements and demand information based on the new supply-demand matrix using the graph neural network algorithm.

[0155] It should be noted that, assuming that a matching feature set has been established through the previous process: {achievement feature data of field A, achievement feature data of field B, ...}, each field has corresponding achievement feature data.

[0156] Step 1: Feature entropy calculation module

[0157] (1) Feature point extraction:

[0158] Use appropriate feature point extraction algorithms (such as SIFT, SURF, etc.) to extract feature points from the achievement feature data of each field.

[0159] (2) Regional division: Divide the characteristic data of each field into several small areas and count the number of characteristic points in each area.

[0160] (3) Normalization: Normalize the number of feature points in each small area so that their sum is 1, and obtain the probability of feature points appearing in each area.

[0161] (4) Calculate entropy value: Use the calculation formula of the entropy value of the outcome characteristic to calculate the entropy value of the outcome characteristic in each field.

[0162] The entropy value of the achievement characteristics in each field is obtained, such as the entropy value of field A is 0.85, the entropy value of field B is 0.70, and so on.

[0163] Step 2: Priority matching module

[0164] (1) Calculate the weight coefficient:

[0165] Use the entropy weight formula to calculate the weight coefficient of each field.

[0166] (2) Adjust priorities:

[0167] According to the weight coefficient, the matching priority of the achievement feature data in different fields is dynamically adjusted; in the field with a higher weight coefficient, the matching priority of the achievement feature data is correspondingly increased, and the weight coefficient of each field and the dynamically adjusted matching priority are obtained.

[0168] Step 3: Supply and Demand Matrix Construction Module

[0169] (1) Data integration:

[0170] Integrate the priority-adjusted outcome characteristic data with the demand characteristic data.

[0171] (2) Constructing the matrix:

[0172] A supply-demand matrix is ​​constructed with the outcome feature data as rows and the demand feature data as columns. The elements in the matrix represent the matching degree or correlation between the outcome feature data and the demand feature data (which can be calculated by a similarity algorithm). The supply-demand matrix is ​​obtained, which contains the matching information between the outcome feature data and the demand feature data.

[0173] Step 4: Supply and Demand Matrix Completion Module

[0174] (1) Non-negative matrix factorization:

[0175] Perform non-negative matrix factorization on the supply and demand matrix to obtain the basis matrix and coefficient matrix.

[0176] (2) Obtaining implicit feature space:

[0177] By analyzing the basis matrix and coefficient matrix, the implicit feature space of the supply and demand matrix is ​​obtained.

[0178] (3) Matrix completion:

[0179] The supply and demand matrix is ​​completed using the matrix completion algorithm to obtain a new supply and demand matrix.

[0180] A new supply-demand matrix is ​​obtained, which contains more complete and accurate matching information.

[0181] Step 5: Supply and Demand Correlation Module

[0182] (1) Data preparation:

[0183] The new supply and demand matrix is ​​used as input data for the graph neural network algorithm.

[0184] (2) Constructing the graph structure:

[0185] According to the matching information in the supply and demand matrix, a graph structure is constructed, in which the nodes represent the outcome feature data and the demand feature data, and the edges represent the matching relationship or correlation between them.

[0186] (3) Training model:

[0187] The constructed graph structure is trained using a graph neural network algorithm to learn the supply-demand correlation between outcome feature data and demand feature data.

[0188] (4) Mining correlation:

[0189] Through the trained model, the supply and demand correlation between technological achievements and demand information is explored.

[0190] The supply-demand correlation between technological achievements and demand information is obtained, providing a basis for subsequent decision-making analysis and application.

[0191] In a specific embodiment, it is assumed that there are the following experimental data:

[0192] The feature data of the results in field A: {Feature 1, Feature 2, ...}.

[0193] The achievement feature data of field B: {Feature 3, Feature 4, ...}.

[0194] Demand characteristic data: {Requirement 1, Requirement 2, ...}.

[0195] After processing by the above modules, the following results are obtained:

[0196] The characteristic entropy value of the results in field A is: 0.85.

[0197] The characteristic entropy value of the results in field B is: 0.70.

[0198] The weight coefficient of field A is 0.6.

[0199] The weight coefficient of field B is 0.4.

[0200] The details are shown in Table 1, Table 2 and Table 3:

[0201] Table 1: Supply and demand matrix (partial)

[0202] ;

[0203] Table 2: New supply and demand matrix (after completion)

[0204] ;

[0205] Table 3: Supply-demand correlation between technological achievements and demand information (partial)

[0206] ;

[0207] The optimal matching result acquisition unit 105 is used to define a multi-objective function based on the mined supply-demand correlation, and to acquire the optimal matching result through a genetic algorithm.

[0208] It should be noted that the specific steps of defining a multi-objective function based on the mined supply-demand correlation and obtaining the optimal matching result through a genetic algorithm are as follows:

[0209] Step 1: Define multi-objective function

[0210] According to business needs, determine the optimization goal, such as maximizing the correlation between supply and demand, minimizing costs, maximizing profits, etc.; integrate multiple objective functions into a comprehensive objective function, and use weighted sum, min-max, Pareto frontier and other methods. For example, the following multi-objective function can be defined:

[0211] ;

[0212] In the formula, F ( x ) represents the value of the multi-objective function; e Represents the weight coefficient of the supply-demand correlation in the objective function; β Represents the weight coefficient of cost in the objective function; f Represents the weight coefficient of transportation time in the objective function; x Indicates the matching situation between suppliers and demanders.

[0213] Step 2: Get the best matching result

[0214] (1) Initialize the population:

[0215] A set of matching schemes is randomly generated as the initial population.

[0216] Each matching solution can be encoded as a chromosome using binary encoding, real number encoding or other suitable encoding methods.

[0217] (2) Fitness evaluation:

[0218] The fitness value of each matching solution is calculated using the defined multi-objective function.

[0219] They can be sorted according to their fitness values, and individuals with higher fitness can be selected for subsequent operations.

[0220] (3) Select an operation:

[0221] Select parent individuals using strategies such as roulette wheel selection, tournament selection, etc.

[0222] Select individuals with higher fitness as parents to preserve excellent genes.

[0223] (4) Crossover operation:

[0224] Perform a crossover operation on the selected parent individuals to generate new offspring individuals.

[0225] You can use strategies such as single-point crossover, double-point crossover, and uniform crossover.

[0226] (5) Mutation operation:

[0227] Perform mutation operations on offspring individuals to increase the diversity of the population.

[0228] Strategies such as bit flipping and real number mutation can be used.

[0229] (6) Update population:

[0230] The generated offspring individuals are merged with the parent individuals to form a new population.

[0231] Select the best individual according to the fitness value and update the population.

[0232] (7) Termination conditions:

[0233] Set termination conditions such as number of iterations and fitness threshold.

[0234] When the termination condition is reached, the iteration stops and the optimal matching solution is output.

[0235] (8) Output the optimal solution:

[0236] The individual with the highest fitness is selected from the population at the time of termination as the optimal matching solution to obtain a specific supply and demand matching result.

[0237] The matching result verification and optimization unit 107 is used to verify the matching result using a game theory algorithm and optimize the matching process through user feedback data in combination with a reinforcement learning mechanism.

[0238] In this optional embodiment, the matching result verification and optimization unit 107 includes: a matching result verification module, an optimization guidance module and a matching process optimization module;

[0239] The matching result verification module is used to use the game theory algorithm to build a game model involving both supply and demand parties, and to verify the stability of the matching results in combination with the mixed strategy Nash equilibrium algorithm, and to mark unstable matching pairs;

[0240] The optimization guidance module is used to build a reinforcement learning model based on the reinforcement learning mechanism, define the adjustment matching strategy as an action space, and formulate a reward function based on the matching stability to guide the optimization process;

[0241] The matching process optimization module is used to obtain user feedback data and optimize the matching process in combination with the deep Q learning algorithm.

[0242] In this optional embodiment, the formula of the deep Q learning algorithm is:

[0243] ;

[0244] In the formula, L ( i ) represents the loss function; E Indicates expected value; s Indicates status; a Indicates action; r Indicates immediate reward; s ´ Indicates the next state; c represents the discount factor; a´ Indicates the next action; Q express Q value; i Represents the set of parameters of a neural network.

[0245] It should be noted that, in a specific embodiment, the details are as follows:

[0246] Step 1: Matching result verification module

[0247] (1) Constructing a game model:

[0248] Set the strategic space for both supply and demand sides, such as the price range that suppliers can choose and the purchase quantity that demanders can choose.

[0249] Construct a profit matrix to represent the profits of both parties under different strategy combinations.

[0250] (2) Calculate the mixed strategy Nash equilibrium:

[0251] Use mathematical methods (such as linear programming) to calculate the mixed strategy Nash equilibrium, that is, the optimal probability distribution of each player under the strategies of other players.

[0252] Verify the stability of the matching results, that is, check whether the mixed strategy Nash equilibrium is reached.

[0253] (3) Mark unstable matching pairs:

[0254] Matching pairs that do not reach the mixed strategy Nash equilibrium are marked as unstable matching pairs.

[0255] Analyze the reasons for unstable matching pairs, such as unreasonable prices, mismatched demands, etc.

[0256] For example: suppose there are 10 suppliers and 10 demanders, and the profit matrix is ​​derived from historical transaction data; through calculation, it is found that 5 pairs of matches have reached the mixed strategy Nash equilibrium, while the other 5 pairs have not; after analyzing the matching pairs that have not reached equilibrium, it is found that the main reason is unreasonable pricing strategy.

[0257] Step 2: Optimize the guidance module

[0258] (1) Building a reinforcement learning model:

[0259] Define the state space, such as the stability of the current matching pair, historical trading data, etc.

[0260] Define the action space, such as adjusting price, changing purchase quantity, etc.

[0261] Define a reward function, such as giving positive or negative rewards based on matching stability.

[0262] (2) Training reinforcement learning model:

[0263] Use historical transaction data as a training set and train the model through reinforcement learning algorithms (such as Q-learning, DQN, etc.).

[0264] During the training process, the matching strategy is adjusted according to the reward function to improve the matching stability.

[0265] (3) Guiding the optimization process:

[0266] Use the trained model to make predictions and optimizations for new matching pairs.

[0267] Adjust the matching strategy according to the optimal strategy output by the model to improve matching stability.

[0268] For example, the DQN model was trained using historical transaction data as a training set; on the test set, the model successfully improved matching stability and reduced the proportion of unstable matching pairs by 20%.

[0269] Step 3: Matching process optimization module

[0270] (1) Collect user feedback data:

[0271] During the matching process, user feedback data is collected through user questionnaires, online reviews, etc.

[0272] Preprocess and classify feedback data, for example, into satisfaction, suggestions, complaints, etc.

[0273] (2) Building a deep Q learning model:

[0274] Define the state space, such as user feedback data, current matching status, etc.

[0275] Define the action space, such as adjusting matching algorithm parameters, optimizing recommendation strategies, etc.

[0276] Use deep neural network as Q-value function approximator.

[0277] (3) Training the deep Q learning model:

[0278] The user feedback data is used as the training set and the model is trained by the deep Q learning algorithm.

[0279] During training, the matching process is adjusted based on user feedback to improve user satisfaction.

[0280] (4) Optimize the matching process:

[0281] Use the trained model to predict and optimize the matching process.

[0282] Adjust the matching process according to the optimal strategy output by the model to improve user satisfaction and matching efficiency.

[0283] For example: 1,000 user feedback data were collected, preprocessed and classified; the deep Q learning model was trained using this data and verified on the test set; the results showed that the model successfully improved user satisfaction and matching efficiency, reducing the proportion of dissatisfied users by 15%.

[0284] The matching result display unit 109 is used to construct a supply-demand matching algorithm framework and display the supply-demand matching results using a visualization tool.

[0285] It should be noted that, in a specific embodiment, the details are as follows:

[0286] Step 1: Construction of supply and demand matching algorithm framework

[0287] (1) Data preparation and preprocessing

[0288] Data collection: Collect multi-dimensional data such as technology level, market level, capital level, etc. required for supply and demand matching from multiple sources (such as scientific and technological achievement database, enterprise database, market research report, investment and financing database, industrial policy, news, etc.).

[0289] Data cleaning: remove duplicate data, process missing values, correct erroneous data, etc. to ensure data accuracy and consistency.

[0290] Data standardization: Standardize data from different sources and in different formats to facilitate subsequent algorithm processing.

[0291] (2) Feature extraction and selection

[0292] Feature extraction: Extract key features from the data of both supply and demand sides, such as technical keywords, achievement team information, enterprise core technology, core product features, enterprise qualifications, investment institutions, investment events, investment amounts, peer competitors, bidding conditions, core component prices, quality, demand, supply, etc.

[0293] Feature selection: Select features that have a greater impact on the supply and demand matching results based on algorithm requirements and business logic.

[0294] (3) Supply and demand matching algorithm design

[0295] Algorithm selection: Select an appropriate supply-demand matching algorithm based on business needs and data characteristics, such as a distance-based matching algorithm, a similarity-based matching algorithm, etc.

[0296] Algorithm implementation: Write code to implement the selected algorithm and optimize it to improve matching efficiency and accuracy.

[0297] (4) Verification and optimization of matching results

[0298] Verification method: Use cross-validation, leave-one-out and other verification methods to verify the matching results.

[0299] Optimization strategy: Adjust algorithm parameters or improve the algorithm based on the verification results to improve the stability and accuracy of the matching results.

[0300] Step 2: Visualization tools display supply and demand matching results

[0301] (1) Visualization tool selection

[0302] Choose appropriate visualization tools, such as Tableau, Power BI, ECharts, etc. These tools can provide a variety of chart types and interactive functions to facilitate the display and analysis of supply and demand matching results.

[0303] (2) Data visualization design

[0304] Supply and demand distribution chart: Use a bar chart or pie chart to show the distribution of supply and demand, such as the comparison of supply and demand for different products.

[0305] Matching result graph: Use a scatter plot or heat map to display the supply and demand matching results, where the position of the point represents the characteristic values ​​of both the supply and demand sides, and the color or size of the point represents the degree of matching.

[0306] Trend analysis chart: Use a time series chart to show the changing trend of supply and demand matching results, such as the matching success rate in different time periods.

[0307] (3) Interactive function design

[0308] Screening and filtering: Screening and filtering functions are provided to enable users to select specific supply and demand parties or characteristics for display as needed.

[0309] Drilling and linkage: Provides drilling and linkage functions, allowing users to deeply explore the details and related information of supply and demand matching results.

[0310] Figure 2 An embodiment of a method for matching supply and demand of achievement transformation of the present invention is shown.

[0311] In this optional embodiment, the method for matching supply and demand of achievement transformation includes:

[0312] The collected technical achievements and demand information are converted into high-dimensional vectors, and matching feature data are extracted based on the high-dimensional vectors to establish a matching feature set; according to the established matching feature set, the entropy weight algorithm is used to dynamically adjust the matching priority of the achievement feature data in different fields, and the supply and demand correlation between the technical achievements and demand information is mined; based on the mined supply and demand correlation, a multi-objective function is defined, and the optimal matching result is obtained through a genetic algorithm; the matching result is verified using a game theory algorithm, and combined with a reinforcement learning mechanism, the matching process is optimized through user feedback data; a supply and demand matching algorithm framework is constructed, and visualization tools are used to display the supply and demand matching results.

[0313] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 3 As shown. The computer device includes a processor, a memory and a network interface connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store static information and dynamic information data. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the steps in the above method embodiment are implemented.

[0314] Those skilled in the art will understand that Figure 3 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0315] In addition, the present invention also provides a computer device, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above method embodiment when executing the computer program.

[0316] In addition, the present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps in the above method embodiment are implemented.

[0317] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present invention can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, etc. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0318] The present invention is not limited to the structures which have been described above and shown in the drawings, and various modifications and changes may be made without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.

Claims

1. A research achievement transformation supply and demand matching system, characterized in that: The system includes: A feature extraction unit is used to convert the collected technical achievements and demand information into a high-dimensional vector, and extract matching feature data based on the high-dimensional vector to establish a matching feature set; The supply-demand correlation mining unit is used to dynamically adjust the matching priority of achievement feature data in different fields based on the established matching feature set using the entropy weight algorithm, and to mine the supply-demand correlation between technical achievements and demand information; The optimal matching result acquisition unit is used to define a multi-objective function based on the mined supply-demand correlation and obtain the optimal matching result through a genetic algorithm; The matching result verification and optimization unit is used to verify the matching results using game theory algorithms and optimize the matching process through user feedback data in combination with reinforcement learning mechanisms; The matching result display unit is used to build a supply-demand matching algorithm framework and use visualization tools to display the supply-demand matching results; The supply-demand correlation mining unit comprises: The supply and demand matrix completion module is used to analyze the supply and demand matrix by using non-negative matrix decomposition technology to obtain the implicit feature space of the supply and demand matrix, and combine it with the matrix completion algorithm to obtain a new supply and demand matrix; The supply and demand matrix completion module includes: The matrix completion submodule is used to combine the matrix completion algorithm to process the missing values ​​in the implicit feature space and obtain a new supply and demand matrix.

2. According to claim 1, a research achievement transformation supply and demand matching system is characterized in that: The feature extraction unit includes: a data acquisition and preprocessing module, a high-dimensional vector generation module, a feature data extraction module and a feature set establishment module; The data acquisition and preprocessing module is used to obtain technical achievements and demand information from data sources using text extraction technology, and to preprocess the obtained technical achievements and demand information; A high-dimensional vector generation module is used to input the pre-processed technical achievements and demand information into the pre-built sentence vector embedding model, generate the corresponding high-dimensional vector, and normalize the high-dimensional vector; The feature data extraction module is used to extract the achievement feature data and demand feature data from the normalized high-dimensional vector according to the deep learning algorithm; The feature set establishment module is used to use a similarity algorithm to calculate the similarity between the result feature data and the demand feature data, and obtain matching feature data based on the similarity to establish a matching feature set.

3. According to claim 2, a research achievement transformation supply and demand matching system is characterized in that: The feature data extraction module includes: a feature extraction model construction submodule, a language information acquisition submodule and a feature data acquisition submodule; The feature extraction model construction submodule is used to select a pre-trained bidirectional encoder representation model as the basic architecture according to the deep learning algorithm, and add a shared layer and a multi-task learning layer to the basic architecture to build a feature extraction model; The language information acquisition submodule is used to input the normalized high-dimensional vector into the feature extraction model, and capture the semantic information in the high-dimensional vector through the multi-layer encoder of the feature extraction model; The feature data acquisition submodule is used to transfer the captured language information to the multi-task learning layer and obtain the result feature data and the requirement feature data from the middle layer of the multi-task learning layer.

4. According to claim 1, a research achievement transformation supply and demand matching system is characterized in that: The supply-demand correlation mining unit also includes: a characteristic entropy value calculation module, a priority matching module, a supply-demand matrix construction module and a supply-demand correlation module; A feature entropy value calculation module is used to calculate the feature entropy value of each field according to the feature data of the results in the established matching feature set; The priority matching module is used to calculate the weight coefficient of each field based on the entropy value of the achievement characteristics of each field using the entropy weight formula, and dynamically adjust the matching priority of the achievement characteristic data in different fields through the weight coefficient; A supply and demand matrix construction module is used to combine the priority-adjusted outcome feature data with the demand feature data to construct a supply and demand matrix; The supply-demand correlation module is used to mine the supply-demand correlation between technological achievements and demand information based on the new supply-demand matrix using the graph neural network algorithm.

5. According to claim 4, a research achievement transformation supply and demand matching system is characterized in that: The supply and demand matrix completion module also includes: a supply and demand matrix decomposition submodule and an implicit feature space acquisition submodule; A supply-demand matrix decomposition submodule is used to randomly initialize two non-negative matrices based on the supply-demand matrix using a non-negative matrix decomposition technique, wherein the two non-negative matrices correspond to a non-negative matrix representing an implicit feature of supply and a non-negative matrix representing an implicit feature of demand, respectively; The implicit feature space acquisition submodule is used to iteratively optimize two non-negative matrices using an alternating minimization algorithm, and output the final two optimized non-negative matrices to obtain the implicit feature space of the supply and demand matrix.

6. A research achievement transformation supply and demand matching system according to claim 5, characterized in that: The combined matrix completion algorithm processes the missing values ​​in the implicit feature space to obtain a new supply and demand matrix including: According to the implicit feature space of the supply and demand matrix, the target matrix is ​​defined, and the low-rank matrix is ​​obtained to fill the missing values ​​by minimizing the nuclear norm of the target matrix; Combined with slack variables, an optimization model with low rank and data fitting is constructed, and the alternating direction multiplier algorithm is used for iterative optimization to update the supply and demand matrix; Extract the complete supply and demand correlation information from the updated supply and demand matrix to generate a new supply and demand matrix.

7. The achievement transformation supply and demand matching system according to claim 6 is characterized in that: The formula of the optimization model is: ； In the formula, X represents a low-rank matrix; E represents the slack variable; represents the nuclear norm of the low-rank matrix; σ represents the regularization parameter; represents the square of the norm of the slack variable; subject to Indicates constraints; P μ represents the projection operator; M Represents the target matrix.

8. The achievement transformation supply and demand matching system according to claim 1 is characterized in that: The matching result verification and optimization unit includes: a matching result verification module, an optimization guidance module and a matching process optimization module; The matching result verification module is used to use the game theory algorithm to build a game model involving both supply and demand parties, and to verify the stability of the matching results in combination with the mixed strategy Nash equilibrium algorithm, and to mark unstable matching pairs; The optimization guidance module is used to build a reinforcement learning model based on the reinforcement learning mechanism, define the adjustment matching strategy as an action space, and formulate a reward function based on the matching stability to guide the optimization process; The matching process optimization module is used to obtain user feedback data and optimize the matching process in combination with the deep Q learning algorithm.

9. The achievement transformation supply and demand matching system according to claim 8 is characterized in that: The formula for the deep Q-learning algorithm is: ； In the formula, L ( θ ) represents the loss function; E Indicates expected value; s Indicates status; Indicates action; r Indicates immediate reward; s´ Indicates the next state; γ represents the discount factor; Indicates the next action; Q express Q value; θ Represents the set of parameters of a neural network.

10. A method for matching supply and demand of achievement transformation, used to implement a system for matching supply and demand of achievement transformation as claimed in any one of claims 1 to 9, characterized in that: The method includes: Convert the collected technical achievements and demand information into high-dimensional vectors, extract matching feature data based on the high-dimensional vectors, and establish a matching feature set; According to the established matching feature set, the entropy weight algorithm is used to dynamically adjust the matching priority of the achievement feature data in different fields, and to explore the supply and demand correlation between technical achievements and demand information; Based on the supply-demand correlation mined, a multi-objective function is defined, and the optimal matching result is obtained through a genetic algorithm; Use game theory algorithms to verify matching results, and combine reinforcement learning mechanisms to optimize the matching process through user feedback data; Build a supply and demand matching algorithm framework and use visualization tools to display the supply and demand matching results.

Images