Technology transfer prediction method, model training method, and related programs and devices

Abstract

Embodiments of the present application provide a technology transfer prediction method, a model training method, and related programs and devices. The technology transfer prediction model training method includes: obtaining target technology data, the target technology data including a plurality of technology modules and technology ownership information corresponding to the technology modules; learning a target technology data embedding representation based on the target technology data; and training a technology transfer prediction model based on the target technology data embedding representation. The technology transfer prediction method includes predicting the probability of technology transfer between the technology modules and the technology ownership subjects using the technology transfer prediction model. The present application improves the technology transfer prediction effect and accuracy by constructing and training a technology transfer prediction model and then using the technology transfer prediction model to predict the probability of technology transfer.

Description

Technical Field

[0001] This application relates to the field of machine learning, specifically to a technology transfer prediction method, model training method, and related programs and devices. Background Technology

[0002] Technology transfer refers to the flow of technology between different organizations, countries, and industries through transfer, licensing, pledging, and other means. In recent years, the rapid development of machine learning technology has made it possible to efficiently and accurately identify technology transfer targets.

[0003] Existing technology transfer prediction methods are mostly based on the similarity calculation of statistical features. These methods rely excessively on the explicit feature information of the input and lack the learning of implicit features. Therefore, the prediction effect of existing technology transfer prediction methods needs to be improved. Summary of the Invention

[0004] In view of this, embodiments of this application provide a technology transfer prediction method, a model training method, and related procedures and equipment to improve the technology transfer prediction effect.

[0005] To achieve the above objectives, the embodiments of this application provide the following technical solutions.

[0006] In a first aspect, embodiments of this application provide a method for training a technology transfer prediction model, including:

[0007] Acquire target technology data, which includes multiple technology modules and technology ownership information corresponding to the technology modules. The technology ownership information includes the technology ownership entity corresponding to the technology module and the transfer and succession relationship of the technology module to different technology ownership entities.

[0008] Based on the target technology data, a target technology data embedding representation is learned using a machine learning algorithm. The target technology data embedding representation is used to indicate the technology ownership information, including the technology ownership subject corresponding to the technology module and the transfer and succession relationship of different technology ownership subjects to the technology module.

[0009] A technology transfer probability model is trained based on the target technology data embedding representation.

[0010] Secondly, embodiments of this application also provide a technology transfer prediction method, including using a technology transfer prediction model to predict the probability of technology transfer between a technology module and the technology owner, wherein the technology transfer prediction model is trained according to the technology transfer prediction model training method described in the first aspect.

[0011] Thirdly, embodiments of this application also provide a technology transfer prediction model training device, comprising: a target technology data acquisition module for acquiring target technology data; a target technology data embedding representation learning module for learning the target technology data to obtain a target technology data embedding representation; a target technology data embedding representation training module for training a technology transfer prediction model based on the target technology data embedding representation; and a technology transfer prediction module for predicting the occurrence of technology transfer between the technology module and the technology owner based on the technology transfer prediction model.

[0012] Fourthly, embodiments of this application also provide a storage medium, wherein the storage medium stores one or more computer-executable instructions, which, when executed, implement the technology transfer prediction model training method as described in the first aspect or the technology transfer prediction method as described in the second aspect.

[0013] Fifthly, embodiments of this application also provide a computer program product, wherein the computer program product includes one or more computer-executable instructions, and when the one or more computer-executable instructions are executed, they implement the technology transfer prediction model training method as described in the first aspect or the technology transfer prediction method as described in the second aspect.

[0014] The technology transfer prediction model training method provided in this application first acquires target technology data, then learns the technology modules in the target technology data, the technology owners corresponding to the technology modules, and the transfer succession relationships of different technology owners to the technology modules, to obtain a target technology data embedding representation. The target technology data embedding representation is used to indicate that the technology ownership information includes the technology owners corresponding to the technology modules and the transfer succession relationships of different technology owners to the technology modules. Then, based on the target technology data embedding representation, a technology transfer prediction model is trained. Finally, the technology transfer prediction model is used to predict the probability of technology transfer between the technology module and the technology owner, thereby providing a direction for training the technology transfer prediction model, realizing the construction of the technology transfer prediction model, and enabling technology transfer prediction based on the trained technology transfer prediction model, thus improving the effectiveness of technology transfer prediction. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0016] Figure 1 This is a flowchart of a technology transfer prediction model training method provided in an embodiment of this application.

[0017] Figure 2 This is a flowchart of obtaining the embedded representation of the target technology data provided in the embodiments of this application.

[0018] Figure 3 This is a schematic diagram of obtaining the embedded representation of the target technology data provided in an embodiment of this application.

[0019] Figure 4 This is another schematic diagram illustrating the embedded representation of the target technology data provided in the embodiments of this application.

[0020] Figure 5 This is another schematic diagram illustrating the embedded representation of the target technical data provided in the embodiments of this application.

[0021] Figure 6 This is a flowchart illustrating the verification of the technology transfer prediction model provided in this application embodiment.

[0022] Figure 7 This is a flowchart of the technology transfer prediction method provided in the embodiments of this application. Detailed Implementation

[0023] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0024] In the process of technology transfer, there are two main entities: technology modules and technology owners. These entities have multiple attributes, and there are various relationships between different entities. How to predict technology transfer based on the attributes of the entities themselves and the relationships between different entities is an important issue in this field.

[0025] This application provides a method for training a technology transfer prediction model, the method being able to... Figure 1 The method flow shown is implemented. (Refer to...) Figure 1 The method includes the following steps:

[0026] Step S100: Obtain target technology data. The target technology data includes multiple technology modules and technology ownership information corresponding to the technology modules. The technology ownership information includes the technology ownership entity corresponding to the technology module and the transfer and succession relationship of the technology module by different technology ownership entities.

[0027] In the process of technology transfer, technology modules are transferred from the party transferring the technology, i.e., one or more technology owners, to the party receiving the technology, i.e., one or more technology owners.

[0028] In step S100, the technology module is the object transferred during the technology transfer process. In a specific example, the technology module may be an invention, a utility model, etc. The technology ownership information describes the ownership relationship between the module and the technology ownership entity. For example, when the technology module is an invention or a utility model, the technology ownership information may include the patent title, abstract, applicant, patentee, applicant's province, etc.

[0029] It should be noted that the method for determining the transfer and succession relationship is that when a technology module is transferred from one or more technology owners of the transferring technology to one or more technology owners of the receiving technology, the one or more technology owners of the transferring technology have a transfer and succession relationship with the technology module. Similarly, the one or more technology owners of the receiving technology also have a transfer and succession relationship with the technology module.

[0030] Furthermore, in step S100, the target technology data is technology data containing technology modules and their corresponding technology attribution information. Here, a technology module and its corresponding technology attribution information can be understood as a sample. The target technology data should include multiple samples to enable prediction of the transfer of different technology modules.

[0031] Furthermore, depending on the number of technology owners of the transferring or receiving technologies, the transfer and succession relationships of the technology modules by different technology owners include: transfer relationships between different technology owners and technology modules, receiving relationships between different technology owners and technology modules, common ownership relationships between different technology owners, common transfer relationships between different technology owners, common receiving relationships between different technology owners, and social relationships between different technology owners. Specifically, when one party transferring the technology is multiple technology owners, all technology owners of the transferring party have a common transfer relationship; when one party receiving the technology is multiple technology owners, all technology owners of the receiving party have a common receiving relationship; when a single technology module corresponds to multiple technology owners, all technology owners corresponding to that technology module have a common ownership relationship; when a single technology module undergoes technology transfer between different technology owners, all technology owners of the transferring and receiving technologies have a social relationship.

[0032] In the technology transfer prediction model training method provided in the embodiments of this application, step S100 is to obtain target technology data to provide training data for the technology transfer prediction model. For example, when the technology module is an invention, the target technology data can be obtained through an invention patent database.

[0033] In some implementations, step S100: the step of acquiring target technical data may include:

[0034] Based on the technology ownership information corresponding to the multiple technology modules, obtain the single technology ownership relationship corresponding to the technology module and the transfer and succession relationship between different technology ownership entities; obtain the target technology data based on the transfer and succession relationship.

[0035] To improve the performance of the technology transfer prediction model, the acquisition of target data also includes further processing of the technology ownership information corresponding to the multiple technology modules. Specifically, the acquisition of the single technology ownership relationship corresponding to the technology module and the transfer succession relationship between different technology ownership entities includes:

[0036] For each technology module belonging to multiple technology ownership entities, the corresponding technology ownership information is split according to the technology ownership entity to obtain the single technology ownership relationship corresponding to the technology module. The single technology ownership relationship corresponding to the technology module indicates a one-to-one ownership relationship between the technology module and the technology ownership entity. For each transfer and continuation of the same technology module by different technology ownership entities, the transfer and continuation relationships between different technology ownership entities are split according to the technology ownership entity to obtain the transfer and continuation relationships between different technology ownership entities. The transfer and continuation relationships between different technology ownership entities indicate a one-to-one transfer and continuation relationship between the technology ownership entities.

[0037] Here, a single technology attribution relationship is a set of one-to-one attribution relationships between the technology module and the technology attribution subject. In the process of learning the target technology data embedding representation based on the target technology data, for some machine learning algorithms used in this process, such as graph convolutional networks, graph attention networks, and other graph neural networks, splitting the technology attribution information corresponding to the technology module into one-to-one attribution relationships between the technology module and the technology attribution subject can effectively improve the efficiency and effectiveness of these machine learning algorithms.

[0038] In some implementations, the technical module is a patent that has been transferred, and the subject of the technology ownership is the applicant, the transferor, and the transferee. For example, when patent A once belonged to transferor B and now belongs to transferee C, that is, when patent A has ownership relationships with both transferor B and transferee C in the technology ownership information corresponding to patent A, it is split into two single technology ownership relationships, namely, patent A has an ownership relationship with transferor B, and patent A has an ownership relationship with transferee C.

[0039] Similarly, for some machine learning algorithms, such as graph convolutional networks, graph attention networks, and other graph neural networks, breaking down the transfer and succession relationships of the different technology owners to the technology modules into one-to-one transfer and succession relationships between the technology owners can effectively improve the efficiency and effectiveness of these machine learning algorithms.

[0040] In some implementations, the technical module is a patent that has been transferred, and the technology owner is the applicant, the transferor, and the transferee. For example, when three technology owners, transferor B, transferor C, and transferor D, jointly own the patent right of patent A, and these three technology owners transfer patent A to other technology owners, that is, when the three technology owners, transferor B, transferor C, and transferor D, have a transfer succession relationship, it is split into a one-to-one transfer succession relationship between the three technology owners, that is, transferor B and transferor C have a one-to-one transfer succession relationship, transferor B and transferor D have a one-to-one transfer succession relationship, and transferor C and transferor D have a one-to-one transfer succession relationship.

[0041] Please continue to refer to this. Figure 1 After step S100, i.e., obtaining the target technology data, step S200 is executed: based on the target technology data, a target technology data embedding representation is learned. The target technology data embedding representation is used to indicate that the technology attribution information includes the technology attribution entity corresponding to the technology module and the transfer and succession relationship of different technology attribution entities to the technology module.

[0042] Machine learning algorithms are a class of algorithms that automatically analyze data to obtain patterns and use these patterns to predict unknown data. In the technology transfer prediction model training method provided in this application, a machine learning algorithm is used to analyze the target technology data and learn an embedding representation of the target technology data. In machine learning, an embedding representation is a mapping from high-dimensional data to a low-dimensional space. Its advantage lies in its ability to transform high-dimensional discrete data into a low-dimensional continuous representation, enabling the model to better learn the relationships between data.

[0043] Figure 2This is a flowchart illustrating the process of obtaining an embedded representation of target technology data according to an embodiment of this application. In some implementations, such as... Figure 2 As shown, step S200 may include the following steps:

[0044] Step S210: Based on the transfer and succession relationship of the technology modules among the different technology ownership entities, a first connection relationship embedding representation is learned using a machine learning algorithm.

[0045] The first connection relationship embedding representation is obtained based on the transfer and succession relationship of the technology modules between the different technology ownership entities, and it is used to indicate the transfer and succession relationship between the different technology ownership entities.

[0046] Specifically, in some embodiments, the technology module is a patent that has been transferred, and the technology owner is the applicant, the transferor, and the transferee. The first connection relationship embedding representation can be obtained by learning through a neural network based on the transfer and succession relationship of the technology module between the different technology owners.

[0047] Obtain the first connection relationship embedding representation The function is defined as follows:

[0048]

[0049] The transfer and succession relationship of the technology modules among the different technology ownership entities is represented by graph data, where A(i) is the technology ownership entity u. i The neighbor set f in the transfer and succession relationship of the technology module among the different technology ownership entities. ik Indicates the subject of technology ownership u i Technology ownership subject u k The transfer and continuation relationship between them, Aggre social It is the first connection aggregation function. Additionally, W and b are the weights and biases of the neural network, respectively, and σ is the non-linear activation function.

[0050] In the process of transferring and continuing the technology module among different technology owners, there are different relationship types, such as joint application, joint transfer, and joint acquisition, denoted as uus. These different relationship types can capture the influence of different neighbors on the technology owners, which is helpful for learning the social embedding representation of the social space. To model the different relationship types uus, an embedding representation e for the transfer and continuation relationship type is introduced. uus ∈R h Regarding the ownership of the technology, u i With the technology owner u k Social interactions involving specific relationship types are fused and embedded using a multilayer perceptron. ik Modeling as the subject of technology ownership uk The original embedding representation x k Embedded representation of transition continuation relationship type e uus The combination of these, this function is denoted as g. uus As shown in the following formula:

[0051]

[0052] In the above formula This represents the operator for concatenating two vectors.

[0053] Furthermore, considering that different neighbors have varying degrees of influence on the technology ownership entity, different weights should be assigned to the transfer succession relationships between different technology ownership entities. Therefore, a graph attention network is used to implement the first connection aggregation function Aggre. social This yields the final first connection embedding representation. As shown in the following formula.

[0054]

[0055]

[0056]

[0057] Among them, a ik For the technology owner u k Regarding the ownership of the technology, u i First connection relationship embedding representation The attention weight of the contribution For the attention score obtained through a two-layer perceptron, Normalization is performed to obtain a ik .

[0058] In the above method for obtaining the first connection relationship embedding representation, in the technical scenario of the patent, the technical module is a patent that has been transferred, and the subject of the technology is the applicant, the transferor and the transferee.

[0059] The purpose of step S210 is to convert the transfer and succession relationships of the technology modules between the different technology ownership entities into data that is easier to train the technology transfer prediction model, thereby improving the model's prediction performance.

[0060] The input data for mainstream graph neural network models consists of user-user social graphs and user-item interaction graphs. This approach suffers from insufficient embedded information for user and item nodes, as well as issues with high-order connectivity between items and item nodes, resulting in performance limitations. Therefore, in some implementations, to enrich the technology attribution information and improve the performance of the technology transfer prediction model, before step S210: learning the first connection relationship embedding representation using a machine learning algorithm based on the transfer succession relationship of the technology modules between different technology attribution entities, the following steps are also included:

[0061] Based on the aforementioned multiple technical modules, the similarity relationships between different technical modules are determined.

[0062] The similarity relationship is used to indicate the degree of similarity between different technical modules.

[0063] In such Figure 1 In the technology transfer prediction model training method implemented by the method flow shown, the technology attribution information only includes the technology attribution subject corresponding to the technology module, and the transfer succession relationship of different technology attribution subjects to the technology module. This limits the effectiveness of the technology transfer prediction method. By supplementing the similarity relationship between different technology modules, the technology attribution information can be enriched, thereby improving the efficiency and prediction accuracy of the technology transfer prediction method.

[0064] Based on the similarity relationships between the different technical modules, a similarity relationship embedding representation is learned using machine learning algorithms.

[0065] The similarity relationship embedding representation is obtained based on the similarity relationship learning between the different technical modules, and it is used to indicate the similarity relationship between the different technical modules.

[0066] In some implementations, the technical module is a patent that has been transferred, and the technology belongs to the applicant, the transferor, and the transferee. Therefore, the title and abstract text data of the patent can be segmented into words, each word can be labeled with its part of speech, and stop words that are useless for text classification can be removed to obtain the terminology for each patent. Then, a pre-trained model is used to process the pre-processed title and abstract text data of the patent to obtain the semantic feature vector of each patent. The cosine similarity calculation method is then used to calculate the similarity between patents, and the similarity between patents is divided into different similarity levels based on the similarity, ultimately obtaining the embedded representation of the similarity relationship.

[0067] Specifically, in some implementations, the similarity relationship embedding representation can be obtained by learning through a neural network based on the similarity relationship between the different technical modules.

[0068] The similarity relationship embedding representation is obtained. The function is defined as follows:

[0069]

[0070] The similarity relationships between the different technical modules are represented by graph data, where D(j) is the similarity between technical module p. j The neighbor set, z, in the similarity relationships between the different technical modules jl Represents the technology module p j For technology module p l Similarity relationships are represented by Aggre similarity It is a similarity aggregation function. Furthermore, W and b are the weights and biases of the neural network, respectively, and σ is a non-linear activation function.

[0071] In the similarity relationships between different technical modules, there are different similarity levels, denoted as pps, of 1, 2, 3, 4, and 5. These different similarity levels can capture the influence of different neighbors on the technical modules, which helps in learning the similarity relationship embedding representation. To model the different similarity levels pps, an embedding representation e for the similarity level is introduced. pps ∈R h For technical module p j With technical module p l A specific similarity level between them is used to represent the similarity relationship z through a multilayer perceptron. jl Modeling as a technology module p l Original embedding representation y l Embedding representation of similarity level e pps The combination of these, this function is denoted as g. pps As shown in the following formula:

[0072]

[0073] in, This represents the operator for concatenating two vectors.

[0074] Furthermore, considering that different neighbors have varying degrees of influence on the similarity relationships between technical modules, different weights should be assigned to the similarity relationships between different technical modules. Therefore, a graph attention network is used to implement the similarity aggregation function Aggre. Similarity This yields the final similarity relation embedding representation. As shown in the following formula:

[0075]

[0076]

[0077]

[0078] Where, η jl For technical module p l Regarding the ownership of the technology by the subject p j Similarity embedding representation The attention weight of the contribution For the attention score obtained through a two-layer perceptron, Normalization is performed to obtain η jl .

[0079] In the above-described method for obtaining the embedded representation of similar relationships, in the context of patent technology, the technology module is a patent that has been transferred, and the technology owner is the applicant, the transferor, and the transferee.

[0080] In some implementations, to further enrich the technology attribution information and improve the performance of the technology transfer prediction model, before step S210: learning the first connection relationship embedding representation using a machine learning algorithm based on the transfer succession relationship of the technology modules between different technology attribution entities, the following steps are also included:

[0081] Based on multiple technical modules, a technical module attribute embedding representation is obtained, which includes: the technical theme of the technical module and the regional attribute of the technical module.

[0082] The obtained technical module attribute embedding representation includes: based on the technical module, using semantic feature extraction and clustering methods to extract the technical topic of the technical module, and taking the region where the technical module is located as the regional attribute of the technical module.

[0083] In the field of machine learning, semantic feature extraction can extract important features from data. Clustering methods are algorithms that involve grouping data. For example, when the technical module is an invention or utility model, one method for extracting the technical theme of the technical module is to first use the SBERT (Sentence-BERT) method to extract semantic sentence vectors from the title and abstract of the invention or utility model, and then use K-means clustering to obtain the technical theme of the invention or utility model. Here, SBERT (Sentence-BERT) is a semantic similarity calculation and sentence vector generation method based on the BERT (Bidirectional Encoder Responses from Transformers) model, and K-means clustering is a commonly used clustering algorithm.

[0084] Based on the technology attribution information, a technology attribution subject attribute embedding representation is obtained, which includes: the technology theme of the technology attribution subject and the regional attribute of the technology attribution subject.

[0085] The process of obtaining the embedded representation of the technology ownership subject attribute includes: based on the technology ownership information, using semantic feature extraction and clustering to extract the technology topic of the technology ownership subject, and taking the location of the technology ownership subject as the regional attribute of the technology ownership subject.

[0086] In a specific example, when the technical module is an invention or utility model, the technical attribution information can be a province, city, district / county, etc. Preferably, the province where the technical attribution entity is located can be used as the regional attribute of the technical attribution entity.

[0087] After step S210, step S220 is executed: based on the technology ownership subject corresponding to the technology module, a machine learning algorithm is used to learn the transition relationship embedding representation and the technology module embedding representation.

[0088] The transfer succession relationship embedding representation and the technology module embedding representation are both learned based on the technology ownership subject corresponding to the technology module. The transfer succession relationship embedding representation is used to indicate the transfer succession relationship of different technology ownership subjects to the technology module, and the technology module embedding representation is used to indicate the ownership relationship between the technology module and its corresponding technology ownership subject.

[0089] The purpose of step S220 is basically the same as that of step S210, which is to convert the technology ownership subject corresponding to the technology module into data that is easier to train for technology transfer prediction model training.

[0090] Given that the similarity relationship embedding representation is introduced into the technology transfer prediction model training method, step S220: based on the transfer succession relationship between different technology ownership entities and technology modules, the transfer succession relationship embedding representation and technology module embedding representation are learned using machine learning algorithms, including the following steps:

[0091] Based on the transfer and succession relationships between the different technology owners and technology modules, machine learning algorithms are used to learn the embedded representation of the transfer and succession relationship and the embedded representation of the second connection relationship.

[0092] The second connection relationship embedding is used to indicate the transfer and succession relationship between the different technology ownership entities and technology modules.

[0093] Specifically, in some implementations, the transfer and continuation relationship embedding representation and the second connection relationship embedding representation can be obtained by learning through a neural network based on the transfer and continuation relationship between the different technology ownership entities and technology modules.

[0094] The embedded representation of the transition continuation relationship is obtained here. The function is defined as follows:

[0095]

[0096] In this context, the transfer and succession relationships between different technology owners and technology modules are represented as graph data, where B(i) represents the technology owner u. i The neighbor set, h, in the transfer and succession relationships between different technology owners and technology modules. ij Indicates the subject of technology ownership u i For technology module p j The transfer and continuation relationship indicates that Aggre patent This is the aggregation function for the technical modules. Additionally, W and b are the weights and biases of the neural network, respectively, and σ is the non-linear activation function.

[0097] In the interaction between a technology subject and a patent, there are different relationship types, such as application, transfer, and acquisition, denoted as upi. By understanding these different relationship types, we can capture the influence of different neighbors on the technology subject, which helps in learning the embedding representation of transfer and succession relationships. To model the different relationship types upi, we introduce an embedding representation e for interaction relationship types. upi ∈R h For the technology subject u i With technical module p j Interactions of specific relationship types between them are represented by a multilayer perceptron to indicate the transitional continuation relationship h. ij Modeling as a technology module p j The original embedding representation y j Embedded representation of interaction relationship type e upi The combination of these, this function is denoted as g. upi As shown in the following formula:

[0098]

[0099] In the above formula, This represents the operator for concatenating two vectors.

[0100] Furthermore, considering that different neighbors have varying degrees of influence on the transfer and succession relationships between technology owners and technology modules, different weights should be assigned to different technology owners and technology modules for these transfer and succession relationships. Therefore, a graph attention network is used to implement the technology module aggregation function Aggre. patentThis yields the final embedding representation of the transition and continuation relationship. As shown in the following formula.

[0101]

[0102]

[0103]

[0104] Where, β ij For technical module p j Regarding the ownership of the technology, u i The embedding representation of the transition and continuation relationship The attention weight of the contribution For the attention score obtained through a two-layer perceptron, Normalization is performed to obtain β ij .

[0105] In the above-described method for obtaining the embedded representation of the transfer and succession relationship, in the technical context of a patent, the technical module is a patent that has been transferred, and the subject of ownership of the technology is the applicant, the transferor, and the transferee.

[0106] And, the second connection relationship embedding representation is obtained. The function is defined as follows:

[0107]

[0108] Among them, the transfer and succession relationships between different technology owners and technology modules are represented by graph data, where C(j) is the patent p. j The neighbor set w in the transfer and succession relationship between different technology owners and technology modules ji Indicates patent p j Regarding the ownership of the technology, u i Attribution relationship, Aggre user This is the aggregation function representing the technology attribution subject. Additionally, W and b are the weights and biases of the neural network, respectively, and σ is the non-linear activation function.

[0109] In the process of technology transfer and patent continuation, there are different relationship types, such as application, assignment, and transferee, denoted as upi. Different relationship types can capture the influence of different neighbors on the patent, which is helpful for learning the second-connection relationship embedding representation. To model the different relationship types upi, an embedding representation e of the interaction relationship type is introduced. upi ∈R h For technical module p j With the technology owner u i Interactions of specific relationship types between them are represented by a multilayer perceptron to indicate attribution relationships.ji Modeling as the subject of technology ownership u i The original embedding representation x i Embedded representation of transition continuation relationship type e upi The combination of these, this function is denoted as g. upi As shown in the following formula:

[0110]

[0111] In the above formula, This represents the operator for concatenating two vectors.

[0112] Furthermore, considering that the influence of transfer and succession relationships between different technology owners and technology modules varies, different weights should be assigned to these relationships. Therefore, a graph attention network is used to implement the second connection aggregation function Aggre. user This yields the final second connection embedding representation. As shown in the following formula.

[0113]

[0114]

[0115]

[0116] in, For the technology owner u i For technology module p j The second connection relationship embedding representation The attention weight of the contribution For the attention score obtained through a two-layer perceptron, Normalization is performed to obtain a ik .

[0117] In the above method for obtaining the second connection relationship embedding representation, in the technical scenario of the patent, the technical module is a patent that has been transferred, and the subject of the technology ownership is the applicant, the transferor and the transferee.

[0118] Given that the technology module attribute embedding representation and technology ownership subject attribute embedding representation are introduced in the technology transfer prediction model training method, step S220: based on the transfer succession relationship between different technology ownership subjects and technology modules, the learning of the transfer succession relationship embedding representation and technology module embedding representation using machine learning algorithms may include:

[0119] Based on the transfer and succession relationships between the different technology owners and technology modules, machine learning algorithms are used to learn the embedded representation of the transfer and succession relationship and the embedded representation of the second connection relationship.

[0120] Wherein, when the similarity relationship embedding representation does not exist, after step S221, the following is performed: the second connection relationship embedding representation and the technology module attribute embedding representation are trained through a third aggregator to obtain the technology module embedding representation; when the similarity relationship embedding representation exists, after step S221, the following is performed: the second connection relationship embedding representation, the technology module attribute embedding representation, and the similarity relationship embedding representation are trained through a third aggregator to obtain the technology module embedding representation.

[0121] The above steps demonstrate that the third aggregator can be trained normally to obtain the embedding representation of the technology module, regardless of whether the similarity relationship embedding representation exists.

[0122] Please continue to refer to this. Figure 2 After step S220, step S230 is executed: the first connection relationship embedding representation and the transfer succession relationship embedding representation are trained by the first aggregator to obtain the technology attribution subject embedding representation.

[0123] During model training, for multimodal data, using an aggregator to merge the embedding representations of different modalities into a more comprehensive multimodal embedding representation is a common method for mapping multimodal data to a low-dimensional space, with the aim of improving the model's prediction performance. In step S220, the first connection relationship embedding representation and the transition continuation relationship embedding representation need to be trained through the first aggregator to obtain the technology attribution subject embedding representation.

[0124] Given that the technology module attribute embedding representation and the technology ownership subject attribute embedding representation are introduced in the technology transfer prediction model training method, step S230: concatenating the first connection relationship embedding representation and the transfer succession relationship embedding representation through a first aggregator to obtain the technology ownership subject embedding representation includes the following steps:

[0125] The first connection relationship embedding representation, the technology ownership subject attribute embedding representation, and the transfer succession relationship embedding representation are trained through the first aggregator to obtain the technology ownership subject embedding representation.

[0126] After step S230, step S240 is executed: the technology attribution subject embedding representation and the technology module embedding representation are trained by the second aggregator to obtain the target technology data embedding representation.

[0127] Similar to the reasoning in step S230, in order to improve the prediction performance of the model, the technology attribution subject embedding representation and the technology module embedding representation need to be trained through the second aggregator to obtain the target technology data embedding representation.

[0128] Figure 3 This is a schematic diagram illustrating how to obtain the embedded representation of the target technology data according to an embodiment of this application. The flowchart provides a clearer understanding of how the embedded representation of the target technology data is obtained. Figure 3 As shown, the first connection relationship embedding representation and the transfer succession relationship embedding representation are trained by the first aggregator to obtain the technology ownership subject embedding representation; the technology ownership subject embedding representation and the technology module embedding representation are trained by the second aggregator to obtain the target technology data embedding representation.

[0129] Furthermore, Figure 4 This is another schematic diagram illustrating the acquisition of the target technology data embedding representation provided in this application embodiment. Based on this flowchart, it can be more clearly understood how the target technology data embedding representation is obtained after introducing the similarity relationship embedding representation. For example... Figure 4 As shown, the first connection relationship embedding representation and the transfer succession relationship embedding representation are trained by the first aggregator to obtain the technology attribution subject embedding representation; the second connection relationship embedding representation and the similarity relationship embedding representation are trained by the third aggregator to obtain the technology module embedding representation; the technology attribution subject embedding representation and the technology module embedding representation are trained by the second aggregator to obtain the target technology data embedding representation.

[0130] Furthermore, Figure 5 This is another schematic diagram illustrating the process of obtaining the target technology data embedding representation provided in this application embodiment. Based on this flowchart, it can be more clearly understood how the target technology data embedding representation is obtained after introducing the technology module attribute embedding representation and the technology attribution subject attribute embedding representation. For example... Figure 5As shown, the first connection relationship embedding representation, the technology attribution subject attribute embedding representation, and the transfer succession relationship embedding representation are trained through a first aggregator to obtain the technology attribution subject embedding representation; the second connection relationship embedding representation, the technology module attribute embedding representation, and the similarity relationship embedding representation are trained through a third aggregator to obtain the technology module embedding representation; the technology attribution subject embedding representation and the technology module embedding representation are trained through a second aggregator to obtain the target technology data embedding representation. Alternatively, the above method may not include the similarity relationship embedding representation. In the absence of the similarity relationship embedding representation, when training through the third aggregator, only the second connection relationship embedding representation and the technology module attribute embedding representation need to be concatenated to obtain the technology module embedding representation.

[0131] Step S300: Based on the target technology data embedding representation, a technology transfer prediction model is trained.

[0132] In processing the target technology data embedding representation, machine learning algorithms were also used for training to obtain a technology transfer prediction model. The target technology data embedding representation includes technology module embedding representation and technology ownership entity embedding representation.

[0133] After step S300 is completed, the technology transfer prediction model has been trained based on the target technology data in this embodiment of the application. Furthermore, with the similarity relationship embedding representation introduced in the technology transfer prediction model training method, the performance of the technology transfer prediction model is stronger than that of the technology transfer prediction model trained without the similarity relationship embedding representation.

[0134] With the introduction of technology module attribute embedding representation and technology ownership subject attribute embedding representation in the training method of the technology transfer prediction model, the performance of the technology transfer prediction model is stronger than that of the technology transfer prediction model trained without introducing technology module attribute embedding representation and technology ownership subject attribute embedding representation.

[0135] With the inclusion of similarity relationship embedding representation, technology module attribute embedding representation, and technology ownership subject attribute embedding representation in the training method of the technology transfer prediction model, the performance of the technology transfer prediction model is stronger than that of the technology transfer prediction model trained by introducing only similarity relationship embedding representation or technology module attribute embedding representation and technology ownership subject attribute embedding representation alone.

[0136] In some implementations, the training method for the technology transfer prediction model further enhances the information richness of the technology transfer prediction system by introducing similarity relationships between different technology modules, embedding representations of technology module attributes, and embedding representations of technology ownership attributes, thereby further improving the prediction effect and accuracy of technology transfer.

[0137] To verify the performance, effectiveness, and universality of the technology transfer prediction model, in some embodiments, the model can also be tested, including the following steps, please refer to [reference needed]. Figure 6 , Figure 6 This is a flowchart illustrating the verification of the technology transfer prediction model provided in this application embodiment.

[0138] Step S500: Select a baseline model, and use the baseline model and the technology transfer prediction model to perform technology transfer prediction on at least two different patent datasets respectively.

[0139] The baseline model mentioned in step S500 is a simple and easy-to-implement benchmark model used for comparison with more complex models. In some implementations, the baseline model can be one or more of the following: a content-based technology transfer prediction model, a collaborative filtering-based technology transfer prediction model, and a hybrid method-based technology transfer prediction model.

[0140] After step S500, continue with step S510.

[0141] Step S510: Adjust the model parameters according to the prediction results, so that the evaluation indicators of the baseline model and the technology transfer prediction model reach the optimal values ​​under the model, wherein the evaluation indicators include at least: accuracy, precision, recall and F1 score.

[0142] The specific content of the evaluation indicators in step S510 is as follows.

[0143] Accuracy is the ratio of the number of correctly predicted samples to the total number of samples. The formula for accuracy is: Accuracy = (True Positives + True Negatives) / (True Positives + False Positives + True Negatives + False Negatives). Accuracy is suitable for situations where the data categories are evenly distributed, and the importance of positive and negative samples is relatively balanced. In this case, accuracy is a commonly used evaluation metric that can intuitively reflect the model's accuracy in classifying the overall sample.

[0144] Precision refers to the proportion of samples that a model predicts to be positive but are actually positive. The formula for precision is: Precision = True Positives / (True Positives + False Positives). Here, true positives are the number of samples correctly predicted as positive, and false positives are the number of negative samples incorrectly predicted as positive.

[0145] Recall is the proportion of positive samples correctly predicted by a model out of all positive samples. The formula for recall is: Recall = True Positives / (True Positives + False Negatives). Recall is suitable for scenarios where the model's coverage of positive samples is important. When the focus is on the proportion of positive samples correctly identified by the model, recall is a key evaluation metric.

[0146] The F1 score combines precision and recall, serving as a comprehensive metric for evaluating model performance. It is the harmonic mean of precision and recall, taking into account both accuracy and coverage. The formula for calculating the F1 score is: F1 score = 2 * (Precision * Recall) / (Precision + Recall). The F1 score is suitable for scenarios requiring a comprehensive evaluation of both model accuracy and coverage. When both precision and recall are important considerations, the F1 score can serve as a more holistic evaluation metric.

[0147] After the evaluation indicators of the baseline model and the technology transfer prediction model reach the optimal values ​​under the model, continue to execute steps S520 and S530.

[0148] Step S520: Based on the different patent datasets, perform paired t-tests on the evaluation metrics of the baseline model and the technology transfer prediction model under different patent datasets to verify the effectiveness of the technology transfer prediction model.

[0149] Step S530: Based on the different patent datasets, perform variance analysis on the evaluation indicators of the baseline model and the technology transfer prediction model under different patent datasets to verify the universality of the technology transfer prediction model.

[0150] In some implementations, based on the above-described verification steps, the technology transfer prediction model trained using the technology transfer prediction model training method provided in the embodiments of this application was verified. The verification results showed that the evaluation index of the technology transfer prediction model was higher than that of the baseline model, especially the recall rate was significantly higher than that of the baseline model, and the effectiveness and universality of the technology transfer prediction model were also verified.

[0151] This application also provides a technology transfer prediction method, please refer to... Figure 7 , Figure 7 This is a flowchart of the technology transfer prediction method provided in the embodiments of this application.

[0152] Step S600: Obtain the trained technology transfer prediction model.

[0153] The technology transfer prediction model is trained according to the technology transfer prediction model training method described in one of the preceding embodiments.

[0154] To perform technology transfer prediction, it is first necessary to obtain the technology transfer prediction model trained according to the technology transfer prediction model training method provided in the embodiments of this application.

[0155] Step S601: Using the technology transfer prediction model, predict the probability of technology transfer between the technology module and the technology owner.

[0156] When using the technology transfer prediction model to predict technology transfer, the prediction results output by the technology transfer prediction model include at least the probability of technology transfer between the technology module and different technology owners.

[0157] After obtaining the trained technology transfer prediction model, the technology modules to be predicted and the information of the technology owners are input into the technology transfer prediction model to obtain the prediction results output by the technology transfer prediction model. The prediction results include at least the probability of technology transfer between the technology modules and different technology owners.

[0158] This application improves the accuracy of technology transfer prediction by constructing and training a technology transfer prediction model and then using the model to predict the probability of technology transfer occurring.

[0159] This application embodiment also provides a technology transfer prediction model training device, including:

[0160] The target technology data acquisition module is used to acquire target technology data;

[0161] The target technology data embedding representation learning module is used to learn the target technology data to obtain the target technology data embedding representation.

[0162] The target technology data embedding representation training module is used to train a technology transfer prediction model based on the target technology data embedding representation.

[0163] The technology transfer prediction module is used to predict the occurrence of technology transfer between the technology module and the technology owner based on the technology transfer prediction model.

[0164] This application also provides a storage medium, wherein the storage medium stores one or more computer-executable instructions, and when the one or more computer-executable instructions are executed, they implement the technology transfer prediction model training method or technology transfer prediction method provided in this application.

[0165] This application also provides a computer program product, wherein the computer program product includes one or more computer-executable instructions, and when the one or more computer-executable instructions are executed, they implement the technology transfer prediction model training method or technology transfer prediction method provided in this application.

[0166] The foregoing describes multiple embodiment schemes provided by the embodiments of this application. The optional methods described in each embodiment scheme can be combined and cross-referenced with each other without conflict, thereby extending to a variety of possible embodiment schemes. These can all be considered as the embodiment schemes disclosed and published by the embodiments of this application.

[0167] While this application discloses the above information, it is not limited thereto. Any person skilled in the art can make various modifications and alterations without departing from the spirit and scope of this application; therefore, the scope of protection of this application shall be determined by the scope defined in the claims.

Claims

1. A method for training a technology transfer prediction model, characterized in that, include: Acquire target technology data, which includes multiple technology modules and corresponding technology ownership information. The technology ownership information includes the technology ownership entity corresponding to each technology module, and the transfer and succession relationships between different technology ownership entities regarding the technology modules. These transfer and succession relationships include: transfer relationships between different technology ownership entities and technology modules; transferee relationships between different technology ownership entities and technology modules; joint ownership relationships between different technology ownership entities; joint transfer relationships between different technology ownership entities; joint transferee relationships between different technology ownership entities; and social relationships between different technology ownership entities. Based on the target technology data, a target technology data embedding representation is learned using a machine learning algorithm. This target technology data embedding representation is used to indicate the technology attribution information, including the technology attribution entity corresponding to the technology module and the transfer and succession relationship of different technology attribution entities to the technology module. The step of learning the target technology data embedding representation using a machine learning algorithm based on the target technology data includes: Based on the transfer and succession relationship of the technology module between the different technology owners, a first connection relationship embedding representation is learned using a machine learning algorithm; wherein, the first connection relationship embedding representation is learned based on the transfer and succession relationship of the technology module between the different technology owners, and is used to indicate the transfer and succession relationship between the different technology owners. Based on the technology ownership entity corresponding to the technology module, a transfer succession relationship embedding representation and a technology module embedding representation are learned using a machine learning algorithm; the technology module embedding representation is used to indicate the ownership relationship of the technology ownership entity corresponding to the technology module. The first connection relationship embedding representation and the transfer succession relationship embedding representation are trained by the first aggregator to obtain the technology attribution subject embedding representation; The target technology data embedding representation is obtained by training the technology attribution subject embedding representation and the technology module embedding representation through a second aggregator. A technology transfer prediction model is trained based on the target technology data embedding representation.

2. The technology transfer prediction model training method according to claim 1, characterized in that, The acquisition of target technology data includes: Based on the technology ownership information corresponding to the multiple technology modules, obtain the single technology ownership relationship corresponding to the technology module and the transfer and succession relationship between different technology ownership entities; Obtain target technology data based on transfer and succession relationships.

3. The technology transfer prediction model training method according to claim 2, characterized in that, The acquisition of the single technology ownership relationship corresponding to the technology module and the transfer and succession relationship between different technology ownership entities includes: The technology attribution information corresponding to each technology module belonging to multiple technology attribution entities is split according to the technology attribution entity to obtain the single technology attribution relationship corresponding to the technology module, wherein the single technology attribution relationship corresponding to the technology module indicates the one-to-one attribution relationship between the technology module and the technology attribution entity; The transfer and continuation relationships of the technology modules corresponding to different technology owners for each technology module transfer and continuation are split according to the technology owner to obtain the transfer and continuation relationships between different technology owners, wherein the transfer and continuation relationships between different technology owners indicate a one-to-one transfer and continuation relationship between the technology owners.

4. The technology transfer prediction model training method according to claim 1, characterized in that, Before learning the first connection relationship embedding representation using a machine learning algorithm based on the transfer and succession relationship of the technology modules between the different technology ownership entities, the method further includes: Based on the multiple technical modules, the similarity relationships between different technical modules are determined, and the similarity relationships are used to indicate the degree of similarity between different technical modules; Based on the similarity relationships between the different technical modules, a similarity relationship embedding representation is learned using machine learning algorithms.

5. The technology transfer prediction model training method according to claim 4, characterized in that, The step of learning, using machine learning algorithms, the embedded representations of the transfer relationships and the embedded representations of the technology modules based on the transfer and succession relationships between the different technology owners and technology modules includes: Based on the transfer and succession relationships between the different technology owners and technology modules, a machine learning algorithm is used to learn the transfer and succession relationship embedding representation and the second connection relationship embedding representation; wherein, the second connection relationship embedding representation is used to indicate the transfer and succession relationships between the different technology owners and technology modules; The second connection embedding representation and the similarity embedding representation are trained by a third aggregator to obtain the technology module embedding representation.

6. The technology transfer prediction model training method according to claim 4, characterized in that, Before learning the first connection embedding representation using a machine learning algorithm, the method further includes: Based on multiple technical modules, a technical module attribute embedding representation is obtained, which includes: the technical theme of the technical module and the regional attribute of the technical module; Based on the technology attribution information, a technology attribution subject attribute embedding representation is obtained, which includes: the technology theme of the technology attribution subject and the regional attribute of the technology attribution subject.

7. The technology transfer prediction model training method according to claim 6, characterized in that: The obtained technical module attribute embedding representation includes: based on the technical module, using semantic feature extraction and clustering methods to extract the technical topic of the technical module, and taking the location of the technical module as the regional attribute of the technical module; The method of obtaining the embedded representation of the technology ownership subject attribute includes: based on the technology ownership information, using semantic feature extraction and clustering methods to extract the technology topic of the technology ownership subject, and taking the location of the technology ownership subject as the regional attribute of the technology ownership subject.

8. The technology transfer prediction model training method according to claim 6, characterized in that, The step of training the first connection relationship embedding representation and the transition continuation relationship embedding representation through the first aggregator to obtain the technology attribution subject embedding representation includes: The first connection relationship embedding representation, the technology ownership subject attribute embedding representation, and the transfer succession relationship embedding representation are trained through the first aggregator to obtain the technology ownership subject embedding representation; The step of learning, using machine learning algorithms, the embedded representations of the transfer relationships and the embedded representations of the technology modules based on the transfer and succession relationships between the different technology owners and technology modules includes: Based on the transfer and succession relationships between the different technology ownership entities and technology modules, machine learning algorithms are used to learn the transfer and succession relationship embedding representation and the second connection relationship embedding representation; Specifically, when the similarity relationship embedding representation does not exist, before training the technology attribution subject embedding representation and the technology module embedding representation through the second aggregator to obtain the target technology data embedding representation, the second connection relationship embedding representation and the technology module attribute embedding representation are trained through the third aggregator to obtain the technology module embedding representation; when the similarity relationship embedding representation exists, before training the technology attribution subject embedding representation and the technology module embedding representation through the second aggregator to obtain the target technology data embedding representation, the second connection relationship embedding representation, the technology module attribute embedding representation, and the similarity relationship embedding representation are trained through the third aggregator to obtain the technology module embedding representation.

9. The method for training a technology transfer prediction model according to any one of claims 1-8, characterized in that, The technical module is a patent that has been transferred, and the subject of ownership of the technology is the applicant, the transferor, and the transferee.

10. The method for training a technology transfer prediction model according to any one of claims 1-8, characterized in that, The machine learning algorithm is a graph attention network.

11. The method for training a technology transfer prediction model according to any one of claims 1-8, characterized in that, The training method for the technology transfer prediction model obtained from the training is a multilayer perceptron.

12. A technology transfer prediction method, characterized in that, This includes using a technology transfer prediction model to predict the probability of technology transfer between a technology module and the technology owner, wherein the technology transfer prediction model is trained using the technology transfer prediction model training method according to any one of claims 1-11.

13. A technology transfer prediction model training device, characterized in that, include: A target technology data acquisition module is used to acquire target technology data. The target technology data includes multiple technology modules and technology ownership information corresponding to the technology modules. The technology ownership information includes the technology ownership entity corresponding to the technology module, and the transfer and succession relationships of different technology ownership entities to the technology module. The transfer and succession relationships of different technology ownership entities to the technology module include: transfer relationships between different technology ownership entities and technology modules, transferee relationships between different technology ownership entities and technology modules, common ownership relationships between different technology ownership entities, common transfer relationships between different technology ownership entities, common transferee relationships between different technology ownership entities, and social relationships between different technology ownership entities. A target technology data embedding representation learning module is used to learn target technology data to obtain a target technology data embedding representation; the target technology data embedding representation is used to indicate the technology attribution information, including the technology attribution entity corresponding to the technology module and the transfer and succession relationship of different technology attribution entities to the technology module; wherein, the target technology data embedding representation learning module, for learning target technology data to obtain a target technology data embedding representation, includes: Based on the transfer and succession relationship of the technology module between the different technology owners, a first connection relationship embedding representation is learned using a machine learning algorithm; wherein, the first connection relationship embedding representation is learned based on the transfer and succession relationship of the technology module between the different technology owners, and is used to indicate the transfer and succession relationship between the different technology owners. Based on the technology ownership entity corresponding to the technology module, a transfer succession relationship embedding representation and a technology module embedding representation are learned using a machine learning algorithm; the technology module embedding representation is used to indicate the ownership relationship of the technology ownership entity corresponding to the technology module. The first connection relationship embedding representation and the transfer succession relationship embedding representation are trained by the first aggregator to obtain the technology attribution subject embedding representation; The target technology data embedding representation is obtained by training the technology attribution subject embedding representation and the technology module embedding representation through a second aggregator. The target technology data embedding representation training module is used to train a technology transfer prediction model based on the target technology data embedding representation. The technology transfer prediction module is used to predict the occurrence of technology transfer between the technology module and the technology owner based on the technology transfer prediction model.

14. A storage medium, wherein, The storage medium stores one or more computer-executable instructions, which, when executed, implement the technology transfer prediction model training method as described in any one of claims 1-11, or execute the technology transfer prediction method as described in claim 12.

15. A computer program product, wherein, The computer program product includes one or more computer-executable instructions, which, when executed, implement the technology transfer prediction model training method as described in any one of claims 1-11, or execute the technology transfer prediction method as described in claim 12.

Images