IPC classification number intelligent recommendation method and system based on multi-source information fusion

Abstract

The invention discloses an IPC classification number intelligent recommendation method and system based on multi-source information fusion, and the method comprises the steps: firstly constructing a corpus through a sorting algorithm based on a patent text, generating the vector representation of each word, and obtaining the text classification score of a patent through a convolutional neural network; secondly, a heterogeneous graph is constructed, a random walk sequence set is obtained through random walk driven by a multi-element structure, and graph structure classification scores are obtained through a classification network; and finally, fusing the text classification score and the graph structure classification score of the patent by adopting multi-modal deep fusion based on a gating mechanism to obtain a final classification prediction score of the patent, outputting a recommendation list of corresponding IPC classification numbers, and performing reverse training. According to the method, adaptive fusion of different modal features is realized, the IPC classification number recommendation list is accurately and efficiently generated, and multi-dimensional decision reference is provided for patent applicants and related institutions.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to an intelligent recommendation method and system for IPC classification numbers that combines textual information and graph structure information. Background Technology

[0002] With the continued growth of global patent applications, not only do patent applicants need to accurately identify the technical subject matter, but patent agencies, corporate intellectual property departments, and local intellectual property protection centers also need to conduct preliminary classification of patent technologies at an early stage in their business processes. Agencies often predict IPC classification based on the received technical disclosure materials and then assign the classification results to agents in the corresponding technical fields. Corporate intellectual property departments pre-classify the disclosure materials submitted by various business units so that they can be processed by patent engineers in different technical areas later. Local protection centers, when providing patent pre-examination services, need to first make a preliminary judgment on the technical field to which the patent belongs, and IPC classification is one of the most direct and effective organizational bases for this. Therefore, at the application stage or the preliminary judgment stage, various agencies responsible for front-end processing urgently need to obtain IPC classifications for patents more accurately and efficiently as a basis.

[0003] However, the IPC architecture exhibits multi-layered and fine-grained characteristics, making it extremely difficult for non-classification experts to understand and correctly select classifications. Current widely used practices still primarily rely on manual database searches or experience-based classification judgments, which have significant limitations. First, manual searches are cumbersome and inefficient, with results highly dependent on search keywords and personal experience. Second, classification relies on the experience of professionals, resulting in high costs for manual intervention. Third, subjective judgments can easily lead to inconsistent classification results, affecting the accuracy of subsequent technology distribution and examination. While existing research has explored deep learning-based automatic patent classification technologies, these technologies are primarily geared towards the examination or verification stage, heavily relying on metadata such as assignee information and examiner review information. This information is simply not available in the initial application assessment or disclosure material stage, making it impossible for existing technologies to be truly implemented on the front-end business side. Furthermore, existing solutions do not fully utilize the many valuable prior information available at the application stage, such as the inventor's historical technical field and the applicant's past patent portfolio structure. Summary of the Invention

[0004] Based on the aforementioned practical problems, this invention proposes an intelligent IPC recommendation method for the application stage and front-end business processes. This invention fully utilizes multi-source information already available at this stage, including patent texts, applicant and inventor's previous patent combinations, etc., and designs a dual-channel feature extraction network: one channel uses a convolutional neural network (CNN) for text semantic modeling, and the other channel uses a multi-layer graph neural network to learn the patent relationship graph structure, and then sorts and displays candidate IPC classification numbers according to the concept of a recommendation system. This invention retains the final human selection right while significantly improving front-end processing efficiency and prediction accuracy through intelligent recommendation.

[0005] To effectively achieve IPC classification number recommendation, this invention proposes an intelligent recommendation method for IPC classification numbers based on multi-source information fusion.

[0006] (1) Text information extraction and preprocessing: from patent The text content of the title, abstract, and description sections of the document is extracted, and the TextRank algorithm is used to automatically extract a predetermined number of semantically most important core sentences from the text; from the core sentences, a predetermined number of words with the highest information entropy are selected to constitute the patent. The text sequence was analyzed, and stop words were removed to construct a corpus.

[0007] (2) Word vector representation generation: The Word2Vec method of the Skip-Gram model is used to generate word vector representations. The text sequence is used for training to generate a vector representation of each word, and all word vectors are concatenated into a two-dimensional matrix. ,in For vector dimensions, This represents the number of words.

[0008] (3) Text feature extraction and classification: extracting and classifying text matrix The input is a convolutional neural network, which performs convolution operations using four sets of convolutional kernels of different sizes in parallel. Max pooling is then performed on the output of each convolutional kernel to extract salient features. These pooled salient features are concatenated and passed through a fully connected layer to output the patent. Text classification score ,in This represents the number of categories.

[0009] (4) Patent Relationship Graph Construction: Construct a heterogeneous graph containing three types of nodes, namely patent nodes. Applicant Node And inventor node And three types of edges, namely the patent-applicant edge. Patent - Inventor Bian Patent-Patent Citation Edge .

[0010] (5) Multi-structure driven random walk: Starting from all patent nodes in the heterogeneous graph, a multi-structure driven random walk strategy is performed to obtain a set of random walk sequences with the same number of elements as the number of patents.

[0011] (6) Graph structure learning and relation modeling: Initial representations of each node are generated using a set of random walk sequences; features of neighboring nodes are aggregated using an aggregation method based on node weights and degree awareness; multi-level neighbor features of nodes are learned through a multi-layer network structure; and the patent is output using hierarchical incremental feature fusion. Graph structure classification score .

[0012] (7) Adaptive fusion of multimodal classification scores: A multimodal deep fusion method based on a gating mechanism is adopted to adaptively fuse patents. Text classification score And graph structure classification score Obtained a patent Final classification prediction score .

[0013] (8) Model training and optimization: Binary cross-entropy is used as the loss function, Adam optimizer is used to update model parameters, multimodal adversarial learning is used to enhance the model's discriminative power, and the entire network is trained through a multi-label classification task.

[0014] (9) Recommendation results output: based on the final classification score Obtained a patent The system provides probability predictions for each technology category and outputs a recommended list of IPC classification numbers.

[0015] In another aspect, this invention proposes an intelligent recommendation system for IPC classification numbers based on multi-source information fusion, comprising the following modules:

[0016] The corpus construction module builds a corpus based on patent texts using a sorting algorithm.

[0017] The text classification module generates a vector representation of each word based on text sequences from a corpus, and obtains the patented text classification score through a convolutional neural network.

[0018] Graph Structure Classification Module: Used to construct heterogeneous graphs. It obtains a set of random walk sequences through multi-structure-driven random walks and obtains graph structure classification scores through a classification network.

[0019] The IPC classification number recommendation module adopts a multimodal deep fusion based on a gating mechanism to adaptively fuse the text classification score and graph structure classification score of the patent to obtain the final classification prediction score of the patent. It obtains the probability prediction result of the patent belonging to each technology category and outputs the corresponding IPC classification number recommendation list, and performs reverse training.

[0020] Compared with traditional manual retrieval or single-source auxiliary methods, the beneficial effects of this invention are: (1) Enhanced decision credibility through multi-dimensional recommendation basis: By integrating multi-source information such as the semantics of patent texts and the applicant's and inventor's historical technical field preferences, an IPC classification number recommendation list is generated accurately and efficiently, providing multi-dimensional decision references for patent applicants and related institutions; (2) Adaptive feature weight allocation: Through the final classification score fusion mechanism based on gating mechanism, the system can dynamically adjust the contribution weight of different information sources in the final recommendation, realizing the adaptive fusion of different modal features; (3) Difference constraint fusion mechanism: Through multi-modal adversarial learning, the learning objective of the fusion channel is compared with the output of the single channel, forcing the model to learn complementary information from the two modalities, greatly improving the model's discrimination ability. Attached Figure Description

[0021] Figure 1 This is a flowchart of the intelligent recommendation method for IPC classification numbers based on multi-source information fusion according to the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the following description is provided in conjunction with the appendix. Figure 1 The present invention provides a detailed description of an intelligent recommendation method for IPC classification numbers based on multi-source information fusion, comprising the following steps:

[0023] (1) Text information extraction and preprocessing:

[0024] 1-1. Core Sentence Generation: From Patents The text content of the title, summary and description sections of the document is extracted, and the TextRank algorithm is used to automatically extract the 10 most semantically important core sentences from the text;

[0025] 1-2. Corpus Construction: Core sentences are segmented into words, and the meaning of each word is calculated. probability of occurrence ; Calculate each word based on information entropy theory The information entropy is given by the following formula:

[0026]

[0027] in for Information entropy; select the 100 words with the highest information entropy to form the text sequence of the patent, and remove stop words to construct a corpus;

[0028] (2) Word vector representation generation: The text sequence is trained using the Skip-Gram model's Word2Vec method to generate a vector representation for each word, and the patent... All word vectors are concatenated into a two-dimensional matrix. ,in For the number of words, For vector dimensions;

[0029] (3) Text feature extraction and classification:

[0030] 3-1. Embedded layer processing: Patented. text matrix Then, it is first fed into an embedding layer, and this matrix is ​​set to be updatable during the training phase to allow the model to adjust the word vector representations according to task requirements. The output of the embedding layer is represented as... ;

[0031] 3-2. Convolution Processing: Using four parallel sets of convolution kernels of different sizes. right Perform a convolution operation to obtain four sets of feature vectors. The formula is as follows:

[0032] ,

[0033]

[0034]

[0035] in, for The number of convolution kernels, represent The first in One convolutional kernel, For the first The width of the group convolution kernel, Representative use The implemented convolution operation, convolution kernel The output feature vector set, represent The first in 1 eigenvector , The dimension of the row vectors in the matrix;

[0036] 3-3. Max pooling processing: For The feature vectors in the convolutional vectors are subjected to max pooling, selecting the maximum value from the output of each convolutional kernel, as shown in the following formula:

[0037]

[0038] in, Indicates to The salient features obtained by performing max pooling are then concatenated into a feature vector from the four sets of max-pooled salient features. The formula is as follows:

[0039]

[0040] 3-4. Text classification score generation: The input was sequentially fed into a dropout layer and a fully connected layer, resulting in a patent. Text-based classification scores , The formula for the number of categories is as follows:

[0041]

[0042] in, Represents a fully connected layer;

[0043] (4) Patent Relationship Graph Construction: Construct a patent relationship heterogeneous graph, which contains three types of nodes, namely patent nodes. Applicant Node And inventor node And three types of edges, namely the patent-applicant edge. Patent - Inventor Bian Patent-Patent Citation Edge ;

[0044] (5) Multi-structure driven random walk: Starting from all patent nodes in the heterogeneous graph, a multi-structure driven random walk strategy is performed to obtain a set of random walk sequences with the same number of elements as the number of patents.

[0045] 5-1. Construction of structural feature vectors: based on patents A random walk is performed starting from the corresponding node; for the node at the current walk position... Calculate each of its neighbor nodes Corresponding node degree PageRank value Clustering coefficient Average degree of neighbors The formula is as follows:

[0046]

[0047]

[0048]

[0049]

[0050] in, It is a node The set of first-order neighbors (i.e., direct neighbors) nodes. This represents the pre-calculated PageRank value. Indicated by node The number of triangles at each vertex; construct the structural feature vector of each neighbor node based on the above information. ;

[0051] 5-2. Structural Feature Embedding Mapping: This involves embedding the above feature vectors... The input linear mapping layer is used to obtain a 32-dimensional structural embedding representation, as shown in the following formula:

[0052]

[0053] in For a trainable parameter matrix, For bias terms;

[0054] 5-3. Next-hop node selection: For the current node The embedding representations of all neighboring nodes are softmax normalized to obtain the transition probability of a neighboring node being selected as the next hop. The formula is as follows:

[0055]

[0056] in The direction vector is learnable; random sampling is performed based on the probability distribution to determine the next hop node;

[0057] 5-4. Single Node Sequence Generation: Repeat steps 5-1–5-3 until the preset walk length of 30 is reached, generating a sequence based on the patent. The corresponding node is the node sequence starting from the node;

[0058] 5-5. Overall sequence generation: Repeat steps 5-1–5-4, and perform random walks starting from all patent nodes in the heterogeneous graph to obtain a set of random walk sequences with the same number of elements as the number of patents. The multi-structure driven random walk strategy proposed in this invention achieves adaptive walk modeling of structural relationships between nodes in the graph by integrating multiple structural statistical features and introducing a trainable probabilistic generation network, thereby improving the expressive power of graph embedding and the performance of downstream tasks.

[0059] (6) Graph structure learning and relation modeling:

[0060] 6-1. Random Walk Sequence Vectorization: Input the set of random walk node sequences obtained in the previous step into the Skip-Gram model for training to obtain the initial vector representation of each node. In the heterogeneous graph of layer 0, the nodes... The initial vector is represented as ;

[0061] 6-1. Neighbor node weight calculation: For each node Calculate the first-order neighbor nodes using the node's own features and degree information. The importance weights are calculated using the following formula:

[0062]

[0063]

[0064] in It is a node Importance weights For a learnable global parameter vector, It is a node In the Vector representation of layers, Hyperparameters for balancing feature and degree information contributions; by introducing node degree information, this invention can dynamically balance the semantic contribution and structural centrality of nodes; and by combining the softmax normalization mechanism, this invention can efficiently perform adaptive modeling of local graph structures.

[0065] 6-2. Feature-weighted aggregation: For each node The node in the next layer of the heterogeneous graph is obtained by aggregating its own features with the weighted features of all its neighboring nodes. Feature Calculation The calculation formula is as follows:

[0066]

[0067] in It is a non-linear activation function. and For learnable parameter matrix, It is a node Importance weights;

[0068] 6-3. Incremental Feature Fusion: Integrating Patents The output features of all layers are weighted and summed to obtain As Classification scores based on graph structures, where The formula for the number of categories is as follows:

[0069]

[0070] in Indicates the floor number of the final floor. These are the weight coefficients corresponding to each layer of vectors, where and The hierarchical incremental feature fusion mechanism proposed in this invention preserves the local structural information of the lower levels while maintaining the high-level semantic abstraction capability, thereby effectively alleviating the oversmoothing problem and improving the model's ability to perceive multi-scale graph structures.

[0071] (7) Adaptive fusion of multimodal classification scores: A multimodal deep fusion method based on gating mechanism is adopted to adaptively fuse classification information of text modality and graph structure modality;

[0072] 7-1. Gating signal generation: First, generate the text score vector. And graph fraction vector The concatenation is performed, and a gated vector is generated by passing the concatenation through a fully connected layer and a sigmoid activation function. This is used to dynamically adjust the contribution weight of each mode in different categories, and the formula is as follows:

[0073]

[0074] in, This represents a vector concatenation operation. and For the trainable parameter matrix and bias terms, Use the Sigmoid activation function;

[0075] 7-2. Cross-modal feature interaction: First, element-wise multiplication of feature interaction terms is used to capture the synergistic effect between text and graph structure features; then, the weighted modal features and interaction features are integrated to obtain a fused feature vector. The formula is as follows:

[0076]

[0077] in, This indicates element-wise multiplication. This is a trainable parameter matrix used to perform linear transformations on the interaction features;

[0078] 7-3. Final classification prediction: The feature vectors will be fused. The patent is obtained through a fully connected layer and a Sigmoid activation function. The final predicted probability score vector for all categories The formula is as follows:

[0079]

[0080] in, and For trainable parameter matrices and bias terms;

[0081] (8) Model training and optimization: Calculate the loss function using multimodal adversarial learning;

[0082] 8-1. Text Channel Loss Calculation: Calculate the binary cross-entropy loss using text classification scores. The formula is as follows:

[0083]

[0084] in and These represent the number of training samples and the number of labels, respectively. This is the actual label value (either 0 or 1). It is the probability value predicted using text information;

[0085] 8-2. Graph Channel Loss Calculation: Calculate binary cross-entropy loss using graph structure classification scores. The formula is as follows:

[0086]

[0087] in and These represent the number of training samples and the number of labels, respectively. This is the actual label value (either 0 or 1). It is the probability value predicted using graph structure information;

[0088] 8-3. Adaptive Fusion Loss Calculation: Calculate the binary cross-entropy loss using the final predicted probability score obtained after adaptive fusion. The formula is as follows:

[0089]

[0090] in and These represent the number of training samples and the number of labels, respectively. This is the actual label value (either 0 or 1). It is the probability value predicted by adaptively fusing textual information and graph structure information;

[0091] 8-4. Multimodal adversarial learning: Combining , and Calculate the final loss function. The fusion process is explicitly constrained by controlling the difference between the final fusion loss and the single-modal loss, as shown in the following formula:

[0092]

[0093] in, As a weighting factor, To prevent tiny constants with zero denominators, the multimodal adversarial learning proposed in this invention enables the fusion module to simultaneously focus on the information contributions of text and graph channels during the optimization process. It is no longer dominated by a single modality, but maintains a dynamic balance in the loss space, thereby obtaining more discriminative fusion features and improving the model's discriminative ability and robustness in complex contract scenarios.

[0094] 8-5. Model Optimization: The Adam optimizer is used to update the model parameters, and the entire network is trained through a multi-label classification task;

[0095] (9) Recommendation results output: based on the final classification score Obtained a patent The probability prediction results for each technology category are used to output the top 10 IPC classification numbers with the highest predicted probabilities as a recommendation list.

[0096] The present invention will further demonstrate the detailed implementation process and technical effects of the intelligent recommendation method for IPC classification numbers based on multi-source information fusion shown in steps (1) to (9) above on a specific dataset through a specific embodiment, so as to facilitate understanding of the technical effects of the present invention.

[0097] In another aspect, the present invention provides an intelligent recommendation system for IPC classification numbers based on multi-source information fusion, comprising the following modules:

[0098] The corpus construction module builds a corpus based on patent texts using a sorting algorithm.

[0099] The text classification module generates a vector representation of each word based on text sequences from a corpus, and obtains the patented text classification score through a convolutional neural network.

[0100] Graph structure classification module: used to construct heterogeneous graphs, obtain a set of random walk sequences driven by multi-structure random walks, and obtain graph structure classification scores through a classification network;

[0101] The IPC classification number recommendation module adopts a multimodal deep fusion based on a gating mechanism to adaptively fuse the text classification score and graph structure classification score of the patent to obtain the final classification prediction score of the patent. It obtains the probability prediction result of the patent belonging to each technology category and outputs the corresponding IPC classification number recommendation list, and performs reverse training.

[0102] Example

[0103] The steps in this embodiment are the same as those in the aforementioned steps (1) to (9) for the intelligent recommendation method of IPC classification numbers based on multi-source information fusion, and will not be repeated here. The main focus is on the specific dataset, some specific parameter settings, and implementation results of this embodiment. For ease of description, the method shown in steps (1) to (9) will be referred to as the method of this invention, and the IPC classification number recommender used will be denoted as MinPac.

[0104] Experimental Dataset: The dataset used in this invention comprises 100,000 patent records from the past five years obtained from the publicly available database of the United States Patent and Trademark Office (USPTO). Each record includes key fields such as title, abstract, applicant, inventor, and citation information, and uses the International Patent Classification (IPC) code as the classification label for supervised learning, specifically for training and evaluating the IPC classification number intelligent recommendation model.

[0105] Evaluation metrics include: precision@k (precision of the first k items), recall@k (recall of the first k items), F1@k (F1 score of the first k items), and NDCG@k (normalized depreciation cumulative gain of the first k items). Precision@k focuses on the "accuracy" of the first k items in the result list, while recall@k measures how many total relevant items the system captured in the first k items, focusing on "recall rate." F1@k, as the harmonic mean of precision@k and recall@k, provides a single comprehensive metric. NDCG@k considers the quality of the ranking order in the result list. Higher values ​​for these metrics indicate better model performance. In this invention patent, k takes values ​​of 1, 3, and 5.

[0106] The specific experimental results of this embodiment are as follows:

[0107] (1) Algorithm comparison results

[0108] Table 1. Comparison results of this invention with current patent classification algorithms.

[0109]

[0110] Table 1 compares the patent classification results of this invention with other comparison methods on the collected dataset. The summary results in Table 1 show that this invention has significant advantages over other comparison methods.

[0111] (2) Ablation test results

[0112] Table 2 The Role of Different Information

[0113]

[0114] In Table 2, w / o graph information indicates the removal of graph structure information, and w / o text information indicates the removal of text information.

[0115] The results in the table show that removing any type of structural information will lead to a decrease in model performance. Removing graph structural information results in a larger decrease in model performance, while removing text information results in a smaller decrease in model performance.

[0116] Table 3. Role of different metadata

[0117]

[0118] In Table 3, "w / o inventor" means removing the inventor's information, and "w / o applicant" means removing the applicant's information.

[0119] The results in the table show that removing any type of metadata will lead to a decrease in model performance. Removing inventor information results in a larger decrease in model performance, while removing applicant information results in a smaller decrease in model performance.

Claims

1. A method for intelligent recommendation of IPC classification numbers based on multi-source information fusion, characterized in that, Includes the following steps: Step 1: Construct a corpus based on the patent text using a sorting algorithm; Step 2: Based on the text sequences in the corpus, generate a vector representation of each word, and obtain the text classification score of the patent through a convolutional neural network; Step 3: Construct a heterogeneous graph, obtain a set of random walk sequences through multi-structure-driven random walks, and obtain the graph structure classification score through a classification network; Step 4: Employ multimodal deep fusion based on a gating mechanism to adaptively fuse the text classification score and graph structure classification score of the patent to obtain the final classification prediction score of the patent. Obtain the probability prediction result of the patent belonging to each technology category and output the corresponding IPC classification number recommendation list, and perform reverse training.

2. The intelligent recommendation method for IPC classification numbers based on multi-source information fusion according to claim 1, characterized in that, Step 1 specifically involves: from the patent The text content of the title, abstract, and description sections of the document is extracted. The TextRank algorithm, based on graph ranking, is used to automatically extract a predetermined number of semantically most important core sentences from the text. From these core sentences, a predetermined number of words with the highest information entropy are selected to constitute the patent. The text sequence was analyzed, and stop words were removed to construct a corpus.

3. The intelligent recommendation method for IPC classification numbers based on multi-source information fusion according to claim 2, characterized in that, The specific implementation process of step 2 is as follows: Word vector representation generation: The Word2Vec word vector generation method is used to generate word vector representations. The text sequence is used for training to generate a vector representation of each word, and all word vectors are concatenated into a two-dimensional matrix. ,in For vector dimensions, For the number of words; Text feature extraction and classification: extracting text matrices... The input is a convolutional neural network, which performs convolution operations using four sets of convolutional kernels of different sizes in parallel. Max pooling is then performed on the output of each convolutional kernel to extract salient features. These pooled salient features are concatenated and passed through a fully connected layer to output the patent. Text classification score ,in This represents the number of categories.

4. The intelligent recommendation method for IPC classification numbers based on multi-source information fusion according to claim 3, characterized in that, The specific implementation process of step 3 is as follows: Construct a heterogeneous graph containing three types of nodes: patent nodes. Applicant Node And inventor node And three types of edges, namely the patent-applicant edge. Patent - Inventor Bian Patent-Patent Citation Edge ; Multi-structure driven random walk: Starting from all patent nodes in the heterogeneous graph, a multi-structure driven random walk strategy is performed to obtain a set of random walk sequences with the same number of elements as the number of patents. Initial representations for each node are generated using a set of random walk sequences; neighbor node features are aggregated using an aggregation method based on node weights and degree awareness; multi-level neighbor features of nodes are learned through a multi-layer network structure; and the patent is output using hierarchical incremental feature fusion. Graph structure classification score .

5. The intelligent recommendation method for IPC classification numbers based on multi-source information fusion according to claim 4, characterized in that, The specific implementation process of the multi-structure driven random walk is as follows: 5-1. Using patents A random walk is performed starting from the corresponding node; for the node at the current walk position... Calculate each of its neighbor nodes Corresponding node degree The node importance measurement algorithm outputs the PageRank value. Clustering coefficient Average degree of neighbors Construct the structural feature vector of each neighbor node. ; 5-2. The feature vector Input a linear mapping layer to obtain a structural embedding representation; 5-3. For the current node The embedding representations of all neighboring nodes are softmax normalized to obtain the transition probability of a neighboring node being selected as the next hop. ; 5-4. Repeat steps 5-1–5-3 until the preset walking length is reached, generating a patent. The corresponding node is the node sequence starting from the node; 5-5. Repeat steps 5-1–5-4, and perform random walks starting from all patent nodes in the heterogeneous graph to obtain a set of random walk sequences with the same number of elements as the number of patents.

6. The intelligent recommendation method for IPC classification numbers based on multi-source information fusion according to claim 5, characterized in that, The specific implementation of obtaining graph structure classification scores through a classification network is as follows: 6-1. Input the set of random walk node sequences into the Skip-Gram model to train and obtain the initial vector representation of each node. In the heterogeneous graph of layer 0, the nodes... The initial features are ; 6-1. For each node Calculate the first-order neighbor nodes using the node's own features and degree information. Importance weights; 6-2. For each node The node in the next layer of the heterogeneous graph is obtained by aggregating its own features with the weighted features of all its neighboring nodes. Feature Calculation ; 6-3. Patent The output features of all layers are weighted and summed to obtain As Classification scores based on graph structures, where This represents the number of categories.

7. The intelligent recommendation method for IPC classification numbers based on multi-source information fusion according to claim 6, characterized in that, The specific implementation process of the multimodal deep fusion based on the gating mechanism is as follows: 7-1. Convert the text score vector And graph fraction vector The concatenation is performed, and a gated vector is generated by passing the concatenation through a fully connected layer and a sigmoid activation function. ; 7-2. Using the feature interaction term of element-wise multiplication Capture the synergistic effect between text and graph structure features; then weight the modal features. , With interaction features Weighted integration is performed to obtain the fused feature vector. ; 7-3. Final classification prediction: The feature vectors will be fused. The patent is obtained through a fully connected layer and a sigmoid activation function. The final predicted probability score vector for all categories .

8. The intelligent recommendation method for IPC classification numbers based on multi-source information fusion according to claim 7, characterized in that, The loss function used in step 3 for training includes: Text channel loss calculation: Calculate binary cross-entropy loss using text classification scores. ; Graph Channel Loss Calculation: Calculate binary cross-entropy loss using graph structure classification scores. ; Adaptive fusion loss calculation: Calculate the binary cross-entropy loss using the final predicted probability scores obtained after adaptive fusion. ; Multimodal adversarial learning: Combining , and Calculate the final loss function. By controlling the final fusion loss , With single-mode loss The difference explicit constraint fusion process. To prevent constants with a denominator of zero.

9. An intelligent recommendation system for IPC classification numbers based on multi-source information fusion, used to implement the intelligent recommendation method for IPC classification numbers as described in any one of claims 1 to 8, characterized in that, Includes the following modules: The corpus construction module builds a corpus based on patent texts using a sorting algorithm. The text classification module generates a vector representation of each word based on text sequences from a corpus, and obtains the patented text classification score through a convolutional neural network. Graph structure classification module: Constructs heterogeneous graphs, obtains a set of random walk sequences through multi-structure-driven random walks, and obtains graph structure classification scores through a classification network; The IPC classification number recommendation module adopts a multimodal deep fusion based on a gating mechanism to adaptively fuse the text classification score and graph structure classification score of the patent to obtain the final classification prediction score of the patent. It obtains the probability prediction result of the patent belonging to each technology category and outputs the corresponding IPC classification number recommendation list, and performs reverse training.

Images