A demand matching method and system

Abstract

The embodiment of the application discloses a demand matching method and system, the method comprises the following steps: receiving a target demand description corpus input by a user; converting the target demand description corpus into a target technology resource corpus based on a generative adversarial network; calculating the matching degree of the target technology resource corpus and technology resource data in a technology resource database based on a convolutional neural network; determining a target technology resource list according to the matching degree ranking and outputting. Through cross-domain conversion of the text corpus and mixed application of supervised learning, efficient demand matching in the field of technology services is realized, and effective support is provided for the technology service industry.

Description

TECHNICAL FIELD

[0001] The embodiment of the present application relates to the technical field of computer application, in particular to a demand matching method and system. BACKGROUND

[0002] Modern science and technology service industry is a new industry emerging and developing rapidly in the global scope since 1980s, with the rapid development of science and technology and the acceleration of industrialization of scientific and technological achievements, and becomes an important part of modern service industry. Science and technology service industry mainly uses emerging technology and professional knowledge to provide intellectual services for the generation, application and diffusion of science and technology, including research and development services, technology transfer and popularization and industrialization services. Science and technology service industry has high interactivity and innovation, and plays a role of bridge and link in science and technology innovation, and is a key industry to promote the upgrading and optimization of industrial structure.

[0003] Due to the professional background of science and technology service resources, the demand of the user of science and technology service is often not a professional description made by an expert in the field, but a demand description conforming to the application background of the user. The traditional matching method often matches simply according to the coincidence degree of words or characters, or needs to design complex matching rules manually, and cannot realize the resource matching task at the semantic level. The matching model using machine learning method needs a large number of correct matching text pairs as training data, but at present, such data is rare and difficult to realize.

[0004] These difficulties make it an urgent problem to be solved in the science and technology service industry how to match the corresponding science and technology service field according to the demand. SUMMARY

[0005] Therefore, the embodiment of the present application provides a demand matching method and system, which realizes efficient demand matching in the science and technology service field through the hybrid application of cross-domain conversion of text corpus and supervised learning, and provides effective support for the science and technology service industry.

[0006] In order to realize the above purpose, the embodiment of the present application provides the following technical scheme:

[0007] According to the first aspect of the embodiment of the present application, a demand matching method is provided, and the method comprises:

[0008] Receiving a target demand description corpus input by a user;

[0009] Converting the target demand description corpus into a target science and technology resource corpus based on a generative adversarial network;

[0010] Calculating the matching degree of the target science and technology resource corpus and science and technology resource data in a science and technology resource database based on a recurrent neural network;

[0011] According to the matching degree ranking, the target resource list of scientific and technological services is determined and output.

[0012] Optionally, the matching degree between the target scientific and technological resource corpus and the scientific and technological resource data in the scientific and technological resource database is calculated based on the convolutional neural network, including:

[0013] The semantic vectors of the target scientific and technological resource corpus and the scientific and technological resource data in the scientific and technological resource database are calculated based on the convolutional neural network respectively.

[0014] The cosine similarity of the semantic vectors of the target scientific and technological resource corpus and each scientific and technological resource data in the scientific and technological resource database is calculated respectively to obtain the matching degree between the target scientific and technological resource corpus and the scientific and technological resource data in the scientific and technological resource database.

[0015] Optionally, before receiving the demand description corpus input by the user, the method further includes:

[0016] Collecting the demand description corpus and the scientific and technological resource description corpus, and dividing the training set and the test set;

[0017] Performing word segmentation and vectorization processing on the demand description corpus and the scientific and technological resource description corpus in the training set;

[0018] Training the generative adversarial network based on the constructed generative adversarial network, the training parameters and the processed corpus.

[0019] Optionally, the construction method of the generative adversarial network includes the following steps:

[0020] Constructing a generator and a discriminator, the generator includes a first generator and a second generator, and the discriminator includes a first discriminator and a second discriminator; the first generator is used to generate a scientific and technological resource corpus according to a demand description corpus, the second generator is used to generate a demand description corpus according to a scientific and technological resource corpus; the first discriminator is used to determine whether the demand description corpus is generated, and the second discriminator is used to determine whether the scientific and technological resource corpus is generated;

[0021] Inputting a first demand description corpus into the first generator to generate a first scientific and technological resource corpus; inputting a second scientific and technological resource corpus into the second generator to generate a second demand description corpus;

[0022] inputting the first technical resource corpus and the second technical resource corpus into the second discriminator, outputting first technical resource corpus classification results and second technical resource corpus classification results, and respectively calculating the adversarial loss values of the second discriminator and the first generator according to the classification results; inputting the first demand description corpus and the second demand description corpus into the first discriminator, outputting first demand description corpus classification results and second demand description corpus classification results, and respectively calculating the adversarial loss values of the first discriminator and the second generator according to the classification results;

[0023] performing gradient back-propagation training on the generative adversarial network according to the total adversarial loss, adjusting parameters of the generative adversarial network to minimize the total adversarial loss, and the total adversarial loss is calculated according to all the adversarial loss values.

[0024] Optionally, the method further comprises:

[0025] collecting data pairs composed of technical service resources in a user-selected target technical service resource list and a user-initial-input target demand description corpus, and performing optimization supervised learning on the generative adversarial network.

[0026] Optionally, the method further comprises:

[0027] collecting several data pairs composed of technical service resources in a user-selected target technical service resource list and technical resource data in a technical resource database, and performing optimization supervised learning on the optimization generative neural network.

[0028] According to a second aspect of the embodiments of the present application, a demand matching system is provided, and the system comprises:

[0029] an information receiving module configured to receive a user-input target demand description corpus;

[0030] a corpus conversion module configured to convert the target demand description corpus into a target technical resource corpus based on a generative adversarial network;

[0031] a matching module configured to calculate a matching degree between the target technical resource corpus and technical resource data in a technical resource database based on an optimization generative neural network;

[0032] an information output module configured to determine a target technical service resource list according to the matching degree and output the target technical service resource list.

[0033] Optionally, the matching module is specifically configured to:

[0034] calculate semantic vectors of the target technical resource corpus and the technical resource data in the technical resource database based on the optimization generative neural network, respectively;

[0035] Calculate the cosine similarity of the semantic vector of each technology resource data in the target technology resource corpus and the technology resource database, respectively, to obtain the matching degree of the target technology resource corpus and the technology resource data in the technology resource database.

[0036] According to a third aspect of the embodiments of the present application, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method of the first aspect.

[0037] According to a fourth aspect of the embodiments of the present application, a computer readable storage medium is provided, having computer readable instructions stored thereon, the computer readable instructions executable by a processor to implement the method of the first aspect.

[0038] In summary, the embodiments of the present application provide a demand matching method and system, which receives a target demand description corpus input by a user; converts the target demand description corpus into a target technology resource corpus based on a generative adversarial network; calculates the matching degree of the target technology resource corpus and technology resource data in a technology resource database based on a convolutional neural network; determines a technology service target resource list according to the matching degree ranking and outputs. Through cross-domain conversion of text corpus and hybrid application of supervised learning, efficient demand matching in the field of technology services is realized, and effective support is provided for the technology service industry. BRIEF DESCRIPTION OF DRAWINGS

[0039] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings needed in the following embodiment or prior art description will be briefly introduced. Obviously, the drawings in the following description are only exemplary, and for those skilled in the art, other drawings can be obtained without creative labor on the basis of the provided drawings.

[0040] The structures, proportions, sizes, etc. shown in the specification are only used to cooperate with the content disclosed in the specification, to be understood and read by those skilled in the art, and do not define the limiting conditions for the implementation of the present application, so they do not have technical significance. Any modification of structure, change of proportion relationship or adjustment of size, without affecting the effect and purpose that the present application can produce, should still fall within the scope of the technical content disclosed by the present application.

[0041] Figure 1 A demand matching method flowchart is provided for the embodiments of the present application;

[0042] Figure 2 A whole embodiment flowchart is provided for the embodiments of the present application;

[0043] Figure 3 An architecture schematic diagram of a generative adversarial network provided by an embodiment of the present application is shown.

[0044] Figure 4 An architecture schematic diagram of a generative adversarial network provided by an embodiment of the present application is shown.

[0045] Figure 5 A block diagram of a demand matching system provided by an embodiment of the present application is shown.

[0046] Figure 6 An architecture schematic diagram of an electronic device provided by an embodiment of the present application is shown.

[0047] Figure 7 An architecture schematic diagram of a computer readable storage medium provided by an embodiment of the present application is shown. DETAILED DESCRIPTION

[0048] The embodiments of the present application will be described in detail by specific embodiments, and those skilled in the art can easily understand other advantages and effects of the present application from the content disclosed in the specification. Obviously, the described embodiments are part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0049] Figure 1 A demand matching method provided by an embodiment of the present application is shown, and the method comprises:

[0050] Step 101: receiving a target demand description corpus input by a user;

[0051] Step 102: converting the target demand description corpus into a target technology resource corpus based on a generative adversarial network;

[0052] Step 103: calculating a matching degree of the target technology resource corpus and technology resource data in a technology resource database based on a generative adversarial network;

[0053] Step 104: determining a target resource list of technology services and outputting according to the matching degree sorting.

[0054] In a possible implementation, in step 103, the calculating of the matching degree of the target technology resource corpus and the technology resource data in the technology resource database based on the generative adversarial network comprises:

[0055] The semantic vectors of the target technology resource corpus and the technology resource data in the technology resource database are calculated based on the convolutional recurrent neural network.

[0056] In a possible implementation, before receiving the user inputted requirement description corpus in step 101, the method further includes:

[0057] The requirement description corpus and the technology resource description corpus are collected, and the training set and the test set are divided; the requirement description corpus and the technology resource description corpus in the training set are subjected to word segmentation and vectorization processing; and the generative adversarial network is trained based on the constructed generative adversarial network, the training parameters and the processed corpus.

[0058] In a possible implementation, the method for constructing the generative adversarial network includes the following steps.

[0059] The generator and the discriminator are constructed, the generator includes a first generator and a second generator, and the discriminator includes a first discriminator and a second discriminator; the first generator is configured to generate a technology resource corpus according to a requirement description corpus, the second generator is configured to generate a requirement description corpus according to a technology resource corpus; the first discriminator is configured to determine whether the requirement description corpus is generated, and the second discriminator is configured to determine whether the technology resource corpus is generated.

[0060] The first requirement description corpus is inputted into the first generator to generate a first technology resource corpus, and the second technology resource corpus is inputted into the second generator to generate a second requirement description corpus.

[0061] The first technology resource corpus and the second technology resource corpus are inputted into the second discriminator to output a first technology resource corpus classification result and a second technology resource corpus classification result, and the second discriminator and the first generator are adjusted according to the classification results to calculate the adversarial loss values of the second discriminator and the first generator, respectively; the first requirement description corpus and the second requirement description corpus are inputted into the first discriminator to output a first requirement description corpus classification result and a second requirement description corpus classification result, and the first discriminator and the second generator are adjusted according to the classification results to calculate the adversarial loss values of the first discriminator and the second generator, respectively.

[0062] The generative adversarial network is trained by gradient back propagation according to the total adversarial loss, the parameters of the generative adversarial network are adjusted to minimize the total adversarial loss, and the total adversarial loss is calculated according to all the adversarial loss values.

[0063] In a possible implementation, the method further includes collecting data pairs composed of the sci-tech service resources in the sci-tech service target resource list selected by the user and the target demand description corpus initially input by the user, and performing supervised learning optimization on the generative adversarial network.

[0064] In a possible implementation, the method further includes collecting several data pairs composed of the sci-tech service resources in the sci-tech service target resource list selected by the user and the sci-tech resource data in the sci-tech resource database, and performing supervised learning optimization on the generative adversarial network.

[0065] The embodiments of the present application are described in detail below. Figures 2-4 The embodiments of the present application are described in detail below. Figure 2 An embodiment provided by the present application is shown, which includes model training, specific application and optimization, and includes the following steps:

[0066] Step 1: Collecting corpus participating in model training, including demand description corpus of sci-tech service and sci-tech service resource description corpus; wherein the two corpora do not need to have a corresponding relationship; the demand description corpus is the text data of the demand description collected by the user; the sci-tech service resource description corpus is the description of the sci-tech service resource, such as the summary of a patent or the introduction of a project, which is also text. The data collected in the two parts does not need to be matched data, but only needs to have certain demand description and certain sci-tech resource description.

[0067] Step 2: Dividing the collected corpus into a training set and a test set, using tokenizer to divide the corpus, and vectorizing the text in the corpus set into a word vector sequence;

[0068] Step 3: Constructing a text corpus processing model based on the CycleGAN structure of the generative adversarial network. This step is based on the idea of the original CycleGAN network, and constructs two generators Gxy and Gyx, and two discriminators Dx and Dy. Among them, Gxy is used to generate sci-tech service description text according to demand text, Gyx is used to generate demand text according to sci-tech service description text, Dx is used to judge whether a text is a demand description, and Dy is used to judge whether a text is a sci-tech service description.

[0069] Step 4: Setting learning rate, training round number and other network hyperparameters: the initial learning rate of the generator is 5e-4, the initial learning rate of the discriminator is 1e-4, and both of them can use AdamW optimizer.

[0070] Step 5: The word vector sequence of the training set is input into the CycleGAN network for training, and the test set is used to test the training effect. The CycleGAN model can obtain the style features of the texts in the two domains and realize the conversion of a piece of corpus in the two domains.

[0071] Step 6: Obtain the user input requirement description, and convert the requirement text to the corresponding technology resource description text through the CycleGAN model.

[0072] Step 7: Load the pre-trained CycleGAN network, calculate the semantic matching degree between the converted resource description text and the description text of the technology service resource in the database, and according to the similarity value, filter and sort the matched technology service resource list.

[0073] Step 8: Collect successful matching (demand corpus and technology service resource) data pairs in the user use process, and use such data pairs to further online supervised learning of the established CycleGAN model to optimize the matching effect. After collecting enough demand-technology service matching data in the production environment, the SBERT network can be further fine-tuned and optimized to strengthen its semantic similarity calculation effect in the technology service field. Through the collected actual data, the SBERT network structure is optimized, and the semantic matching effect is improved.

[0074] The model based on the CycleGAN structure is as shown in Figure 4 The CycleGAN model construction method can be as follows:

[0075] Step 1: input the requirement text x1 in the requirement domain into the generator Gxy to generate the technology service resource description text y1; that is, convert the description text from the requirement domain to the description text in the technology service domain; similarly, input the technology service description text y2 in the technology service domain into the generator Gyx to generate the requirement text x2.

[0076] Step 2: y1 and y2 in step 1 are input into discriminator Dy at the same time, Dy is used to determine whether yi (i = 1, 2, 3,..., n) is a real text or a text generated by the generator, this problem is regarded as a classification problem, if it is a real text, y2 is identified as 1, if it is a generated text, y1 is identified as 0, and the classification loss of this classification task is used as the adversarial loss of Gxy and Dy network in CycleGAN; Similarly, x1 and x2 are input into the discriminator Dx at the same time, whether x1 and x2 are real texts is determined, and the classification loss is calculated as the adversarial loss of Gyx and Dx network in CycleGAN. Further, the model is trained by gradient back propagation according to the total loss value, and the model parameters are optimized. The total adversarial loss can be shown as the GAN loss function mentioned later. The total loss value is the sum of all losses, which can be weighted. The specific weight is set before training, which is a hyperparameter and can be adjusted according to the actual experimental situation. The other two loss functions are a technique in the training process of CycleGAN, which can make the trained model better and more stable.

[0077] The loss function here is actually a loss function of a binary classification problem, which can be a binary classification loss function cross entropy. For a real text s, the goal of the discriminator is to determine it as 1, the closer the actual output value of the discriminator is to 1, the smaller the cross entropy output, the farther away from 1, the larger, indicating that the discriminator is too wrong; Conversely, for a generated text ss, the target output of the discriminator is 0, so the closer to 0, the smaller the loss.

[0078] Step 3: when the adversarial loss in step 2 is gradually stagnant during the optimization of the training process, stop the training of the CycleGAN model; At this time, it is indicated that the technology service resource description text y1 generated by the generator Gxy according to the demand text x1 is the same or similar in style to the real resource description text y2, and the demand text x2 generated by Gyx according to the technology service resource description text y2 is the same or similar in style to the real demand description text x1.

[0079] In one possible implementation, the word vector sequence of the training set is input into the CycleGAN network for training. The training loss function used is as follows:

[0080] GAN loss

[0081]

[0082] Cycle consistency loss:

[0083]

[0084] Full mapping loss:

[0085]

[0086] Total loss function:

[0087] L = L GAN (G, D, X, Y) + l c L cycle (G xy , G yx , X, Y) + l I L Identity (G xy , G yx , X, Y)

[0088] After training, save the model and parameters, and apply the generator Gxy in the network to the production environment, so that the user's demand description can be converted into the corresponding technology service resource description.

[0089] In a possible implementation, a Siamese neural network is used to calculate the semantic similarity of two texts. The subnetworks of the Siamese network (SBERT) model both use a BERT model, and the two BERT models share parameters. The semantic vectors of the two texts are calculated through the BERT network model, and then the semantic similarity of the two original texts is obtained by calculating the cosine similarity of the two semantic vectors.

[0090] As Figure 4 shown, when comparing the similarity of two sentences A and B, they are respectively input into the BERT network, and the output is two sets of vectors representing the sentences. Then the similarity of the two is calculated. Vector clustering can also be used to achieve unsupervised learning tasks using this principle. In addition to setting hyperparameters, the SBERT network can also be optimized by constructing training data.

[0091] It should be noted that the method for calculating the semantic matching degree of the text provided in the embodiments of the application includes but is not limited to using the SBERT network.

[0092] The method provided in the embodiments of the application mainly aims to solve the problem that paired (demand description and technology resource) data is difficult to collect and there is no ready-made data set. If the original demand description is directly used for retrieval, the effect is poor, because the demand is often not proposed by experts and scholars, and there is often a problem of different word usage from the professional technology resource description. Therefore, the method provided in the embodiments of the application converts the original user demand description into a text close to the technology service resource description, and then performs similarity operation to match. Through the mixed application of unsupervised learning and supervised learning, a demand matching system in the technology service field is established, which provides effective support for the technology service industry.

[0093] In summary, the embodiment of the present application provides a demand matching method, which receives a target demand description corpus input by a user; converts the target demand description corpus into a target technology resource corpus based on a generative adversarial network; calculates a matching degree of the target technology resource corpus and technology resource data in a technology resource database based on a convolutional generative neural network; and determines a target resource list of technology services according to a matching degree ranking and outputs the target resource list of technology services. The demand matching in the field of technology services is efficiently performed through cross-domain conversion of text corpora and mixed application of unsupervised learning and supervised learning, thereby providing effective support for the technology service industry.

[0094] Based on the same technical concept, the embodiment of the present application also provides a demand matching system, as shown in Figure 5 The system comprises:

[0095] An information receiving module 501 is configured to receive a target demand description corpus input by a user.

[0096] A corpus conversion module 502 is configured to convert the target demand description corpus into a target technology resource corpus based on a generative adversarial network.

[0097] A matching module 503 is configured to calculate a matching degree of the target technology resource corpus and technology resource data in a technology resource database based on a convolutional generative neural network.

[0098] An information output module 504 is configured to determine a target resource list of technology services according to a matching degree ranking and output the target resource list of technology services.

[0099] In a possible implementation, the matching module 503 is specifically configured to calculate semantic vectors of the target technology resource corpus and technology resource data in the technology resource database based on a convolutional generative neural network; and calculate a cosine similarity of the semantic vectors of the target technology resource corpus and each technology resource data in the technology resource database, respectively, to obtain the matching degree of the target technology resource corpus and the technology resource data in the technology resource database.

[0100] The embodiment of the present application also provides an electronic device corresponding to the method provided in the foregoing embodiments. Please refer to Figure 6 which shows a schematic diagram of an electronic device provided by some embodiments of the present application. The electronic device 20 can comprise a processor 200, a memory 201, a bus 202 and a communication interface 203, wherein the processor 200, the communication interface 203 and the memory 201 are connected through the bus 202; the memory 201 stores a computer program which can be run on the processor 200, and the processor 200 executes the computer program to perform the method provided in any one of the foregoing embodiments of the present application.

[0101] The memory 201 can include a random access memory (RAM) and can also include a non-volatile memory, such as at least one disk memory. The communication connection between the system network element and at least one other network element is realized through at least one physical port 203 (which can be wired or wireless), and the Internet, a wide area network, a local area network, a metropolitan area network, etc. can be used.

[0102] The bus 202 can be an ISA bus, a PCI bus, an EISA bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. The memory 201 is used to store programs, and the processor 200 executes the programs after receiving execution instructions. The method disclosed in any of the embodiments of the present application can be applied to the processor 200 or implemented by the processor 200.

[0103] The processor 200 can be an integrated circuit chip with signal processing capability. In the implementation process, each step of the above method can be completed by integrated logic circuits or instructions in the form of software in the processor 200. The processor 200 described above can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a ready-to-program gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. Each method, step and logic block disclosed in the embodiments of the present application can be implemented or executed. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in combination with the embodiments of the present application can be directly embodied as a hardware code processor for execution, or a combination of hardware and software modules in the code processor for execution. The software module can be located in a random access memory, a flash memory, a read-only memory, a programmable read-only memory, an electrically erasable programmable memory, a register, etc. The storage medium in the art. The storage medium is located in the memory 201, and the processor 200 reads the information in the memory 201 and combines the hardware to complete the steps of the above method.

[0104] The electronic device provided by the embodiments of the present application and the method provided by the embodiments of the present application have the same beneficial effects as the method they adopt, run or implement.

[0105] The embodiments of the present application also provide a computer readable storage medium corresponding to the method provided in the foregoing embodiments. Please refer to Figure 7The computer readable storage medium shown in the figure is an optical disc 30, on which a computer program (i.e. a program product) is stored, which, when run by a processor, will execute the method provided by any of the preceding embodiments.

[0106] It should be noted that examples of the computer readable storage medium can also include, but are not limited to, a phase change memory (PRAM), a static random access memory (SRAM), a dynamic random access memory (DRAM), other types of random access memory (RAM), a read only memory (ROM), an electrically erasable programmable read only memory (EEPROM), a flash memory, or other optical, magnetic storage medium, which will not be listed one by one here.

[0107] The computer readable storage medium provided by the above embodiments of the present application has the same beneficial effects as the method adopted, run or implemented by the application program stored therein, based on the same inventive concept.

[0108] It should be noted that:

[0109] The algorithms and displays presented herein are not inherently related to any particular computer, virtual apparatus, or other apparatus. Various general purpose apparatus can be used with these teachings or with variations thereof, depending on the particular application. The structure required to be taken for constructing such apparatus will be apparent to those of skill in the art from the above description. In addition, the present application is not described with reference to any particular programming language. It will be appreciated that a variety of programming languages can be used to implement the teachings of the application as described herein, and any references below to specific languages are provided for disclosure of enablement only.

[0110] In the description provided herein, numerous specific details are set forth. However, it is understood that embodiments of the application can be practiced without these specific details. In some instances, well-known methods, structures and techniques have not been described in detail in order to avoid obscuring the understanding of this description.

[0111] Similarly, it is to be understood that the above description is only illustrative of the application and certain examples thereof, and is subject to the prior art. Therefore, the techniques are not intended to be limited to the specific embodiments disclosed. Rather, it is intended that the scope of the application be defined by the claims appended hereto, which fully accrue to patent to be granted.

[0112] Those skilled in the art will appreciate that the modules in the apparatuses in the embodiments can be adapted and placed in one or more apparatuses other than the embodiments. The modules or units or components in the embodiments can be combined into one module or unit or component, and furthermore can be split into multiple sub-modules or sub-units or sub-components. Any combination of all the features disclosed in the specification (including the accompanying claims, abstract and drawings), and any method or device so disclosed, can be made unless expressly stated otherwise. Each feature disclosed in the specification (including the accompanying claims, abstract and drawings) can be replaced by alternative features providing the same, equivalent or similar functionality unless expressly stated otherwise.

[0113] Furthermore, those skilled in the art will appreciate that different embodiments of the application have different features and that some features of one embodiment can not be present in another embodiment. It should be noted that any of the features described in this specification (including the accompanying claims, abstract and drawings) can be replaced by alternative features serving the same, equivalent or similar purpose, unless expressly stated otherwise. Thus, unless expressly stated otherwise, each feature disclosed in this specification (including the accompanying claims, abstract and drawings) is considered to be an independent embodiment disclosable under this application.

[0114] The various component embodiments of the application can be implemented in hardware, or as software modules running in one or more processors, or in combinations thereof. Those skilled in the art will appreciate that a microprocessor or digital signal processor (DSP) can be used in practice to implement some or all of the functionality of some or all of the components of the apparatus for creating a virtual machine according to embodiments of the application. The application can also be implemented as a program (for example, a computer program and a computer program product) for executing any or all of the methods described herein on a device or apparatus. Such program(s) of the application can be stored on a computer readable medium, or can be in the form of one or more signals. Such signals can be downloaded from an Internet website, or provided on a carrier medium, or in any other form.

[0115] It should be noted that the foregoing examples have been provided merely for the purpose of explanation and are in no way to be construed as limiting of the present application. While the application has been described with reference to preferred embodiments and illustrations, the preferred embodiments and illustrations provided herein are not the only ways in which the present application can be implemented. Those skilled in the art will recognize that changes can be made in the preferred embodiments and illustrations without departing from the scope and spirit of the present application. In the claims, any reference signs placed between parentheses shall not be construed as limiting the claim. The word comprising does not exclude the presence of elements or steps not listed in a claim. The word a or an shall not exclude the presence of a plurality of elements, articles or steps, as used herein. The use of the words first, second and third, and / or the use of other numerals does not connote any order, but is used for naming purposes only. The use of the words "a" and "an" and "the" and "said" and "at least one" and "one or more" and "at least one member" and "one or more members" and "at least one element" and "one or more elements" and "at least one step" and "one or more steps" and "at least one component" and "one or more components" and "at least one part" and "one or more parts" and "at least one element" and "one or more elements" and "at least one member" and "one or more members" and "at least one" and "one or more" does not exclude the presence of zero or more elements or steps or components, articles or parts. The use of the words first and second to describe a variety of constituents does not imply that the first constituent must precede the second constituent. The use of the indefinite article does not preclude the presence of additional and / or earlier elements than can be explicitly mentioned.

[0116] The above description is implemented only as a preferred embodiment of the present application. Therefore, any changes or modifications made by those skilled in the art within the technical scope of the present application should be covered by the scope of the present application. Therefore, the scope of the present application should be defined by the scope of the claims.

Claims

1. A demand matching method, characterized in that, The method includes: Collect demand description corpora and technology resource description corpora, and divide them into training set and test set; The demand description corpus and the science and technology resource description corpus in the training set were processed by word segmentation and vectorization. The generative adversarial network is trained based on the constructed generative adversarial network, training parameters, and processed corpus. Receive user-input corpus describing target requirements; The target demand description corpus is converted into a target science and technology resource corpus based on generative adversarial networks; The matching degree between the target science and technology resource corpus and the science and technology resource data in the science and technology resource database is calculated based on the synaptic neural network. The list of target technology service resources is determined and output based on the matching degree. The method for constructing a generative adversarial network includes the following steps: constructing a generator and a discriminator, wherein the generator includes a first generator and a second generator, and the discriminator includes a first discriminator and a second discriminator; the first generator is used to generate a science and technology resource corpus based on a demand description corpus, and the second generator is used to generate a demand description corpus based on the science and technology resource corpus; the first discriminator is used to determine whether the demand description corpus is generated, and the second discriminator is used to determine whether the science and technology resource corpus is generated; the first demand description corpus is input into the first generator to generate a first science and technology resource corpus; the second science and technology resource corpus is input into the second generator to generate a second demand description corpus; the first science and technology resource corpus and the second... The second discriminator inputs a corpus of scientific and technological resources, outputting classification results for the first and second corpora of scientific and technological resources. Adversarial loss values ​​for the second discriminator and the first generator are calculated based on these classification results. The first and second demand description corpora are input into the first discriminator, outputting classification results for the first and second demand description corpora. Adversarial loss values ​​for the first discriminator and the second generator are calculated based on these classification results. The generative adversarial network (GAN) is trained using gradient backpropagation based on the total adversarial loss, adjusting its parameters to minimize this total adversarial loss. The total adversarial loss is calculated based on all adversarial loss values.

2. The method as described in claim 1, characterized in that, The calculation of the matching degree between the target science and technology resource corpus and the science and technology resource data in the science and technology resource database based on the spastic neural network includes: The semantic vectors of the target science and technology resource corpus and the science and technology resource data in the science and technology resource database are calculated based on the synaptic neural network. The cosine similarity between the semantic vector of the target science and technology resource corpus and the semantic vector of each science and technology resource data in the science and technology resource database is calculated to obtain the matching degree between the target science and technology resource corpus and the science and technology resource data in the science and technology resource database.

3. The method as described in claim 1, characterized in that, The method further includes: The generative adversarial network is optimized and supervised by collecting data pairs consisting of technology service resources from the user's selected technology service target resource list and the user's initial input target demand description corpus.

4. The method as described in claim 1, characterized in that, The method further includes: Several data pairs are collected, consisting of technology service resources from the user-selected technology service target resource list and technology resource data from the technology resource database, and then used to optimize and supervise the learning of the synaptic neural network.

5. A demand matching system, characterized in that, The system includes: A training module is used to collect demand description corpora and technology resource description corpora, and divide them into training and testing sets; the demand description corpora and technology resource description corpora in the training set are processed by word segmentation and vectorization; the generative adversarial network is trained based on the constructed generative adversarial network, training parameters, and processed corpora; the construction method of the generative adversarial network includes the following steps: constructing a generator and a discriminator, the generator including a first generator and a second generator, the discriminator including a first discriminator and a second discriminator; the first generator is used to generate technology resource corpora based on demand description corpora, and the second generator is used to generate demand description corpora based on technology resource corpora; the first discriminator is used to determine whether the demand description corpora are generated, and the second discriminator is used to determine whether the technology resource corpora are generated; the first demand description corpora are input into the first generator to generate the first demand description corpora. A first corpus of scientific and technological resources is used; a second corpus of scientific and technological resources is input into a second generator to generate a second corpus of demand descriptions; the first and second corpora of scientific and technological resources are input into a second discriminator, which outputs classification results for the first and second corpora of scientific and technological resources, and calculates adversarial loss values ​​for the second discriminator and the first generator based on the classification results; the first and second demand description corpora are input into the first discriminator, which outputs classification results for the first and second demand description corpora, and calculates adversarial loss values ​​for the first discriminator and the second generator based on the classification results; the generative adversarial network is trained by gradient backpropagation based on the total adversarial loss, and the parameters of the generative adversarial network are adjusted with the goal of minimizing the total adversarial loss; the total adversarial loss is calculated based on all adversarial loss values. The information receiving module is used to receive user-input corpus describing target requirements. The corpus conversion module is used to convert the target demand description corpus into target scientific and technological resource corpus based on generative adversarial networks; The matching module is used to calculate the matching degree between the target science and technology resource corpus and the science and technology resource data in the science and technology resource database based on the synaptic neural network. The information output module is used to determine and output a list of target technology service resources based on matching degree.

6. The system as described in claim 5, characterized in that, The matching module is specifically used for: The semantic vectors of the target science and technology resource corpus and the science and technology resource data in the science and technology resource database are calculated based on the synaptic neural network. The cosine similarity between the semantic vector of the target science and technology resource corpus and the semantic vector of each science and technology resource data in the science and technology resource database is calculated to obtain the matching degree between the target science and technology resource corpus and the science and technology resource data in the science and technology resource database.

7. An electronic device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the method as claimed in any one of claims 1-4.

8. A computer-readable storage medium, characterized in that, It stores computer-readable instructions that can be executed by a processor to implement the method as described in any one of claims 1-4.

Images