A collaborative filtering recommendation method for analyzing the coupling relationship between users and experts in expert recommendation systems

Abstract

The present invention belongs to the technical field of professional talent selection, and in particular, relates to a collaborative filtering recommendation method for analyzing the coupling relationship between users and experts applied to an expert recommendation system. The present invention designs two aggregation analysis models based on the coupling relationship between users and expert selection processes and a relative analysis model based on the coupling relationship between users and expert selection processes; then the outputs of the two models are used as input to train a top-level DNN; finally, the trained models are integrated and fine-tuned. The accuracy of the method of the present invention is better than that of the currently disclosed methods, and has good practical value. It can effectively target the outstanding problems existing in the selection of talents and the employment of scientific and technological experts in large state-owned enterprises, greatly improve the effect and accuracy of talent recommendation in various professional fields, and promote the overall improvement of the company's business construction and production level. It plays a vital role in many fields such as scientific and technological project evaluation, technical consultation, project construction and implementation for large state-owned enterprises.

Description

Technical Field

[0001] The present invention belongs to the technical field of professional talent selection, and in particular relates to a collaborative filtering recommendation method for analyzing the coupling relationship between users and experts in an expert recommendation system. Background Art

[0002] Due to the huge business volume of state-owned enterprises and the wide range of professional fields, the evaluation of professional talents and the selection and recommendation of talents are arduous. Especially with the continuous development of emerging technologies in various industries, multi-field cross-technology is changing with each passing day, which puts higher requirements on talent selection and recommendation, and also increases the difficulty. At present, actual work mostly uses manual methods to select and recommend talents. Due to the limited understanding of field knowledge by actual staff and certain subjective tendencies, and with the continuous increase in the number of talent reserves, the efficiency and accuracy of manual selection gradually decrease. At the same time, the information of the professional talent pool lacks an effective real-time optimization mechanism, and the information structure and content are unreasonable, which cannot meet the ever-changing talent selection needs.

[0003] The method based on collaborative filtering is currently widely used in expert recommendation systems. It mainly relies on the historical interaction between users and experts to analyze users' preferences for experts. The interaction process includes users' ratings of experts, browsing traces, and actual selection behavior. The rating matrix between users and experts shows the user's preference for experts, and each element in the matrix shows this one-to-one relationship. Therefore, the rating matrix is ​​usually used as the main data source in the field of expert recommendation.

[0004] In actual work, because there are too many experts in the database, this rating matrix is ​​very sparse. It is difficult to obtain reasonable expert recommendation accuracy if we rely solely on such a data source. In addition, if we use popular data dimensionality reduction methods, it is difficult to fundamentally change this problem. At present, the common solution is to introduce additional user and expert information to participate in the estimation of the rating, but this idea has limited effect on improving accuracy. Summary of the invention

[0005] In view of the shortcomings of the above-mentioned prior art, the present invention provides a collaborative filtering recommendation method for analyzing the coupling relationship between users and experts applied to an expert recommendation system. The purpose is to achieve the invention purpose of effectively improving the accuracy of domain expert recommendations by integrating multiple user-expert correlation features and obtaining the advantages of multiple models.

[0006] The technical solution adopted by the present invention to achieve the above-mentioned purpose is:

[0007] A collaborative filtering recommendation method for analyzing the coupling relationship between users and experts in an expert recommendation system, comprising the following steps:

[0008] Step 1: Acquisition and preprocessing of expert attribute information and user-expert selection interaction information;

[0009] Step 2: Divide the dataset into three parts: the training set is used to train the model, and the validation set is used to verify the training results;

[0010] Step 3: Construct an aggregation analysis model based on the coupling relationship between the user and expert selection process, a relativity analysis model based on the coupling relationship between the user and expert selection process, and a fusion model of the two. Take the outputs of the aggregation analysis model based on the coupling relationship between the user and expert selection process and the relativity analysis model based on the coupling relationship between the user and expert selection process as input, train a top-level DNN+LSTM model, fuse the trained models, perform fine-tuning training, and obtain the expert recommendation results for talent selection;

[0011] Step 4: Use the training set to train the models in step 3 respectively, and use the validation set to observe the training effects of the two models. Finally, complete the testing of the test set to prove the robustness of the two models.

[0012] Step 5: Arrange the trained models in step 3 in parallel, connect the output nodes of each individual model to the DNN hidden layer as the input of the top-level DNN, and finally reduce the learning rate of the fusion model for fine-tuning;

[0013] Step 6: Use the validation set to verify the trained model. By comparing the accuracy and error of the test set and validation set, adjust the model parameters. After multiple trainings, select the model with the best performance on the validation set as the training result.

[0014] Step 7: Run the model in the actual environment. When there is a large deviation between the predicted value and the actual value, add the latest data to the training set and train the model again.

[0015] Furthermore, the acquisition and preprocessing of the expert attribute information and the user and expert selection interaction information in step 1 include the following steps:

[0016] Step 1.1: Collect expert attribute information and user and expert selection interaction information;

[0017] Step 1.2: Preprocess the data, including but not limited to: missing value mean interpolation, data standardization, one-hot encoding, and timestamp conversion;

[0018] Step 1.3: Form the user and expert selection interaction information matrix and the expert attribute information matrix;

[0019] Step 1.4: Take the selection demand information represented by the user's selected keywords as the key point to obtain the expert recommendation list to verify the effectiveness of the method.

[0020] Furthermore, step 3 constructs a polymerization analysis model based on the coupling relationship between the user and the expert selection process, a relativity analysis model based on the coupling relationship between the user and the expert selection process, and a fusion model of the two, including the following steps:

[0021] Step 3.1: Based on the coupling relationship aggregation analysis model between the user and expert selection process, the user and expert selection interaction information matrix and the expert attribute information matrix are aggregated and used as the input of the network to extract features to complete the expert recommendation;

[0022] Step 3.2: Based on the relativity analysis model of the coupling relationship between the user and the expert selection process, the row vectors of the above two matrices are polled, and the two vectors are merged as the data of the model. The model analyzes the relative characteristics of the two vectors to complete the recommendation work of the domain experts;

[0023] Step 3.3: The fusion model combines the results of step 3.1 and step 3.2 to enhance the prediction accuracy of the model, and finally obtains the expert's recommendation results for talent selection.

[0024] Furthermore, the step 4 uses the training set to train the model of step 3 and the step 5 arranges the trained model of step 3 in parallel, both of which include: the models of step 3.1 and step 3.2.

[0025] Furthermore, the coupling relationship aggregation analysis model diagram of the user and expert selection process includes the following steps:

[0026] Step 1: Data preprocessing;

[0027] Step 2: Design the fusion matrix of user-expert selection interaction information and expert attribute information;

[0028] Step 3: The fusion matrix extracts significant features through the convolution layer and forms a fixed-length vector through the local pooling layer to express the local fusion matrix feature relationship;

[0029] Step 4: The fusion matrix directly forms a global feature vector through the pooling layer;

[0030] Step 5: The local pooling vector is fused with the global pooling vector to form a coupling aggregation relationship analysis of the user-expert interaction process;

[0031] Step 6: Divide the dataset into two parts: the training set is used to train the model, and the validation set is used to verify the training results.

[0032] Furthermore, the relativity analysis model based on the coupling relationship between the user and the expert selection process includes the following steps:

[0033] Step 1: Data preprocessing;

[0034] Step 2: Extract historical user-expert selection interaction information and expert attribute information by item and conduct correlation analysis;

[0035] Step 3: Use ROIPooling technology to perform fixed-dimensional spatial pooling on the user-expert selection interaction information and expert attribute information vector;

[0036] Step 4: Merge the two pieces of user and expert selection interaction information with the expert attribute feature vector;

[0037] Step 5: Perform convolutional layer learning on the feature fusion vector to form coupling analysis results;

[0038] Step 6: Divide the dataset into two parts: the training set is used to train the model, and the validation set is used to verify the training results.

[0039] Furthermore, the relativity analysis model based on the coupling relationship between the user and the expert selection process includes the following steps:

[0040] Step 1: Combine the coupling relationship aggregation analysis results with the relative analysis results;

[0041] Step 2: Use the top-level DNN deep neural network + LSTM long short-term memory model hidden layer for learning analysis;

[0042] Step 3: Use the output of the branch model as input to train a top-level DNN+LSTM model to obtain expert recommendation results;

[0043] Step 4: Arrange the trained individual models in parallel, connect the output nodes of each individual model to the DNN hidden layer as the input of the top-level DNN, and finally reduce the learning rate of the fusion model for fine-tuning;

[0044] Step 5: Use the validation set to verify the trained model. By comparing the accuracy and error of the test set and validation set, adjust the model parameters. After multiple trainings, select the model with the best performance on the validation set as the training result.

[0045] Step 6: Run the model in the actual environment. When there is a large deviation between the predicted value and the actual value, add the latest data to the training set and train the model again.

[0046] Furthermore, the learning analysis using the top DNN+LSTM hidden layer described in step 2 includes the following steps:

[0047] Step 2.1: Construct a separate 3-layer DNN model and train it separately using the training set;

[0048] A single DNN model contains a hidden layer, an input layer, and an output layer, and uses a fully connected structure and tanh activation function to obtain the relationship between any input nodes;

[0049] Step 2.2: Construct a separate LSTM model and train it separately using the training set;

[0050] RNN takes as input the sequence {x1,x2,…,x n} Input a loop unit with the following function:

[0051] h t =f(h t-1 ,x t )

[0052] In the above formula, x t is the input at time t, h t is the hidden state. t It can be considered as the representation vector of all previous inputs up to time t. To solve the problem of gradient explosion or gradient disappearance, LSTM introduces a variety of gate structures in the recurrent unit, which are calculated as follows:

[0053] i t =σ(W i ·[h t-1 ,x t ]+b i )

[0054] f t =σ(W f ·[h t-1 ,x t ]+b f )

[0055]

[0056] o t =σ(W o ·[h t-1 ,x t ]+b o )

[0057] h t =o t *tanh(C t )

[0058] In the above formula: i t 、f t and t They are input gate, forget gate and output gate respectively; C t is the cell state, represents the candidate unit state, i t Decide whether to update it to the cell state Ct The cell state helps the model to better transmit gradient information; W and b are the parameters and biases in the cell; by converting the input sequence {x1, x2, …, x n} is input into LSTM to obtain a column of corresponding hidden states {h1,h2,…,h n}; These hidden states are used as feature representations of the sequence to generate outputs and as inputs for the next cycle.

[0059] A computer storage medium stores a computer program, which, when executed by a processor, implements the steps of a collaborative filtering recommendation method for analyzing the coupling relationship between users and experts in an expert recommendation system.

[0060] The present invention has the following beneficial effects and advantages:

[0061] The present invention integrates multiple user and expert correlation features, obtains the advantages of multiple models, and effectively improves the accuracy of domain expert recommendation. The expert recommendation system plays a vital role in many fields such as scientific and technological project review, technical consultation, project construction and implementation for large state-owned enterprises. The effective expert recommendation system has brought great improvement to the development and economic benefits of enterprises. The present invention first designs two aggregation analysis models based on the coupling relationship between the user and the expert selection process and a relative analysis model based on the coupling relationship between the user and the expert selection process; then the outputs of the two models are used as input to train a top-level DNN; finally, the trained models are integrated and fine-tuned. The experimental results show that the accuracy of this method is better than the currently disclosed methods and has good practical value. It can effectively target the outstanding problems existing in the selection of talents and the employment of scientific and technological experts in large state-owned enterprises, greatly improve the recommendation effect of talents in various professional fields, and promote the overall improvement of the company's business construction and production level. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0063] Figure 1 This is a graph of the aggregation analysis model based on the coupling relationship between the user and the expert selection process of the present invention;

[0064] Figure 2 This is a relative analysis model diagram of the coupling relationship between the user and the expert selection process of the present invention;

[0065] Figure 3 for Figure 1 Model and Figure 2 The final fusion model of the model is the schematic diagram of the final method proposed by this model. DETAILED DESCRIPTION

[0066] In order to more clearly understand the above-mentioned objectives, features and advantages of the present invention, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0067] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0068] Refer to the following Figure 1-Figure 3 The technical solutions of some embodiments of the present invention are described.

[0069] Example 1

[0070] The present invention is a collaborative filtering recommendation method for analyzing the coupling relationship between users and experts in an expert recommendation system, comprising the following steps:

[0071] Step 1: Acquisition and preprocessing of expert attribute information and user-expert selection interaction information;

[0072] Step 1.1: Collect expert attribute information and user-expert selection interaction information;

[0073] Step 1.2: Preprocess the data, including but not limited to: missing value mean interpolation, data standardization, one-hot encoding, timestamp conversion, etc.;

[0074] Step 1.3: Form the user and expert selection interaction information matrix and the expert attribute information matrix;

[0075] Step 1.4: Take the selection demand information represented by the user's selected keywords as the key point to obtain the expert recommendation list, so as to verify the effectiveness of the method.

[0076] Step 2: Divide the dataset into three parts: the training set is used to train the model, and the validation set is used to verify the training results;

[0077] Step 3: Construct a polymerization analysis model based on the coupling relationship between the user and the expert selection process, a relativity analysis model based on the coupling relationship between the user and the expert selection process, and a fusion model of the two;

[0078] Step 3.1: Based on the coupling relationship aggregation analysis model between the user and expert selection process, the user and expert selection interaction information matrix and the expert attribute information matrix are aggregated and used as the input of the network to extract features to complete the expert recommendation;

[0079] Step 3.2: Based on the relativity analysis model of the coupling relationship between the user and the expert selection process, the row vectors of the above two matrices are polled, and the two vectors are merged as the data of the model. The model analyzes the relative characteristics of the two vectors to complete the recommendation work of the domain experts;

[0080] Step 3.3: The fusion model combines the results of step 3.1 and step 3.2 to enhance the prediction accuracy of the model.

[0081] Step 4: Use the training set to train the models of step 3.1 and step 3.2 respectively, and use the validation set to observe the training effects of the two models to ensure that the parameters of the model reach a good level during training. Finally, complete the testing of the test set to prove the robustness of the two models.

[0082] Step 5: Arrange the trained models of step 3.1 and step 3.2 in parallel, connect the output nodes of each individual model as the input of the top-level DNN to the hidden layer of the DNN, and finally reduce the learning rate of the fusion model for fine-tuning;

[0083] Step 6: Use the validation set to verify the trained model. By comparing the accuracy and error of the test set and validation set, adjust the model parameters. After multiple trainings, select the model with the best performance on the validation set as the training result.

[0084] Step 7: Run the model in the actual environment. When there is a large deviation between the predicted value and the actual value, add the latest data to the training set and train the model again.

[0085] Example 2

[0086] The present invention further provides an embodiment, which is a collaborative filtering recommendation method for analyzing the coupling relationship between users and experts in an expert recommendation system. Figure 1 As shown, Figure 1 This is a graph of the aggregation analysis model based on the coupling relationship between the user and the expert selection process of the present invention;

[0087] The specific steps include:

[0088] Step 1: Data preprocessing;

[0089] Step 2: Design the fusion matrix of user-expert selection interaction information and expert attribute information;

[0090] Step 3: The fusion matrix extracts significant features through the convolution layer and forms a fixed-length vector through the local pooling layer, which expresses the local fusion matrix feature relationship;

[0091] Step 4: The fusion matrix directly forms a global feature vector through the pooling layer;

[0092] Step 5: The local pooling vector is fused with the global pooling vector to form a coupling aggregation relationship analysis of the user-expert interaction process;

[0093] Step 6: Split the dataset into two parts: the training set is used to train the model, and the validation set is used to verify the training results. In this case, the first 80% of the dataset is used as the training set, and the last 20% is used as the validation set.

[0094] like Figure 2 As shown, Figure 2 The relativity analysis model diagram based on the coupling relationship between the user and the expert selection process adopted in this embodiment specifically includes the following steps:

[0095] Step 1: Data preprocessing;

[0096] Step 2: Extract historical user-expert selection interaction information and expert attribute information by item and conduct correlation analysis;

[0097] Step 3: Use ROIPooling technology to perform fixed-dimensional spatial pooling on the user-expert selection interaction information and the expert attribute information vector; ROIPooling technology refers to the region of interest pooling technology, which is an existing technology. Step 4: Fusion of the two user-expert selection interaction information and the expert attribute feature vector;

[0098] Step 5: Perform convolutional layer learning on the feature fusion vector to form coupling analysis results;

[0099] Step 6: Divide the dataset into two parts: the training set is used to train the model, and the validation set is used to verify the training results.

[0100] In this embodiment, the first 80% of the data set is used as a training set, and the last 20% is used as a validation set.

[0101] like Figure 3 As shown, Figure 3 for Figure 1 Model and Figure 2 The final fusion model of the model is the schematic diagram of the final proposed method of this model. Based on the relative analysis model of the coupling relationship between the user and expert selection process, it specifically includes the following steps:

[0102] Step 1: Combine the coupling relationship aggregation analysis results with the relative analysis results;

[0103] Step 2: Use the top-level DNN deep neural network + LSTM long short-term memory model hidden layer for learning analysis. Both DNN and LSTM are existing technologies.

[0104] Step 2.1: Construct a separate 3-layer DNN model and train it separately using the training set;

[0105] A single DNN model contains a hidden layer, an input layer, and an output layer. It uses a fully connected structure and tanh activation function to obtain the relationship between any input nodes.

[0106] Theoretically, as long as there are enough hidden neurons, a three-layer DNN can also fit any complex function. The main reason for using a single hidden layer structure here is to enable the DNN model to obtain more information about the original data as a whole, and each neuron in a single hidden layer is connected to every input node.

[0107] Step 2.2: Construct a separate LSTM model and train it separately using the training set;

[0108] The LSTM model is a type of recurrent neural network. Recurrent neural networks are very effective in modeling sequence data because they can remember information from past inputs. However, when the sequence length is long, the simple RNN structure will encounter the problem of gradient explosion or gradient vanishing. The LSTM with gated recurrent units solves this problem well. LSTM can effectively avoid the problem of gradient explosion or gradient vanishing, so it has been widely used in modeling time series with long-term dependencies.

[0109] The LSTM model takes as input the sequence {x1,x2,…,x n} Input a loop unit with the following function:

[0110] h t =f(h t-1 ,x t )

[0111] In the above formula, x t is the input at time t, h t is the hidden state. t It can be considered as the representation vector of all previous inputs up to time t. In order to solve the problem of gradient explosion or gradient disappearance, LSTM introduces a variety of gate structures in the recurrent unit, which are calculated as follows:

[0112] i t =σ(W i ·[h t-1 ,x t ]+b i )

[0113] f t =σ(W f ·[h t-1 ,x t ]+b f )

[0114]

[0115] o t =σ(W o ·[h t-1 ,x t ]+b o )

[0116] h t =o t *tanh(C t )

[0117] In the above formula: i t 、f t and t They are input gate, forget gate and output gate respectively; C t is the cell state, represents the candidate unit state, i t Decide whether to update it to the cell state C t The cell state helps the model to better transmit gradient information; W and b are the parameters and biases in the cell; by converting the input sequence {x1, x2, …, x n} is input into LSTM to obtain a column of corresponding hidden states {h1,h2,…,h n These hidden states are used as feature representations of the sequence to generate outputs and as inputs for the next cycle.

[0118] LSTM can solve the dependency problem between long-term continuous time steps and achieve good results. However, when facing a very long time series, LSTM may still lose key information in the long history. This makes it difficult for LSTM to obtain an overall perception of historical data when facing a very long input sequence, thus affecting the improvement of accuracy.

[0119] In order to improve the prediction accuracy of the model, the present invention adopts a 5-layer LSTM unit stacking structure.

[0120] Step 3: Use the output of the branch model as input to train a top-level DNN+LSTM model to obtain expert recommendation results.

[0121] Step 4: Arrange the trained individual models in parallel, connect the output nodes of each individual model to the DNN hidden layer as the input of the top-level DNN, and finally reduce the learning rate of the fusion model for fine-tuning;

[0122] Step 5: Use the validation set to verify the trained model. By comparing the accuracy and error of the test set and validation set, adjust the model parameters. After multiple trainings, select the model with the best performance on the validation set as the training result.

[0123] Step 6: Run the model in the actual environment. When there is a large deviation between the predicted value and the actual value, add the latest data to the training set and train the model again.

[0124] Example 3

[0125] Based on the same inventive concept, an embodiment of the present invention further provides a computer storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of a collaborative filtering recommendation method for analyzing the coupling relationship between users and experts applied to an expert recommendation system as described in Embodiment 1 or 2 are implemented.

[0126] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0127] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0128] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0129] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0130] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A collaborative filtering recommendation method for analyzing the coupling relationship between users and experts in an expert recommendation system, characterized by: The following steps are involved: Step 1: Acquisition and preprocessing of expert attribute information and user-expert selection interaction information, including: Step 1.1: Collect expert attribute information and user-expert selection interaction information; Step 1.2: Preprocess the data, including but not limited to: missing value mean interpolation, data standardization, one-hot encoding, timestamp conversion; Step 1.3: Form a user-expert selection interaction information matrix and an expert attribute information matrix; Step 1.4: Use the selection demand information represented by the user's selected keywords as the key point to obtain the expert recommendation list, so as to verify the effectiveness of the method; Step 2: Divide the data set into three parts: the training set is used to train the model, and the validation set is used To verify the training results; Step 3: Construct an aggregation analysis model based on the coupling relationship between the user and the expert selection process, a relative analysis model based on the coupling relationship between the user and the expert selection process, and a fusion model of the two. The outputs of the aggregation analysis model based on the coupling relationship between the user and the expert selection process and the relative analysis model based on the coupling relationship between the user and the expert selection process are used as input to train a top-level DNN+LSTM model, fuse the trained models, and perform fine-tuning training to obtain the expert recommendation results for talent selection; including the following steps: Step 3.1: The aggregation analysis model based on the coupling relationship between the user and the expert selection process aggregates the interaction between the user and the expert selection process. The information matrix and the expert attribute information matrix are used as the input of the network to extract features to complete the recommendation of experts; Step 3.2: Based on the relativity analysis model of the coupling relationship between the user and the expert selection process, the row vectors of the above two matrices are polled, and the two vectors are merged as the data of the model. The model analyzes the relative characteristics of the two vectors to complete the recommendation work of the domain experts; Step 3.3: The fusion model combines the results of the two models in step 3.1 and step 3.2 to enhance the prediction accuracy of the model, and finally obtains the expert recommendation results for the selection of talents; Step 4: Use the training set to train the model of step 3 respectively, and use the validation set to observe the two models The training effect of each model is evaluated, and finally the test set is tested to prove the robustness of the two models. Step 5: Arrange the trained models of step 3 in parallel, connect the output nodes of each individual model to the hidden layer of the DNN as the input of the top-level DNN, and finally reduce the learning rate of the fusion model for fine-tuning. Step 6: Use the validation set to verify the trained model, adjust the model parameters by comparing the accuracy and error of the test set and the validation set, and select the model with the best performance on the validation set as the training result after multiple trainings. Step 7: Run the model in the actual environment. When there is a large deviation between the predicted value and the actual value, add the latest data to the training set and train the model again.

2. The collaborative filtering recommendation method for analyzing the coupling relationship between users and experts in an expert recommendation system according to claim 1, characterized in that: The step 4 uses the training set to train the model of step 3 and the step 5 arranges the trained model of step 3 in parallel, both of which include: the models of step 3.1 and step 3.

2.

3. The collaborative filtering recommendation method for analyzing the coupling relationship between users and experts in an expert recommendation system according to claim 1, characterized in that: The coupling relationship aggregation analysis model diagram of the user and expert selection process includes the following steps: Step 1: Data preprocessing; Step 2: Design the fusion matrix of user-expert selection interaction information and expert attribute information; Step 3: The fusion matrix extracts significant features through the convolution layer and forms a fixed-length vector through the local pooling layer to express the local fusion matrix feature relationship; Step 4: The fusion matrix directly forms a global feature vector through the pooling layer; Step 5: The local pooling vector is fused with the global pooling vector to form a coupling aggregation relationship analysis of the user-expert interaction process; Step 6: Divide the dataset into two parts: the training set is used to train the model, and the validation set is used to verify the training results.

4. The collaborative filtering recommendation method for analyzing the coupling relationship between users and experts in an expert recommendation system according to claim 1, characterized in that: The relativity analysis model based on the coupling relationship between the user and the expert selection process includes the following steps: Step 1: Data preprocessing; Step 2: Extract historical user-expert selection interaction information and expert attribute information by item and conduct correlation analysis; Step 3: Use ROIPooling technology to perform fixed-dimensional spatial pooling on the user-expert selection interaction information and expert attribute information vector; Step 4: Merge the two pieces of user and expert selection interaction information with the expert attribute feature vector; Step 5: Perform convolutional layer learning on the feature fusion vector to form coupling analysis results; Step 6: Divide the dataset into two parts: the training set is used to train the model, and the validation set is used to verify the training results.

5. The collaborative filtering recommendation method for analyzing the coupling relationship between users and experts in an expert recommendation system according to claim 1, characterized in that: The relativity analysis model based on the coupling relationship between the user and the expert selection process includes the following steps: Step 1: Combine the coupling relationship aggregation analysis results with the relative analysis results; Step 2: Use the top-level DNN deep neural network + LSTM long short-term memory model hidden layer for learning analysis; Step 3: Use the output of the branch model as input to train a top-level DNN+LSTM model to obtain expert recommendation results; Step 4: Arrange the trained individual models in parallel, connect the output nodes of each individual model to the DNN hidden layer as the input of the top-level DNN, and finally reduce the learning rate of the fusion model for fine-tuning; Step 5: Use the validation set to verify the trained model. By comparing the accuracy and error of the test set and validation set, adjust the model parameters. After multiple trainings, select the model with the best performance on the validation set as the training result. Step 6: Run the model in the actual environment. When there is a large deviation between the predicted value and the actual value, add the latest data to the training set and train the model again.

6. The collaborative filtering recommendation method for analyzing the coupling relationship between users and experts in an expert recommendation system according to claim 5, characterized in that: Step 2 uses the top DNN+LSTM hidden layer for learning analysis, including the following steps: Step 2.1: Construct a separate 3-layer DNN model and train it separately using the training set; A single DNN model contains a hidden layer, an input layer, and an output layer, and uses a fully connected structure and tanh activation function to obtain the relationship between any input nodes; Step 2.2: Construct a separate LSTM model and train it separately using the training set; RNN takes as input the sequence {x1,x2,…,x n } Input a loop unit with the following function: h t =f(h t-1 ,x t ) In the above formula, x t is the input at time t, h t is the hidden state; h t It can be considered as the representation vector of all previous inputs up to time t. To solve the problem of gradient explosion or gradient disappearance, LSTM introduces a variety of gate structures in the recurrent unit, which are calculated as follows: i t =σ(W i ·[h t-1 ,x t ]+b i ) f t =σ(W f ·[h t-1 ,x t ]+b f ) the t =σ(W o ·[h t-1 ,x t ]+b o ) h t =o t *tanh(C t ) In the above formula: i t 、f t and t They are input gate, forget gate and output gate respectively; C t is the cell state, represents the candidate unit state, i t Decide whether to update it to the cell state C t The cell state helps the model to better transmit gradient information; W and b are the parameters and biases in the cell; by converting the input sequence {x1, x2, …, x n } is input into LSTM to obtain a column of corresponding hidden states {h1,h2,…,h n }; These hidden states are used as feature representations of the sequence to generate outputs and as inputs for the next cycle.

7. A computer storage medium, characterized in that: The computer storage medium stores a computer program, which, when executed by a processor, implements the steps of a collaborative filtering recommendation method for analyzing the coupling relationship between users and experts in an expert recommendation system as described in any one of claims 1 to 6.

Images