Technology promotion system and method based on Internet

Abstract

The invention relates to the technical field of technology promotion, and discloses a technology promotion system and method based on the Internet, and the method comprises the steps: obtaining the technology information of historical transaction promotion, carrying out the data preprocessing of the technology information, extracting a key information point, and making a historical sample data set; the method comprises the following steps: constructing a DeepFM preliminary model by using TensorFlow; the DeepFM preliminary model is optimized, and an optimized DeepFM model is obtained; inputting the historical sample data set into the optimized DeepFM model for training to obtain a technology promotion model; stock technical information is obtained, the stock technical information is input into the technical promotion model to be processed, and weighted technical information of different weight factors is output; sequencing the weighted technical information according to the weight factors from high to low, and performing exposure promotion on the technical information according to the sequence of sequencing; the technical information is accurately recommended, the problem of effective popularization of different kinds of technical information is solved, and the recommendation efficiency and the technical matching degree are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of technology promotion, and in particular to a technology promotion system and method based on the Internet. Background Art

[0002] With the rapid development and popularization of information technology in recent years, the technology promotion industry has also ushered in new development opportunities. The current situation of technology promotion shows a trend of diversification and rapid development. The market size of the technology promotion service industry has continued to grow in recent years and is expected to maintain a steady growth trend in the next few years. The technology promotion industry helps enterprises improve production efficiency, reduce operating costs and enhance market competitiveness by continuously introducing new technologies and new methods. All sectors of society encourage technological innovation and transformation of results, providing strong support for the development of the technology promotion industry. At the same time, the influx of social capital has also injected new vitality into the technology promotion industry and promoted the rapid development of the industry.

[0003] In terms of technology promotion, although all sectors of society attach great importance to it, there are still many problems in actual operation. For example, some new technologies lack corresponding publicity, and traditional publicity methods cannot enable technology suppliers and demand enterprises to match information and establish communication channels. At the same time, the unreasonable composition of technology promotion personnel, incomplete configuration, poor hardware configuration, and the need to improve the overall business quality have also affected the quality and efficiency of technology promotion. Summary of the invention

[0004] The purpose of the present invention is to solve the above problems and to design a technology promotion system and method based on the Internet.

[0005] The first aspect of the present invention provides a technology promotion method based on the Internet, the method comprising the following steps:

[0006] Obtain technical information of historical transaction promotions, perform data preprocessing on the technical information, extract key information points, and create a historical sample data set;

[0007] Use TensorFlow to build a preliminary DeepFM model. The model consists of two parts: the deep neural network Deep and the factorization machine FM. The outputs of the two parts are connected through a fully connected layer for prediction.

[0008] Optimize the DeepFM preliminary model to obtain the optimized DeepFM model;

[0009] Input the historical sample data set into the optimized DeepFM model for training to obtain the technology promotion model;

[0010] Obtain the existing technology information, input the existing technology information into a technology promotion model for processing, and output weighted technology information with different weight factors;

[0011] Sort the weighted technology information according to the weight factors from high to low, and promote the technology information in the order of sorting.

[0012] Optionally, in the first implementation manner of the first aspect of the present invention, obtaining the technology information of historical transaction promotion, performing data preprocessing on the technology information, extracting key information points, and making a historical sample data set, including:

[0013] Obtain the technology information of historical transaction promotion, including technology basic information, technology transaction information, technology evaluation information, intellectual property information, technology transfer implementation information, legal and policy information, market competition information, and risk information;

[0014] Process the technology information using a hash algorithm to remove duplicate data, so that each technology information is unique in the data set;

[0015] Use natural language processing technology to denoise the technology information and remove irrelevant information;

[0016] According to business requirements and technology characteristics, select and extract features that have an important impact on the technology promotion effect;

[0017] Divide the extracted key information points into a training set, a validation set, and a test set according to a certain proportion.

[0018] Optionally, in the second implementation manner of the first aspect of the present invention, using TensorFlow to build a preliminary DeepFM model, including:

[0019] Define the input layer of the model, and the input shape matches the number of features of the data set;

[0020] For sparse features, convert them into dense vector representations through an embedding layer;

[0021] Build an FM layer, including first-order features and second-order interaction features. The first-order features are implemented through a simple linear layer, and the second-order interaction features are obtained through dot product operations on the embedding vectors;

[0022] Build a DNN layer, a multi-layer neural network, which is implemented by stacking multiple fully connected layers. The number of neurons and activation functions in each layer are hyperparameters;

[0023] Merge the outputs of the FM layer and the DNN layer, which is achieved by adding them together. The merged output can be used as the final output of the model;

[0024] Define the output layer, use the sigmoid activation function and one neuron to generate the predicted value.

[0025] Optionally, in the third implementation manner of the first aspect of the present invention, the optimization of the DeepFM preliminary model includes:

[0026] Add an scSE model introducing visual attention between the embedding layer and the DNN input layer of the DeepFM preliminary model;

[0027] The scSE model adds the output results of the cSE module and the sSE module along the channel dimension to obtain the final information correction result;

[0028] The scSE model enhances the ability of the convolutional neural network to capture the channel and spatial importance of the feature map.

[0029] Optionally, in the fourth implementation manner of the first aspect of the present invention, the inputting the historical sample data set into the optimized DeepFM model for training to obtain the technology promotion model includes:

[0030] Obtain the historical sample data set, use the Pandas tool of Python to build a data preprocessing pipeline, convert the cleaned and formatted data into the model input format, and input it into the optimized DeepFM model;

[0031] Train the model on the training set and update the model parameters through the backpropagation algorithm;

[0032] Adopt a three-layer neural network structure, with a batch size of 512, the number of epoch iterations set to 40, a learning rate of 0.001, the optimizer selected as adam, and the embedding dimension of the embedding layer being 6;

[0033] Monitor the training process of the model through the validation set;

[0034] Export the trained model into a deployable format to obtain the technology promotion model.

[0035] Optionally, in the fifth implementation manner of the first aspect of the present invention, the sorting of the weighted technical information according to the weight factor from high to low, and the exposure and promotion of the technical information in the order of sorting include:

[0036] Sort the weighted technical information according to the weight factor from high to low, rank the technical information with a high weight factor first, and gradually arrange the technical information in the order of decreasing weight factor;

[0037] Formulate a promotion strategy according to the characteristics of the technical information and the needs of the target audience;

[0038] Promote the technologies with high weight factors in a way of high exposure, high traffic and multiple platforms.

[0039] The second aspect of the present invention provides an Internet-based technology promotion system, which includes a data acquisition module, a model construction module, a model optimization module, a model training module, a technology output module, and a technology promotion module, where:

[0040] The data acquisition module is used to acquire the technology information of historical transaction promotion, perform data preprocessing on the technology information, extract key information points, and make a historical sample data set;

[0041] The model construction module is used to build a preliminary DeepFM model using TensorFlow. The model is divided into two parts: a deep neural network Deep and a factorization machine FM. The outputs of the two parts are connected through a fully connected layer for prediction;

[0042] The model optimization module is used to optimize the preliminary DeepFM model to obtain an optimized DeepFM model;

[0043] The model training module is used to input the historical sample data set into the optimized DeepFM model for training to obtain a technology promotion model;

[0044] The technology output module is used to acquire the stock technology information, input the stock technology information into the technology promotion model for processing, and output weighted technology information with different weight factors;

[0045] The technology promotion module is used to sort the weighted technology information according to the weight factor from high to low, and promote the technology information for exposure in the order of the sorting.

[0046] Optionally, in the first implementation manner of the second aspect of the present invention, the data acquisition module includes a historical data collection sub-module, a data processing sub-module, a denoising processing sub-module, a feature extraction sub-module, and a data set generation sub-module, where:

[0047] The historical data collection sub-module is used to acquire the technology information of historical transaction promotion, including technology basic information, technology transaction information, technology evaluation information, intellectual property information, technology transfer implementation information, legal and policy information, market competition information, and risk information;

[0048] The data processing sub-module is used to process the technology information using a hash algorithm, perform data deduplication processing, and make each technology information unique in the data set;

[0049] The denoising processing sub-module is used to use natural language processing technology to perform denoising processing on the technology information and remove irrelevant information;

[0050] A feature extraction sub-module, configured to extract features that have an important impact on the technology promotion effect according to business requirements and technical characteristics;

[0051] A dataset generation sub-module, configured to divide the extracted key information points into a training set, a validation set, and a test set according to a ratio.

[0052] Optionally, in the second implementation manner of the second aspect of the present invention, the model building module includes an input layer creation sub-module, an embedding layer building sub-module, an FM layer building sub-module, a DNN layer building sub-module, an output merging sub-module, and an output layer definition sub-module, where

[0053] The input layer creation sub-module is configured to define the input layer of the model, and the shape of the input matches the number of features of the dataset;

[0054] The embedding layer building sub-module is configured to convert sparse features into dense vector representations through the embedding layer;

[0055] The FM layer building sub-module is configured to build an FM layer, including first-order features and second-order interaction features. The first-order features are implemented through a simple linear layer, and the second-order interaction features are obtained through dot product operations on the embedding vectors;

[0056] The DNN layer building sub-module is configured to build a DNN layer, a multi-layer neural network, which is implemented by stacking multiple fully connected layers. The number of neurons and activation functions in each layer are hyperparameters;

[0057] The output merging sub-module is configured to merge the outputs of the FM layer and the DNN layer, which is implemented by addition. The merged output can be used as the final output of the model;

[0058] The output layer definition sub-module is configured to define the output layer, use a sigmoid activation function and a neuron to generate prediction values.

[0059] Optionally, in the third implementation manner of the second aspect of the present invention, the model training module includes a data input sub-module, a training sub-module, a parameter setting sub-module, a validation sub-module, and an export sub-module, where

[0060] The data input sub-module is configured to obtain a historical sample dataset, build a data preprocessing pipeline using the Pandas tool in Python, convert the cleaned and formatted data into the model input format, and input it into the optimized DeepFM model;

[0061] The training sub-module trains the model on the training set and updates the model parameters through the backpropagation algorithm;

[0062] A parameter setting sub-module, which uses a three-layer neural network structure, with a batch size of 512, the number of epoch iterations set to 40, a learning rate of 0.001, the optimizer selected as adam, and the embedding dimension of the embedding layer is 6;

[0063] A verification sub-module, which monitors the training process of the model through a validation set;

[0064] An export sub-module, which exports the trained model into a deployable format to obtain a technology promotion model.

[0065] In the technical solution provided by the present invention, by obtaining the technical information of historical transaction promotion, performing data preprocessing on the technical information, extracting key information points, and making a historical sample data set; using TensorFlow to build a preliminary DeepFM model; optimizing the preliminary DeepFM model to obtain an optimized DeepFM model; inputting the historical sample data set into the optimized DeepFM model for training to obtain a technology promotion model; obtaining the stock technical information, inputting the stock technical information into the technology promotion model for processing, and outputting weighted technical information with different weight factors; sorting the weighted technical information from high to low according to the weight factors, and promoting the technical information for exposure in the order of sorting; the present invention makes accurate recommendations for technical information, solves the problem of effective promotion of different technical information, and improves the recommendation efficiency and technical matching degree. Description of the Drawings

[0066] By reading the detailed description of the preferred embodiments below, various other advantages and benefits will become clear to those of ordinary skill in the art. The drawings are only for the purpose of showing the preferred embodiments and are not considered to be a limitation of the present invention.

[0067] Figure 1 It is a schematic diagram of the first embodiment of the Internet-based technology promotion method provided by the embodiment of the present invention;

[0068] Figure 2 It is a schematic diagram of the second embodiment of the Internet-based technology promotion method provided by the embodiment of the present invention;

[0069] Figure 3 It is a schematic diagram of a structure of the Internet-based technology promotion system provided by the embodiment of the present invention. Detailed Embodiments

[0070] In the description of the present invention, the claims, and the above-mentioned drawings, the terms "first", "second", "third", "fourth", etc. (if any) are used to distinguish similar objects and do not necessarily describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the term "comprising" or "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, device, product, or equipment that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products, or equipment.

[0071] For ease of understanding, the specific process of the embodiments of the present invention will be described below. Please refer to Figure 1 The schematic diagram of the first embodiment of the Internet-based technology promotion method provided by the embodiments of the present invention specifically includes the following steps:

[0072] Step 101: Obtain the technical information of historical successful promotions, perform data preprocessing on the technical information, extract key information points, and create a historical sample data set;

[0073] Step 102: Use TensorFlow to build a preliminary DeepFM model. The model is divided into two parts: a deep neural network Deep and a factorization machine FM. The outputs of the two parts are connected through a fully connected layer for prediction;

[0074] In this embodiment, the input layer of the model is defined, and the shape of the input matches the number of features of the data set; for sparse features, they are converted into dense vector representations through an embedding layer; an FM layer is built, including first-order features and second-order interaction features. The first-order features are implemented through a simple linear layer, and the second-order interaction features are obtained by performing a dot product operation on the embedding vectors; a DNN layer is built, a multi-layer neural network implemented by stacking multiple fully connected layers. The number of neurons and activation functions in each layer are hyperparameters; the outputs of the FM layer and the DNN layer are merged by addition, and the merged output can be used as the final output of the model; the output layer is defined, using a sigmoid activation function and a neuron to generate a predicted value.

[0075] Step 103: Optimize the preliminary DeepFM model to obtain an optimized DeepFM model;

[0076] In this embodiment, an scSE model introducing visual attention is added between the embedding layer of the initial DeepFM model and the input layer of the DNN; the scSE model adds the output results of the cSE module and the sSE module along the channel dimension to obtain the final information correction result; the scSE model enhances the ability of the convolutional neural network to capture the channel and spatial importance of the feature map.

[0077] Step 104: Input the historical sample dataset into the optimized DeepFM model for training to obtain a technology promotion model.

[0078] In this embodiment, a historical sample dataset is obtained, and the data preprocessing pipeline is constructed using the Pandas tool in Python. The cleaned and formatted data is converted into the model input format and input into the optimized DeepFM model; the model is trained on the training set, and the model parameters are updated through the backpropagation algorithm; a three-layer neural network structure is adopted, the batch size batchsize is 512, the epoch iteration times are set to 40, the learning rate is 0.001, the optimizer is selected as adam, and the embedding dimension of the embedding layer is 6; the training process of the model is monitored through the validation set; the trained model is exported in a deployable format to obtain a technology promotion model.

[0079] Step 105: Obtain the stock technology information, input the stock technology information into the technology promotion model for processing, and output the weighted technology information with different weight factors.

[0080] Step 106: Sort the weighted technology information in descending order of the weight factor, and promote the technology information in the order of the sorting.

[0081] In this embodiment, the weighted technology information is sorted in descending order of the weight factor, and the technology information with a high weight factor is ranked first, and the technology information is gradually arranged in the order of decreasing weight factor.

[0082] According to the characteristics of the technology information and the needs of the target audience, formulate a promotion strategy.

[0083] Promote the technology with a high weight factor in a way of high exposure, high traffic, and multiple platforms.

[0084] In the embodiments of the present invention, by obtaining the technical information of historical transaction promotion, performing data preprocessing on the technical information, extracting key information points, and making a historical sample data set; using TensorFlow to build a preliminary DeepFM model; optimizing the preliminary DeepFM model to obtain an optimized DeepFM model; inputting the historical sample data set into the optimized DeepFM model for training to obtain a technology promotion model; obtaining the stock technical information, inputting the stock technical information into the technology promotion model for processing, and outputting weighted technical information with different weight factors; sorting the weighted technical information according to the weight factors from high to low, and promoting the exposure of the technical information in the order of sorting; the present invention makes accurate recommendations for technical information, solves the problem of effective promotion of different technical information, and improves the recommendation efficiency and technical matching degree.

[0085] Please refer to Figure 2 , the schematic diagram of the second embodiment of the Internet-based technology promotion method provided by the embodiments of the present invention, the method includes:

[0086] Step 201, obtain the technical information of historical transaction promotion, including technical basic information, technical transaction information, technical evaluation information, intellectual property information, technology transfer implementation information, legal and policy information, market competition information, and risk information;

[0087] Step 202, process the technical information by using the hash algorithm, perform deduplication processing on the data, and make each technical information unique in the data set;

[0088] Step 203, use natural language processing technology to denoise the technical information and remove irrelevant information;

[0089] Step 204, select and extract the features that have an important impact on the technology promotion effect according to business requirements and technical characteristics;

[0090] Step 205, divide the extracted key information points into a training set, a validation set, and a test set according to a ratio.

[0091] Please refer to Figure 3 , the schematic diagram of a structure of the Internet-based technology promotion system provided by the embodiments of the present invention, the system includes a data acquisition module, a model building module, a model optimization module, a model training module, a technology output module, and a technology promotion module, wherein:

[0092] The data acquisition module 301 is used to obtain the technical information of historical transaction promotion, perform data preprocessing on the technical information, extract key information points, and make a historical sample data set;

[0093] The model building module 302 is used to build a preliminary DeepFM model using TensorFlow. The model is divided into two parts: a deep neural network Deep and a factorization machine FM. The outputs of the two parts are connected through a fully connected layer for prediction;

[0094] The model optimization module 303 is used to optimize the preliminary DeepFM model to obtain an optimized DeepFM model;

[0095] The model training module 304 is used to input the historical sample dataset into the optimized DeepFM model for training to obtain a technology promotion model;

[0096] The technology output module 305 is used to obtain the stock technology information, input the stock technology information into the technology promotion model for processing, and output the weighted technology information with different weight factors;

[0097] The technology promotion module 306 is used to sort the weighted technology information in descending order of the weight factors, and expose and promote the technology information in the order of the sorting.

[0098] In this embodiment, the data acquisition module 301 includes a historical data collection sub-module, a data processing sub-module, a denoising processing sub-module, a feature extraction sub-module, and a dataset generation sub-module, where:

[0099] The historical data collection sub-module is used to obtain the technology information of historical transaction promotion, including technology basic information, technology transaction information, technology evaluation information, intellectual property information, technology transfer implementation information, law and policy information, market competition information, and risk information;

[0100] The data processing sub-module is used to process the technology information using a hash algorithm, remove duplicates from the data, and make each technology information unique in the dataset;

[0101] The denoising processing sub-module is used to use natural language processing technology to denoise the technology information and remove irrelevant information;

[0102] The feature extraction sub-module is used to select and extract features that have an important impact on the technology promotion effect according to business requirements and technology characteristics;

[0103] The dataset generation sub-module is used to divide the extracted key information points into a training set, a validation set, and a test set according to a ratio.

[0104] In this embodiment, the model building module 302 includes a create input layer sub-module, a build embedding layer sub-module, a build FM layer sub-module, a build DNN layer sub-module, a merge output sub-module, and a define output layer sub-module, where,

[0105] Create an input layer sub-module to define the input layer of the model, where the shape of the input matches the number of features in the dataset;

[0106] Build an embedding layer sub-module to convert sparse features into dense vector representations through the embedding layer for sparse features;

[0107] Build an FM layer sub-module to build the FM layer, including first-order features and second-order interaction features. The first-order features are implemented through a simple linear layer, and the second-order interaction features are obtained through dot product operations on the embedding vectors;

[0108] Build a DNN layer sub-module to build the DNN layer, a multi-layer neural network implemented by stacking multiple fully connected layers. The number of neurons and activation functions in each layer are hyperparameters;

[0109] Merge the output sub-module to merge the outputs of the FM layer and the DNN layer, which is achieved by addition. The merged output can be used as the final output of the model;

[0110] Define an output layer sub-module to define the output layer, which uses a sigmoid activation function and one neuron to generate prediction values.

[0111] In this embodiment, 304 includes a data input sub-module, a training sub-module, a parameter setting sub-module, a validation sub-module, and an export sub-module. Among them,

[0112] The data input sub-module is used to obtain the historical sample dataset, build a data preprocessing pipeline using the Pandas tool in Python, convert the cleaned and formatted data into the model input format, and input it into the optimized DeepFM model;

[0113] The training sub-module trains the model on the training set and updates the model parameters through the backpropagation algorithm;

[0114] The parameter setting sub-module is used to adopt a three-layer neural network structure, with a batch size of 512, the number of epoch iterations set to 40, the learning rate of 0.001, the optimizer selected as adam, and the embedding dimension of the embedding layer is 6;

[0115] The validation sub-module monitors the training process of the model through the validation set;

[0116] The export sub-module exports the trained model into a deployable format to obtain a technology promotion model.

[0117] Precisely recommend technical information, solve the problem of effective promotion of different technical information, and improve the recommendation efficiency and technical matching degree.

[0118] The foregoing has shown and described the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments, and the above embodiments and the descriptions in the specification are only preferred examples of the present invention, which are not used to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of the present invention claimed is defined by the appended claims and their equivalents.

Claims

1. A technology promotion method based on the Internet, characterized in that The Internet-based technology promotion method includes the following steps: Obtain the technical information of historical successful promotions, perform data preprocessing on the technical information, extract key information points, and produce a historical sample data set; Use TensorFlow to build a preliminary DeepFM model. The model is divided into two parts: a deep neural network Deep and a factorization machine FM. The outputs of the two parts are connected through a fully connected layer for prediction; Optimize the preliminary DeepFM model to obtain an optimized DeepFM model; Input the historical sample data set into the optimized DeepFM model for training to obtain a technology promotion model; Obtain the inventory technical information, input the inventory technical information into the technology promotion model for processing, and output weighted technical information with different weight factors; Sort the weighted technical information in descending order of the weight factors, and promote the exposure of the technical information in the order of the sorting.

2. The method for promoting technology based on the Internet according to claim 1, characterized in that, The obtaining of the technical information of historical successful promotions, performing data preprocessing on the technical information, extracting key information points, and producing a historical sample data set includes: Obtain the technical information of historical successful promotions, including basic technical information, technical transaction information, technical evaluation information, intellectual property information, technology transfer implementation information, legal and policy information, market competition information, and risk information; Use the hash algorithm to process the technical information, remove duplicate data, and ensure that each piece of technical information in the data set is unique; Use natural language processing technology to denoise the technical information and remove irrelevant information; According to business requirements and technical characteristics, select and extract features that have an important impact on the technology promotion effect; Divide the extracted key information points into a training set, a validation set, and a test set according to a certain proportion.

3. The method for promoting technology based on the Internet according to claim 1, wherein The building of the preliminary DeepFM model using TensorFlow includes: Define the input layer of the model, and the input shape matches the number of features in the data set; For sparse features, convert them into dense vector representations through an embedding layer; Build an FM layer, including first-order features and second-order interaction features. The first-order features are implemented through a simple linear layer, and the second-order interaction features are obtained through dot product operations on the embedding vectors; Build a DNN layer, a multi-layer neural network implemented by stacking multiple fully connected layers. The number of neurons and activation functions in each layer are hyperparameters; Merge the outputs of the FM layer and the DNN layer, which is achieved by adding them together. The merged output can be used as the final output of the model; Define the output layer, use the sigmoid activation function and a single neuron to generate a prediction value.

4. A method for promoting technology based on the Internet according to claim 1, characterized in that The optimization of the preliminary DeepFM model includes: Add an scSE model with visual attention between the embedding layer and the DNN input layer of the preliminary DeepFM model; The scSE model adds the output results of the cSE module and the sSE module along the channel dimension to obtain the final information correction result; The scSE model enhances the ability of the convolutional neural network to capture the channel and spatial importance of the feature map.

5. A method for promoting technology based on the Internet according to claim 1, characterized in that, Inputting the historical sample data set into the optimized DeepFM model for training to obtain a technology promotion model, including: Obtain the historical sample data set, use the Pandas tool in Python to build a data preprocessing pipeline, convert the cleaned and formatted data into the model input format, and input it into the optimized DeepFM model; Train the model on the training set and update the model parameters through the backpropagation algorithm; Adopt a three-layer neural network structure, with a batch size of 512, the number of epoch iterations set to 40, a learning rate of 0.001, the optimizer selected as adam, and the embedding dimension of the embedding layer is 6; Monitor the training process of the model through the validation set; Export the trained model into a deployable format to obtain a technology promotion model.

6. A method for promoting technologies based on the Internet according to claim 1, characterized in that, Sorting the weighted technical information from high to low according to the weight factor, and promoting the technical information for exposure in the order of sorting, including: Sort the weighted technical information from high to low according to the weight factor, rank the technical information with a high weight factor first, and gradually arrange the technical information in the order of decreasing weight factor; Formulate a promotion strategy according to the characteristics of the technical information and the needs of the target audience; Promote the technology with a high weight factor in a way of high exposure, high traffic, and multiple platforms.

7. An Internet-based technology promotion system, characterized in that, The system includes a data acquisition module, a model construction module, a model optimization module, a model training module, a technology output module, and a technology promotion module, where: The data acquisition module is used to obtain the technical information of historical transaction promotion, perform data preprocessing on the technical information, extract key information points, and produce a historical sample data set; The model construction module is used to build a preliminary DeepFM model using TensorFlow. The model is divided into two parts: a deep neural network Deep and a factorization machine FM. The outputs of the two parts are connected through a fully connected layer for prediction; The model optimization module is used to optimize the preliminary DeepFM model to obtain an optimized DeepFM model; The model training module is used to input the historical sample data set into the optimized DeepFM model for training to obtain a technology promotion model; The technology output module is used to obtain the stock technical information, input the stock technical information into the technology promotion model for processing, and output the weighted technical information with different weight factors; The technology promotion module is used to sort the weighted technical information from high to low according to the weight factor, and promote the technical information for exposure in the order of sorting.

8. A technology promotion system based on the Internet according to claim 7, characterized in that, The data acquisition module includes a historical data collection sub-module, a data processing sub-module, a denoising processing sub-module, a feature extraction sub-module, and a data set generation sub-module, where: The historical data collection sub-module is used to obtain the technical information of historical transaction promotion, including technical basic information, technical transaction information, technical evaluation information, intellectual property information, technology transfer implementation information, legal and policy information, market competition information, and risk information; A data processing sub-module for processing the technical information using a hash algorithm, removing duplicate data, and ensuring that each piece of technical information is unique in the dataset; A denoising processing sub-module for using natural language processing techniques to denoise the technical information and remove irrelevant information; A feature extraction sub-module for selecting and extracting features that have an important impact on the technical promotion effect according to business requirements and technical characteristics; A dataset generation sub-module for dividing the extracted key information points into a training set, a validation set, and a test set according to a certain proportion.

9. The Internet-based technology promotion system according to claim 7, characterized in that, The model building module includes a create input layer sub-module, a build embedding layer sub-module, a build FM layer sub-module, a build DNN layer sub-module, a merge output sub-module, and a define output layer sub-module. Among them, The create input layer sub-module is used to define the input layer of the model, and the input shape matches the number of features in the dataset; The build embedding layer sub-module is used to convert sparse features into dense vector representations through the embedding layer; The build FM layer sub-module is used to build the FM layer, including first-order features and second-order interaction features. The first-order features are implemented through a simple linear layer, and the second-order interaction features are obtained through dot product operations on the embedding vectors; The build DNN layer sub-module is used to build the DNN layer, which is a multi-layer neural network implemented by stacking multiple fully connected layers. The number of neurons and activation functions in each layer are hyperparameters; The merge output sub-module is used to merge the outputs of the FM layer and the DNN layer, which is achieved by adding them together. The merged output can be used as the final output of the model; The define output layer sub-module is used to define the output layer, using the sigmoid activation function and one neuron to generate prediction values.

10. A technology promotion system based on the Internet according to claim 7, characterized in that, The model training module includes a data input sub-module, a training sub-module, a parameter setting sub-module, a validation sub-module, and an export sub-module. Among them, The data input sub-module is used to obtain the historical sample dataset, build a data preprocessing pipeline using Python's Pandas tool, convert the cleaned and formatted data into the model input format, and input it into the optimized DeepFM model; The training sub-module trains the model on the training set and updates the model parameters through the backpropagation algorithm; The parameter setting sub-module is used to adopt a three-layer neural network structure, with a batch size of 512, an epoch iteration number of 40, a learning rate of 0.001, the adam optimizer is selected, and the embedding dimension of the embedding layer is 6; The validation sub-module monitors the training process of the model through the validation set; The export sub-module exports the trained model into a deployable format to obtain the technical promotion model.

Images