Patent data processing method and device and electronic equipment

Abstract

The invention provides a patent data processing method and device and electronic equipment. The method comprises the steps of obtaining text information and non-text information of data of each patent in a to-be-processed patent data set; processing the abstract data of each patent to obtain an abstract embedding vector of each patent; determining keyword information of each patent by using a word frequency inverse text frequency algorithm; establishing a text association relationship and a non-text association relationship between any two patents; utilizing the established text association relationship and non-text association relationship to construct a graph structure which takes the patent as a node and takes the association relationship as an edge; generating a training set of an initial neural network word embedding model based on the graph structure, training the initial neural network word embedding model, and obtaining a feature vector of each patent in the to-be-processed patent data set based on the trained neural network word embedding model; and processing the to-be-processed patent data set based on the obtained feature vector of each patent.

Description

Technical Field

[0001] The present disclosure relates to the field of computer technology, and in particular to a patented data processing method, device, and electronic device. Background Art

[0002] Prior art patent analysis typically involves processing patent data using artificial intelligence (AI). For example, machine learning is used to classify and match patents, followed by analysis based on the machine learning results. However, current AI-based patent processing often relies solely on the textual features of patents, ignoring other features. This results in results that fail to meet the actual needs of analysts. Summary of the Invention

[0003] The purpose of the present disclosure is to provide a patent data processing method, device and electronic device.

[0004] According to a first aspect of the present disclosure, a patent data processing method is provided, the method comprising:

[0005] Acquire text information and non-text information of each patent data in the patent data set to be processed; wherein the text information at least includes: abstract text data and claim text data;

[0006] The abstract data of each patent is processed using a pre-trained embedding model to obtain the abstract embedding vector of each patent; the keyword information of each patent is determined using the word frequency inverse text frequency algorithm and the patent claim data;

[0007] Based on the abstract embedding vector and keyword information of each patent, a textual correlation relationship between any two patents is established; based on the obtained non-textual information, a non-textual correlation relationship between any two patents is established;

[0008] Utilize the established textual and non-textual relationships to construct a graph structure with patents as nodes and relationships as edges;

[0009] A training set of an initial neural network word embedding model is generated based on the graph structure, and the initial neural network word embedding model is trained. A feature vector of each patent in the patent data set to be processed is obtained based on the trained neural network word embedding model; and the patent data set to be processed is processed based on the obtained feature vector of each patent.

[0010] According to a second aspect of the present application, a patent data processing device is provided, the device comprising: an information acquisition module for acquiring text information and non-text information of each patent data in a patent data set to be processed; wherein the text information comprises at least abstract text data and claim text data;

[0011] The information extraction module is used to process the abstract data of each patent using a pre-trained embedding model to obtain the abstract embedding vector of each patent; and to determine the keyword information of each patent using the word frequency inverse text frequency algorithm and the patent claim data;

[0012] A relationship determination module is used to establish a textual association relationship between any two patents based on the abstract embedding vector and keyword information of each patent; and to establish a non-textual association relationship between any two patents based on the obtained non-text information;

[0013] A graph construction module is used to construct a graph structure with patents as nodes and associations as edges using the established text association relationships and non-text association relationships;

[0014] A data processing module is used to generate a training set of an initial neural network word embedding model based on the graph structure, train the initial neural network word embedding model, obtain a feature vector for each patent in the patent data set to be processed based on the trained neural network word embedding model; and process the patent data set to be processed based on the obtained feature vector of each patent.

[0015] According to a third aspect of the present disclosure, an electronic device is provided, comprising a processor and a memory, wherein the memory stores a computer program, and the processor executes the computer program stored in the memory to implement the steps in the method described in the first aspect above.

[0016] According to a fourth aspect of the present disclosure, a computer program is provided, comprising computer instructions, which implement the steps of the method described in the first aspect when executed by a processor.

[0017] According to a fifth aspect of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the method described in the first aspect are implemented. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 A flowchart of a patent data processing method provided in accordance with one embodiment of the present disclosure;

[0019] Figure 2 A schematic diagram of a line segment in a graph structure provided by one embodiment of the present disclosure;

[0020] Figure 3 A flowchart of a specific patent data processing method provided for one embodiment of the present disclosure;

[0021] Figure 4 A schematic diagram of the structure of a patented data processing device provided in accordance with one embodiment of the present disclosure;

[0022] Figure 5 A schematic structural diagram of an electronic device provided in one embodiment of the present disclosure. DETAILED DESCRIPTION

[0023] Before introducing the embodiments of the present disclosure, it should be noted that:

[0024] Some embodiments of the present disclosure are described as processing flows. Although the various operation steps of the flow may be given sequential step numbers, the operation steps therein may be implemented in parallel, concurrently, or simultaneously.

[0025] In the embodiments of the present disclosure, the terms "first", "second", etc. may be used to describe various features, but these features should not be limited by these terms. These terms are used only to distinguish one feature from another.

[0026] The term “and / or” may be used in embodiments of the present disclosure. “And / or” includes any and all combinations of one or more of the listed associated features.

[0027] It should be understood that when describing the connection relationship or communication relationship between two components, unless it is explicitly stated that the two components are directly connected or directly communicating, the connection or communication between the two components can be understood as direct connection or communication, or as indirect connection or communication through an intermediate component.

[0028] In order to make the technical solutions and advantages of the embodiments of the present disclosure more clearly understood, the exemplary embodiments of the present disclosure are further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present disclosure, and are not an exhaustive list of all the embodiments. It should be noted that the embodiments and features in the embodiments of the present disclosure can be combined with each other unless they conflict.

[0029] Currently, when analyzing patents, it is usually necessary to process patent data using artificial intelligence (AI), such as using machine learning to classify and match patents, and then analyze them based on the machine learning processing results. However, when processing patents based on AI, processing often relies solely on the text features of the patents. For example, when matching patents, matching is performed based solely on the patent's text information, while ignoring other features of the patents. This results in processing results that cannot meet the actual needs of analysts. For example, when analyzing patents, in addition to considering the content recorded in the patent itself, it is also necessary to consider information such as the patent's applicant, inventor, patent family, agency, application location, citation relationship, etc. However, when processing patents, it is currently impossible to take all of the above into account, resulting in analysts' analysis based on AI processing results being relatively one-sided.

[0030] Based on the above problems, Figure 1 As shown, this application proposes a patented data processing method, including:

[0031] S101, obtaining text information and non-text information of each patent data in the patent data set to be processed; wherein the text information at least includes: abstract data and claim data;

[0032] S102, using a pre-trained embedding model to process the abstract data of each patent to obtain an embedding vector for the abstract of each patent; using a word frequency inverse text frequency algorithm and the patent claim data to determine the keyword information of each patent;

[0033] S103, based on the obtained abstract embedding vector and keyword information of each patent, establishing a text association relationship between any two patents; based on the obtained non-text information, establishing a non-text association relationship between any two patents;

[0034] S104, using the established text association relationships and non-text association relationships to construct a graph structure with patents as nodes and association relationships as edges;

[0035] S105, generating a training set of an initial neural network word embedding model based on the graph structure, and training the initial neural network word embedding model, obtaining a feature vector for each patent in the patent data set to be processed based on the trained neural network word embedding model; and processing the patent data set to be processed based on the obtained feature vector of each patent.

[0036] Using the above method, the text information and non-text information of each patent data are obtained, and the association relationship between the patent data is established based on the text information and non-text information, and a graph structure is further generated based on the association relationship. There is an association relationship of text information or non-text information between the nodes linked by edges in the graph structure, that is, patent data related to text information and non-text information will be connected by edges, and their positions in the graph structure will be closer. Further, based on the graph structure, a training set of an initial neural network word embedding model is constructed and trained. When the trained neural network word embedding model processes the patent data, the feature vector output contains both the text information and the non-text information of the patent. The neural network word embedding model is trained based on the graph structure data. In the feature space output by the neural network word embedding model, the feature vectors with multi-dimensional association relationships will be closer, and the feature vectors without association relationships will be farther away, so subsequent processing can be performed based on the feature vector.

[0037] Each step in the above patent data processing method is described in detail below.

[0038] In S101 above, the "unprocessed patent data set" refers to the patent data set used by the analyst for analysis, which includes several items of pre-collected patent data. The text information obtained for each patent data item may include the abstract text data, claim text data, and specification text data of each patent, and the non-text information obtained may include the patent's bibliographic information, citation relationships, and drawing information. The bibliographic information includes at least one of the following: applicant, inventor, right holder, application date, patent classification number, application location, and agency. The citation relationships include at least one of the following: family relationships and citations in documents (references explicitly stated in the specification).

[0039] In the above S102, the abstract data of each patent is processed using the pre-trained embedding model to obtain the abstract embedding vector of each patent, which can be:

[0040] For any patent, its abstract data is segmented to obtain a set of abstract segmentation words. A pre-trained word embedding model is used to embed each word in the abstract segmentation set to obtain a word vector for each word. Each word vector in the abstract segmentation set is concatenated to obtain a summary embedding vector for the abstract data.

[0041] Among them, when training the word embedding model, the abstract information of multiple patents can be obtained from the patent database, and a training sample set of the word embedding model can be generated based on the obtained abstract information to train the word embedding model. Since the description of patent abstract data has its own characteristics and features, the word embedding model is trained using patent abstract information rather than directly using a general word embedding model. In this way, the feature vector of the abstract output by the pre-trained word embedding model is more in line with the characteristics of the patent abstract itself. The embedding model can be implemented using models such as the n-gram model, the bag-of-words model, the neural network language model NNLM, the Word2Vec model, etc., and this application does not limit this.

[0042] Furthermore, for any patent, a pre-trained embedding model can be used to embed the abstract text to obtain a summary embedding vector for the abstract text data. That is, the complete text of the abstract is directly embedded to obtain a summary embedding vector. This embedding model can be implemented using models such as Sentence-BERT, Doc2Vec, and GloVe, and this application does not limit this. Similarly, this embedding model also obtains abstract information of multiple patents from a patent database, and generates a training sample set for the embedding model based on the obtained abstract information, thereby training the embedding model.

[0043] In the above S102, the keyword information of each patent is determined using the word frequency inverse text frequency algorithm and the patent claim data, which can be:

[0044] For any patent, perform word segmentation on its claim data to obtain a claim data word segmentation set;

[0045] For each segmentation word in the claim data segmentation set, determine the word frequency of the segmentation word in the patent, and determine the inverse text frequency of the segmentation word in the patent data set to be processed; obtain a criticality score for each segmentation word based on the word frequency and inverse text frequency;

[0046] Sorting the words in the claim data word set according to the criticality score, wherein the higher the criticality score, the higher the ranking;

[0047] The preset number of segmented words with the highest ranking are determined as the keyword information of the patent.

[0048] Specifically, term frequency TF(t) = (the number of times term t appears in the patent's claims) / (the total number of terms in the patent's claims);

[0049] Inverse document frequency (IDF) = log((the total number of patents in the patent dataset) / (the number of patents in the patent dataset whose claims contain the word t));

[0050] Keyness score = TF(t)*IDF(t).

[0051] For example, the information of word segmentation and key score is shown in Table 1.

[0052] Participle Crucial score A 100 B 91 C 89 ...... ......

[0053] Table 1

[0054] As shown in Table 1, the segmentations in the claim data segmentation set are sorted according to the criticality score, wherein the higher the criticality score, the higher the ranking. If the preset number is 3, the segmentations A, B, and C are determined as the keyword information of the patent.

[0055] In another embodiment, the segmented words with scores exceeding a preset score threshold may be determined as the keyword information of the patent. Still taking Table 1 as an example, for example, if the preset score threshold is 90, A and B are determined as the keyword information of the patent.

[0056] Using the above approach, keyword extraction from claim text data is performed. On the one hand, the claim data volume is large. Using an embedding model to embed the claim data would consume significant processing resources and result in low processing efficiency. Determining only keyword information from the claim text data can significantly conserve processing resources and improve overall processing efficiency. On the other hand, since the claims are the core of a patent, encompassing the entire technical solution protected by the patent, determining keyword information from only the claim text, rather than the entire patent text, can largely determine the core keyword information of the patent as a whole.

[0057] In another embodiment, when using the word frequency inverse text frequency algorithm and the patent claim data to determine the keyword information of each patent, the word frequency inverse text frequency algorithm can also be used to determine the keyword information of the specification text data of each patent; the keyword information of the specification text data and the keyword information of the claim data are combined to obtain the keyword information of each patent, wherein the above-mentioned combination can be to determine the intersection of the keyword information of the specification text data and the keyword information of the claim data as the keyword information of each patent.

[0058] That is, in this embodiment, keyword information is determined not only for the claim text data, but also for the specification text data, so that keyword information for each patent can be obtained to the greatest extent possible. The method for determining keyword information for the specification text data can be referred to above and will not be repeated here. Those skilled in the art can select any of the above embodiments to obtain patent keyword information according to actual needs.

[0059] After obtaining the abstract embedding vector and keyword information of each patent, a text association relationship between any two patents can be established based on the obtained abstract embedding vector and keyword information of each patent. In the above S103, the text association relationship between any two patents can be established based on the obtained abstract embedding vector and keyword information of each patent in the following manner.

[0060] For any two patent data, if the similarity of the abstract embedding vectors of the two patents is greater than a first preset similarity threshold, and / or the number of identical keywords included in the keyword information exceeds a preset number threshold, a text association relationship between the two patents is established.

[0061] That is, if the similarity between the abstract embedding vectors of two patents is greater than a first preset similarity threshold, it indicates that the abstracts of the two patents are substantially identical in terms of content and semantics. Abstracts often contain information about the core inventions and core technical effects of the patents, thus confirming a correlation between the two patents. Any similarity algorithm can be used to calculate the similarity of abstract embedding vectors, and this application does not limit this.

[0062] If the number of identical keywords included in the keyword information of two patents exceeds a preset threshold, it means that there is a lot of identical keyword information in the invention point information of the two patents, and then it can be determined that there is a correlation between the two patents.

[0063] This application proposes that, in order to establish text association relationships to the greatest extent possible, a text association relationship between two patents may be established when the similarity of the abstract embedding vectors of the two patents is greater than a first preset similarity threshold, or when the number of identical keywords included in the keyword information exceeds a preset threshold. In addition, in order to find the most relevant patents in terms of text, a text association relationship may be established when the similarity of the abstract embedding vectors of the two patents is greater than a first preset similarity threshold, and when the number of identical keywords included in the keyword information exceeds a preset threshold. This application does not impose any restrictions on this.

[0064] The above describes the method of establishing text association relationships. The following describes the specific method of establishing non-text association relationships in S103.

[0065] In one embodiment, the non-text information includes at least: bibliographic information and citation relationship. Based on the obtained non-text information, a non-text patent relationship is established between any two patents, including: for any two patent data, if the same information exists in the bibliographic items of the two patents, and / or there is a citation relationship, then a non-text association relationship is established between the two patents.

[0066] As described above, when conducting analysis, patent professionals need to consider not only the textual information of patents but also other information. Patent bibliographic data and citation relationships are fundamental to many patent analyses. For example, if multiple patents are filed by the same inventor on the same application date, they are often related (e.g., influencing each other's novelty). Furthermore, due to technological interplay between patents, there are often citation relationships between them. Therefore, this application proposes establishing non-textual relationships between patents based on bibliographic data and citation relationships. Bibliographic data here includes at least one of the following: applicant, inventor, right holder, application date, application location, agency, and patent classification number; citation relationships include family relationships and citations within documents. Specifically, to establish non-textual relationships across the broadest range, for any two patents, if their bibliographic data contain identical information or a citation relationship, a non-textual relationship is established between the two patents. For example, if two patents have the same applicant, a non-textual relationship is established between them; or if two patents belong to the same patent family, a non-textual relationship is established between them. In addition, in order to determine the most relevant patent in non-textual aspects, a non-textual association relationship between the two patents can be established only when there is the same information in the bibliographic items of the two patents and a citation relationship exists. This application does not limit this.

[0067] In another embodiment, the non-textual information includes not only bibliographic information and citation relationships but also patent drawing information. When establishing a non-textual association relationship, the similarity of each drawing information of any two patents may be calculated based on a preset similarity algorithm. For any two patent data, if the bibliographic information of the two patents contains the same information, there is a citation relationship, and / or the similarity between any drawing information exceeds a second preset similarity threshold, then a non-textual association relationship is established between the two patents.

[0068] That is, in addition to bibliographic information and citation relationships, the patent's drawing information is also taken into account. If the similarity between the drawings of two patents is high, then the two patents are likely to be related. Therefore, when the similarity between any drawing information is greater than a second preset similarity threshold, a non-textual association relationship between the two patents is also established. The preset similarity algorithm can adopt a general image similarity algorithm. This application does not limit the algorithm itself, but when training the preset similarity algorithm, it is necessary to use patent drawings for training so that the trained preset similarity algorithm has better performance when identifying and comparing similarities with patent drawings.

[0069] After establishing textual and non-textual relationships, a graph structure can be constructed, with patents as nodes and relationships as edges. Specifically, for a pending patent dataset, any two patents in the dataset are used as nodes. If there is a textual or non-textual relationship between the two patents, an edge is constructed between the two patents to form a graph structure.

[0070] In the graph structure constructed using the above method, each node in the graph structure is a patent. If two nodes are connected by an edge, it means that there is a textual association relationship or a non-textual association relationship between the two nodes. Therefore, the overall graph structure constructed can establish connections between the patents in the patent data set to be processed in two dimensions.

[0071] In the above S105, a training set of an initial neural network word embedding model is generated based on the graph structure, including:

[0072] The graph structure is split into multiple line segments. Each line segment includes two endpoint nodes and several intermediate nodes between the two endpoint nodes. Adjacent nodes are connected by edges. The number of nodes on any line segment is the sum of the number of input words and the number of output words of the initial neural network word embedding model.

[0073] For any line segment, according to the input and output word quantity requirements of the initial neural network word embedding model, the one-hot encoding of some of the nodes in it is used as a training sample feature value of the initial neural network word embedding model, and the one-hot encoding of other nodes in it is used as the label value of the training sample of the initial neural network word embedding model;

[0074] The training set of the initial neural network word embedding model is composed of multiple training samples corresponding to multiple line segments.

[0075] In this application, the line segments composed of connected nodes in the graph structure are regarded as a text sentence, and each node is equivalent to a word segment in the text sentence. Therefore, the training set of the initial neural network word embedding model can be constructed using multiple line segments of the graph structure.

[0076] Specifically, if the initial neural network word embedding model is a NNLM (Neural Network Language model) neural network language model, the neural network language model defines a window and uses a preset number of words in the window except the last word to predict the last word. In this application, a window is defined, for example, the window includes 5 nodes, and the first 4 nodes are used to predict the last node. Figure 2As shown in the figure, a line segment is split from the graph structure. The line segment includes two endpoints 1 and 5, and three intermediate nodes 2, 3, and 4. Adjacent nodes are connected by edges. The features of a training sample (NNLM model) for training the initial neural network word embedding model constructed with this line segment are the one-hot encodings of nodes 1, 2, 3, and 4, and the label is the one-hot encoding of node 5. The entire or partial structure of the graph structure is split into several line segments consisting of 5 nodes using the above method. Each line segment is allowed to have repeated nodes, that is, the same node can appear in multiple line segments. After generating training samples using the above method, the training sample set of the initial neural network word embedding model can be obtained.

[0077] If the initial neural network word embedding model is a Word2Vec model, the Word2Vec model defines a window, which includes a preset number of words, and uses the middle word to predict other words before and after the middle word in the window, or predicts the middle word before and after other words. In this application, a window is defined, for example, Figure 2 As shown, the window includes five nodes, and nodes 1, 2, 4, and 5 are used to predict node 3, or node 3 is used to predict 1, 2, 4, and 5. The training sample for an initial neural network word embedding model constructed using this line segment (the model is the Word2Vec model) has the features of the one-hot encoding of nodes 1, 2, 4, and 5, and the label is the one-hot encoding of node 3; or the features of the training sample are the one-hot encoding of node 3, and the labels are the one-hot encoding of nodes 1, 2, 4, and 5. After all or part of the graph structure is split into several line segments consisting of five nodes using the above method, and training samples are generated using the above method, the training sample set for the initial neural network word embedding model can be obtained.

[0078] This application does not limit the specific type of the initial neural network word embedding model.

[0079] In one implementation, the graph structure is split to obtain multiple line segments in the following manner.

[0080] The first method is:

[0081] The graph structure is split to obtain a preset number of line segments, wherein each line segment is obtained as follows: any node in the graph that has an adjacent node is taken as the first endpoint node, wherein having an adjacent node means that the node is connected to at least one other node through an edge; for the first endpoint node, the node with the highest degree of association with the first endpoint node is obtained as the next intermediate node, and for each intermediate node, the node with the highest degree of association with the intermediate node, except the previous node in the line segment, is obtained as the next intermediate node, until the second endpoint node is found, wherein the number of all nodes from the first endpoint node to the second endpoint node satisfies the sum of the number of input words and the number of output words of the initial neural network word embedding model.

[0082] The degree of association between nodes is determined based on the text association relationship and the non-text association relationship, that is, it can be measured by the text association degree and the non-text association degree. Specifically, it can be determined in the following way.

[0083] If, when establishing a text association relationship between any two patents, for any two patent data, if the similarity of the abstract embedding vectors of the two patents is greater than a first preset similarity threshold, then a text association relationship between the two patents is established.

[0084] Then, when determining the degree of text association, the similarity of the abstract embedding vectors of the two patents can be used as the degree of text association.

[0085] If, when establishing a text association relationship between any two patents, the number of identical keywords included in the keyword information exceeds a preset threshold, a text association relationship between the two patents is established.

[0086] Then, when determining the degree of text correlation, the number of identical keywords included in the two patents can be used as the degree of text correlation.

[0087] If, when establishing a text association relationship between any two patents, for any two patent data, if the similarity of the abstract embedding vectors of the two patents is greater than a first preset similarity threshold, and the number of identical keywords included in the keyword information exceeds a preset number threshold, then a text association relationship between the two patents is established.

[0088] Then, when determining the degree of text relevance, the following formula can be used for calculation.

[0089] C1 / (C A +C B )+B

[0090] Where C1 is the number of the same keywords included in the two patents, (C A +C B) is the sum of the number of keywords of the two patents, and B is the similarity of the abstract embedding vectors of the two patents.

[0091] If, when establishing a non-textual association relationship between any two patents, for any two patent data, if there is the same information in the bibliographic items of the two patents, then a non-textual association relationship between the two patents is established.

[0092] Therefore, when determining the degree of non-textual relevance, the number of identical information in the bibliographic items can be used as the degree of non-textual relevance.

[0093] If, when establishing a non-textual association relationship between any two patents, for any two patent data, if there is a citation relationship between the two patents, then a non-textual association relationship between the two patents is established.

[0094] Therefore, when determining the degree of non-textual association, the number of citations can be used as the degree of non-textual association. For example, for two patents, Patent 1 and Patent 2, it may be that Patent 1 cites Patent 2, or Patent 2 cites Patent 1, then the number of citations is 1. If the two patents cite each other, the number of citations is 2. Among them, the same family relationship can be understood as mutual citation. If two patents are both in the same family relationship (considered to have a citation number of 2) and the mutual citation relationship is clearly stated in the specification (the citation number is 2), then the number of citations of these two patents is 4.

[0095] If, when establishing a non-textual association relationship between any two patents, it is for any two patent data, if there is the same information in the bibliographic items of the two patents, and there is a citation relationship, then a non-textual association relationship between the two patents is established.

[0096] Therefore, when determining the degree of textual relevance, the sum of the number of identical information and the number of citations in the bibliographic items can be used as the degree of non-textual relevance.

[0097] Finally, the degree of association between nodes can be obtained by normalizing the obtained text association degree and non-text association degree and summing them up.

[0098] The first method is used to split the graph structure and obtain a preset number of line segments. Nodes with stronger correlation can be preferentially selected to form line segments. Subsequently, a training set of the initial neural network word embedding model is generated based on the line segments, and the initial neural network word embedding model is trained. This can make the embedding effect of the trained neural network word embedding model better.

[0099] The second method is to split the graph structure to obtain a preset number of line segments, where each line segment is obtained by taking any node in the graph structure that has an adjacent node as the first endpoint node, where having an adjacent node means that the node is connected to at least one other node through an edge. For the first endpoint node, randomly obtain the node connected to the first endpoint node as the next intermediate node. For each intermediate node, randomly obtain the node connected to the intermediate node, excluding the previous node in the line segment, as the next intermediate node, until the second endpoint node is found. The total number of nodes from the first endpoint node to the second endpoint node satisfies the sum of the number of input words and the number of output words of the initial neural network word embedding model.

[0100] The second method is used to split the graph structure to obtain a preset number of line segments. Since there is no need to calculate the degree of association, the efficiency of obtaining the line segments can be higher, and the training efficiency of the overall initial neural network word embedding model can also be higher.

[0101] After generating the training sample set of the initial neural network word embedding model, the training sample set can be used to perform supervised training on the initial neural network word embedding model. The trained neural network word embedding model is used to embed the one-hot encoding of the input patent data into the feature vector of the patent data.

[0102] In S105 above, the one-hot encoding of each patent may be input into the trained neural network word embedding model to obtain a feature vector for each patent in the patent data set to be processed. The patent data set to be processed may then be processed based on the obtained feature vector for each patent. When creating the one-hot encoding for each patent, each patent in the patent data set to be processed may be represented by one bit, resulting in a binary feature vector corresponding to the number of patents in the patent data set to be processed, thereby achieving one-hot encoding for each patent.

[0103] When processing the patent data set to be processed based on the obtained characteristic vector of each patent, the characteristic vector of each patent data in the patent data set to be processed can be used to classify using a preset classification method to obtain the category of each patent data, or, the characteristic vectors of any two patent data can be matched using a preset matching method to determine the degree of matching between the two patent data.

[0104] The neural network word embedding model is trained based on graph structure data. In the feature space output by the neural network word embedding model, the feature vectors with multi-dimensional associations (text and non-text associations) will be closer, and the feature vectors without associations will be farther away. Therefore, subsequent matching and classification processing can be performed based on this feature vector. The processing results take into account both the text information and the non-text information of the patent. Subsequent patent personnel can perform subsequent analysis based on this processing result, making the analysis results more comprehensive.

[0105] like Figure 3 As shown in FIG, it is a flowchart of a specific implementation method of the patent data processing method proposed in this application.

[0106] First, the text information and non-text information of each patent data in the patent data set to be processed are obtained.

[0107] For text information, the pre-trained embedding model is used to process the abstract data of each patent to obtain the abstract embedding vector of each patent, and the word frequency inverse text frequency algorithm is used to determine the keyword information of the claim data of each patent.

[0108] Based on the abstract embedding vector and keyword information obtained for each patent, a text association relationship is established between any two patents. The text association relationship is judged and determined in the following manner: if the similarity of the abstract embedding vectors of the two patents is greater than a first preset similarity threshold, or if the number of identical keywords included in the keyword information exceeds a preset number threshold, then a text association relationship is established between the two patents.

[0109] For non-text information, determine whether there is a citation relationship between patents and whether there are identical entries. If so, establish a non-text association relationship.

[0110] The graph structure is constructed using the established text association relationships and non-text association relationships.

[0111] The graph structure is used to generate a training set for the initial neural network word embedding model, and the initial neural network word embedding model is trained.

[0112] The trained neural network word embedding model is used to obtain the feature vector of each patent in the patent data set to be processed.

[0113] The patent data set to be processed is processed based on the obtained feature vector of each patent.

[0114] Based on the same inventive concept, this application also proposes a patented data processing device, which is as follows Figure 4 The device includes:

[0115] The information acquisition module 410 is used to acquire text information and non-text information of each patent data in the patent data set to be processed; wherein the text information at least includes: abstract text data and claim text data;

[0116] The information extraction module 420 is used to process the abstract data of each patent using a pre-trained embedding model to obtain an embedding vector for the abstract of each patent; and to determine the keyword information of each patent using a word frequency inverse text frequency algorithm and the patent claim data;

[0117] Relationship determination module 430 is used to establish a textual association relationship between any two patents based on the abstract embedding vector and keyword information of each patent obtained; and to establish a non-textual association relationship between any two patents based on the obtained non-text information;

[0118] A graph construction module 440 is used to construct a graph structure with patents as nodes and associations as edges using the established text association relationships and non-text association relationships;

[0119] The data processing module 450 is used to generate a training set of an initial neural network word embedding model based on the graph structure, train the initial neural network word embedding model, obtain a feature vector for each patent in the patent data set to be processed based on the trained neural network word embedding model; and process the patent data set to be processed based on the obtained feature vector of each patent.

[0120] In one embodiment, the information extraction module 420 is specifically used to perform word segmentation processing on the abstract data of any patent to obtain a set of abstract word segments, and use a pre-trained word embedding model to perform word embedding processing on each word in the abstract word segmentation set to obtain a word vector for each word; and splice each word vector in the abstract word segmentation set to obtain a summary embedding vector of the summary text data.

[0121] In one embodiment, the information extraction module 420 is specifically configured to perform word segmentation processing on the claim data of any patent to obtain a claim data word segmentation set;

[0122] For each segmentation word in the claim data segmentation set, determine the word frequency of the segmentation word in the patent, and determine the inverse text frequency of the segmentation word in the patent data set to be processed; obtain a criticality score for each segmentation word based on the word frequency and inverse text frequency;

[0123] Sorting the words in the claim data word set according to the criticality score, wherein the higher the criticality score, the higher the ranking;

[0124] The preset number of segmented words with the highest ranking are determined as the keyword information of the patent.

[0125] In one embodiment, the text information of each patent data also includes specification text data. The information extraction module 420 is also used to determine the keyword information of the specification text data of each patent using a word frequency inverse text frequency algorithm; and to obtain the keyword information of each patent by combining the keyword information of the specification text data and the keyword information of the claim data.

[0126] In one embodiment, the relationship determination module 430 is specifically used to establish a text association relationship between any two patent data if the similarity of the abstract embedding vectors of the two patents is greater than a first preset similarity threshold, and / or the number of identical keywords included in the keyword information exceeds a preset number threshold.

[0127] In one embodiment, the relationship determination module 430 is specifically used to establish a non-textual association relationship between any two patent data if there is identical information in the bibliographic items of the two patents and / or there is a citation relationship between the two patents.

[0128] In one embodiment, the relationship determination module 430 is also used to calculate the similarity of each drawing information of any two patents based on a preset similarity algorithm; for any two patent data, if there is the same information in the bibliographic items of the two patents, there is a citation relationship, and / or the similarity between any drawing information is greater than a second preset similarity threshold, then a non-text association relationship is established between the two patents.

[0129] In one embodiment, the graph construction module 440 is specifically used to take any two patents in the patent data set to be processed as nodes, and if there is a textual association relationship or a non-textual association relationship between the two patents, then construct an edge between the two patents to obtain the graph structure.

[0130] In one embodiment, the data processing module 450 is specifically configured to split the graph structure to obtain a plurality of line segments, wherein each line segment includes two endpoint nodes and a plurality of intermediate nodes between the two endpoint nodes, and adjacent nodes are connected by edges; the number of nodes on each line segment corresponds to the sum of the number of input words and the number of output words of the initial neural network word embedding model;

[0131] For any line segment, according to the input and output word quantity requirements of the initial neural network word embedding model, the one-hot encoding of some of the nodes therein is used as a feature value of a training sample of the initial neural network word embedding model, and the one-hot encoding of other nodes therein is used as the label value of the training sample of the initial neural network word embedding model;

[0132] A plurality of training samples corresponding to the plurality of line segments are used to form a training set of the initial neural network word embedding model.

[0133] In one embodiment, the data processing module 450 is specifically used to train the initial neural network word embedding model using the training set, and the trained neural network word embedding model is used to embed the one-hot encoding of the input patent data into the feature vector of the patent data.

[0134] In one embodiment, the data processing module 450 is specifically used to use the feature vector of each patent data in the patent data set to be processed, and use a preset classification method to classify to obtain the category of each patent data, or, for the feature vectors of any two patent data, use a preset matching method to match to determine the degree of matching between the two patent data.

[0135] The solutions in the embodiments of the present application can be implemented using various computer languages, for example, the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0136] In addition, an embodiment of the present invention further provides an electronic device, including a bus, a transceiver, a memory, a processor, and a computer program stored in the memory and executable on the processor. The transceiver, the memory, and the processor are respectively connected via a bus. When the computer program is executed by the processor, the various processes of the various embodiments of the patented data processing method are implemented, and the same technical effects can be achieved. To avoid repetition, they will not be described here.

[0137] For details, see Figure 5 As shown, the electronic device includes a bus 1110 , a processor 1120 , a transceiver 1130 , a bus interface 1140 , a memory 1150 , and a user interface 1160 .

[0138] In an embodiment of the present invention, the electronic device further includes: a computer program stored in the memory 1150 and executable on the processor 1120, wherein the computer program implements the above-mentioned patented data processing method when executed by the processor 1120.

[0139] The transceiver 1130 is configured to receive and send data under the control of the processor 1120 .

[0140] In an embodiment of the present invention, a bus architecture (represented by bus 1110) may include any number of interconnected buses and bridges, and bus 1110 connects various circuits including one or more processors represented by processor 1120 and a memory represented by memory 1150.

[0141] Bus 1110 represents one or more of any of several types of bus structures, including a memory bus and memory controller, a peripheral bus, an Accelerated Graphical Port (AGP), a processor, or a local bus using any of a variety of bus architectures. By way of example and not limitation, such architectures include an Industry Standard Architecture (ISA) bus, a Micro Channel Architecture (MCA) bus, an Enhanced ISA (EISA) bus, a Video Electronics Standards Association (VESA), and a Peripheral Component Interconnect (PCI) bus.

[0142] The processor 1120 can be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method embodiment can be completed by an integrated logic circuit of hardware in the processor or an instruction in the form of software. The above-mentioned processor includes: a general-purpose processor, a central processing unit (CPU), a network processor (NP), a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), a complex programmable logic device (CPLD), a programmable logic array (PLA), a microcontroller unit (MCU) or other programmable logic devices, discrete gates, transistor logic devices, discrete hardware components. The various methods, steps and logic block diagrams disclosed in the embodiments of the present invention can be implemented or executed. For example, the processor can be a single-core processor or a multi-core processor, and the processor can be integrated into a single chip or located on multiple different chips.

[0143] The processor 1120 can be a microprocessor or any conventional processor. The method steps disclosed in conjunction with the embodiments of the present invention can be directly executed by a hardware decoding processor, or can be executed by a combination of hardware and software modules in the decoding processor. The software module can be located in a readable storage medium known in the art, such as a random access memory (RAM), a flash memory (Flash Memory), a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), or a register. The readable storage medium is located in a memory, and the processor reads the information in the memory and performs the steps of the above method in conjunction with its hardware.

[0144] The bus 1110 may also connect various other circuits, such as peripheral devices, voltage regulators, or power management circuits. The bus interface 1140 provides an interface between the bus 1110 and the transceiver 1130. These are all well known in the art and are therefore not further described in this embodiment of the present invention.

[0145] The transceiver 1130 can be a single component or multiple components, such as multiple receivers and transmitters, providing a means for communicating with various other devices over a transmission medium. For example, the transceiver 1130 receives external data from other devices and transmits data processed by the processor 1120 to other devices. Depending on the nature of the computer system, a user interface 1160 may also be provided, such as a touch screen, physical keyboard, display, mouse, speaker, microphone, trackball, joystick, or stylus.

[0146] It should be understood that in an embodiment of the present invention, the memory 1150 may further include a memory remotely located relative to the processor 1120, and these remotely located memories may be connected to a server via a network. One or more parts of the aforementioned network may be an ad hoc network, an intranet, an extranet, a virtual private network (VPN), a local area network (LAN), a wireless local area network (WLAN), a wide area network (WAN), a wireless wide area network (WWAN), a metropolitan area network (MAN), the Internet, a public switched telephone network (PSTN), a plain old telephone service network (POTS), a cellular telephone network, a wireless network, a wireless fidelity (Wi-Fi) network, or a combination of two or more of the aforementioned networks. For example, the cellular telephone network and the wireless network can be a Global System for Mobile Communications (GSM) system, a Code Division Multiple Access (CDMA) system, a Worldwide Interoperability for Microwave Access (WiMAX) system, a General Packet Radio Service (GPRS) system, a Wideband Code Division Multiple Access (WCDMA) system, a Long Term Evolution (LTE) system, an LTE Frequency Division Duplex (FDD) system, an LTE Time Division Duplex (TDD) system, an Advanced Long Term Evolution (LTE-A) system, a Universal Mobile Telecommunications (UMTS) system, an Enhanced Mobile Broadband (eMBB) system, a Massive Machine Type of Communication (mMTC) system, an Ultra Reliable Low Latency Communications (uRLLC) system, and the like.

[0147] It should be understood that the memory 1150 in the embodiment of the present invention may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. Non-volatile memories include read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory.

[0148] Volatile memory includes random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct RAM bus random access memory (DRRAM). The memory 1150 of the electronic device described in the embodiments of the present invention includes, but is not limited to, the above and any other suitable types of memory.

[0149] In the embodiment of the present invention, the memory 1150 stores the following elements of the operating system 1151 and the application 1152: executable modules, data structures, or subsets thereof, or extended sets thereof.

[0150] Specifically, the operating system 1151 includes various system programs, such as a framework layer, a core library layer, and a driver layer, which are used to implement various basic services and process hardware-based tasks. The application 1152 includes various application programs, such as a media player and a browser, which are used to implement various application services. The program that implements the method of the embodiment of the present invention may be included in the application 1152. The application 1152 includes applets, objects, components, logic, data structures, and other computer system executable instructions that perform specific tasks or implement specific abstract data types.

[0151] In addition, an embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the various processes of the various embodiments of the above-mentioned patent data processing method are implemented, and the same technical effects can be achieved. To avoid repetition, they will not be described here.

[0152] An embodiment of the present invention also provides a computer program comprising one or more computer instructions, which, when executed by a processor, implement the steps of the above-mentioned patented data processing method and generate, in whole or in part, the process or function described in the embodiment of the present application.

[0153] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present application.

[0154] Obviously, those skilled in the art may make various changes and modifications to this application without departing from the spirit and scope of this application. Thus, if these modifications and variations of this application fall within the scope of the claims of this application and their equivalents, this application is intended to include these modifications and variations.

Claims

1. A patent data processing method, characterized in that: include: Obtaining text information and non-text information of each patent data in the patent data set to be processed; The text information at least includes: abstract text data, claim text data; The abstract data of each patent is processed using a pre-trained embedding model to obtain the abstract embedding vector of each patent; the keyword information of each patent is determined using the word frequency inverse text frequency algorithm and the patent claim data; Based on the abstract embedding vector and keyword information of each patent, a textual correlation relationship between any two patents is established; based on the obtained non-textual information, a non-textual correlation relationship between any two patents is established; Utilize the established textual and non-textual relationships to construct a graph structure with patents as nodes and relationships as edges; A training set of an initial neural network word embedding model is generated based on the graph structure, and the initial neural network word embedding model is trained. A feature vector of each patent in the patent data set to be processed is obtained based on the trained neural network word embedding model; and the patent data set to be processed is processed based on the obtained feature vector of each patent.

2. The method according to claim 1, characterized in that The abstract text data of each patent is processed using the pre-trained embedding model to obtain the abstract embedding vector of each patent, including: For any patent, its abstract data is segmented to obtain a set of abstract segmentation words, and each segmentation word in the abstract segmentation word set is embedded using a pre-trained word embedding model to obtain a word vector for each word; each word vector in the abstract segmentation word set is concatenated to obtain a summary embedding vector of the abstract text data.

3. The method according to claim 1, characterized in that The keyword information of each patent claim data is determined by using the word frequency inverse text frequency algorithm, including: For any patent, perform word segmentation on its claim data to obtain a claim data word segmentation set; For each segmentation word in the claim data segmentation set, determine the word frequency of the segmentation word in the patent, and determine the inverse text frequency of the segmentation word in the patent data set to be processed; obtain a criticality score for each segmentation word based on the word frequency and inverse text frequency; Sorting the words in the claim data word set according to the criticality score, wherein the higher the criticality score, the higher the ranking; The preset number of segmented words with the highest ranking are determined as the keyword information of the patent.

4. The method according to claim 1, wherein The text information of each patent data also includes specification text data, and the method further includes: Determining keyword information of the specification text data of each patent using a word frequency inverse text frequency algorithm; said determining keyword information of each patent using a word frequency inverse text frequency algorithm and patent claim data includes: The keyword information of each patent is obtained by integrating the keyword information of the specification text data and the keyword information of the claim data.

5. The method according to claim 1 or 4, characterized in that Based on the abstract embedding vector and keyword information of each patent, a textual association relationship between any two patents is established, including: For any two patent data, if the similarity of the abstract embedding vectors of the two patents is greater than a first preset similarity threshold, and / or the number of identical keywords included in the keyword information exceeds a preset number threshold, a text association relationship between the two patents is established.

6. The method according to claim 1, characterized in that The non-text information includes at least: bibliographic information and citation relationships; The non-textual patent relationship between any two patents is established based on the obtained non-textual information, including: For any two patent data, if there is the same information in the bibliographic items of the two patents and / or there is a citation relationship, a non-textual association relationship between the two patents is established.

7. The method according to claim 1, characterized in that The non-text information includes: bibliographic information, citation relationships, and patent drawing information; the method includes: Calculate the similarity of each drawing information of any two patents based on a preset similarity algorithm; The non-textual patent relationship between any two patents is established based on the obtained non-textual information, including: For any two patent data, if there is identical information in the bibliographic items of the two patents, there is a citation relationship, and / or the similarity between any drawing information is greater than a second preset similarity threshold, a non-text association relationship is established between the two patents.

8. The method according to claim 1, characterized in that The method of utilizing the established text association relationships and non-text patent relationships to construct a graph structure with patents as nodes and association relationships as edges includes: For the patent data set to be processed, any two patents in the set are taken as nodes. If there is a textual association relationship or a non-textual association relationship between the two patents, an edge between the two patents is constructed to obtain the graph structure.

9. The method according to claim 1, characterized in that The generating of a training set of an initial neural network word embedding model based on the graph structure includes: Splitting the graph structure to obtain multiple line segments, each line segment includes two endpoint nodes and a plurality of intermediate nodes between the two endpoint nodes, and adjacent nodes are connected by edges; the number of nodes on each line segment is the sum of the number of input words and the number of output words of the initial neural network word embedding model; For any line segment, according to the input and output word quantity requirements of the initial neural network word embedding model, the one-hot encoding of some of the nodes therein is used as a feature value of a training sample of the initial neural network word embedding model, and the one-hot encoding of other nodes therein is used as the label value of the training sample of the initial neural network word embedding model; A plurality of training samples corresponding to the plurality of line segments are used to form a training set of the initial neural network word embedding model.

10. The method according to claim 9, characterized in that The graph structure is split to obtain multiple line segments, including: A preset number of line segments are obtained, wherein each line segment is obtained as follows: any node in the graph that has an adjacent node is taken as the first endpoint node, wherein having an adjacent node means that the node is connected to at least one other node through an edge; for the first endpoint node, the node with the highest degree of association with the first endpoint node is obtained as the next intermediate node; for each intermediate node, the node with the highest degree of association with the intermediate node, excluding the previous node in the line segment, is obtained as the next intermediate node, until the second endpoint node is found, wherein the number of all nodes from the first endpoint node to the second endpoint node satisfies the sum of the number of input words and the number of output words of the initial neural network word embedding model; the degree of association between nodes is obtained based on text association and non-text association.

11. The method according to claim 9, characterized in that The training of the initial neural network word embedding model includes: The initial neural network word embedding model is supervisedly trained using the training set. The trained neural network word embedding model is used to embed the one-hot encoding of the input patent data into the feature vector of the patent data.

12. The method according to claim 1, characterized in that The processing of the patent data set to be processed based on the obtained feature vector of each patent includes: The characteristic vector of each patent data in the patent data set to be processed is used to classify the data using a preset classification method to obtain the category of each patent data, or the characteristic vectors of any two patent data are matched using a preset matching method to determine the matching degree of the two patent data.

13. A patented data processing device, characterized in that, include: An information acquisition module, used to acquire text information and non-text information of each patent data in the patent data set to be processed; The text information at least includes: abstract text data, claim text data; The information extraction module is used to process the abstract data of each patent using a pre-trained embedding model to obtain the abstract embedding vector of each patent; and to determine the keyword information of each patent using the word frequency inverse text frequency algorithm and the patent claim data; A relationship determination module is used to establish a textual association relationship between any two patents based on the abstract embedding vector and keyword information of each patent; and to establish a non-textual association relationship between any two patents based on the obtained non-text information; A graph construction module is used to construct a graph structure with patents as nodes and associations as edges using the established text association relationships and non-text association relationships; A data processing module is used to generate a training set of an initial neural network word embedding model based on the graph structure, train the initial neural network word embedding model, obtain a feature vector for each patent in the patent data set to be processed based on the trained neural network word embedding model; and process the patent data set to be processed based on the obtained feature vector of each patent.

14. An electronic device comprising a processor and a memory, wherein the memory stores a computer program, wherein: The processor executes the computer program stored in the memory to implement the steps in the method according to any one of claims 1 to 12.

15. A computer program comprising computer instructions, characterized in that When the computer instructions are executed by a processor, the steps of the method according to any one of claims 1 to 12 are implemented.

16. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the computer program implements the steps of the method according to any one of claims 1 to 12.

Images