CN117455421B

CN117455421B - Subject classification method and device for scientific research projects, computer equipment and storage medium

Info

Publication number: CN117455421B
Application number: CN202311791838.2A
Authority: CN
Inventors: 林世清; 郑晓雯; 赵崇旭
Original assignee: Hangzhou Qingta Technology Co ltd
Current assignee: Hangzhou Qingta Technology Co ltd
Priority date: 2023-12-25
Filing date: 2023-12-25
Publication date: 2024-04-16
Anticipated expiration: 2043-12-25
Also published as: CN117455421A

Abstract

The application provides a subject classification method, a subject classification device, computer equipment and a storage medium for scientific research projects, wherein the method comprises the following steps: acquiring target content information and target associated information corresponding to a scientific research project to be classified based on the scientific research project data information to be classified from a user; the target associated information comprises at least one of the item type of the scientific research item to be classified, the scientific research item plan of the scientific research item to be classified and the unit-based scientific site distribution information; inputting the target content information into a pre-constructed target text classification network model based on an attention mechanism to obtain an initial subject classification result output by the target text classification network model; and inputting the initial subject classification result and the target association information into a pre-constructed target full-connection classification network to obtain a target subject classification result output by the target full-connection classification network, and transmitting the target subject classification result to a user side. The application improves the accuracy and the classification efficiency of subject classification of scientific research projects.

Description

Subject classification method and device for scientific research projects, computer equipment and storage medium

Technical Field

The application relates to the technical field of intelligent decision making of artificial intelligence, in particular to a subject classification method, a subject classification device, computer equipment and a storage medium of scientific research projects.

Background

The subject classification of the scientific research project refers to dividing the scientific research project into different first-level subject categories according to the research content and the characteristics of the scientific research project, and the subject classification of the scientific research project is very important for the management of the scientific research project and the data analysis depending on the subject to which the scientific research project belongs.

In the related prior art, a scientific research person determines the type of the item to which the scientific research item belongs according to the content of the scientific research item, or automatically classifies the scientific research item by using a machine according to a comparison table of keywords and primary subjects. However, the manual classification method by using scientific researchers and the like is high in accuracy, but takes a long time, and the automatic classification method by using a machine according to the comparison table of the keywords and the primary subjects mechanically splits and compares the keywords, so that the classification accuracy is lower although the speed is improved compared with the manual classification.

Disclosure of Invention

The embodiment of the application provides a subject classification method, a subject classification device, computer equipment and a storage medium for scientific research projects, and aims to improve the speed of subject classification of the scientific research projects, ensure the accuracy of subject classification of the scientific research projects and improve the subject classification efficiency of the scientific research projects.

In a first aspect, an embodiment of the present application provides a subject classification method for a scientific research project, including:

Acquiring target content information and target associated information corresponding to a scientific research project to be classified based on the scientific research project data information to be classified from a user; the target content information indicates item content information of the scientific research item to be classified; the target associated information comprises at least one of the item type of the scientific research item to be classified, the scientific research item plan of the scientific research item to be classified and the unit-based scientific site distribution information;

Inputting the target content information into a pre-constructed target text classification network model based on an attention mechanism, and obtaining an initial subject classification result output by the target text classification network model;

and inputting the initial subject classification result and the target associated information into a pre-constructed target full-connection classification network to obtain a target subject classification result output by the target full-connection classification network, and sending the target subject classification result to the user side.

In a second aspect, an embodiment of the present application provides a subject classification apparatus for a scientific research project, including:

The acquisition unit is used for acquiring target content information and target associated information corresponding to the scientific research projects to be classified based on the scientific research project data information to be classified from the user side; the target content information indicates item content information of the scientific research item to be classified; the target associated information comprises at least one of the item type of the scientific research item to be classified, the scientific research item plan of the scientific research item to be classified and the unit-based scientific site distribution information;

The first classification processing unit is used for inputting the target content information into a pre-constructed target text classification network model based on an attention mechanism to obtain an initial subject classification result output by the target text classification network model;

and the second classification processing unit is used for inputting the initial subject classification result and the target association information into a pre-constructed target full-connection classification network to obtain a target subject classification result output by the target full-connection classification network, and sending the target subject classification result to the user side.

In a third aspect, an embodiment of the present application further provides a computer device, which includes a memory, a processor, and a computer program stored in the memory and capable of running on the processor, where the processor implements the subject classification method of the scientific research project in the first aspect when executing the computer program.

In a fourth aspect, an embodiment of the present application further provides a computer readable storage medium, where the computer readable storage medium stores a computer program, which when executed by a processor, causes the processor to perform the subject classification method of the scientific research project of the first aspect.

According to the method, the target content information and the target associated information (comprising at least one of the item type, the scientific research item plan and the unit-based scientific locus distribution information of the scientific research item to be classified) of the indication item content information corresponding to the scientific research item to be classified are obtained based on the scientific research item data information to be classified from the user side, the target content information is classified by using the target text classification network based on the attention mechanism, an initial subject classification result is obtained, the initial subject classification result and the target associated information are input into the target full-connection classification network on the basis, and therefore the target subject classification result output by the target full-connection classification network is obtained, and is used as the final classification result of the scientific research item to be classified and sent to the user side. Therefore, the application utilizes the target text classification network based on the attention mechanism to classify the target content information of the scientific research project to obtain an initial subject classification result, and inputs the target associated information of the scientific research project to be classified and the initial subject classification result into the target full-connection classification network to obtain the target subject classification result.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for the description of the embodiments will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present application, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.

Fig. 1 is an application scenario schematic diagram of a subject classification method of a scientific research project according to an embodiment of the present application;

Fig. 2 is a flow chart of a subject classification method of a scientific research project according to an embodiment of the present application;

FIG. 3 is a schematic sub-flowchart of a subject classification method for scientific research projects according to an embodiment of the present application;

FIG. 4 is a schematic diagram of an attachment tree between units of support according to an embodiment of the present application;

FIG. 5 is a schematic diagram of another sub-flow of the subject classification method for scientific research projects according to an embodiment of the present application;

FIG. 6 is a schematic diagram of another sub-flow of the subject classification method for scientific research projects according to an embodiment of the present application;

FIG. 7 is a schematic diagram of another sub-flow of the subject classification method for scientific research projects according to an embodiment of the present application;

FIG. 8 is a schematic block diagram of a subject classification device for scientific research projects provided by an embodiment of the application;

fig. 9 is a schematic block diagram of a computer device according to an embodiment of the present application.

Detailed Description

The following description of the embodiments of the present application will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are some, but not all embodiments of the application. All other embodiments, which can be made by those skilled in the art based on the embodiments of the application without making any inventive effort, are intended to be within the scope of the application.

It should be understood that the terms "comprises" and "comprising," when used in this specification and the appended claims, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof.

It is also to be understood that the terminology used in the description of the application herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. As used in this specification and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise.

It should be further understood that the term "and/or" as used in the present specification and the appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes such combinations.

The application provides a subject classification method, a device, computer equipment and a storage medium for scientific research projects, which can improve the speed of the subject classification of the scientific research projects, ensure the accuracy of the subject classification of the scientific research projects and improve the subject classification efficiency of the scientific research projects. The subject of the subject classification method of the scientific research project can be the subject classification device of the scientific research project provided by the embodiment of the application, and can be a computer device integrated with the subject classification device of the scientific research project. The subject classification device of the scientific research project can be realized in a hardware or software mode; the computer device may be a terminal or a server, and the terminal may be a smart phone, a tablet computer, a palmtop computer, a notebook computer, or the like.

For example, taking a server as an execution body as an example, fig. 1 is a schematic view of an application scenario provided in an embodiment of the present application. In the scene, a user side sends scientific research project data information to be classified to a server, the server obtains target content information and target associated information corresponding to the scientific research project to be classified based on the scientific research project data information to be classified, a text classification network based on an attention mechanism is input for the target content information to obtain an initial subject classification result output by the text classification network, the initial subject classification result and the target associated information are input into a target full-connection classification network on the basis, and accordingly a target subject classification result output by the target full-connection classification network is obtained, and the target subject classification result is sent to the user side.

Referring to fig. 2, fig. 2 is a flowchart illustrating a subject classification method for a scientific research project according to an embodiment of the application, and the method specifically includes the following steps S101 to S103.

Step S101, acquiring target content information and target associated information corresponding to a scientific research project to be classified based on the scientific research project data information to be classified from a user; the target content information indicates item content information of the scientific research item to be classified; the target associated information comprises at least one of the item type of the scientific research item to be classified, the scientific research item plan of the scientific research item to be classified and the unit-based scientific site distribution information.

The data information of the scientific research projects to be classified is related data information corresponding to the scientific research projects to be classified. Specifically, the scientific research item to be classified is a scientific research item to be classified according to the subject, and the relevant data information corresponding to the scientific research item to be classified specifically may include identification information (such as an item number, an item name, etc.) of the scientific research item to be classified and relevant data information of the scientific research item to be classified. The related data information may be preset related information.

In this embodiment, a user may send the to-be-classified scientific research project data information to an execution subject of the present application through a user side, and after receiving the to-be-classified scientific research project data information, the execution subject of the present application may obtain target content information and target association information corresponding to the to-be-classified scientific research project based on the to-be-classified scientific research project data information.

The target content information indicates the item content of the scientific research item to be classified, which is specifically related information capable of reflecting the item content of the scientific research item to be classified, and when implemented, may be one or more of the item name, the item abstract, the research content and the like of the scientific research item to be classified.

In a specific implementation, the target content information may be obtained by performing word segmentation preprocessing on item contents (for example, item names, item abstracts, and research contents) of scientific research items to be classified, so as to facilitate subsequent text classification.

The target association information indicates related association information of the scientific research project to be classified, which can reflect subject association information of the scientific research project to be classified directly or indirectly, and specifically may include at least one of a project type, a project plan of the scientific research project to be classified, and unit-based scientific site distribution information.

Specifically, the scientific project plan to which the scientific project to be classified belongs may be a national-level, provincial-level, and college-level scientific project plan. For example, national-level scientific project plans may be national emphasis development plans, national social science foundation, education sector personality project, national natural science foundation, and the like. In this embodiment, since each of the scientific project plans is generally applicable to scientific projects in a specific field, for example, projects funded by national natural science foundation are more prone to scientific projects belonging to the primary subjects in the natural science field; the education part project of the human focuses on the scientific research project belonging to the field of human ; national social science foundation aims at supporting research projects and the like in the field of social science, so that a belonging scientific project plan of a scientific project to be classified can be obtained as the target associated information.

The types of the scientific research items to be classified are types of classifications corresponding to the scientific research items to be classified, and may include, for example, theoretical types, experimental types, comprehensive types, basic types, application types, development types, and the like. In a specific embodiment, it may specifically refer to a corresponding item classification in the generic scientific project plan, for example, taking a national emphasis development plan as an example, where the corresponding item classification includes: basic research (including basic theoretical research, basic technical research, new material research, new technical research, new technological research, and the like), application research (including application basic research, new product development, new technical development, new equipment development, new technological development, and the like), and social development (including social development research, social development technical support, social development technical popularization, social development technical service, and the like). In this embodiment, the item type of the scientific research item can reflect the first-level subject (or the first-level subject range) to which the scientific research item belongs to a certain extent, so that the item type of the scientific research item to be classified can be obtained as the target associated information.

The unit-based scientific site distribution information is the scientific site distribution information of the target unit-based scientific site corresponding to the scientific research project to be classified. The target supporting unit is a direct supporting unit or an indirect supporting unit of the scientific research project to be classified, the direct supporting unit is a unit supported by the scientific research project to be classified, and the indirect supporting unit is a unit supported directly or indirectly by the direct supporting unit of the scientific research project to be classified; the science site distribution information may indicate the first-class science site of the target unit of support, which may specifically include the first-class science site, the filling science site, and the doctor site. For example, taking the a medical college as the target support unit, the first-class academic loci of a medical college may indicate that the first-class academic loci of a medical college include: biological (first-order subject doctor point), basic medicine (first-order subject doctor point), clinical medicine (first-order subject master, doctor point), oral medicine (first-order subject master, doctor point), pharmacy (first-order subject doctor point), public management (first-order subject doctor point). In this embodiment, since an entity has a learning site in a certain level of discipline (e.g., physics) field, the entity may be more prone to develop related scientific projects in the first level of discipline (physics) field, so as to obtain the distribution information of the learning sites of the entity corresponding to the scientific projects to be classified as the target association information.

In order to improve accuracy of subject classification of the scientific research project to be classified, in some specific embodiments, the target associated information of the scientific research project to be classified may include the type of the project to which the scientific research project to be classified belongs, the project plan of the scientific research to which the scientific research project to be classified belongs, and the distribution information of the scientific sites depending on the unit.

In a specific implementation, the target associated information of the scientific research project to be classified may specifically be characteristic information of one-hot coding (single hot coding), where the one-hot coding includes a plurality of coding values corresponding to the types, and the coding value is 0 or 1.

For example, the item type scientific research of the to-be-classified scientific research item is specifically an item type one-hot feature, the scientific research item plan is specifically a scientific research item plan one-hot feature, and the depending unit scientific site distribution information is specifically a depending unit scientific site one-hot feature. In the item type one-hot characteristics, the value of the item type to which the scientific research item to be classified belongs is 1, and the rest is 0; the one-hot feature of the scientific project plan comprises coding values corresponding to a plurality of scientific project plans respectively, wherein the coding value corresponding to the scientific project plan to which the scientific project to be classified belongs is 1, and the rest is 0; the one-hot characteristic of the science site of the support unit comprises coding values corresponding to a plurality of first-level subjects, and the coding value corresponding to the first-level science site of the support unit (the target support unit is described below) is 1.

Specifically, in order to facilitate rapid acquisition of the distribution information of the learning sites of the unit of science corresponding to the scientific research item to be classified, in an embodiment, as shown in fig. 3, the method may be implemented based on the following steps S201 to S203.

Step S201, obtaining a direct support unit of the scientific research project to be classified.

In this embodiment, the unit directly supported by the scientific research project to be classified is a unit supported by the scientific research project to be classified, and the supported unit (i.e., the supported unit) can provide the required resource for executing the scientific research project.

Step S202, acquiring a target support unit corresponding to the scientific research project to be classified based on a pre-constructed support unit attachment relationship and the direct support unit.

Wherein the depending unit affiliation indicates an affiliation between each depending unit; the target support unit is a support unit with a scientific site, which is directly or indirectly supported by the scientific projects to be classified. In this embodiment, since the direct support units of the scientific research projects to be classified do not have any scientific sites (for example, a hospital F with the ability to develop the scientific research projects), for quickly determining the target support units with the scientific sites corresponding to the scientific research projects to be classified, the attachment relationship between the support units is pre-established, so that the determination of the target support units based on the attachment relationship of the support units and the direct support units is facilitated.

In the specific implementation, the attachment relation of the support units can be constructed by relying on the attachment relation of the support units in pairs, and the specific expression form of the attachment relation of the support units can be determined according to actual conditions.

To graphically represent the affiliations between the units of support, in some embodiments, the units of support affiliations described above may be represented in a manner that builds an affiliation tree. Specifically, there is a direct or indirect attachment relationship between units (depending units) in the same attachment tree, and the parent node of the child depending unit is the direct attachment unit (i.e., the direct depending unit) of the child depending unit.

For example, one affiliation tree in the G province may be as shown in fig. 4, where G1, G2, G3 are 3 urban areas of the G province. In fig. 4, "G university" is located at the uppermost node of the affiliation tree, i.e., "G university" is the uppermost depending unit in the affiliation tree; for the first people hospital in G2, the "college of medical science" is the father node, i.e. the directly superior subordinate unit of the "first people hospital in G2" is the "college of medical science".

And step 203, determining the science site distribution information corresponding to the target depending unit as the depending unit science site distribution information.

In this embodiment, after a target supporting unit is obtained based on a pre-constructed supporting unit attachment relationship, the scientific site distribution information corresponding to the target supporting unit is determined to be the supporting unit scientific site distribution information, and subject classification is performed by taking the target association information, so that the rapid and scientific determination of the supporting unit scientific site distribution information is realized.

In order to uniformly and rapidly acquire the target supporting units corresponding to the scientific research projects to be classified, in one embodiment, the supporting unit affiliation comprises one or more of the above-mentioned affiliated trees, and in the affiliated trees, the supporting unit positioned at the uppermost layer is the supporting unit with the scientific site; the step S202 may be specifically implemented by the following steps:

determining a target affiliation tree where the direct support unit is located in the support unit affiliation;

And determining the uppermost supporting unit in the target affiliation tree as the target supporting unit.

In this embodiment, since the attachment unit attachment relationship includes one or more attachment relationship trees, it is necessary to determine a target attachment relationship tree where the direct attachment unit is located, where the direct attachment unit is used as a node in the target attachment relationship tree.

In this embodiment, in the affiliation tree, the supporting unit at the uppermost layer refers to a supporting unit without a father node, and since the supporting unit at the uppermost layer has a scientific site, after determining the target affiliation tree where the direct supporting unit is located, the supporting unit at the uppermost layer in the target affiliation is determined as the target supporting unit for uniformly and quickly acquiring the target supporting unit corresponding to the scientific research item to be classified. Wherein the uppermost supported unit may or may not be the direct supported unit.

In some embodiments, in order to reduce the range of the academic loci included in the scientific locus distribution information of the support units, improve the accuracy of the classification of the subjects of the scientific research projects to be classified, and determine the target support units with the scientific loci in a step-by-step judgment manner.

For example, in an embodiment, the step S202 may be specifically implemented by the following steps.

And step A, judging whether the direct support unit has a scientific site, if so, executing the step B, otherwise, executing the step C.

And B, determining the direct support unit as the target support unit.

And C, judging whether the directly upper auxiliary unit of the directly supported unit in the supported unit auxiliary relationship has a scientific site, if so, executing the step D, otherwise, executing the step F.

And D, determining the direct upper-level auxiliary unit as the target supporting unit.

And F, taking the direct upper-level auxiliary unit as the direct support unit, and returning to execute the step C.

In this embodiment, the attachment unit relationship may be specifically formed by one or more attachment relationship trees, and when the attachment unit relationship is formed by an attachment relationship tree, the steps a to F all determine the node unit in the attachment relationship tree (target attachment relationship tree) where the direct attachment unit is located. Thus, the above-mentioned "directly upper-level subordinate unit of directly supported unit" refers to the supported unit corresponding to the parent node of the node where the directly supported unit is located in the target subordinate relationship tree.

In this embodiment, whether the direct support unit has a science site is determined first, if the direct support unit has a science site, the direct support unit is directly determined as a target support unit, if the direct support unit does not have a science site, whether a direct upper auxiliary unit of the direct support unit has a science site is further determined, if the direct upper auxiliary unit has a science site, the direct upper auxiliary unit is determined as a target support unit, otherwise, whether the direct upper auxiliary unit of the direct upper auxiliary unit has a science site is further determined upwardly, until the support unit with a science site is determined, and the direct upper auxiliary unit is taken as a target support unit, so that step-by-step determination of the target support unit is realized, the range of distribution information of the science site of the support unit is conveniently reduced, and therefore, the accuracy of classification of subjects is facilitated to be improved.

And step S102, inputting the target content information into a pre-constructed target text classification network model based on an attention mechanism, and obtaining an initial subject classification result output by the target text classification network model.

The attention mechanism-based target text classification network model is a pre-constructed model, classifies the input text information based on the attention mechanism, and obtains the initial subject classification result. In this embodiment, for natural language processing, the attention mechanism simulates the human attention distribution process, and can automatically learn important parts in input data, so as to improve the performance and efficiency of the model, and compared with RNN (recurrent neural network) networks, the method has the advantages of fewer parameters, faster speed and better effect.

In particular implementations, the attention-based target text classification network model is determined based on an attention-based underlying target text classification network model, which may be, for example, a BART (Bidirectional and Auto-REGRESSIVE TRANSFORMERS) model, a BERT (Bidirectional Encoder Representations from Transformers) model, or the like.

For example, in one embodiment, the above-mentioned attention mechanism-based target text classification network model may be constructed based on the BERT model, and as shown in fig. 5, a specific construction method includes the following steps S301 to S304.

Step S301, acquiring a BERT model based on an attention mechanism.

Wherein the BERT model is a pre-trained language representation model that can generate deep bi-directional language representations that can capture complex semantic information, including word sense, syntax, context, in the input text. From a network architecture perspective, the BERT model does use a multi-layer Transformer architecture. In contrast to conventional RNNs (recurrent neural networks) and CNNs (convolutional neural networks), the Transformer structure can handle words at any two positions in a sequence through Self-Attention mechanisms (Self-Attention).

Step S302, adding a full connection layer and a Softmax layer after the BERT model to obtain a classification model to be trained based on an attention mechanism;

In this embodiment, the BERT model is mainly used to capture complex semantic information in an input text, characterize the complex semantic information as multidimensional semantic features, and when the BERT is required to be used for classification, a full-connection layer and a Softmax layer are added after the BERT model, so that the classification model to be trained for text classification is constructed.

The main function of the fully-connected layer is to convert the multidimensional semantic features output by the BERT model into a vector with a fixed length and suitable for a classifier, and the vector can capture global information of input data, including correlations among the features.

The main function of the Softmax layer is to convert the output of the fully connected layer into a probability distribution vector through the Softmax function, and the sum of the probabilities is 1, so that the category with the highest probability can be used as a prediction result.

Specifically, the softmax function has the expression:

Wherein Z is a vector and Z _K is an element in the vector Z.

In a specific implementation, the output dimension of the Softmax layer is determined according to the total number of all the categories actually classified, for example, the output dimension of the Softmax layer may be a vector with K probabilities, where K is the total number of primary disciplines, and each probability in the vector represents a probability that a scientific research item to be classified belongs to a corresponding primary discipline.

Step S303, acquiring a plurality of first scientific research project training samples; the first scientific research project training sample comprises target content information and an initial subject classification label corresponding to a scientific research project.

In this embodiment, the plurality of first training samples for scientific research projects are used for training and optimizing the classification model to be trained. Specifically, each first scientific research project training sample comprises the target content information and an initial discipline classification label corresponding to a scientific research project, wherein the target content information is used for inputting the classification model to be trained to obtain a discipline classification result of the classification model to be trained for the target content information; the initial subject classification label is used as a label of the subject classification result, and parameter adjustment is performed on the classification model to be trained, so that the classification accuracy of the classification model to be trained is gradually improved.

Step S304, carrying out label training on the classification model to be trained based on a preset target model training strategy and a plurality of first scientific research project training samples to obtain a target text classification network model; the target model training strategy is a first model training strategy.

The target model training strategy indicates how to perform labeled training on the classification model to be trained by using a plurality of first scientific research project training samples, wherein labeled training refers to training on the classification model to be trained based on a supervised learning mode. In the step S303, the target model training strategy is specifically a first model training strategy.

And step S103, inputting the initial subject classification result and the target associated information into a pre-constructed target full-connection classification network to obtain a target subject classification result output by the target full-connection classification network, and sending the target subject classification result to the user side.

In this embodiment, after obtaining an initial subject classification result obtained by subject classification by the target text classification network according to the target content information, in order to further improve accuracy of subject classification, the initial subject classification result and the target association information are input into the target full-connection classification network by referring to more relevant information, and the target full-connection classification network performs synthesis and further classification based on the existing initial subject classification result and the target association information, so as to obtain an output target subject classification result.

In the process of inputting the initial subject classification result and the target association information into the target full-connection classification network, the initial subject classification result and the target association information may be spliced and then input into the target full-connection classification network.

Specifically, the target full-connection classification network is a pre-constructed network model, and is used for comprehensively processing the initial subject classification result and the target association information, capturing the interrelation between the information, and outputting the target subject classification result. In this embodiment, the target subject classification result is a final classification result of the scientific research item to be classified, and the target subject classification result is sent to the user side.

In particular implementations, the target fully-connected classification network may include an input layer, a fully-connected layer, and a Softmax layer.

For example, in an embodiment, as shown in fig. 6, the specific method for constructing the target fully-connected classification network may specifically include the following steps S401 to S403.

S401, constructing a full-connection network to be trained; the fully-connected network to be trained comprises an input layer, a fully-connected layer and a Softmax layer.

In this embodiment, the to-be-trained fully-connected network is used for classifying based on the input information, and the to-be-trained fully-connected network needs to be trained and optimized to obtain the target fully-connected classified network. Specifically, the fully-connected network to be trained comprises an input layer, a fully-connected layer and a Softmax layer. The input layer is used for receiving input data; the full connection layer converts input into vectors with a certain length in a full connection mode; the Softmax layer is used for converting the output of the full-connection layer into a probability distribution, and then the category with the highest probability can be used as the target subject classification result.

Step S402, a plurality of second scientific research project training samples are obtained; the second scientific research project training sample comprises target associated information corresponding to the scientific research project, the initial subject classification result and a target subject classification result label.

In this embodiment, the plurality of second training samples for scientific research projects are used for training and optimizing the to-be-trained fully-connected network. Specifically, each second scientific research project training sample includes the target associated information, the initial subject classification result and the target subject classification result label corresponding to one scientific research project. The target association information and the initial subject classification result are used for inputting the to-be-trained full-connection network to obtain a subject classification result output by the to-be-trained full-connection network; the target subject classification label is used as a label of the subject classification result, and parameter adjustment is performed on the to-be-trained fully-connected network, so that the classification accuracy of the to-be-trained fully-connected network is gradually improved.

The initial subject classification result corresponding to the scientific research project included in the second scientific research project training sample is an output result obtained by inputting the target content information corresponding to the scientific research project into the target text classification network model (trained network) based on the attention mechanism.

Step S403, carrying out label training on the to-be-trained fully-connected network based on a preset target model training strategy and a plurality of second scientific research project training samples to obtain the target fully-connected classification network; the target model training strategy is a second model training strategy.

The target model training strategy indicates how to perform labeled training on the classification model to be trained by using a plurality of second scientific research project training samples, wherein labeled training refers to training on the fully-connected network to be trained based on a supervised learning mode. In the above step S403, the target model training strategy is specifically the second model training strategy.

In the specific implementation, the first model training strategy used in the step S304 and the second model training strategy used in the step S403 may be the same or different.

In one embodiment, as shown in fig. 7, the first model training strategy and/or the second model training strategy (i.e., the target model training strategy) may specifically include the following steps S501 to S504.

Step S501, dividing the plurality of training samples of the scientific research project into a training sample set, a verification sample set and a test sample set.

When the first model training strategy includes steps S501 to S504, the plurality of training samples for the scientific research project specifically refer to the plurality of training samples for the first scientific research project; when the second model training strategy includes steps S501 to S504, the plurality of training samples of the scientific research project specifically refer to the plurality of training samples of the second scientific research project.

In this embodiment, for a plurality of training samples of scientific research projects, the training samples are divided into three parts, one part is used as a training sample set, one part is used as a verification sample set, and the remaining part is used as a test sample set. In specific implementations, the ratio of the number of samples included in the training sample set, the verification sample set, and the test sample set may be specifically determined according to actual situations. For example, in a specific embodiment, the ratio of the number of samples included in the training sample set, the verification sample set, and the test sample set may be 3:7:1.

And step S502, adjusting internal parameters of the network model to be trained by using the training sample set, and adjusting super parameters of the network model to be trained by using the verification sample set to obtain a training result model.

Wherein, when the first model training strategy includes steps S501 to S504, the network model to be trained specifically refers to the classification model to be trained; when the second model training strategy includes steps S501 to S504, the network model to be trained specifically refers to the fully connected network to be trained.

In this embodiment, the training sample set is used to train the network model to be trained, so as to adjust internal parameters of the classification model to be trained, where the internal parameters refer to parameters learned by the model in the training process. For example, weight parameters and bias; the verification sample set is used for verifying the classification effect of the classification model to be trained on the classification model to be trained so as to adjust the super-parameters of the network model to be trained, wherein the super-parameters are initially set before the training of the model is started and are used for guiding the training process. For example, learning rate, batch size, number of rounds of training, optimizers, etc. are super parameters.

And step S503, performing classification accuracy test on the training result model by using the test sample set to obtain a classification accuracy result.

In this embodiment, after the parameter adjustment of the network model to be trained is completed by using a training sample set and a verification sample set, the training result model is obtained, and in order to verify the generalization capability of the training result model, the classification accuracy test is performed on the training result model by using a test sample set, so as to obtain a classification accuracy result.

Specifically, the training result model is utilized to carry out classification test on each scientific research project training sample in the test sample set, so that the classification accuracy corresponding to each scientific research project training sample is obtained, and the classification accuracy results are obtained by integrating the classification accuracy corresponding to each scientific research project training sample.

And step S504, when the classification accuracy result is larger than a preset classification accuracy threshold, taking the training result model as a target network model.

Wherein, when the first model training strategy includes steps S501 to S504, the target network model specifically refers to the target text classification network model; when the second model training strategy includes steps S501 to S504, the target network model specifically refers to the target fully-connected classification network.

In this embodiment, when the classification accuracy result is greater than a preset classification accuracy threshold, it is indicated that the generalization capability of the training result model is better, and there is no overfitting, so that the training result model can be used as a target network model for the primary subject classification of the actual scientific research project.

In some embodiments, when the classification accuracy result is not greater than a preset classification accuracy threshold, it is indicated that the generalization capability of the training result model is poor, and there is a fitting, at this time, the hyper-parameters of the training result model may be modified, and a new training sample set may be obtained again, and the training result model may be trained by the verification sample set until the classification accuracy result is greater than the preset classification accuracy threshold; optionally, the internal parameters of the network model to be trained may be reinitialized, a new training sample set may be obtained again, and the training may be performed on the network model to be trained by the verification sample set until the classification accuracy result is greater than the preset classification accuracy threshold.

In the embodiment, the model parameters are adjusted and optimized by setting the training sample set and the verification sample set, and the generalization capability of the model is verified by setting the test sample set, so that the classification accuracy of the obtained target network model is improved.

In summary, the method acquires the target content information and the target associated information (including at least one of the item type, the belonging scientific research item plan and the depending unit scientific science site distribution information of the scientific research item to be classified) corresponding to the indicated item content information based on the scientific research item data information to be classified from the user side, classifies the target content information by using the target text classification network based on the attention mechanism to obtain an initial subject classification result, and inputs the initial subject classification result and the target associated information into the target full-connection classification network on the basis of the initial subject classification result, so that the target subject classification result output by the target full-connection classification network is obtained, and the target subject classification result is used as the final classification result of the scientific research item to be classified and is sent to the user side. Therefore, the application utilizes the text classification network based on the attention mechanism to classify the target content information of the scientific research project to obtain an initial subject classification result, and inputs the target associated information of the scientific research project to be classified and the initial subject classification result into the target full-connection classification network to obtain the target subject classification result.

The embodiment of the application also provides a subject classification device for the scientific research project, which is used for executing the steps in any embodiment of the subject classification method for the scientific research project. Specifically, referring to fig. 8, fig. 8 shows a schematic structural diagram of a subject classification device 600 for a scientific research project according to an embodiment of the present application, where the subject classification device 600 for a scientific research project specifically includes a receiving unit 601, a publishing unit 602, and an authorized use unit 603.

The acquiring unit 601 is configured to acquire target content information and target association information corresponding to a scientific research project to be classified based on data information of the scientific research project to be classified from a user terminal; the target content information indicates item content information of the scientific research item to be classified; the target associated information comprises at least one of the item type of the scientific research item to be classified, the scientific research item plan of the scientific research item to be classified and the unit-based scientific site distribution information;

A first classification processing unit 602, configured to input the target content information into a pre-constructed target text classification network model based on an attention mechanism, and obtain an initial subject classification result output by the target text classification network model;

the second classification processing unit 603 is configured to input the initial discipline classification result and the target association information into a pre-constructed target fully-connected classification network, obtain a target discipline classification result output by the target fully-connected classification network, and send the target discipline classification result to the user side.

In some embodiments, the obtaining unit 601 may be further configured to obtain a BERT model based on an attention mechanism; the subject classification device 600 of the scientific research project further includes a construction unit, configured to add a full connection layer and a Softmax layer after the BERT model, to obtain a classification model to be trained based on an attention mechanism; the obtaining unit 601 may be further configured to obtain a plurality of first training samples for scientific research projects; the first scientific research project training sample comprises target content information and an initial subject classification label corresponding to a scientific research project; the subject classification device 600 of the scientific research project further includes a training unit, configured to perform labeled training on the classification model to be trained based on a preset target model training strategy and a plurality of training samples of the first scientific research project, so as to obtain the target text classification network model; the target model training strategy is a first model training strategy.

In some embodiments, the building unit may be further configured to build a fully connected network to be trained; the to-be-trained full-connection network comprises an input layer, a full-connection layer and a Softmax layer; the obtaining unit 601 may be further configured to obtain a plurality of second training samples for scientific research projects; the second scientific research project training sample comprises target associated information, the initial subject classification result and a target subject classification result label corresponding to a scientific research project; the training unit is further configured to perform labeled training on the to-be-trained fully-connected network based on a preset target model training strategy and a plurality of training samples of the second scientific research project, so as to obtain the target fully-connected classification network; the target model training strategy is a second model training strategy.

In some embodiments, the training unit may be specifically configured to divide the plurality of training samples of the scientific research project into a training sample set, a verification sample set and a test sample set; adjusting internal parameters of the network model to be trained by using the training sample set, and adjusting super parameters of the network model to be trained by using the verification sample set to obtain a training result model; performing classification accuracy test on the training result model by using the test sample set to obtain a classification accuracy result; and when the classification accuracy result is larger than a preset classification accuracy threshold, taking the training result model as a target network model.

In some embodiments, the obtaining unit 601 may be specifically configured to obtain a direct support unit of the scientific research item to be classified; acquiring a target support unit corresponding to the scientific research project to be classified based on a pre-constructed support unit attachment relationship and the direct support unit; wherein the depending unit affiliation indicates an affiliation between each depending unit; the target support unit is a support unit with a scientific site; and determining the scientific site distribution information corresponding to the target depending unit as the depending unit scientific site distribution information.

In some embodiments, the depending unit affiliations include one or more affiliation trees that indicate affiliations between the depending units; in the affiliation tree, the support unit at the uppermost layer is a support unit with a scientific site; the obtaining unit 601 may be specifically configured to determine, in the attachment relationship of the unit of support, a target attachment relationship tree where the unit of direct support is located; and determining the uppermost supporting unit in the target affiliation tree as the target supporting unit.

In some embodiments, the obtaining unit 601 may be specifically configured to determine whether the direct-support unit has a scientific site; when the direct support unit has a scientific site, determining the direct support unit as the target support unit; when the direct support unit does not have a scientific site, judging whether a direct upper-level auxiliary unit of the direct support unit in the support unit auxiliary relationship has the scientific site or not; if the direct superior accessory unit has a scientific site, determining the direct superior accessory unit as the target supporting unit; and if the direct superior accessory unit does not have a academic site, taking the direct superior accessory unit as the direct support unit, and returning to the step of executing the judgment of whether the direct superior accessory unit of the direct support unit in the support unit accessory relationship has the academic site.

It should be noted that, as a person skilled in the art can clearly understand the specific implementation process of the subject classification device 600 and each unit of the above scientific research project, reference may be made to the corresponding description in the foregoing method embodiments, and for convenience and brevity of description, the description is omitted here.

The subject classification means of the scientific research project described above may be implemented in the form of a computer program which may be run on a computer device as shown in fig. 9.

Referring to fig. 9, fig. 9 is a schematic block diagram of a computer device according to an embodiment of the present application. The computer device 700 may be a terminal device such as a smart phone, tablet computer, personal computer, smart wearable device, server, etc. With reference to FIG. 9, the computer device 700 includes a processor 702, a memory, and a network interface 705, which are connected by a device bus 701, wherein the memory may include a storage medium 703 and an internal memory 704.

The storage medium 703 may store an operating system 7031 and a computer program 7032. The computer program 7032, when executed, can cause the processor 702 to perform a subject classification method for a scientific research project.

The processor 702 is used to provide computing and control capabilities to support the operation of the overall computer device 700.

The internal memory 704 provides an environment for the execution of a computer program 7032 in the storage medium 703, which computer program 7032, when executed by the processor 702, causes the processor 702 to perform a subject classification method for a scientific research project.

The network interface 705 is used for network communication, such as providing transmission of data information, etc. It will be appreciated by those skilled in the art that the structure shown in FIG. 9 is merely a block diagram of some of the structures associated with the present inventive arrangements and does not constitute a limitation of the computer device 700 to which the present inventive arrangements may be applied, and that a particular computer device 700 may include more or fewer components than shown, or may combine certain components, or may have a different arrangement of components.

The processor 702 is configured to execute a computer program 7032 stored in the memory, so as to implement the subject classification method of the scientific research project disclosed in the embodiment of the application.

Those skilled in the art will appreciate that the embodiment of the computer device shown in fig. 9 is not limiting of the specific construction of the computer device, and in other embodiments, the computer device may include more or less components than those shown, or certain components may be combined, or a different arrangement of components. For example, in some embodiments, the computer device may include only a memory and a processor, and in such embodiments, the structure and function of the memory and the processor are consistent with the embodiment shown in fig. 9, and will not be described again.

It should be appreciated that in embodiments of the present application, the processor 702 may be a central processing unit (Central Processing Unit, CPU), the processor 702 may also be other general purpose processors, digital signal processors (DIGITAL SIGNAL processors, DSPs), application SPECIFIC INTEGRATED Circuits (ASICs), off-the-shelf programmable gate arrays (Field-programmable GATE ARRAY, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or the like. Wherein the general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.

In another embodiment of the application, a computer-readable storage medium is provided. The computer readable storage medium may be a nonvolatile computer readable storage medium or a volatile computer readable storage medium. The computer readable storage medium stores a computer program, wherein the computer program when executed by a processor implements the subject classification method of scientific research projects disclosed in the embodiments of the present application.

It will be clearly understood by those skilled in the art that, for convenience and brevity of description, specific working procedures of the apparatus, device and unit described above may refer to corresponding procedures in the foregoing method embodiments, which are not repeated herein. Those of ordinary skill in the art will appreciate that the elements and algorithm steps described in connection with the embodiments disclosed herein may be embodied in electronic hardware, in computer software, or in a combination of the two, and that the elements and steps of the examples have been generally described in terms of function in the foregoing description to clearly illustrate the interchangeability of hardware and software. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present application.

In the several embodiments provided by the present application, it should be understood that the disclosed apparatus, device and method may be implemented in other manners. For example, the apparatus embodiments described above are merely illustrative, for example, the division of units is merely a logical function division, there may be another division manner in actual implementation, or units having the same function may be integrated into one unit, for example, multiple units or components may be combined or may be integrated into another apparatus, or some features may be omitted, or not performed. In addition, the coupling or direct coupling or communication connection shown or discussed with each other may be an indirect coupling or communication connection via some interfaces, devices, or elements, or may be an electrical, mechanical, or other form of connection.

The units described as separate units may or may not be physically separate, and units shown as units may or may not be physical units, may be located in one place, or may be distributed over a plurality of network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the embodiment of the present application.

In addition, each functional unit in the embodiments of the present application may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit. The integrated units may be implemented in hardware or in software functional units.

The integrated units may be stored in a storage medium if implemented in the form of software functional units and sold or used as stand-alone products. Based on such understanding, the technical solution of the present application may be essentially or a part contributing to the prior art, or all or part of the technical solution may be embodied in the form of a software product stored in a storage medium, comprising several instructions for causing a computer device (which may be a personal computer, a background server, or a network device, etc.) to perform all or part of the steps of the methods of the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a magnetic disk, an optical disk, or other various media capable of storing program codes.

The present application is not limited to the above embodiments, and various equivalent modifications and substitutions can be easily made by those skilled in the art within the technical scope of the present application, and these modifications and substitutions are intended to be included in the scope of the present application. Therefore, the protection scope of the application is subject to the protection scope of the claims.

Claims

1. The subject classification method for the scientific research projects is characterized by comprising the following steps of:

Inputting the target content information into a pre-constructed target text classification network model based on an attention mechanism, and obtaining an initial subject classification result output by the target text classification network model; the target text classification network model comprises a BERT model, a full-connection layer and a Softmax layer which are spliced in sequence; the Softmax layer is used for outputting probability distribution vectors, and the probability distribution vectors indicate the prediction probabilities respectively corresponding to the scientific research items to be classified belonging to the subjects; the initial subject classification result indicates a first subject category to which the scientific research item to be classified belongs;

Inputting the initial subject classification result and the target associated information into a pre-constructed target full-connection classification network to obtain a target subject classification result output by the target full-connection classification network, and sending the target subject classification result to the user side; the target full-connection classification network comprises an input layer, a full-connection layer and a Softmax layer; and the target subject classification result indicates a target subject category to which the scientific research item to be classified belongs.

2. The method of claim 1, wherein the inputting the target content information into a pre-constructed attention-based target text classification network model results in an initial subject classification result output by the target text classification network model, the method further comprising:

acquiring a BERT model based on an attention mechanism;

adding a full connection layer and a Softmax layer after the BERT model to obtain a classification model to be trained based on an attention mechanism;

Acquiring a plurality of first scientific research project training samples; the first scientific research project training sample comprises target content information and an initial subject classification label corresponding to a scientific research project;

Performing label training on the classification model to be trained based on a preset target model training strategy and a plurality of first scientific research project training samples to obtain a target text classification network model; the target model training strategy is a first model training strategy.

3. The method according to claim 1, wherein before the initial subject classification result and the target association information are input into a pre-constructed target fully-connected classification network to obtain a target subject classification result corresponding to the scientific research item to be classified output by the target fully-connected classification network, the method further comprises:

constructing a full-connection network to be trained; the to-be-trained full-connection network comprises an input layer, a full-connection layer and a Softmax layer;

acquiring a plurality of second scientific research project training samples; the second scientific research project training sample comprises target associated information, the initial subject classification result and a target subject classification result label corresponding to a scientific research project;

Carrying out label training on the fully-connected network to be trained based on a preset target model training strategy and a plurality of second scientific research project training samples to obtain the target fully-connected classification network; the target model training strategy is a second model training strategy.

4. A method according to claim 2 or 3, wherein the target model training strategy comprises:

dividing a plurality of scientific research project training samples into a training sample set, a verification sample set and a test sample set;

Adjusting internal parameters of the network model to be trained by using the training sample set, and adjusting super parameters of the network model to be trained by using the verification sample set to obtain a training result model;

performing classification accuracy test on the training result model by using the test sample set to obtain a classification accuracy result;

and when the classification accuracy result is larger than a preset classification accuracy threshold, taking the training result model as a target network model.

5. The method of claim 1, wherein the method for obtaining information on the distribution of the scientific loci of support unit comprises:

Acquiring a direct support unit of the scientific research project to be classified;

acquiring a target support unit corresponding to the scientific research project to be classified based on a pre-constructed support unit attachment relationship and the direct support unit; wherein the depending unit affiliation indicates an affiliation between each depending unit; the target support unit is a support unit with a scientific site;

and determining the scientific site distribution information corresponding to the target depending unit as the depending unit scientific site distribution information.

6. The method of claim 5, wherein the depending units affiliations include one or more affiliation trees that indicate affiliations between the depending units; in the affiliation tree, the support unit at the uppermost layer is a support unit with a scientific site; the obtaining the target support unit corresponding to the scientific research project to be classified based on the pre-constructed support unit attachment relationship and the direct support unit comprises the following steps:

7. The method of claim 5, wherein the obtaining the target unit of support corresponding to the scientific research item to be classified based on the pre-constructed unit of support attachment and the direct unit of support comprises:

Judging whether the direct support unit has a scientific site or not;

When the direct support unit has a scientific site, determining the direct support unit as the target support unit;

When the direct support unit does not have a scientific site, judging whether a direct upper-level auxiliary unit of the direct support unit in the support unit auxiliary relationship has the scientific site or not;

if the direct superior accessory unit has a scientific site, determining the direct superior accessory unit as the target supporting unit;

And if the direct superior accessory unit does not have a academic site, taking the direct superior accessory unit as the direct support unit, and returning to the step of executing the judgment of whether the direct superior accessory unit of the direct support unit in the support unit accessory relationship has the academic site.

8. The subject classification device for scientific research projects is characterized by comprising:

the first classification processing unit is used for inputting the target content information into a pre-constructed target text classification network model based on an attention mechanism to obtain an initial subject classification result output by the target text classification network model; the target text classification network model comprises a BERT model, a full-connection layer and a Softmax layer which are spliced in sequence; the Softmax layer is used for outputting probability distribution vectors, and the probability distribution vectors indicate the prediction probabilities respectively corresponding to the scientific research items to be classified belonging to the subjects; the initial subject classification result indicates a first subject category to which the scientific research item to be classified belongs;

The second classification processing unit is used for inputting the initial subject classification result and the target association information into a pre-constructed target full-connection classification network to obtain a target subject classification result output by the target full-connection classification network, and sending the target subject classification result to the user side; the target full-connection classification network comprises an input layer, a full-connection layer and a Softmax layer; and the target subject classification result indicates a target subject category to which the scientific research item to be classified belongs.

9. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the method of any one of claims 1 to 7 when executing the computer program.

10. A computer readable storage medium storing a computer program, characterized in that the computer program when executed by a processor implements the method according to any one of claims 1 to 7.