Scientific achievement processing method and device, electronic equipment and readable medium

Abstract

The embodiment of the invention provides a scientific research achievement processing method and device, electronic equipment and a readable medium, and is applied to a scientific research achievement management system, and the method comprises the steps: obtaining related data and at least one piece of market demand information of a to-be-processed scientific research achievement, and determining at least one achievement feature of the to-be-processed scientific research achievement based on the related data; for any one piece of market demand information, judging whether the result features are related to the market demand information by using a preset deep transfer learning method; and if the achievement characteristics are related to the market demand information, generating a target application achievement of the to-be-processed scientific research achievement matched with the market demand information by using a preset knowledge graph completion algorithm, generating a target conversion path between the to-be-processed scientific research achievement and the target application achievement, and converting the to-be-processed scientific research achievement to the target application achievement by generating an achievement conversion path. Mining of scientific and technological achievement application values is optimized, and achievement transformation support is provided for users.

Description

Technical Field

[0001] The present application relates to the field of information technology, and particularly to a method for processing scientific research achievements, an apparatus for processing scientific research achievements, an electronic device, and a computer-readable medium. Background Art

[0002] In related technologies, scientific research achievement management systems are usually used for the management and application of scientific research achievements, including providing support for the transformation of scientific research achievements into practical applications. However, currently, the method of manually analyzing scientific research achievements and communicating the analysis results to determine the methods or suggestions for transforming scientific research achievements into practical applications results in low transformation efficiency of scientific research achievements. Summary of the Invention

[0003] Embodiments of the present application provide a method, an apparatus, an electronic device, and a computer-readable storage medium for processing scientific research achievements, so as to solve the problem that the transformation efficiency of scientific research achievements is low due to manually analyzing scientific research achievements and communicating the analysis results to determine the methods or suggestions for transforming scientific research achievements into practical applications.

[0004] Embodiments of the present application disclose a method for processing scientific research achievements, which is applied to a scientific research achievement management system. The method includes:

[0005] Obtaining relevant materials of the scientific research achievement to be processed and at least one piece of market demand information, and determining at least one achievement feature of the scientific research achievement to be processed based on the relevant materials;

[0006] For any one of the pieces of market demand information, using a preset deep transfer learning method to determine whether there is a correlation between the achievement feature and the market demand information;

[0007] If there is a correlation between the achievement feature and the market demand information, using a preset knowledge graph completion algorithm to generate a target application achievement of the scientific research achievement to be processed that matches the market demand information, and generating a target transformation path between the scientific research achievement to be processed and the target application achievement.

[0008] Optionally, the determining at least one achievement feature of the scientific research achievement to be processed based on the relevant materials includes:

[0009] Using a preset multimodal knowledge graph technology to extract target fields from the relevant materials;

[0010] Based on the relevant materials, using the multimodal knowledge graph technology to determine the association relationship between the target fields;

[0011] Based on the target fields and the association relationships between the target fields, use the multi-modal knowledge graph technology to generate a feature graph of the scientific research result to be processed.

[0012] Optionally, the relevant materials include at least one of text materials, image materials, and numerical materials; determining at least one result feature of the scientific research result to be processed based on the relevant materials includes:

[0013] For the text materials, use a preset language processing model to determine the text features of the text materials;

[0014] For the image materials, use a preset cross-modal contrast learning method to determine the image features of the image materials;

[0015] For the numerical materials, use a dynamic feature decomposition method based on causal discovery to generate data features of the numerical materials.

[0016] Optionally, the scientific research result to be processed includes the current scientific research result of the scientific research corresponding to the scientific research result to be processed, and the historical scientific research results of the scientific research at at least one historical time; the method includes:

[0017] Based on the current result features of the current scientific research result and the historical result features of the historical scientific research results, use a preset dynamic spatio-temporal graph network technology to construct a dynamic time series graph of the scientific research result to be processed.

[0018] Optionally, using a preset knowledge graph completion algorithm to generate a target application result of the scientific research result to be processed that matches the market demand information, and generating a target conversion path between the scientific research result to be processed and the target application result includes:

[0019] Use the knowledge graph completion algorithm to generate an initial conversion path graph between the scientific research result to be processed and the target application result; at least one conversion step is included in the initial conversion path graph; any one of the conversion steps has at least one object to be applied;

[0020] For any one of the conversion steps, according to the preset preference information of the user, determine a target application object from at least one object to be applied in the conversion step;

[0021] Based on the target application object, adjust the initial conversion path graph to obtain a target conversion path graph.

[0022] Optionally, the preset preference information includes: a user demand model and / or the unlabeled demand information of the user; according to the preset preference information of the user, determining a target application object from at least one object to be applied in the conversion step includes:

[0023] Generate the potential preference features of the unlabeled demand information by using a preset self-supervised learning method;

[0024] Use a preset recommendation algorithm based on contrastive learning to match the at least one object to be applied with the user demand model to determine the target application object, and / or,

[0025] Match the at least one object to be applied with the potential preference features to determine the target application object.

[0026] Optionally, relevant materials of at least one award-winning scientific research achievement are stored in the scientific research achievement management system; the method includes:

[0027] Generate a reward recommendation for the scientific research achievement to be processed according to the relevant materials of the award-winning scientific research achievement;

[0028] Use a preset document generation model to convert the reward recommendation and the generation process of the reward recommendation into a reward recommendation document for the scientific research achievement to be processed.

[0029] Optionally, the relevant materials of the award-winning scientific research achievement include the historical award-winning conditions of the award-winning scientific research achievement; generating the reward recommendation for the scientific research achievement to be processed according to the relevant materials of the award-winning scientific research achievement includes:

[0030] Use a preset neural network model based on time series to extract the long-term dependence features of the historical award-winning conditions;

[0031] Use a preset semantic similarity evaluation model to determine the first matching degree between the achievement features of the scientific research achievement to be processed and the long-term dependence features.

[0032] Optionally, the relevant materials of the award-winning scientific research achievement include the achievement features of the award-winning scientific research achievement; generating the reward recommendation for the scientific research achievement to be processed according to the relevant materials of the award-winning scientific research achievement includes:

[0033] Use a preset sequence-to-sequence generation model to analyze the achievement features of the award-winning scientific research achievement to determine the common features of the award-winning scientific research achievement;

[0034] Use the semantic similarity evaluation model to determine the second matching degree between the achievement features of the scientific research achievement to be processed and the common features.

[0035] Optionally, generating the reward recommendation for the scientific research achievement to be processed according to the relevant materials of the award-winning scientific research achievement includes:

[0036] Determine the award level of the scientific research result to be processed based on the first matching degree and the second matching degree.

[0037] Optionally, the method includes:

[0038] Use a preset graph neural network to display the determination process of the target application object in the target transformation path graph.

[0039] Optionally, the method includes:

[0040] Based on the relevant materials and dynamic time series graph of at least one scientific research result to be processed, use a preset hash coding technology to evaluate the similarity between at least one scientific research result to be processed.

[0041] Optionally, the using a preset cross-modal contrast learning method to determine the image features of the image class materials includes:

[0042] Use the cross-modal contrast learning method to train a preset image processing model to obtain an image feature extraction model;

[0043] Use the image feature extraction model to determine the image features of the image class materials.

[0044] Optionally, the method includes:

[0045] Use a preset regularization technique and / or a preset semantic error correction algorithm based on a language model to detect missing fields in the relevant materials and fill in the missing fields.

[0046] Optionally, the method includes:

[0047] Use a preset scoring mechanism based on Bayesian optimization to determine the completeness score of the relevant materials;

[0048] If the completeness score is lower than a preset score threshold, use a preset generative adversarial network to generate filling fields corresponding to the missing fields in the relevant materials;

[0049] Use the filling fields to improve the completeness score of the scientific research result to be processed.

[0050] An embodiment of the present application also discloses a processing device for scientific research results, which is applied to a scientific research result management system. The device includes:

[0051] A data acquisition module, configured to acquire relevant materials of a scientific research result to be processed and at least one market demand information, and determine at least one result feature of the scientific research result to be processed based on the relevant materials;

[0052] A judgment module, configured to determine, for any one of the market demand information, whether there is a correlation between the result feature and the market demand information by using a preset deep transfer learning method;

[0053] A path generation module, configured to, if there is a correlation between the result feature and the market demand information, generate a target application result of the to-be-processed scientific research result that matches the market demand information by using a preset knowledge graph completion algorithm, and generate a target transformation path between the to-be-processed scientific research result and the target application result.

[0054] Optionally, the data acquisition module includes:

[0055] A target field extraction sub-module, configured to extract target fields from the relevant data by using a preset multi-modal knowledge graph technology;

[0056] An association relationship determination sub-module, configured to determine the association relationship between the target fields based on the relevant data by using the multi-modal knowledge graph technology;

[0057] A feature graph generation sub-module, configured to generate a feature graph of the to-be-processed scientific research result by using the multi-modal knowledge graph technology based on the target fields and the association relationship between the target fields.

[0058] Optionally, the relevant data includes at least one of text data, image data, and numerical data; the data acquisition module includes:

[0059] A text feature determination sub-module, configured to determine the text features of the text data by using a preset language processing model for the text data;

[0060] An image feature determination sub-module, configured to determine the image features of the image data by using a preset cross-modal contrast learning method for the image data;

[0061] A data feature determination sub-module, configured to generate data features of the numerical data by using a dynamic feature decomposition method based on causal discovery for the numerical data.

[0062] Optionally, the to-be-processed scientific research result includes the current scientific research result of the scientific research corresponding to the to-be-processed scientific research result, and the historical scientific research results of the scientific research at at least one historical time; the device includes:

[0063] A dynamic time series graph construction module, configured to construct a dynamic time series graph of the to-be-processed scientific research result by using a preset dynamic spatio-temporal graph network technology based on the current result features of the current scientific research result and the historical result features of the historical scientific research results.

[0064] Optionally, the path generation module includes:

[0065] An initial transformation path graph generation sub-module, configured to generate an initial transformation path graph between the scientific research result to be processed and the target application result by using the knowledge graph completion algorithm; at least one transformation step is included in the initial transformation path graph; any one of the transformation steps has at least one object to be applied;

[0066] A target application object determination sub-module, configured to, for any one of the transformation steps, determine a target application object from at least one object to be applied in the transformation step according to the preset preference information of the user;

[0067] A target transformation path graph obtaining sub-module, configured to adjust the initial transformation path graph based on the target application object to obtain a target transformation path graph.

[0068] Optionally, the preset preference information includes: a user demand model and / or the unannotated demand information of the user; the target application object determination sub-module includes:

[0069] A feature generation unit, configured to generate potential preference features of the unannotated demand information by using a preset self-supervised learning method;

[0070] A matching unit, configured to use a preset recommendation algorithm based on contrast learning to match at least one object to be applied with the user demand model to determine the target application object, and / or,

[0071] Match at least one object to be applied with the potential preference features to determine the target application object.

[0072] Optionally, relevant materials of at least one award-winning scientific research result are stored in the scientific research result management system; the device includes:

[0073] An award nomination generation module, configured to generate an award nomination for the scientific research result to be processed according to the relevant materials of the award-winning scientific research result;

[0074] A document conversion module, configured to convert the award nomination and the generation process of the award nomination into an award recommendation document for the scientific research result to be processed by using a preset document generation model.

[0075] Optionally, the relevant materials of the award-winning scientific research result include the historical award-winning conditions of the award-winning scientific research result; the award nomination generation module includes:

[0076] A feature extraction sub-module, configured to extract long-term dependence features of the historical award-winning conditions by using a preset neural network model based on time series;

[0077] The first matching degree determination sub-module is used to determine the first matching degree between the result features of the scientific research result to be processed and the long-term dependence features by using a preset semantic similarity evaluation model.

[0078] Optionally, the relevant materials of the awarded scientific research result include the result features of the awarded scientific research result; the award recommendation generation module includes:

[0079] The common feature determination sub-module is used to analyze the result features of the awarded scientific research result by using a preset sequence-to-sequence generation model to determine the common features of the awarded scientific research result;

[0080] The second matching degree determination sub-module is used to determine the second matching degree between the result features of the scientific research result to be processed and the common features by using the semantic similarity evaluation model.

[0081] Optionally, the award recommendation generation module includes:

[0082] The award level determination sub-module is used to determine the award level of the scientific research result to be processed based on the first matching degree and the second matching degree.

[0083] Optionally, the device includes:

[0084] The display module is used to display the determination process of the target application object in the target transformation path diagram by using a preset graph neural network.

[0085] Optionally, the device includes:

[0086] The evaluation module is used to evaluate the similarity between at least one scientific research result to be processed based on the relevant materials of at least one scientific research result to be processed and a dynamic time series graph by using a preset hash coding technology.

[0087] Optionally, the image feature determination sub-module includes:

[0088] The model obtaining unit is used to train a preset image processing model by using the cross-modal contrast learning method to obtain an image feature extraction model;

[0089] The image feature determination unit is used to determine the image features of the image-type materials by using the image feature extraction model.

[0090] Optionally, the device includes:

[0091] The filling module is used to detect the missing fields in the relevant materials and fill the missing fields by using a preset regularization technology and / or a preset semantic error correction algorithm based on a language model.

[0092] Optionally, the apparatus includes:

[0093] A scoring determination module, configured to determine a completeness score of the relevant materials by using a preset scoring mechanism based on Bayesian optimization;

[0094] A field generation module, configured to, if the completeness score is lower than a preset score threshold, generate a filling field corresponding to a missing field in the relevant materials by using a preset generative adversarial network;

[0095] A score improvement module, configured to use the filling field to improve the completeness score of the scientific research achievement to be processed.

[0096] An embodiment of the present application also discloses an electronic device, including a processor, a communication interface, a memory, and a communication bus. Among them, the processor, the communication interface, and the memory complete communication with each other through the communication bus;

[0097] The memory is used to store a computer program;

[0098] The processor is configured to implement the method described in the embodiment of the present application when executing the program stored in the memory.

[0099] An embodiment of the present application also discloses one or more computer-readable media, on which instructions are stored. When executed by one or more processors, the instructions cause the processors to execute the method described in the embodiment of the present application.

[0100] The embodiments of the present application include the following advantages:

[0101] In the embodiment of the present application, a scientific research achievement management system can obtain relevant materials of a scientific research achievement to be processed and at least one market demand information, and determine at least one achievement feature of the scientific research achievement to be processed based on the relevant materials; for any one of the market demand information, use a preset deep transfer learning method to determine whether the achievement feature is relevant to the market demand information; if the achievement feature is relevant to the market demand information, use a preset knowledge graph completion algorithm to generate a target application achievement of the scientific research achievement to be processed that matches the market demand information, and generate a target transformation path between the scientific research achievement to be processed and the target application achievement. By generating an achievement transformation path, the excavation of the application value of scientific and technological achievements is optimized, and achievement transformation support is provided for users. BRIEF DESCRIPTION OF THE DRAWINGS

[0102] Figure 1 is a flowchart of steps of a method for processing a scientific research achievement provided in an embodiment of the present application;

[0103] Figure 2 is a flowchart of steps of another method for processing a scientific research achievement provided in an embodiment of the present application;

[0104] Figure 3 It is a structural block diagram of a processing device for scientific research achievements provided in an embodiment of the present application;

[0105] Figure 4 It is a block diagram of an electronic device provided in an embodiment of the present application;

[0106] Figure 5 It is a schematic diagram of a computer-readable medium provided in an embodiment of the present application. Detailed implementation manners

[0107] To make the above objects, features, and advantages of the present application more obvious and understandable, the present application will be further described in detail below with reference to the accompanying drawings and specific implementation manners.

[0108] To facilitate the understanding of the technical solutions and technical effects of the embodiments of the present application, the related technologies of the present application will be briefly described below.

[0109] In the related technologies, a scientific research achievement management system is usually used for the management and application of scientific research achievements, including: sorting and archiving the relevant materials of scientific research achievements, providing support for the transformation of scientific research achievements into practical applications, and assisting in the declaration and recommendation of scientific and technological awards for scientific research achievements, etc.

[0110] The scientific research achievement management system usually relies on database storage and manual operations to manage and apply scientific research achievements, mainly including the following steps:

[0111] 1) Enter scientific research achievements: Complete the preliminary registration of scientific research achievements in the system by manually inputting or uploading scientific research achievements.

[0112] 2) Data sorting: Manually check the integrity and consistency of the data related to scientific research achievements, and perform basic classification and tagging of scientific research achievements.

[0113] 3) Transformation management: Manually analyze scientific research achievements and communicate the analysis results to determine the methods or suggestions for transforming scientific research achievements into practical applications, which may lead to low transformation efficiency of scientific research achievements.

[0114] 4) Award recommendation: Experts construct historical condition screening rules based on the historical experience of awarding scientific research achievements, and screen award recommendation objects from numerous scientific research achievements based on these rules, and this process has a high degree of subjectivity.

[0115] Therefore, the scientific research achievement management system has at least the following problems:

[0116] 1) High cost of manual intervention: The scientific research achievement management system relies on manual labor for the input, data collation, transformation management, and award recommendation of scientific research achievements. Its level of intelligence is relatively low, time-consuming, laborious, and inefficient.

[0117] 2) Insufficient consistency detection: The scientific research achievement management system lacks systematic checks for the consistency and completeness of the scientific research achievement data entered into the system, which easily leads to data omission or errors, affecting subsequent management processes such as transformation management and award recommendation.

[0118] 3) Low recommendation accuracy: Static matching of scientific research achievements and rules is performed based on simple historical condition screening rules, without fully utilizing the historical experience of scientific research achievement awards. This may result in lower scientificity and success rate of award recommendation.

[0119] Refer to Figure 1 and shows the step flowchart of a method for processing scientific research achievements provided in an embodiment of the present application, which is applied to a scientific research achievement management system and specifically may include the following steps:

[0120] Step 101, obtain relevant materials of the scientific research achievement to be processed and at least one piece of market demand information, and based on the relevant materials, determine at least one achievement feature of the scientific research achievement to be processed;

[0121] In an embodiment of the present application, the scientific research achievement management system can be used for the management and application of scientific research achievements, such as sorting and archiving relevant materials of scientific research achievements, detecting the completeness and consistency of relevant materials, providing support for the transformation of scientific research achievements into practical applications, and assisting in the declaration and recommendation of scientific research achievements for scientific and technological awards.

[0122] In an embodiment of the present application, the user can input relevant materials of the scientific research achievement to be processed and at least one piece of market demand information into the scientific research achievement management system. The scientific research achievement management system can determine at least one achievement feature of the scientific research achievement to be processed based on the relevant materials of the scientific research achievement to be processed.

[0123] In some embodiments of the present application, the method includes:

[0124] Detect missing fields in the relevant materials by using a preset regularization technique and / or a preset semantic error correction algorithm based on a language model, and fill in the missing fields.

[0125] In an embodiment of the present application, after the scientific research achievement management system obtains the relevant materials of the scientific research achievement to be processed, it is necessary to perform data standardization processing on the relevant materials. Among them, the relevant materials of the scientific research achievement to be processed include relevant texts, pictures, tables of the scientific research achievement to be processed, and relevant materials of historical scientific research achievements corresponding to the scientific research to be processed.

[0126] In the embodiments of the present application, data standardization processing refers to automatically cleaning the relevant materials of the scientific research results to be processed by using regularization techniques and / or semantic error correction algorithms based on language models, detecting and filling in the missing values in the relevant materials, and unifying the formats of the relevant materials of the scientific research results to be processed.

[0127] Among them, the regularization technique refers to using regular expressions to match, search for, and replace specific patterns in the materials. The semantic error correction algorithm based on the language model can use the language model to evaluate the rationality of the materials, identify and correct the abnormal parts that do not conform to the predictions of the language model. The language model includes T5 (Text-to-Text Transfer Transformer) and / or GPT (Generative Pre-trained Transformer). Both T5 and GPT are pre-trained language models based on the Transformer architecture. The Transformer architecture is a deep learning model architecture.

[0128] In some embodiments of the present application, determining at least one result feature of the scientific research results to be processed based on the relevant materials includes:

[0129] Extracting target fields from the relevant materials by using a preset multi-modal knowledge graph technique;

[0130] Based on the relevant materials, determining the association relationship between the target fields by using the multi-modal knowledge graph technique;

[0131] Generating a feature graph of the scientific research results to be processed by using the multi-modal knowledge graph technique based on the target fields and the association relationship between the target fields.

[0132] In the embodiments of the present application, the result features of the scientific research results to be processed may include a feature graph. The scientific research result management system can use the multi-modal knowledge graph (Multi-modal Knowledge Graph) technique to extract target fields from the relevant materials of the scientific research results to be processed; the target fields are also called key fields, including the name, type, timestamp, and person in charge of the scientific research results to be processed, etc.

[0133] In the embodiments of the present application, based on the relevant materials of the scientific research results to be processed, the scientific research result management system can use the multi-modal knowledge graph technique to determine the association relationship between the target fields. Based on the target fields and the association relationship between the target fields, the scientific research result management system can use the multi-modal knowledge graph technique to generate a feature graph of the scientific research results to be processed.

[0134] In some embodiments of the present application, the relevant materials include at least one of text materials, image materials, and numerical materials; determining at least one result feature of the scientific research result to be processed based on the relevant materials includes:

[0135] For the text materials, use a preset language processing model to determine the text features of the text materials;

[0136] For the image materials, use a preset cross-modal contrastive learning method to determine the image features of the image materials;

[0137] For the numerical materials, use a dynamic feature decomposition method based on causal discovery to generate the data features of the numerical materials.

[0138] In the embodiments of the present application, the relevant materials of the scientific research result to be processed include at least one of text materials, image materials, and numerical materials. The result features of the scientific research result to be processed may also include at least one of text features, image features, and data features.

[0139] For the text materials of the scientific research result to be processed, the scientific research result management system can use a language processing model to extract the deep semantic vectors of the text materials, that is, extract the complex meanings contained in the text and convert them into vector forms. The text features of the text materials are the complex meanings in vector form. Among them, the language processing model is ChatGPT (Conversational Generative Pre-trained Transformer) or FLAN-T5 (Fine-tuned Language Net with T5).

[0140] For the image materials of the scientific research result to be processed, the cross-modal contrastive learning method (CLIP, Contrastive Language-Image Pre-training) can be used to extract the image features of the image materials.

[0141] For the numerical materials of the scientific research result to be processed, through a dynamic feature decomposition method based on causal discovery, the data features of the data value materials can be generated, that is, statistical description indicators related to the data value materials, such as change rate, seasonal influence, etc.

[0142] In some embodiments of the present application, the using a preset cross-modal contrastive learning method to determine the image features of the image materials includes:

[0143] Using the cross-modal contrastive learning method, train a preset image processing model to obtain an image feature extraction model;

[0144] Use the image feature extraction model to determine the image features of the image-based materials.

[0145] In the embodiments of the present application, by combining the cross-modal contrastive learning method with an image processing model, the image features in the image-based materials can be extracted. The image features are the image semantics and context-related features, referring to the objects, scenes, behaviors, and their mutual relationships in the image. Among them, the image processing model can be a Transformer model. Due to its self-attention mechanism, the Transformer model can effectively capture the long-range dependencies and context information in the image.

[0146] Specifically, the scientific research achievement management system can use the cross-modal contrastive learning method to train the image processing model, improve the model's understanding and representation ability of the image materials of scientific research achievements, and obtain an image feature extraction model. The scientific research achievement management system can use the image feature extraction model to determine the image features of the image-based materials.

[0147] In some embodiments of the present application, the scientific research achievement to be processed includes the current scientific research achievement of the scientific research corresponding to the scientific research achievement to be processed, and the historical scientific research achievements of the scientific research at at least one historical time; the method includes:

[0148] Based on the current achievement features of the current scientific research achievement and the historical achievement features of the historical scientific research achievements, use the preset dynamic spatio-temporal graph network technology to construct a dynamic time series graph of the scientific research achievement to be processed.

[0149] In the embodiments of the present application, the scientific research achievement to be processed includes the current scientific research achievement of the scientific research corresponding to the scientific research achievement to be processed, and the historical scientific research achievements of the scientific research at at least one historical time. The scientific research achievement management system can determine the current achievement features of the current scientific research achievement and the historical achievement features of the historical scientific research achievements.

[0150] In the embodiments of the present application, the scientific research achievement management system uses the dynamic spatio-temporal graph network (DST-GN, Dynamic Spatial-Temporal Graph Network) technology. Based on the current achievement features and the historical achievement features at at least one historical time point, a dynamic time series graph of the scientific research achievement to be processed can be constructed.

[0151] The scientific research achievement management system can also generate a relationship table between achievement features and time, that is, a time dynamic feature table. It should be noted that in the dynamic time series graph and the time dynamic feature table, multiple achievement features corresponding to the same time point are sorted according to their importance order.

[0152] Step 102, for any one of the market demand information, use a preset deep transfer learning method to determine whether there is a correlation between the achievement feature and the market demand information;

[0153] In the embodiments of the present application, the achievement feature and the market demand information can be in text format. The scientific research achievement management system can use the deep transfer learning method to analyze and determine the correlation between the achievement feature of the scientific research achievement to be processed and the market demand information, that is, the cross-domain correlation between the achievement feature and the market demand, and determine whether there is a correlation between the achievement feature of the scientific research achievement to be processed and the market demand information. Among them, the achievement feature can refer to the current achievement feature.

[0154] Step 103, if there is a correlation between the achievement feature and the market demand information, use a preset knowledge graph completion algorithm to generate a target application achievement of the scientific research achievement to be processed that matches the market demand information, and generate a target transformation path between the scientific research achievement to be processed and the target application achievement.

[0155] In the embodiments of the present application, the knowledge graph completion algorithm includes industry information, company information, product information, etc. related to the scientific research achievement to be processed.

[0156] For any one of the market demand information, if the achievement feature of the scientific research achievement is relevant to the market demand information, the scientific research achievement management system can use the knowledge graph completion algorithm to generate a target application achievement of the scientific research achievement to be processed that matches the market demand information, and generate a target transformation path between the scientific research achievement to be processed and the target application achievement. Among them, the target transformation path can be in text format or a target transformation path graph.

[0157] In some embodiments of the present application, the step of using a preset knowledge graph completion algorithm to generate a target application achievement of the scientific research achievement to be processed that matches the market demand information, and generate a target transformation path between the scientific research achievement to be processed and the target application achievement includes:

[0158] Use the knowledge graph completion algorithm to generate an initial transformation path graph between the scientific research achievement to be processed and the target application achievement; at least one transformation step is included in the initial transformation path graph; any one of the transformation steps has at least one object to be applied;

[0159] For any of the above transformation steps, according to the preset preference information of the user, determine a target application object from at least one object to be applied in the transformation step;

[0160] Based on the target application object, adjust the initial transformation path diagram to obtain a target transformation path diagram.

[0161] In the embodiments of the present application, the scientific research achievement management system can generate an initial transformation path diagram between the scientific research achievements to be processed and the target application achievements by using a knowledge graph completion algorithm. Among them, the initial transformation path diagram can be a flowchart, including at least one transformation step for the transformation between the scientific research achievements to be processed and the target application achievements. The transformation steps can include: steps of determining the industry situation related to market demand information, steps of selecting partners, and steps of determining target application achievements that meet market demand information, etc.

[0162] In the embodiments of the present application, any transformation step has at least one object to be applied. For example: in the initial transformation path diagram, there can be at least one potential partner.

[0163] In the embodiments of the present application, for any transformation step in the initial transformation path diagram, the scientific research achievement management system can determine a target application object from at least one object to be applied in the transformation step according to the preset preference information of the user, and then adjust the initial transformation path diagram based on the target application object to obtain a target transformation path diagram.

[0164] In the embodiments of the present application, since the target transformation path diagram can include steps of determining the industry situation related to market demand information, steps of selecting partners, and steps of determining target application achievements that meet market demand information, etc., a transformation suggestion report can be generated based on the target transformation path diagram. The transformation suggestion report can include industry situation analysis, potential partner identification, and technology implementation scenarios, etc.

[0165] In the embodiments of the present application, in the process of adjusting the initial transformation path diagram to obtain a target transformation path diagram, reinforcement learning (RL) technology can also be used to optimize the initial transformation path diagram.

[0166] In some embodiments of the present application, the preset preference information includes: a user demand model and / or the unlabeled demand information of the user; the determining of a target application object from at least one object to be applied in the transformation step according to the preset preference information of the user includes:

[0167] Using a preset self-supervised learning method to generate potential preference features of the unlabeled demand information;

[0168] Using a preset recommendation algorithm based on contrastive learning, match the at least one object to be applied with the user demand model to determine the target application object, and / or,

[0169] Match the at least one object to be applied with the potential preference features to determine the target application object.

[0170] In the embodiments of the present application, the preset preference information of the user may include a user demand model and / or the unlabeled demand information of the user. Among them, the user demand model usually includes a large amount of user preference information. The unlabeled demand information refers to the user demand-related information that has not been manually marked or classified.

[0171] In the embodiments of the present application, the scientific research achievement management system may adopt a recommendation algorithm based on contrastive learning (CLR, Contrastive Learning for Recommendation), match at least one object to be applied in the transformation step with the user demand model to determine the target application object. The scientific research achievement management system may also use the self-supervised learning method to generate potential preference features of the unlabeled demand information, match at least one object to be applied in the transformation step with the potential preference features to determine the target application object, and improve the collaborative filtering effect.

[0172] In some embodiments of the present application, the method includes:

[0173] Use a preset graph neural network to display the determination process of the target application object in the target transformation path graph.

[0174] In the embodiments of the present application, the scientific research achievement management system may input the processes of "matching at least one object to be applied in the transformation step with the user demand model to determine the target application object" and "matching at least one object to be applied in the transformation step with the potential preference features to determine the target application object", as well as the target application object into the graph neural network, and display the determination process of the target application object in the target transformation path graph through the graph neural network (GNN, Graph Neural Network), and display the recommendation result in the form of a dynamic chart to enhance the recommendation interpretability.

[0175] In some embodiments of the present application, relevant materials of at least one award-winning scientific research achievement are stored in the scientific research achievement management system; the method includes:

[0176] Generate a reward recommendation for the scientific research achievement to be processed according to the relevant materials of the award-winning scientific research achievement;

[0177] Using a preset document generation model, convert the award recommendation and the generation process of the award recommendation into an award recommendation document for the scientific research achievement to be processed.

[0178] In the embodiment of the present application, relevant materials of at least one award-winning scientific research achievement are stored in the scientific research achievement management system, and the relevant materials may include the award-winning conditions, evaluation criteria of the award-winning scientific research achievement, and the achievement characteristics of the award-winning scientific research achievement.

[0179] In the embodiment of the present application, according to the relevant materials of the award-winning scientific research achievement, the scientific research achievement to be processed can be evaluated to generate an award recommendation for the scientific research achievement to be processed.

[0180] In the embodiment of the present application, the Transformer model can be used as the document generation model. The scientific research achievement management system can convert the award recommendation of the scientific research achievement to be processed and the process of "generating the award recommendation of the scientific research achievement to be processed according to the relevant materials of the award-winning scientific research achievement" into a personalized award recommendation document through the Transformer model. In the embodiment of the present application, an automated document generation tool and natural language generation technology (NLG, Natural Language Generation) can also be used to optimize the personalized award recommendation document to complete high-quality document output.

[0181] In some embodiments of the present application, the relevant materials of the award-winning scientific research achievement include the historical award-winning conditions of the award-winning scientific research achievement; the generating the award recommendation of the scientific research achievement to be processed according to the relevant materials of the award-winning scientific research achievement includes:

[0182] Using a preset neural network model based on time series to extract the long-term dependence features of the historical award-winning conditions;

[0183] Using a preset semantic similarity evaluation model to determine the first matching degree between the achievement characteristics of the scientific research achievement to be processed and the long-term dependence features.

[0184] In the embodiment of the present application, the relevant materials of the award-winning scientific research achievement include the historical award-winning conditions of the award-winning scientific research achievement. In the embodiment of the present application, the scientific research achievement management system can use a neural network model based on time series (TCN, Temporal Convolutional Network) to extract the long-term dependence features of the historical award-winning conditions. The scientific research achievement management system can also calculate the first matching degree between the achievement characteristics of the scientific research achievement to be processed and the long-term dependence features of the historical award-winning conditions based on the semantic similarity of BERT. Among them, BERT is the semantic similarity evaluation model.

[0185] In some embodiments of the present application, the relevant materials of the award-winning scientific research achievements include the achievement characteristics of the award-winning scientific research achievements; generating a reward recommendation for the scientific research achievements to be processed according to the relevant materials of the award-winning scientific research achievements includes:

[0186] Using a preset sequence-to-sequence generation model to analyze the achievement characteristics of the award-winning scientific research achievements and determine the common characteristics of the award-winning scientific research achievements;

[0187] Using the semantic similarity evaluation model to determine the second matching degree between the achievement characteristics of the scientific research achievements to be processed and the common characteristics.

[0188] In the embodiments of the present application, the relevant materials of the award-winning scientific research achievements may also include the achievement characteristics of the award-winning scientific research achievements. In the embodiments of the present application, a sequence-to-sequence generation model (Seq2Seq) is used to analyze the achievement characteristics of the award-winning scientific research achievements to determine the common characteristics of the award-winning scientific research achievements, that is, the common characteristics. The scientific research achievement management system may also calculate the second matching degree between the achievement characteristics of the scientific research achievements to be processed and the common characteristics of the award-winning scientific research achievements based on the semantic similarity of BERT.

[0189] In some embodiments of the present application, generating a reward recommendation for the scientific research achievements to be processed according to the relevant materials of the award-winning scientific research achievements includes:

[0190] Based on the first matching degree and the second matching degree, determine the award level of the scientific research achievements to be processed.

[0191] In the embodiments of the present application, the first preset weight of the first matching degree and the second preset weight of the second matching degree may be used to perform weighted summation of the first matching degree and the second matching degree according to their respective preset weights to determine the target matching degree.

[0192] In the embodiments of the present application, the scientific research achievement management system may introduce a matching evaluation method based on multi-objective optimization (Multi-objective Optimization) to generate a level classification for the reward recommendation. Specifically, the award level of the scientific research achievements to be processed may be determined according to the target matching degree. In a specific example, if the target matching degree is greater than the preset matching degree threshold, the award level is high; if the target matching degree is not greater than the preset matching degree threshold, the award level is low.

[0193] In some embodiments of the present application, the method includes:

[0194] Based on the relevant materials of at least one of the scientific research achievements to be processed and the dynamic time series graph, using a preset hash coding technology to evaluate the similarity between at least one of the scientific research achievements to be processed.

[0195] In the embodiments of the present application, the scientific research achievement management system can evaluate the similarity between the scientific research achievements to be processed and determine whether the content of the scientific research achievements to be processed is consistent by using the hash coding technology (MinHash) based on the relevant materials of at least one scientific research achievement to be processed and the dynamic time series diagram of the scientific research achievement to be processed.

[0196] In some embodiments of the present application, the method includes:

[0197] Determining the completeness score of the relevant materials by using a preset scoring mechanism based on Bayesian optimization;

[0198] If the completeness score is lower than a preset score threshold, using a preset generative adversarial network to generate filling fields corresponding to the missing fields in the relevant materials;

[0199] Using the filling fields to improve the completeness score of the scientific research achievements to be processed.

[0200] In the embodiments of the present application, the scientific research achievement management system can use a scoring mechanism based on Bayesian optimization to determine the completeness score of the relevant materials. If the completeness score is lower than a preset score threshold, it uses a generative adversarial network (GAN) to generate filling fields corresponding to the missing fields in the relevant materials; and uses the filling fields to improve the completeness score of the scientific research achievements to be processed.

[0201] In the embodiments of the present application, there may be inconsistencies in the relevant materials of the scientific research achievements to be processed. Inconsistencies refer to format inconsistencies and fuzzy matching problems existing in the relevant materials of the scientific research achievements to be processed. The fuzzy matching problem refers to the problem that there are differences between two fields but they are actually the same field. In the embodiments of the present application, the scientific research achievement management system can optimize the data repair strategy through semantic correction and reinforcement learning methods to automatically process format deviations and fuzzy matching problems.

[0202] In the embodiments of the present application, the scientific research achievement management system can obtain the relevant materials of the scientific research achievements to be processed and at least one piece of market demand information, and determine at least one achievement feature of the scientific research achievements to be processed based on the relevant materials; for any piece of market demand information, use a preset deep transfer learning method to determine whether the achievement feature is relevant to the market demand information; if the achievement feature is relevant to the market demand information, use a preset knowledge graph completion algorithm to generate the target application achievements of the scientific research achievements to be processed that match the market demand information, and generate the target transformation path between the scientific research achievements to be processed and the target application achievements, and optimize the excavation of the application value of scientific and technological achievements through achievement transformation path analysis and personalized recommendation functions, providing users with comprehensive achievement management support.

[0203] In the embodiments of the present application, dynamic feature extraction is achieved through a time series embedding model, which significantly reduces the time of traditional manual screening and matching, realizes the rapid analysis of scientific and technological achievement features and historical conditions, and has a fast response speed. The achievement feature extraction method based on the BERT model and the time series evolution network significantly improves the accuracy of achievement consistency detection and award application condition matching. The AI (Artificial Intelligence) - driven dynamic recommendation algorithm ensures the scientificity and reliability of award application recommendations, and the award application recommendation accuracy is high. The scientific research achievement management system integrates full - process functions such as data pre - processing, dynamic feature extraction, achievement transformation management, and award application recommendation. Users can complete complex operations through the visual interface without a deep technical background. The scientific research achievement management system supports multi - modal data processing such as text, images, and numerical values, can adapt to different fields and various types of scientific and technological achievements, and is widely applicable to the achievement management and recommendation scenarios of universities, enterprises, and scientific research institutions. Through the matching analysis of historical award conditions and current achievements, combined with intelligent award application suggestions, the scientific research achievement management system improves the scientificity and success rate of award application decisions; through automated achievement consistency checks, transformation process management, and award application material generation, it significantly reduces manual intervention and repetitive labor, improves the overall efficiency, and solves problems such as incomplete and inconsistent materials in the process of managing pending scientific research achievements, and the low efficiency of checking the novelty and creativity of pending scientific research achievements. The scientific research achievement management system also integrates functions such as intelligent reminder, process visualization, and path optimization, enabling users to quickly understand the current situation and improvement direction of achievements, and enhancing the operation experience and job satisfaction.

[0204] In the embodiments of the present application, through the method of dynamic extraction and analysis of achievement features based on time series embedding, the intelligent level of managing and recommending pending scientific research achievements is improved. The present application realizes the accurate extraction and dynamic analysis of the features of pending scientific research achievements, supports the detection of the completeness and consistency of achievement materials, and significantly improves the efficiency and accuracy of achievement duplication checking. In addition, by introducing artificial intelligence technology, the present application optimizes the management process of the transformation of pending scientific research achievements, provides personalized suggestions for the achievement transformation path, and effectively promotes the market application of pending scientific research achievements. At the same time, the present application constructs an award application condition analysis model based on historical award data, supports the award application recommendation of pending scientific research achievements and the optimization of the award process, and improves the scientificity and accuracy of achievement recommendation. The present application improves the efficiency of managing and analyzing pending scientific research achievements, strengthens the intelligent capabilities of achievement transformation and recommendation, and provides strong technical support for the full - process management of scientific and technological innovation.

[0205] In the embodiments of the present application, a complete dynamic management and recommendation system for scientific research results to be processed is constructed by combining time series analysis and artificial intelligence technology. First, the present application uses the graph extraction time series evolution network technology to perform hierarchical and multi-dimensional dynamic analysis on the characteristics of scientific research results to be processed, capture the key nodes of the results' evolution over time, and thus generate time series feature representations. Subsequently, based on a large-scale time embedding model, the variation law of the result characteristics in the time dimension is further extracted to provide high-quality data support for subsequent analysis and recommendation.

[0206] In the feature extraction stage, the present application supports multiple methods, including text-based feature extraction, statistical feature analysis, and BERT-based semantic feature extraction, to ensure the comprehensiveness and accuracy of the result feature representation. For the completeness and consistency of the scientific research result materials to be processed, the present application designs an automatic detection and correction mechanism to effectively reduce data deviation caused by human intervention. For result duplicate checking, the present application combines time series feature similarity analysis and a dynamic update mechanism, greatly improving the duplicate checking efficiency and accuracy.

[0207] In terms of the transformation management of scientific research results to be processed, the present application combines an artificial intelligence model to construct a module for optimizing the result transformation path and recommending market-oriented applications, generating personalized transformation suggestions based on result characteristics and market demands. At the same time, to support the recommendation of scientific research results to be processed for awards, the present application designs a condition matching model by analyzing historical award data, which can generate accurate award recommendation suggestions according to the result characteristics and award conditions.

[0208] Through this systematic design idea, the present application realizes the intelligent upgrade of the entire process of managing, transforming, and recommending scientific research results to be processed, which is both innovative and ensures the feasibility and scalability of practical applications. It solves the problems that due to the diverse types, complex characteristics, and frequent updates of scientific research results to be processed, the existing management methods cannot efficiently meet the needs of dynamic analysis and intelligent recommendation of results, and the traditional result management system lacks the ability to dynamically extract and analyze result characteristics, making it difficult to capture the changes in characteristics during the development process of results, and the recommendation methods based on simple rules or static matching cannot fully utilize the historical data characteristics, resulting in low scientificity and success rate of award recommendation results, as well as the lack of artificial intelligence support for the result transformation support and recommendation functions, making it difficult to achieve automated processing and intelligent decision-making.

[0209] Refer to Figure 2 , which shows the step flowchart of another method for processing scientific research results provided in the embodiments of the present application, and specifically may include the following steps:

[0210] Step 201, data preprocessing.

[0211] Data standardization, initial feature screening, and timeline construction are performed on the relevant materials of the scientific research results to be processed. Among them, data standardization refers to automatically cleaning the data, repairing missing values, and unifying the format by using regularization techniques and semantic error correction algorithms based on language models. Initial feature screening refers to extracting key fields from the relevant materials through a multimodal knowledge graph, constructing associations between key fields, and constructing a feature graph. Timeline construction refers to constructing a dynamic time series graph by using dynamic spatio-temporal graph network technology.

[0212] Step 202, dynamic feature extraction.

[0213] Based on the TFT (Temporal Fusion Transformer) model with a temporal attention mechanism, different time points and the corresponding result features in the dynamic time series graph of the scientific research results to be processed are encoded, and the time points and the corresponding result features are converted into a specific format that the model can understand and operate on.

[0214] The dynamic time series graph of the scientific research results to be processed has at least one time feature, where the time feature refers to the result features corresponding to different time points. The scientific research result management system can use Time2Vec to map at least one time feature in the dynamic time series graph of the result features to a high-dimensional space vector. Among them, Time2Vec is an improved time enhancement embedding technology. Through the TFT model with a temporal attention mechanism and Time2Vec, the modeling ability for complex time-dependent relationships can be improved.

[0215] For text materials, ChatGPT or FLAN-T5 is used to extract deep semantic vectors from the text. For image materials, a cross-modal contrast learning method combined with Transformer is used to extract image semantic and context-related features. For numerical materials, statistical description indicators (such as change rate, seasonal influence) are generated through a dynamic feature decomposition method based on causal discovery. The multi-modal features of the extracted text materials, image materials, and numerical materials are integrated to generate a time dynamic feature table, and the representation structure is optimized through feature importance ranking.

[0216] Step 203, result consistency and completeness detection.

[0217] Based on the achievement characteristics and relevant materials of the scientific research achievements to be processed, quickly detect the similarity of achievements and evaluate the content consistency based on hash coding technology. For the relevant materials of the scientific research achievements to be processed, automatically calculate the completeness based on the scoring mechanism optimized by Bayesian optimization, and use generative adversarial network technology to generate missing fields. For the relevant materials of the scientific research achievements to be processed, optimize the data repair strategy through semantic correction and reinforcement learning methods, and automatically process format deviation and fuzzy matching problems. Step 201, step 202, and step 203 involve the extraction and management of the dynamic characteristics of scientific and technological achievements by the scientific research achievement management system.

[0218] Step 204, analysis of the achievement transformation path.

[0219] Using deep transfer learning technology, mine the cross-domain correlation between the achievement characteristics of the scientific research achievements to be processed and the market demand information, and determine whether they are relevant. If relevant, use the knowledge graph completion algorithm to generate the transformation path between the scientific research achievements to be processed and the target application achievements that meet the market demand information, and combine the reinforcement learning optimization technology to implement the plan. Based on the transformation path, automatically generate a transformation recommendation report, including target industry analysis, identification of potential partners, and technology implementation scenarios.

[0220] Step 205, personalized recommendation.

[0221] Adopt a recommendation algorithm based on contrastive learning, match the transformation path with the user demand model, determine the target application object among at least one of the application objects of the transformation steps of the transformation path, and optimize the transformation path.

[0222] Use self-supervised learning methods to generate potential preference features of unlabeled demand data, match the transformation path with the potential preference features, determine the target application object among at least one of the application objects of the transformation steps of the transformation path, optimize the transformation path, and improve the collaborative filtering effect.

[0223] In the target transformation path graph through the graph neural network, display the optimization process of the transformation path. Step 204 and step 205 involve the transformation of scientific and technological achievements and market-oriented recommendations.

[0224] Step 206, historical data learning.

[0225] Use a preset neural network model based on time series to extract the long-term dependence features of historical award-winning conditions;

[0226] Use a sequence-to-sequence generation model to analyze the achievement characteristics of award-winning scientific research achievements and determine the common characteristics of award-winning scientific research achievements.

[0227] Step 207, matching of award application conditions.

[0228] Using the BERT-based semantic similarity, determine the first matching degree between the achievement features of the scientific research achievement to be processed and the long-term dependence features, and determine the second matching degree between the achievement features of the scientific research achievement to be processed and the common features. Weighted sum the first matching degree and the second matching degree according to their respective preset weights to determine the target matching degree, and introduce a matching evaluation method based on multi-objective optimization to generate a hierarchical classification for the award recommendation.

[0229] Step 208, award recommendation output.

[0230] Generate a personalized award recommendation document through the Transformer model, including the execution processes of steps 206 and 207. Use an automated document generation tool and natural language generation technology (Natural Language Generation, NLG) to optimize the personalized award recommendation document to complete high-quality document output. Steps 206, 207, and 208 are related to the award recommendation of scientific and technological achievements.

[0231] It should be noted that for the method embodiments, for the sake of simple description, they are all expressed as a series of action combinations. However, those skilled in the art should know that the embodiments of the present application are not limited by the described action sequence, because according to the embodiments of the present application, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions involved are not necessarily essential for the embodiments of the present application.

[0232] Refer to Figure 3 , which shows the structural block diagram of a processing device for scientific research achievements provided in the embodiments of the present application, applied to a scientific research achievement management system, and specifically may include the following modules:

[0233] A data acquisition module 301, configured to acquire relevant data of the scientific research achievement to be processed and at least one piece of market demand information, and determine at least one achievement feature of the scientific research achievement to be processed based on the relevant data;

[0234] A judgment module 302, configured to use a preset deep transfer learning method to judge whether there is a correlation between the achievement feature and the market demand information for any one of the market demand information;

[0235] A path generation module 303, configured to, if there is a correlation between the achievement feature and the market demand information, generate a target application achievement of the scientific research achievement to be processed that matches the market demand information by using a preset knowledge graph completion algorithm, and generate a target conversion path between the scientific research achievement to be processed and the target application achievement.

[0236] In an alternative embodiment of the present application, the data acquisition module includes:

[0237] A target field extraction sub-module, configured to extract target fields from the relevant data by using a preset multi-modal knowledge graph technology;

[0238] An association relationship determination sub-module, configured to determine the association relationship between the target fields based on the relevant data by using the multi-modal knowledge graph technology;

[0239] A feature graph generation sub-module, configured to generate a feature graph of the scientific research result to be processed based on the target fields and the association relationship between the target fields by using the multi-modal knowledge graph technology.

[0240] In an alternative embodiment of the present application, the relevant data includes at least one of text data, image data, and numerical data; the data acquisition module includes:

[0241] A text feature determination sub-module, configured to determine the text features of the text data by using a preset language processing model for the text data;

[0242] An image feature determination sub-module, configured to determine the image features of the image data by using a preset cross-modal contrast learning method for the image data;

[0243] A data feature determination sub-module, configured to generate the data features of the numerical data by using a dynamic feature decomposition method based on causal discovery for the numerical data.

[0244] In an alternative embodiment of the present application, the scientific research result to be processed includes the current scientific research result of the scientific research corresponding to the scientific research result to be processed, and the historical scientific research results of the scientific research at at least one historical time; the device includes:

[0245] A dynamic time series graph construction module, configured to construct a dynamic time series graph of the scientific research result to be processed based on the current result features of the current scientific research result and the historical result features of the historical scientific research results by using a preset dynamic spatio-temporal graph network technology.

[0246] In an alternative embodiment of the present application, the path generation module includes:

[0247] An initial transformation path graph generation sub-module, configured to generate an initial transformation path graph between the scientific research result to be processed and the target application result by using the knowledge graph completion algorithm; at least one transformation step is included in the initial transformation path graph; any one of the transformation steps has at least one object to be applied;

[0248] A target application object determination sub-module, configured to, for any one of the conversion steps, determine a target application object from at least one to-be-applied object of the conversion step according to the preset preference information of the user;

[0249] A target conversion path diagram obtaining sub-module, configured to adjust the initial conversion path diagram based on the target application object to obtain a target conversion path diagram.

[0250] In an alternative embodiment of the present application, the preset preference information includes: a user demand model and / or the unannotated demand information of the user; the target application object determination sub-module includes:

[0251] A feature generation unit, configured to generate potential preference features of the unannotated demand information by using a preset self-supervised learning method;

[0252] A matching unit, configured to use a preset recommendation algorithm based on contrastive learning to match the at least one to-be-applied object with the user demand model to determine the target application object, and / or,

[0253] match the at least one to-be-applied object with the potential preference features to determine the target application object.

[0254] In an alternative embodiment of the present application, relevant materials of at least one award-winning scientific research achievement are stored in the scientific research achievement management system; the device includes:

[0255] An award recommendation generation module, configured to generate an award recommendation for the to-be-processed scientific research achievement according to the relevant materials of the award-winning scientific research achievement;

[0256] A document conversion module, configured to use a preset document generation model to convert the award recommendation and the generation process of the award recommendation into an award recommendation document for the to-be-processed scientific research achievement.

[0257] In an alternative embodiment of the present application, the relevant materials of the award-winning scientific research achievement include the historical award-winning conditions of the award-winning scientific research achievement; the award recommendation generation module includes:

[0258] A feature extraction sub-module, configured to extract long-term dependence features of the historical award-winning conditions by using a preset neural network model based on time series;

[0259] A first matching degree determination sub-module, configured to use a preset semantic similarity evaluation model to determine a first matching degree between the achievement features of the to-be-processed scientific research achievement and the long-term dependence features.

[0260] In an alternative embodiment of the present application, the relevant materials of the awarded scientific research achievements include the achievement characteristics of the awarded scientific research achievements; the award recommendation generation module includes:

[0261] A common feature determination sub-module, configured to analyze the achievement characteristics of the awarded scientific research achievements by using a preset sequence-to-sequence generation model, and determine the common features of the awarded scientific research achievements;

[0262] A second matching degree determination sub-module, configured to determine the second matching degree between the achievement characteristics of the scientific research achievement to be processed and the common features by using the semantic similarity evaluation model.

[0263] In an alternative embodiment of the present application, the award recommendation generation module includes:

[0264] An award level determination sub-module, configured to determine the award level of the scientific research achievement to be processed based on the first matching degree and the second matching degree.

[0265] In an alternative embodiment of the present application, the device includes:

[0266] A display module, configured to display the determination process of the target application object in the target transformation path diagram by using a preset graph neural network.

[0267] In an alternative embodiment of the present application, the device includes:

[0268] An evaluation module, configured to evaluate the similarity between at least one of the scientific research achievements to be processed based on the relevant materials and the dynamic time series diagram of at least one of the scientific research achievements to be processed by using a preset hash coding technology.

[0269] In an alternative embodiment of the present application, the image feature determination sub-module includes:

[0270] A model obtaining unit, configured to train a preset image processing model by using the cross-modal contrast learning method to obtain an image feature extraction model;

[0271] An image feature determination unit, configured to determine the image features of the image type materials by using the image feature extraction model.

[0272] In an alternative embodiment of the present application, the device includes:

[0273] A filling module, configured to detect missing fields in the relevant materials by using a preset regularization technology and / or a preset semantic error correction algorithm based on a language model, and fill the missing fields.

[0274] In an alternative embodiment of the present application, the device includes:

[0275] A scoring determination module, configured to determine the completeness score of the relevant materials by using a preset scoring mechanism based on Bayesian optimization;

[0276] A field generation module, configured to, if the completeness score is lower than a preset score threshold, generate a filling field corresponding to a missing field in the relevant materials by using a preset generative adversarial network;

[0277] A score improvement module, configured to use the filling field to improve the completeness score of the scientific research result to be processed.

[0278] For the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and for the related parts, please refer to the partial description of the method embodiment.

[0279] In addition, an embodiment of the present application further provides an electronic device, as Figure 4 shown, including a processor 401, a communication interface 402, a memory 403, and a communication bus 404. Among them, the processor 401, the communication interface 402, and the memory 403 complete mutual communication through the communication bus 404.

[0280] The memory 403 is used to store a computer program;

[0281] The processor 401, when executing the program stored on the memory 403, implements the following steps:

[0282] Obtain relevant materials of the scientific research result to be processed and at least one piece of market demand information, and based on the relevant materials, determine at least one result feature of the scientific research result to be processed;

[0283] For any one of the market demand information, use a preset deep transfer learning method to determine whether there is a correlation between the result feature and the market demand information;

[0284] If there is a correlation between the result feature and the market demand information, use a preset knowledge graph completion algorithm to generate a target application result of the scientific research result to be processed that matches the market demand information, and generate a target conversion path between the scientific research result to be processed and the target application result.

[0285] In an alternative embodiment of the present application, the determining at least one result feature of the scientific research result to be processed based on the relevant materials includes:

[0286] Use a preset multi-modal knowledge graph technology to extract target fields in the relevant materials;

[0287] Based on the relevant materials, use the multi-modal knowledge graph technology to determine the association relationship between the target fields;

[0288] Based on the target fields and the association relationships between the target fields, use the multi-modal knowledge graph technology to generate a feature graph of the scientific research result to be processed.

[0289] In an alternative embodiment of the present application, the relevant materials include at least one of text materials, image materials, and numerical materials; determining at least one result feature of the scientific research result to be processed based on the relevant materials includes:

[0290] For the text materials, use a preset language processing model to determine the text features of the text materials;

[0291] For the image materials, use a preset cross-modal contrast learning method to determine the image features of the image materials;

[0292] For the numerical materials, use a dynamic feature decomposition method based on causal discovery to generate data features of the numerical materials.

[0293] In an alternative embodiment of the present application, the scientific research result to be processed includes the current scientific research result corresponding to the scientific research result to be processed, and the historical scientific research results of the scientific research at at least one historical time; the method includes:

[0294] Based on the current result features of the current scientific research result and the historical result features of the historical scientific research results, use a preset dynamic spatio-temporal graph network technology to construct a dynamic time series graph of the scientific research result to be processed.

[0295] In an alternative embodiment of the present application, using a preset knowledge graph completion algorithm to generate a target application result of the scientific research result to be processed that matches the market demand information, and generating a target transformation path between the scientific research result to be processed and the target application result includes:

[0296] Use the knowledge graph completion algorithm to generate an initial transformation path graph between the scientific research result to be processed and the target application result; at least one transformation step is included in the initial transformation path graph; any one of the transformation steps has at least one object to be applied;

[0297] For any one of the transformation steps, according to the preset preference information of the user, determine a target application object from at least one object to be applied in the transformation step;

[0298] Based on the target application object, adjust the initial transformation path graph to obtain a target transformation path graph.

[0299] In an alternative embodiment of the present application, the preset preference information includes: a user demand model and / or the unlabeled demand information of the user; determining a target application object from at least one object to be applied in the conversion step according to the preset preference information of the user includes:

[0300] Using a preset self-supervised learning method to generate potential preference features of the unlabeled demand information;

[0301] Using a preset recommendation algorithm based on contrast learning to match the at least one object to be applied with the user demand model to determine the target application object, and / or,

[0302] Matching the at least one object to be applied with the potential preference features to determine the target application object.

[0303] In an alternative embodiment of the present application, relevant materials of at least one award-winning scientific research achievement are stored in the scientific research achievement management system; the method includes:

[0304] Generating a reward recommendation for the scientific research achievement to be processed according to the relevant materials of the award-winning scientific research achievement;

[0305] Using a preset document generation model to convert the reward recommendation and the generation process of the reward recommendation into a reward recommendation document for the scientific research achievement to be processed.

[0306] In an alternative embodiment of the present application, the relevant materials of the award-winning scientific research achievement include the historical award-winning conditions of the award-winning scientific research achievement; generating a reward recommendation for the scientific research achievement to be processed according to the relevant materials of the award-winning scientific research achievement includes:

[0307] Using a preset neural network model based on time series to extract long-term dependence features of the historical award-winning conditions;

[0308] Using a preset semantic similarity evaluation model to determine the first matching degree between the achievement features of the scientific research achievement to be processed and the long-term dependence features.

[0309] In an alternative embodiment of the present application, the relevant materials of the award-winning scientific research achievement include the achievement features of the award-winning scientific research achievement; generating a reward recommendation for the scientific research achievement to be processed according to the relevant materials of the award-winning scientific research achievement includes:

[0310] Using a preset sequence-to-sequence generation model to analyze the achievement features of the award-winning scientific research achievement to determine the common features of the award-winning scientific research achievement;

[0311] Using the semantic similarity evaluation model, determine the second matching degree between the result features of the scientific research result to be processed and the common features.

[0312] In an alternative embodiment of the present application, generating a reward recommendation for the scientific research result to be processed according to the relevant materials of the awarded scientific research result includes:

[0313] Based on the first matching degree and the second matching degree, determine the award level of the scientific research result to be processed.

[0314] In an alternative embodiment of the present application, the method includes:

[0315] Using a preset graph neural network, display the determination process of the target application object in the target transformation path graph.

[0316] In an alternative embodiment of the present application, the method includes:

[0317] Based on the relevant materials and dynamic time series graphs of at least one scientific research result to be processed, use a preset hash coding technique to evaluate the similarity between at least one scientific research result to be processed.

[0318] In an alternative embodiment of the present application, the using a preset cross-modal contrast learning method to determine the image features of the image-type materials includes:

[0319] Using the cross-modal contrast learning method to train a preset image processing model to obtain an image feature extraction model;

[0320] Using the image feature extraction model to determine the image features of the image-type materials.

[0321] In an alternative embodiment of the present application, the method includes:

[0322] Using a preset regularization technique and / or a preset semantic error correction algorithm based on a language model to detect missing fields in the relevant materials and fill in the missing fields.

[0323] In an alternative embodiment of the present application, the method includes:

[0324] Using a preset scoring mechanism based on Bayesian optimization to determine the completeness score of the relevant materials;

[0325] If the completeness score is lower than a preset score threshold, use a preset generative adversarial network to generate filling fields corresponding to the missing fields in the relevant materials;

[0326] Using the filling fields to improve the completeness score of the scientific research result to be processed.

[0327] The communication bus mentioned in the above terminal may be a Peripheral Component Interconnect (PCI) bus, an Extended Industry Standard Architecture (EISA) bus, or the like. The communication bus can be divided into an address bus, a data bus, a control bus, etc. For the sake of convenience of representation, only a thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0328] The communication interface is used for communication between the above terminal and other devices.

[0329] The memory may include a Random Access Memory (RAM), or may also include a non-volatile memory, such as at least one disk memory. Optionally, the memory may also be at least one storage device located far from the aforementioned processor.

[0330] The above-mentioned processor may be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it may also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0331] As Figure 5 shown, in another embodiment provided by the present application, a computer-readable storage medium 501 is further provided. Instructions are stored in the computer-readable storage medium. When it runs on a computer, it causes the computer to execute a processing method for a scientific research achievement described in the above embodiment.

[0332] In another embodiment provided by the present application, a computer program product containing instructions is further provided. When it runs on a computer, it causes the computer to execute a processing method for a scientific research achievement described in the above embodiment.

[0333] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center by wire (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that includes one or more integrated available media. The available medium can be a magnetic medium (such as a floppy disk, hard disk, magnetic tape), an optical medium (such as a DVD), or a semiconductor medium (such as a solid state disk (SSD)).

[0334] It should be noted that, in this document, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including", or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article, or device including a series of elements not only includes those elements but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article, or device. Without further limitation, an element defined by the statement "including a..." does not exclude the existence of additional identical elements in the process, method, article, or device including the element.

[0335] Each embodiment in this specification is described in a related manner. The same or similar parts among the embodiments can be referred to each other, and the differences between each embodiment and other embodiments are emphasized. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the description of the method embodiment.

[0336] The above are only the preferred embodiments of the present application and are not intended to limit the protection scope of the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present application are all included in the protection scope of the present application.

Claims

1. A method for processing scientific research achievements, characterized in that, Applied to a scientific research achievement management system, the method includes: Obtain relevant materials of the scientific research achievement to be processed and at least one piece of market demand information, and based on the relevant materials, determine at least one achievement feature of the scientific research achievement to be processed; For any one of the market demand information, use a preset deep transfer learning method to determine whether there is a correlation between the achievement feature and the market demand information; If there is a correlation between the achievement feature and the market demand information, use a preset knowledge graph completion algorithm to generate a target application achievement of the scientific research achievement to be processed that matches the market demand information, and generate a target transformation path between the scientific research achievement to be processed and the target application achievement.

2. The method according to claim 1, characterized in that, The determining at least one achievement feature of the scientific research achievement to be processed based on the relevant materials includes: Use a preset multimodal knowledge graph technology to extract target fields from the relevant materials; Based on the relevant materials, use the multimodal knowledge graph technology to determine the association relationship between the target fields; Based on the target fields and the association relationship between the target fields, use the multimodal knowledge graph technology to generate a feature graph of the scientific research achievement to be processed.

3. The method according to claim 1, characterized in that, The relevant materials include at least one of text materials, image materials, and numerical materials; the determining at least one achievement feature of the scientific research achievement to be processed based on the relevant materials includes: For the text materials, use a preset language processing model to determine the text features of the text materials; For the image materials, use a preset cross-modal contrast learning method to determine the image features of the image materials; For the numerical materials, use a dynamic feature decomposition method based on causal discovery to generate data features of the numerical materials.

4. The method according to any one of claims 1-3, characterized in that, The scientific research achievement to be processed includes the current scientific research achievement of the scientific research corresponding to the scientific research achievement to be processed, and the historical scientific research achievements of the scientific research at at least one historical time; the method includes: Based on the current achievement features of the current scientific research achievement and the historical achievement features of the historical scientific research achievements, use a preset dynamic spatio-temporal graph network technology to construct a dynamic time series graph of the scientific research achievement to be processed.

5. The method according to claim 1, characterized in that, The using a preset knowledge graph completion algorithm to generate a target application achievement of the scientific research achievement to be processed that matches the market demand information, and generate a target transformation path between the scientific research achievement to be processed and the target application achievement includes: Use the knowledge graph completion algorithm to generate an initial transformation path graph between the scientific research achievement to be processed and the target application achievement; at least one transformation step is included in the initial transformation path graph; any one of the transformation steps has at least one object to be applied; For any one of the transformation steps, determine a target application object from at least one object to be applied in the transformation step according to the preset preference information of the user; Based on the target application object, adjust the initial transformation path graph to obtain a target transformation path graph.

6. The method according to claim 5, characterized in that, The preset preference information includes: a user demand model and / or the unlabeled demand information of the user; determining a target application object from at least one object to be applied in the conversion step according to the preset preference information of the user includes: Generating potential preference features of the unlabeled demand information by using a preset self-supervised learning method; Using a preset recommendation algorithm based on contrast learning to match the at least one object to be applied with the user demand model to determine the target application object, and / or, Matching the at least one object to be applied with the potential preference features to determine the target application object.

7. The method according to claim 1, wherein Relevant materials of at least one award-winning scientific research achievement are stored in the scientific research achievement management system; the method includes: Generating a reward recommendation for the scientific research achievement to be processed according to the relevant materials of the award-winning scientific research achievement; Using a preset document generation model to convert the reward recommendation and the generation process of the reward recommendation into a reward recommendation document for the scientific research achievement to be processed.

8. The method according to claim 7, wherein The relevant materials of the award-winning scientific research achievement include the historical award-winning conditions of the award-winning scientific research achievement; generating a reward recommendation for the scientific research achievement to be processed according to the relevant materials of the award-winning scientific research achievement includes: Extracting long-term dependence features of the historical award-winning conditions by using a preset neural network model based on time series; Using a preset semantic similarity evaluation model to determine a first matching degree between the achievement features of the scientific research achievement to be processed and the long-term dependence features.

9. The method according to claim 8, characterized in that, The relevant materials of the award-winning scientific research achievement include the achievement features of the award-winning scientific research achievement; generating a reward recommendation for the scientific research achievement to be processed according to the relevant materials of the award-winning scientific research achievement includes: Analyzing the achievement features of the award-winning scientific research achievement by using a preset sequence-to-sequence generation model to determine the common features of the award-winning scientific research achievement; Using the semantic similarity evaluation model to determine a second matching degree between the achievement features of the scientific research achievement to be processed and the common features.

10. The method according to claim 9, wherein Generating a reward recommendation for the scientific research achievement to be processed according to the relevant materials of the award-winning scientific research achievement includes: Determining the reward level of the scientific research achievement to be processed based on the first matching degree and the second matching degree.

11. The method according to claim 5, characterized in that, The method includes: Using a preset graph neural network to display the determination process of the target application object in the target conversion path graph.

12. The method according to claim 4, wherein The method includes: Evaluating the similarity between at least one scientific research achievement to be processed based on the relevant materials of at least one scientific research achievement to be processed and a dynamic time series graph by using a preset hash coding technique.

13. The method according to claim 3, characterized in that, The method of using a preset cross-modal contrast learning method to determine the image features of the image-type materials includes: Training a preset image processing model by using the cross-modal contrast learning method to obtain an image feature extraction model; Determining the image features of the image-type materials by using the image feature extraction model.

14. The method according to claim 1, wherein The method includes: Detecting missing fields in the relevant materials by using a preset regularization technique and / or a preset semantic error correction algorithm based on a language model, and filling the missing fields.

15. The method according to claim 1, characterized in that, The method includes: Determine the completeness score of the relevant materials by using a preset scoring mechanism based on Bayesian optimization; If the completeness score is lower than a preset score threshold, use a preset generative adversarial network to generate filling fields corresponding to the missing fields in the relevant materials; Use the filling fields to improve the completeness score of the scientific research results to be processed.

16. A processing device for scientific research achievements, characterized in that, Applied to a scientific research result management system, the device includes: A data acquisition module, configured to acquire relevant materials of the scientific research results to be processed and at least one piece of market demand information, and determine at least one result feature of the scientific research results to be processed based on the relevant materials; A judgment module, configured to, for any one of the market demand information, use a preset deep transfer learning method to judge whether there is a correlation between the result feature and the market demand information; A path generation module, configured to, if there is a correlation between the result feature and the market demand information, use a preset knowledge graph completion algorithm to generate a target application result of the scientific research results to be processed that matches the market demand information, and generate a target conversion path between the scientific research results to be processed and the target application result.

17. An electronic device, characterized in that, It includes a processor, a communication interface, a memory, and a communication bus. Among them, the processor, the communication interface, and the memory complete communication with each other through the communication bus; The memory is used to store a computer program; When the processor is used to execute the program stored on the memory, it implements the method according to any one of claims 1-15.

18. One or more computer-readable media, on which instructions are stored, and when executed by one or more processors, cause the processors to execute the method according to any one of claims 1-15.

Images