Patent intelligent modification method, auxiliary reply method and system based on DIKW graph

Abstract

The present invention discloses a patent intelligent modification method based on the DIKW graph, which can correct or modify the patent application document according to the provisions of patent laws and regulations, so that the patent application document complies with the provisions of patent laws and regulations, and includes: collecting tripartite data; importing the public data and the tripartite data into a patent input relational database; converting the relational patent input data into a DIKW graph through data initialization, real-time processing and incremental import framework; storing the DIKW graph patent information in a patent graph database; reconstructing the content of the patent graph data through a distributed graph processing framework; all the reconstructed contents of the patent application document to be modified form the modified patent application document. The present invention also discloses a patent intelligent modification device based on the DIKW graph, and a patent intelligent auxiliary reply method and system based on the DIKW graph. The present invention achieves that the reply to the examination opinion complies with the legal provisions and the reply speed is fast.

Description

Technical Field

[0001] The present invention relates to the field of DIKW graph technology, and in particular to a patent intelligent modification method, an auxiliary reply method and a system thereof based on a DIKW graph. Background Art

[0002] The DIKW hierarchy is a system of data, information, knowledge and wisdom, which explains the relationship between the four elements. It has also been discussed in the field of information science and knowledge management. It is one of the most basic, widely recognized and most taken-for-granted models.

[0003] Data, Information, Knowledge and Wisdom Model, also known as DIKW model. The DIKW graph architecture includes data graph, information graph and knowledge graph.

[0004] Data graphs are various data structures, such as stacks, queues, etc. They can record the frequency attributes of objects, such as time frequency, space frequency, and structure frequency, and then calculate their comprehensive frequency to serve the information cleaning and semantic computing of information graphs and knowledge graphs.

[0005] Information Graph is used to store various interactive relationships between patent content resources, such as the combination of product components, product design steps, etc., and visualize them in the form of a directed graph.

[0006] Knowledge Graph is a directed graph used to record rules and can be used for various semantic calculations.

[0007] The prior art does not provide a technique for using the DIKW spectrum for analyzing patent application texts. Therefore, using the DIKW spectrum for analyzing patent application texts is still a research direction for those skilled in the art. Summary of the invention

[0008] To this end, one of the purposes of the present invention is to propose a patent intelligent modification method based on the DIKW map, which can correct or modify the patent application documents according to the provisions of patent laws and regulations, so that the patent application documents comply with the provisions of patent laws and regulations, which is conducive to saving a lot of manpower costs.

[0009] The second purpose of the present invention is to propose a patent intelligent auxiliary reply method based on DIKW graph, which can correct or modify the patent application documents according to the provisions of patent laws and regulations and the requirements of patent examination opinion notices, so that the patent application documents comply with the provisions of patent laws and regulations and the requirements of patent examination opinion notices, and assist the patent examination opinion notice responder to respond to the examination opinion. This is equivalent to equipping the patent examination opinion notice responder with an artificial intelligence assistant, saving working time and improving work efficiency.

[0010] The third purpose of the present invention is to propose a patent intelligent modification device based on the DIKW graph, which can be used as a professional artificial intelligence assistant in the patent field.

[0011] The fourth objective of the present invention is to propose a patent intelligent auxiliary reply system based on the DIKW graph, which can be used as a professional artificial intelligence assistant in the patent field.

[0012] One of the purposes of the present invention is achieved by the following technical solution:

[0013] The patent intelligent modification method based on the DIKW map can correct or modify the patent application documents according to the provisions of patent laws and regulations, so that the patent application documents comply with the provisions of patent laws and regulations, including:

[0014] Collecting three-party data, wherein the three parties are a sender, an intermediary, and a receiver; the three-party data at least includes the patent application document to be modified of the sender, the modification requirements of the patent application document to be modified of the receiver, and the patent data of the intermediary;

[0015] Importing the three-party data into a patent input relational database, wherein the data stored in the patent input relational database is referred to as relational patent input data;

[0016] The relational patent input data is converted into a DIKW graph through data initialization, real-time processing and incremental import framework, wherein the DIKW graph is called DIKW graph patent information, and the DIKW graph patent information includes a sender DIKW graph, a receiver DIKW graph P and an intermediate DIKW graph N;

[0017] The DIKW graph patent information is stored in a patent graph database, wherein the DIKW graph patent information stored in the patent graph database is referred to as patent graph data;

[0018] The patent image data is reconstructed through a distributed graph processing framework, wherein the patent image data after content reconstruction is referred to as patent content reconstruction data, and the content reconstruction includes the following sub-steps:

[0019] Sub-step S101: According to the entity S that the receiver needs to reconstruct, based on the intermediate party DIKW graph N and the receiver DIKW graph P, find the i-degree node set A of the entity S in the intermediate party DIKW graph = {a 1 ,a 2 ,…,a n}, where i is initially 1;

[0020] Sub-step S102: According to the node set A={a 1 ,a 2 ,…,a n}, matching node set B in the receiver DIKW graph = {b 1 ,b 2 ,…,b m}, where b j ∈{A,P}, if B=ε, then i=i+1, and the process returns to the sub-step S101; otherwise, the process proceeds to sub-step S103;

[0021] Sub-step S103: Each element b in the set B obtained in sub-step S102 j The entity S is passed into a previously trained Embedding layer to obtain two word vectors ξ 1 and 2 , calculate the cosine similarity, as shown in formula (4-8):

[0022]

[0023] Sub-step S104: input the calculated cosine similarity result into the sigmoid function, convert it into probability, select entities with probability values ​​greater than a set threshold, and replace the original entities to be reconstructed, which is the reconstructed content;

[0024] The entire reconstructed content of the patent application document to be modified forms the modified patent application document.

[0025] The second object of the present invention is achieved by adopting the following technical solution:

[0026] The patent intelligent auxiliary reply method based on the DIKW map can correct or modify the patent application documents according to the provisions of patent laws and regulations and the requirements of the patent examination opinion notice, so that the patent application documents comply with the provisions of patent laws and regulations and the requirements of the patent examination opinion notice, and assist the patent examination opinion notice respondent to respond to the examination opinion.

[0027] The patent intelligent assisted reply method based on the DIKW map includes the patent intelligent modification method based on the DIKW map, wherein the modification requirement of the patent application document to be modified is the examination opinion notice of the patent application document to be modified.

[0028] The third object of the present invention is achieved by adopting the following technical solution:

[0029] The patent intelligent modification device based on the DIKW map can correct or modify the patent application documents according to the provisions of the patent laws and regulations, so that the patent application documents comply with the provisions of the patent laws and regulations, including:

[0030] A data collection module is used to collect three-party data, wherein the three parties are a sender, an intermediary, and a receiver; the three-party data at least includes the patent application document to be modified of the sender, the modification requirements of the patent application document to be modified of the receiver, and the patent data of the intermediary;

[0031] A first relational database system is used to import the three-party data into a patent input relational database, wherein the data stored in the patent input relational database is referred to as relational patent input data;

[0032] A DIKW graph generation module is used to convert the relational patent input data into a DIKW graph through data initialization, real-time processing and incremental import framework, wherein the DIKW graph is called DIKW graph patent information, and the DIKW graph patent information includes a sender DIKW graph, a receiver DIKW graph P and an intermediate DIKW graph N;

[0033] A first graph database system is used to store the DIKW graph patent information in a patent graph database, wherein the DIKW graph patent information stored in the patent graph database is referred to as patent graph data;

[0034] The graph data content reconstruction module reconstructs the content of the patent graph data through a distributed graph processing framework, wherein the patent graph data after content reconstruction is referred to as patent content reconstruction data. The graph data content reconstruction module includes the following submodules:

[0035] Submodule M101: According to the entity S that needs to be reconstructed by the receiver, based on the intermediate party DIKW graph N and the receiver DIKW graph P, find the i-degree node set A of the entity S in the intermediate party DIKW graph = {a 1 ,a 2 ,…,a n}, where i is initially 1;

[0036] Submodule M102: According to the node set A={a 1 ,a 2 ,…,a n}, matching node set B in the receiver DIKW graph = {b 1 ,b 2 ,…,b m}, where b j ∈{A,P}, if B=ε, then i=i+1, and return to the submodule M101; otherwise, enter the submodule M103;

[0037] Submodule M103: Each element b in the set B obtained by the submodule S102 is j The entity S is passed into a previously trained Embedding layer to obtain two word vectors ξ 1 and 2 , calculate the cosine similarity, as shown in formula (4-8):

[0038]

[0039] Submodule M104: Input the calculated cosine similarity result into the sigmoid function, convert it into probability, select entities with probability values ​​greater than the set threshold, and replace the original entities to be reconstructed, which is the reconstructed content;

[0040] The entire reconstructed content of the patent application document to be modified forms the modified patent application document.

[0041] The fourth object of the present invention is achieved by adopting the following technical solution:

[0042] The patent intelligent auxiliary reply system based on the DIKW map can correct or modify the patent application documents according to the provisions of patent laws and regulations and the requirements of the patent examination opinion notice, so that the patent application documents comply with the provisions of patent laws and regulations and the requirements of the patent examination opinion notice, and assist the patent examination opinion notice respondent to respond to the examination opinion;

[0043] The patent intelligent assisted reply method based on the DIKW graph includes the patent intelligent modification device based on the DIKW graph, wherein the modification requirement of the patent application document to be modified is the examination opinion notice of the patent application document to be modified.

[0044] According to other technical solutions of the present invention, it may also include one or more technical features described herein. As long as such a combination of technical features is feasible, the new technical solutions thus formed belong to a part of the present invention.

[0045] The present invention relies on and combines the modeling management of the entire software engineering development life cycle, and is oriented to the entire process of intellectual property handling. For invention patents within the life cycle, the application and response of invention patents are considered from the perspective of content transmission, and the differences in content understanding caused by factors such as differences in personnel personality and domain knowledge are considered. The content is modeled from three levels: data, information, and knowledge. The present invention studies the storage, processing, and conversion of data, information, and knowledge. Real-time analysis and evaluation of invention patents are achieved, thereby achieving real-time understanding of the applicant's intentions. In response to the demand for modeling patent content resources under uncertain circumstances, research is conducted based on the relevant content stored in data graphs, information graphs, and knowledge graphs.

[0046] Compared with the prior art, the present invention has the following beneficial effects:

[0047] 1) Retrieve various documents, analyze the differences in understanding of various types of knowledge among different groups of people, and store them in the DIKW map. Research on modeling and conversion mapping of relevant content resources from the perspective of the DIKW conceptual framework. By introducing the DIKW map architecture, the storage, processing, transmission modeling and analysis of content resources are carried out;

[0048] 2) Build a user DIKW profile based on the differences in understanding of patent technology content between applicants and examiners, and reconstruct content resources to improve the communication efficiency between the two parties;

[0049] 3) When modifying the invention patent technical documents, adopt the principle of advantage, and carry out competitive filling based on different rules, combining purpose-driven and value-driven;

[0050] 4) According to the requirements of patent laws and regulations or the modification requirements of the examination opinions, the automatic content reconstruction of the patent application documents is realized, that is, the modification of the patent application documents, thereby reducing the workload of the staff, ensuring that the patent application documents meet the legal requirements, reducing unnecessary repeated modifications in the subsequent review process, and shortening the review process time.

[0051] 5) According to the modification requirements of the examination opinions, the automatic content reconstruction of the patent application documents is realized, that is, the modification of the patent application documents, thereby reducing the workload of the staff and improving the quality and speed of responding to the examination opinions. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] The features, advantages and characteristics of the present invention will be better understood through the following description of specific embodiments with reference to the accompanying drawings, in which:

[0053] Figure 1 : Schematic diagram of the DIKW graph architecture in the prior art;

[0054] Figure 2: A schematic diagram of the graph database content of the invention patent application document in the preferred embodiment of the patent intelligent modification method based on the DIKW graph of the present invention;

[0055] Figure 3 : A schematic diagram of the graph database content of the opinion statement in the preferred embodiment of the patent intelligent modification method based on the DIKW graph of the present invention;

[0056] Figure 4 : A multimodal semantic structure diagram based on a DIKW graph of a preferred embodiment of the patent intelligent modification method based on a DIKW graph of the present invention;

[0057] Figure 5 : A schematic diagram of the association relationship among the data graph, information graph and knowledge graph of the preferred embodiment of the patent intelligent modification method based on the DIKW graph of the present invention;

[0058] Figure 6 : A multimodal semantic modeling flow chart based on a DIKW graph of a preferred embodiment of the patent intelligent modification method based on a DIKW graph of the present invention;

[0059] Figure 7 : A schematic diagram of a content transmission path solution of a preferred embodiment of the patent intelligent modification method based on the DIKW graph of the present invention;

[0060] Figure 8 : Schematic diagram of content reconstruction scheme under different transmission paths of the preferred embodiment of the patented intelligent modification method based on DIKW graph of the present invention;

[0061] Fig. 9 : A flow chart of interactive area division and transmission optimization based on DIKW graphs of a preferred embodiment of the patent intelligent modification method based on DIKW graphs of the present invention;

[0062] Fig.10 : Schematic diagrams of six situations of blocks of the type to which the content sub-graph belongs according to the preferred embodiment of the patent intelligent modification method based on the DIKW graph of the present invention;

[0063] Fig.11 : A flow chart of a content reconstruction algorithm based on a DIKW graph according to a preferred embodiment of the patent intelligent modification method based on a DIKW graph of the present invention;

[0064] Fig.12 : The content semantic value calculation framework diagram of the redundant and error-oriented imitation blockchain node and node network of the present invention;

[0065] Fig.13 : A schematic diagram of the value distribution of nodes under different dimensions of a preferred embodiment of the patent intelligent modification method based on the DIKW graph of the present invention;

[0066] Fig.14 : A comparative diagram of the semantic value calculation at the knowledge graph level of the present invention and the two-dimensional Boolean operation in graphics;

[0067] Fig.15 : Schematic diagram of a node network with two central nodes of the present invention;

[0068] Fig.16 : A DIKW graph construction framework diagram of a preferred embodiment of the patent intelligent modification method based on the DIKW graph of the present invention;

[0069] Fig.17 : A schematic diagram of the logical structure of a relational database table of a preferred embodiment of the patent intelligent modification method based on the DIKW graph of the present invention;

[0070] Fig.18 : Patent modification display interface of the preferred embodiment of the patent intelligent modification method based on DIKW graph of the present invention; DETAILED DESCRIPTION

[0071] Hereinafter, the present invention will be further described in conjunction with the accompanying drawings and specific implementation modes.

[0072] Patent application documents: refers to the documents in the patent application stage, including the documents submitted by the patent applicant during the patent application process, and various notices issued by the Patent Office, such as correction notices, examination opinion notices, etc.

[0073] Where permitted by law and consistent with semantic logic, the word "include" in this article should be understood as "at least include", used to express an open-ended inclusion semantics.

[0074] Fig.17 The "participle part of speech" in this article also corresponds to the phrase part of speech table. It should be understood that the "participle" in this article expresses the meaning of the phrase separated from the text, and should not be understood as the meaning of the grammatical unit participle in English.

[0075] Based on the inventor's conception, the following describes in detail a preferred implementation of the patented intelligent modification method based on the DIKW map. Figure 1 It is a schematic diagram of the DIKW graph architecture in the prior art.

[0076] 1. Constructing a data map of invention patent application documents

[0077] 1.1、Application data map construction

[0078] When applying for a new invention patent, you generally only need to submit a description, claims, abstract, drawings, and invention patent application.

[0079] like Figure 2 The following is the graph database content of the application.

[0080] 1.2. Construction of data map of opinion statement

[0081] During the substantive examination stage of the patent application process, the examiner will examine whether the patent application documents meet the authorization conditions. If the authorization conditions are not met, the examiner will issue a notice. The applicant can make an opinion statement or modify the patent application documents based on the notice until the patent is authorized or rejected.

[0082] The content of the opinion statement includes: patent application information, statement matters, etc. Fig.13 Shown is the content of the opinion statement database.

[0083] like Figure 3 Shown is the graph database content of the opinion statement.

[0084] 2. Content reconstruction and content semantic value calculation based on DIKW graph

[0085] 2.1 Multimodal semantic modeling based on DIKW graph

[0086] The intelligent assisted reply and examination of invention patents have great requirements for natural language semantic understanding. Only by modeling the semantics of natural language, especially the semantics of multimodal content, can the machine understand redundant and complex semantics. Knowledge graphs have great potential in processing natural language expressions in an interpretable way. However, existing knowledge graph technologies face challenges in the content of semantic relationship fusion expressions, and the processing efficiency of multimodal mixed content in knowledge graphs needs to be improved. This paper maps the mixed-form semantic expression carrier to the data, information, knowledge and wisdom model (DIKW model), and through the metamodel and executable formal definition of the DIKW architecture, maps the mixed-form expressions of natural language to typed elements and subgraphs of typed element expressions, and provides improved means to improve the efficiency of processing distributed resources.

[0087] Multimodal semantic modeling based on DIKW graph is mainly used to obtain the maximum benefit improvement and value of strategic investment practice from efficiency improvement and strategic investment practice. Figure 4 It is a multimodal semantic modeling structure diagram based on the DIKW graph.

[0088] In the DIKW graph, discrete things are generally recorded on the data graph, usually expressed as three frequencies: time, space, and structure. In the invention patent application document, the data graph can record the operation sequence of each step, the spacing, size, and combination of each component in the product, etc. The information graph records the interaction relationship of each entity. New nodes can be inferred on the information graph by calculating the interaction frequency and re-included in the data graph. For example, when the size of n-1 accessories of a product is known, the size of the last accessory can be inferred and judged, and if it meets the requirements, it is recorded as a new node; the relationship between classes on the knowledge graph is further abstracted using relationship extraction rules, and the integrity of the technical solution expression is supplemented to improve the efficiency of communication. Figure 5 The following is a schematic diagram of the relationship between data graph, information graph and knowledge graph. The meanings of the arrows in the figure are shown in Table 1.

[0089] Table 1 Symbols

[0090]

[0091]

[0092] The main process of multimodal semantic modeling based on DIKW graph is as follows: Figure 6 As shown:

[0093] Step 1: Obtain the applicant's demand description for the invention patent claims, which is described by one or more applicants. There is redundancy and inconsistency in the multi-source demand. If the agent collects the applicant's demand for patent claims in a research manner, when there are many applicants or many claim demands, the applicant's description of the claim demand will be very confusing;

[0094] Step 2: Extract the keywords describing the applicant's requirements for the claims through information extraction technology, and record the keywords in the form of time, space and structure frequency on the data map. The time frequency Tem_f is the time scale of the keyword's appearance, the space frequency Spa_f is the coordinate scale of the keyword, and the structure frequency Str_f is the number of relationships between the keyword in the entity relationship model;

[0095] Step 3: Mark the claim keywords with a time frequency of 1 as the target object, and calculate the total frequency of the data according to formula (4-1).

[0096] frequency()=Str_f()*Tem_f()*Spa_f() formula (4-1)

[0097] Step 4: Calculate the support and confidence of the claim keywords according to formulas (4-2) and (4-3). Based on the results, clean the discrete data on the data map to remove marginal data and redundant data. DG Represents the sum of the frequencies of all nodes on the data graph, r i 、r j Refers to two different nodes.

[0098]

[0099]

[0100] Step 5: Perform semantic analysis on the keyword objects recorded in the data graph, taking nouns as entities and verbs as relationships to form an information graph.

[0101] Step 6: Calculate the interaction frequency between nodes on the information graph according to formula (4-4). The frequency of interaction between nodes indicates the number of interactions between nodes. E1 and E2 refer to any two computing nodes in the calculation process, and P represents all complete interaction paths between nodes.

[0102] frequency(E1,E2)=∑ P(E1→E2) 1Formula (4-4)

[0103] Step 7: Calculate the total frequency of all non-discrete nodes on the information graph according to formula (4-5) frequency , Frequency IG It represents the interaction frequency of entities on the information graph, and α and β are the weights of the data graph frequency and the information graph frequency.

[0104] Total frequency =αFrequency DG *βFrequency IG Formula (4-5)

[0105] Step 8: Calculate the cohesion of each node in the information graph according to formula (4-6). Integrate the different connected entities with the largest entity cohesion in the form of attributes and operations, and express them in the form of new nodes on the information graph to enhance the cohesion of the claim and improve the abstractness. EI Indicates external interaction, Degree II Represents internal interactions.

[0106]

[0107] Step 9: Update the frequency recorded on the data graph based on the new nodes and new relationships generated by the information graph.

[0108] Step 10: Based on the semantic relationship on the information graph, the knowledge graph is assisted to infer new relationships, and the accuracy Cr is calculated according to formula (4-7), where π represents a path between node 1 and node 2, Q(E1→E2) represents all paths between node E1 and node E2, R represents the relationship between the two nodes, Q represents all paths in the circled graph, and θ(π) represents the training weight. When the accuracy exceeds a certain set threshold, the new relationship is considered to be established:

[0109]

[0110] Step 11: Optimize the knowledge graph and delete redundant nodes and loops.

[0111] 2.2 Content transmission and optimization based on DIKW graph

[0112] 2.21 Interaction area division and transmission optimization based on DIKW graph

[0113] The accuracy and transmission speed of patents are important criteria for measuring the quality of graphs. In order to eliminate the misunderstanding of patent content by relevant personnel, the graph based on the DIKW graph proposed in this paper relies on the traditional patent application scheme and introduces a series of optimization measures, such as classifying and integrating patent content to enhance the referenceability of the content, adding personalized and interactive semantics on the basis of ensuring professionalism, and reducing misunderstandings.

[0114] By building a DIKW content library, the content of the delivered patent application scheme can be reconstructed to solve the problem of differences in understanding of patent content between applicants, agents and other technical personnel and examiners. The reconstructed patent content can achieve personalized expression on the basis of retaining the key semantics of the original content. At the same time, the system also integrates and optimizes the delivered patent content to improve the accuracy and efficiency of content delivery. Figure 7 This is a schematic diagram of the content transmission path solution. Figure 8 Schematic diagram of content reconstruction solutions under different transmission paths, where S, R, and D represent content reconstruction at the applicant, intermediate party, and receiver, respectively.

[0115] There may be a situation where a certain concept does not exist in the content library of one party, or the same concept may correspond to different understandings in the content libraries of both parties. In order to achieve consistency in the understanding of the same concept between the two parties, this article reconstructs the transferred concept so that the two parties can achieve differentiated understanding, and the reconstruction can be performed by any of the three parties.

[0116] The content libraries of both applicants and examiners constructed in this paper include private libraries and public libraries. The private library records user privacy information, and the public library is the features uploaded by users. The data source of the public library is the features in the private library, and users can choose to upload them according to their wishes.

[0117] The specific process of interactive area division and transmission optimization based on DIKW graph is as follows: Fig. 9 As shown:

[0118] Step 1: Applicant S sends the patent content and performs semantic analysis on the edge device to obtain the patent content subgraphs and subgraph keywords.

[0119] Step 2: Based on the content libraries of the applicant, the intermediary and the recipient, analyze the patent content sub-graphs and sub-graph keywords, determine the type of block to which they belong, and determine the transmission plan and reconstruction plan accordingly. The type of block to which the sub-graph belongs is shown in Figure 10. This includes situations where the three-party content libraries completely overlap, are independent of each other, and overlap. Among them, situation B1 can be fully understood without reconstruction, and has the highest transmission efficiency; situation B2 has no overlapping public library, and has the lowest transmission efficiency.

[0120] Step 3: Screen the reconstruction plan through the integration of storage, conversion, and computing. The screening is based on three aspects: storage cost, computing cost, and conversion cost.

[0121] Step 4: Reconstruct patent content through sub-graph migration.

[0122] 2.2.2 Content reconstruction based on DIKW graph

[0123] This section proposes a solution to the problem of transmission content reconstruction. The flow chart of the content reconstruction algorithm based on the DIKW atlas is as follows: the sender sends data to the receiver, and the receiver extracts key information and encapsulates the information into a package, and confirms whether the package can be verified by the receiver's DIKW atlas library. If it can be verified, it will be forwarded directly. Otherwise, detailed information such as the receiver's DIKW atlas content library and the third-party DIKW atlas content library will be obtained, and a reconstruction algorithm will be used to reconstruct the unrecognizable content and forward the reconstructed content to the receiver. Fig.11 shown.

[0124] In the reconstruction process, the most important part is to reconstruct the content based on the receiver's DIKW graph and the third-party DIKW graph. This system uses an entity similarity-based method to reconstruct the content. Its basic idea is based on the Word2Vec model, using the word vector model previously established in entity relationship extraction to match the entity that is most similar to the reconstructed content. The basic process of the algorithm is:

[0125] Step 1: According to the entity S that the receiver needs to reconstruct, based on the third-party DIKW graph N and the receiver's DIKW graph P, find the node set A of the entity S with degree i (i is initially 1) in the third-party DIKW graph = {a 1 ,a 2 ,…,a n}.

[0126] Step 2: According to the node set A = {a 1 ,a 2 ,…,a n}, matching node set B in the receiver DIKW graph = {b 1 ,b 2 ,…,b m}, where b j ∈{A,P}, if B=ε, then i=i+1, and return to step 1; otherwise, go to step 3.

[0127] Step 3: For each element b in the set B obtained in step 2 j The entity S is passed into a previously trained Embedding layer to obtain two word vectors ξ 1 and 2 , calculate the cosine similarity. As shown in formula (4-8):

[0128]

[0129] Step 4: Input the calculated cosine similarity result into the sigmoid function, convert it into probability, select the entity with a probability value greater than the set threshold, and replace the original entity that needs to be reconstructed, which is the reconstructed content.

[0130] 2.2.3 Calculation of semantic value of multimodal content

[0131] In order to ensure the consistency of the graph after migration, it is necessary to calculate the value of multimodal content semantics. This paper imitates the blockchain technology and proposes a content semantic value calculation algorithm for nodes and node networks to process multimodal content semantics. Fig.12 This is a diagram of the content semantic value calculation framework for redundant and fault-oriented blockchain-like nodes and node networks.

[0132] This paper performs semantic value calculation on the DIKW three-layer graph architecture.

[0133] (1) Semantic value calculation on data graph

[0134] In the data graph, the value calculation formula of the data node without considering the multi-dimensional and multi-modal situation is shown in (4-9), where Amount_DY is the number of concrete nodes (existing things), and Amount_D is the number of all nodes. When the result is greater than 51%, the data in the data graph is considered to be credible.

[0135]

[0136] Since the statistical units of data are different under different task indicators, the value calculation formula considering multiple dimensions is shown in (4-10), where V_dim() is the value function under a single dimension.

[0137]

[0138] Since different data may have different distribution intervals, the value calculation formula of the data node in the multimodal dimension is shown in (4-11), where θ k Expressed as weight.

[0139]

[0140] Fig.13 This is a diagram showing the value distribution of different dimensions when the data is clustered.

[0141] (2) Semantic value calculation on information graph

[0142] In the information graph, due to the semantic relationship between nodes, there will inevitably be duplicate nodes, error nodes and useless nodes. Therefore, when calculating the value of nodes on the information graph, the frequency calculation formula on the data graph will no longer apply. This article classifies the nodes counted on the information graph into positive nodes, negative nodes and neutral nodes. Among them, positive nodes represent nodes with positive semantic trends, negative nodes represent nodes with reverse semantic trends, and neutral nodes represent useless nodes with neither positive semantic trends nor reverse semantic trends. In the calculation of semantic value, neutral nodes need to be deleted to avoid the semantic value information being covered up. Therefore, the calculation of node semantic value is shown in formula (4-12), where Amount_I is the number of all nodes, Amount_IYes is the number of positive nodes, and Amount_IIrr is the number of neutral nodes.

[0143] Value_IR=Amount_IYes / (Amount_I-Amount_IIrr) formula (4-12)

[0144] Similar to the calculation of semantic value of data graph, the value calculation formula under multi-dimensional conditions is shown in (4-13). Under the multi-modal dimension, the value calculation formula of data nodes is shown in (4-14).

[0145]

[0146]

[0147] (3) Semantic value calculation on knowledge graph

[0148] The knowledge graph records rules and existence characteristics, and has no original thing nodes. Therefore, this paper combines the node statistics of the knowledge graph with the two-dimensional cloth operation to summarize the nodes. Table 3 shows the implementation method of semantic value calculation on the knowledge graph. Among them, Value_KR i is the value of a single node in the knowledge graph.

[0149] Table 3. Implementation method of semantic value calculation

[0150]

[0151] For example, assuming that all claims have at most ten claims, they are summarized as nodes in the knowledge graph, and a rule is established that there are at most ten claims in the claims. If there is a fact that there are eleven claims in a statistical node, then the original rule is directly broken, and the knowledge of the claims can only retain the part other than the ten claims, that is, Fig.14 The C1 part.

[0152] (4) Semantic value calculation on the DIKW graph

[0153] The semantic value of the DIKW graph is calculated comprehensively based on the semantic value calculation results on the data graph, the semantic value calculation results on the information graph, and the semantic value calculation results on the knowledge graph. R The calculation formula for _DIK is shown in (4-15).

[0154] Value R _DIK=α*Value R _D+β*Value R _I*μ DI +γ*Value R _K*μ DK *μ IK Formula(4-15)

[0155] Among them, α, β, γ are equalization coefficients; μ DI is the influence coefficient of the calculation based on the data graph on the calculation based on the information graph, μDK and μ IK And so on.

[0156] This paper also gives the calculation of semantic value when multiple graphs are interrelated, that is, when there are multiple central points and multiple keywords in a node network. This paper classifies the central nodes in the node network into cooperative nodes and competitive nodes, and assumes that the calculation of semantic value is related to the in-and-out degree of the central node. Fig.15 As shown, CP1 and CP2 are competing nodes.

[0157] By calculating the in-and-out degree of the central node of the node network, we can determine the influence factor of the central node on other nodes, and use the influence factor as the value of other nodes. The calculation is shown in formula (4-16):

[0158] V_dim IFL =(deg + +deg - ) / 2 formula (4-16)

[0159] Then the value of the node network Value_DNR is calculated as shown in formula (4-17):

[0160]

[0161] Among them, m is the total number of nodes in the node network, and n is the number of forward nodes.

[0162] 3. Realization of patent modification and reply functions

[0163] This system uses Neo4j as the graph database for storing DIKW graphs (Li et al., 2019). According to the structure of the relational database and the interface provided by Neo4j, a Binlog-based import architecture model is designed to build the DIKW graph, such as Fig.16 shown.

[0164] Binlog is a binary log that records all updates to table structures (such as creating and updating tables...) and modifications to table data (inserting, updating, deleting...) in the MySQL database. It does not record queries and displays such behaviors that do not modify the data itself. Based on this mechanism, the Canal Server middleware is used to monitor Binlog. When the MySQL database changes, the Canal Server sends the data to the Canal Client. There will be a Kafka Producer in the Canal Client, which is a message producer that can transmit messages to the Kafka message queue. The KafkaConsumer consumer retrieves the data from the message queue and finally calls an interface provided by Neo4j to write the data to Neo4j.

[0165] Since a database is needed to store entity relationship triples, the table structure of the database needs to be designed. By analyzing the characteristics of the data to be stored, the following relationship can be obtained:

[0166] A: A text message can be divided into multiple phrases, so there is a one-to-many relationship between the text message and the phrases.

[0167] B: A phrase may exist in two forms of entity relationship in different contexts, so the relationship between phrase and entity is a one-to-many relationship.

[0168] Design the logical structure of the database table structure based on these relationships. Fig.17 shown.

[0169] After completing the design of the logical structure of the table, the specific structure of each table can be specifically designed based on the structure, including the type of each field, parameter size, and constraints. The design results are shown in Table 4, Table 5, and Table 6.

[0170] Table 4 Attribute design table of the "text" table

[0171] Table 9 Attribute design table of "Text" table

[0172]

[0173] Table 5 Attribute design table of "participle part of speech" table

[0174] Table 10 Attribute design table of "Part of Speech" table

[0175]

[0176] Table 6 Attribute design table of "Entity Relationship" table

[0177] Table 11 Attribute design table of "Entity Relationship" table

[0178]

[0179] Based on the designed attribute table and logical model, the physical model can be obtained. Define the types, parameter sizes and constraints of each field designed previously in ERwin, generate SQL statement code through ERwin's forward engineering, copy the code to MySQL for execution, and generate the database and the table in the database.

[0180] Based on the above design, the patent modification and reply functions are realized.

[0181] The user clicks the patent modification button and selects the specification, claims, specification summary, etc. from the drop-down options to submit. After the upload is successful, a new page will be displayed to display the marked document and provide a download button for the document. Fig.18 Shown is the patent modification display interface.

[0182] For more relevant content, please refer to the other three patent application documents applied for by the applicant on the same day. The names of the inventions are "A patent intelligent assisted examination method and system based on DIKW", "A patent intelligent application method and system based on DIKW", and "A patent content retrieval method and system based on DIKW".

[0183] Based on the above invention concept, referring to Fig.16 According to a preferred embodiment of the patent intelligent modification method based on the DIKW map of the present invention, it can correct or modify the patent application document according to the provisions of the patent laws and regulations, so that the patent application document complies with the provisions of the patent laws and regulations, which includes:

[0184] Collect three-party data, where the three parties are a sender, an intermediary, and a receiver; the three-party data at least includes the sender's patent application document to be modified, the receiver's modification requirements for the patent application document to be modified, and the intermediary's patent data; it can be understood that in the patent application stage, the sender is, for example, the patent applicant, the receiver is, for example, the patent examination authority, that is, the patent office, and the intermediary or third party is, for example, other applicants in the same technical field; the intermediary or third-party patent data is, for example, other patent documents derived from the patent electronic application system that are different from the patent application document to be modified. In general, the patent application document of the patent applicant needs to be modified according to the modification requirements of the patent office, and the patent applicant can refer to the correct modifications made by other applicants in the same technical field for similar problems. The key point of this technical solution is to automatically find such modification instructions to assist the sender to modify quickly and correctly. Of course, the sender can also be a patent writing agent, the receiver can be a patent reviewing agent, and the intermediary is the holder of many other drafts and final drafts of patent application documents. The draft of the patent application document of the patent writing agent is automatically modified and improved by this technical solution according to the writing requirements stipulated by the reviewing patent agent, with the help of the guidance of the relevant examples that meet the writing requirements of the intermediary, so that it is close to the final draft. Similarly, the draft of the technical briefing book of the technical engineer can also be automatically modified and improved according to the company's writing requirements. In a patent electronic application system such as the CPC offline patent application system, if patent application A is the patent application document to be modified, then other patent applications X that have been downloaded in the CPC offline patent application system can be understood as the patent data of the intermediary.

[0185] Import the three-party data into a patent input relational database, wherein the data stored in the patent input relational database is referred to as relational patent input data; it is understandable that relational database technology is prior art. Common relational database systems include MySql, Oracle, Microsoft SQL server, etc. Import the patent application document as a whole or by statement into the data table of the relational database to facilitate subsequent fast and efficient data storage. The relational patent input data is stored in the database in the form of data records, which facilitates the reading, updating or insertion of large amounts of data. The patent input relational database is a database file in a relational database system used to store imported three-party patent data.

[0186] The relational patent input data is converted into a DIKW graph through data initialization, real-time processing and incremental import framework, wherein the DIKW graph is called DIKW graph patent information, and the DIKW graph patent information includes the sender DIKW graph, the receiver DIKW graph P and the intermediary DIKW graph N; it can be understood that the DIKW graph includes three layers as described above: data graph, information graph, knowledge graph. Therefore, the sender also has three layers of graphs: sender data graph, sender information graph, sender knowledge graph; the receiver and the intermediary also have them. The data initialization method varies depending on the relational database, and is usually executed through SQL statements or stored procedures. For example, refer to Fig.16 , the data initialization is performed through MySql Connector and Neo4j's APOCJava stored procedure package. The real-time processing and incremental import framework includes, for example, Kafka and Neo4j Driver. Kafka was originally developed by Linkedin and is a distributed, partitioned, multi-copy, multi-subscriber, zookeeper-coordinated distributed log system (also known as an MQ system), which can be commonly used for web / nginx logs, access logs, message services, etc. Of course, if the relational database is Micorsoft SQL Server, the data initialization can be performed through Ms SQLConnector and Neo4j's APOCJava stored procedure package.

[0187] The DIKW graph patent information is stored in a patent graph database, wherein the DIKW graph patent information stored in the patent graph database is referred to as patent graph data; it can be understood that the graph database is a prior art, and the graph database (Graph Database) is a new type of NoSQL database based on graph theory. Its data storage structure and data query method are based on graph theory. The basic elements of a graph in graph theory are nodes and edges, which correspond to nodes and relationships in a graph database. In a graph database, the relationship between data and data forms a graph structure through nodes and relationships and implements all the characteristics of the database on this structure, such as the ability to create, read, update, and delete (Create, Read, Update, Delete, referred to as CRUD) operations on graph data objects, as well as the ability to process transactions and high availability. Commonly used graph database systems include Neo4j, HyperGraphDB, Titan distributed graph database, ArangoDB, and Apache TinkerPop. Therefore, the real-time processing and incremental import framework includes, for example, Kafka and HyperGraphDBDriver. The patent graph database is a database file in a graph database system used to store the DIKW graph patent information described in this article.

[0188] For more information about the DIKW map, please refer to the patent documents CN201810590693.2ORCN202110043702.8, CN201710394197.5, CN201810023920.3, CN201710490217.9, CN202110043010.3, CN201710989006.X, CN201810910609.0, CN202011099503.0, CN202011103480.6, CN20 1810109766.1, CN201810037199.3, CN201710416328.5, CN201710395067.3, CN201810590852.9, CN201710363378.1, CN 201710374772.5, CN201710376105.0, CN201810590851.4, CN201710435186.7, CN201710745709.8, CN201810590798.8.

[0189] The patent graph data is reconstructed through a distributed graph processing framework, wherein the patent graph data after content reconstruction is referred to as patent content reconstruction data. It can be understood that a distributed graph processing framework is a set of program functions for processing graph data, such as Fig.16 As shown, the distributed graph processing framework is, for example, Spark GraphX. Spark GraphX ​​is a distributed graph processing framework that provides simple, easy-to-use and rich interfaces for graph computing and graph mining based on the Spark platform, greatly facilitating the demand for distributed graph processing. The distributed graph processing framework can also be a processing framework based on ArangoDB, Apache TinkerPop, or HyperGraphDB.

[0190] The content reconstruction includes the following sub-steps:

[0191] Sub-step S101: According to the entity S that the receiver needs to reconstruct, based on the intermediate party DIKW graph N and the receiver DIKW graph P, find the i-degree node set A of the entity S in the intermediate party DIKW graph = {a 1 ,a 2 ,…,a n}, where i is initially 1.

[0192] Sub-step S102: According to the node set A={a 1 ,a 2 ,…,a n}, matching node set B in the receiver DIKW graph = {b 1 ,b 2 ,…,b m}, where b j ∈{A,P}, if B=ε, then i=i+1, and the process returns to the sub-step S101; otherwise, the process goes to sub-step S103.

[0193] Sub-step S103: Each element b in the set B obtained in sub-step S102 j The entity S is passed into a previously trained Embedding layer to obtain two word vectors ξ 1 and 2 , calculate the cosine similarity, as shown in formula (4-8):

[0194]

[0195] Sub-step S104: input the calculated cosine similarity result into the sigmoid function, convert it into probability, select entities with probability values ​​greater than the set threshold, and replace the original entities that need to be reconstructed, that is, reconstruct the content; it can be understood that the content reconstruction algorithm can be implemented by Java language.

[0196] The entire reconstructed content of the patent application document to be modified forms the modified patent application document.

[0197] The embedding layer and sigmoid function used in this technical solution are layers and functions in the entity similarity algorithm based on the Word2Vec model. For the entity similarity algorithm of the Word2Vec model, see patent documents CN111178059, CN111027315, CN109376352, CN107436864, CN109062892, CN110661875.

[0198] It is understandable that the recipient may need to reconstruct multiple entities S, and these multiple entities S that need to be reconstructed may be all or part of the entities of the sender's patent application document. In other words, all entities in the DIKW graph decomposed from the part of the sender's patent application document string that needs to be modified need to be reconstructed. One of the technical points of the present invention is to convert the patent application document into DIKW graph data. Fig.11 , compare the DIKW graph of the sender with the DIKW graph of the receiver. If all entities, relationships and other data information of both parties exist in the DIKW graph, it means that both parties can understand each other, that is, the patent application document does not need to be modified. If all entities, relationships and other data information of both parties do not exist in the DIKW graph, then the DIKW graph of the intermediary party is used to complete the entities and relationships that do not exist on the receiver or sender, and the corresponding DIKW graph is reconstructed so that the receiver can understand the content represented by the reconstructed DIKW graph.

[0199] The entirety of the reconstructed part and the unreconstructed part of the DIKW map corresponding to the patent application document to be modified received by the receiving party forms the modified patent application document.

[0200] For more information about the interactive area division and transmission optimization based on the DIKW graph, please refer to the patent announcement text CN110442734B of the same applicant, whose invention is named "Interactive area division and transmission optimization processing method based on data graph, information graph and knowledge graph", and patent announcement text CN110457488B, whose invention is named "Content transmission modeling and processing optimization method based on data graph, information graph and knowledge graph".

[0201] For more information about content reconstruction based on DIKW graph and multimodal semantic modeling of DIKW graph, please refer to patent documents CN107038261B and CN107038262B.

[0202] like Fig.18 As shown, the embodiment basically realizes modifying the patent application document to be modified into a patent application document that meets the modification requirements. In the accompanying drawings, the text marked in red is text that does not meet the requirements of patent regulations and should be deleted.

[0203] Preferably, if the above technical solution simultaneously screens out multiple content reconstruction versions, the content reconstruction version with the highest probability is selected.

[0204] Advantageously, if the above technical solution simultaneously screens out multiple content reconstruction versions, or there are two or less content reconstruction versions with the same maximum probability, the content reconstruction version with the largest value is selected according to the total semantic value ValueR_DIK calculated by the multimodal content semantic value calculation method described above.

[0205] For more information about the calculation of multimodal content semantic value, please refer to the patent disclosure text CN110471996A of the same applicant, the invention name is "Content semantic value calculation mechanism for redundant imitation blockchain nodes and node networks".

[0206] Preferably, if Fig.18 As shown, the patent intelligent modification method based on the DIKW map also includes displaying the patent application document to be modified on a graphical interface in a visual manner, wherein the modified part is marked with different fonts, colors or annotations. It can be understood that, Fig.18 As shown, the text marked in red is not in compliance with the requirements of patent regulations and should be deleted.

[0207] Preferably, before the content reconstruction, the real-time processing and incremental import framework performs multimodal semantic modeling based on the DIKW graph according to the following sub-steps:

[0208] Sub-step S201: extracting the keywords of the requirement description of the patent application document to be modified by information extraction technology, and recording the keywords in the form of time, space and structure frequency on the data map, wherein the time frequency Tem_f is the time scale of the keyword appearance, the space frequency Spa_f is the coordinate scale of the keyword, and the structure frequency Str_f is the number of relationships of the keyword in the entity relationship model;

[0209] Sub-step S202: Mark the keywords with a time frequency of 1 as target objects, and calculate the total frequency of the data according to formula (4-1);

[0210] frequency()=Str_f()*Tem_f()*Spa_f() formula (4-1)

[0211] Sub-step S203: Calculate the support and confidence of the keyword according to formulas (4-2) and (4-3); and based on the obtained results, clean the discrete data on the data map to delete marginal data and redundant data; where Frequency DG Represents the sum of the frequencies of all nodes on the data graph, ri 、r j Refers to two different nodes.

[0212]

[0213]

[0214] Sub-step S204: semantic analysis is performed on the keyword objects recorded on the data graph, taking nouns as entities and verbs as relations to form an information graph;

[0215] Sub-step S205: Calculate the interaction frequency between nodes on the information graph according to formula (4-4), where the interaction frequency between nodes represents the number of interactions between nodes, E1 and E2 refer to any two computing nodes in the calculation process, and P represents all complete interaction paths between nodes;

[0216] frequency(E1,E2)=∑ P(E1→E2) 1Formula (4-4)

[0217] Sub-step S206: Calculate the comprehensive frequency Total of all non-discrete nodes on the information graph according to formula (4-5) frequency , Frequency IG represents the interaction frequency of entities on the information graph, α and β are the weights of the data graph frequency and the information graph frequency;

[0218] Total frequency =αFrequency DG *βFrequency IG Formula (4-5)

[0219] Sub-step S207: Calculate the cohesion of each node in the information graph according to formula (4-6); integrate different connected entities with the largest entity cohesion in the form of attributes and operations, and express them in the form of new nodes on the information graph to enhance the cohesion of the patent application document to be modified and improve the abstractness; where Degree EI Indicates external interaction, Degree II represents internal interactions;

[0220]

[0221] Sub-step S208: updating the frequency recorded on the data graph according to the new nodes and new relationships generated by the information graph;

[0222] Sub-step S209: Based on the semantic relationship on the information graph, the knowledge graph is assisted to infer new relationships, and the correctness Cr is calculated according to formula (4-7), where π represents a path between node 1 and node 2, Q(E1→E2) represents all paths between node E1 and node E2, R represents the relationship between two nodes or two entities, Q represents all paths in the circled graph, θ(π) represents the training weight, and when the correctness exceeds a certain set threshold, the new relationship is considered to be established:

[0223]

[0224] Sub-step S210: Optimize the knowledge graph and delete redundant nodes and loops.

[0225] It can be understood that the above-mentioned “node” can be understood as an “entity”.

[0226] The DIKW graph formed through the above-mentioned multimodal semantic modeling based on the DIKW graph is concise, accurate and efficient.

[0227] Preferably, refer to Fig.16 , the patent intelligent modification method based on DIKW graph also includes collecting public data of patents, wherein the public data of patents at least includes the specification and / or claims of authorized invention patents, the examination opinion notice of authorized invention patents, and the examination opinion reply statement of authorized invention patents; the public data of patents is used as the patent data of the intermediary. It can be understood that the richer the patent data of the intermediary, the higher the accuracy of the content reconstruction, and of course, there may be more versions of content reconstruction.

[0228] Preferably, refer to Fig.16 , in the patented intelligent modification method based on the DIKW atlas:

[0229] The patent input relational database is established in a MYSQL database system, wherein the MySql includes a Binlog binary log file; the patent graph database is established in a Neo4j graph database system;

[0230] The data initialization is performed through MySql Connector and Neo4j's APOCJava stored procedure package; the real-time processing and incremental import framework obtains the relational patent input data through the Binlog binary log file, and processes it in real time through the Kafka distributed log system and Neo4j Driver, and imports the real-time processed data into the patent graph database; the distributed graph processing framework includes Spark Graphx; the public data is collected through a crawler program; the third-party data is collected through the API interface of the patent application and management program.

[0231] Preferably, refer to Fig.16 In the real-time processing and incremental import framework, the Binlog binary log file is a binary log that records all updates to the table structure and modifications to the table data in the relational patent input data; the Canal Server middleware is used to monitor the Binlog binary log file, and when the relational patent input data changes, the Canal Server sends the data to the Canal Client; the Canal Client includes a Kafka Producer, which is a message producer that can transmit messages to the Kafka message queue, and the Kafka Consumer takes data from the message queue; an interface provided by the Neo4j is called to write data to the patent graph database.

[0232] Preferably, refer to Fig.17 , in the patent intelligent modification method based on the DIKW graph: the DIKW graph includes entity relationship triple data, wherein the entity relationship triple data includes a text information table, a phrase part-of-speech table and an entity relationship table;

[0233] The text information table at least includes text ID, text content, and acquisition time fields;

[0234] The phrase part-of-speech table at least includes a segmentation ID, a part-of-speech, an entity, a relationship, and the text ID field;

[0235] The entity relationship table at least includes an entity relationship ID, a relationship ID, and the word segmentation ID field.

[0236] wherein the text information table forms a one-to-many data constraint relationship with the phrase part-of-speech table through the text ID; the phrase part-of-speech table forms a one-to-many data constraint relationship with the entity relationship table through the segmentation ID, and the entity relationship table forms a data constraint relationship with the text information table through the text ID in the phrase part-of-speech table; and

[0237] Design the field attribute table and logical model of the text information table, phrase part-of-speech table, and entity relationship table, and obtain their physical models, generate SQL statement code through forward engineering of the Erwin data modeling tool, execute the SQL statement code, and generate the text information table, phrase part-of-speech table, and entity relationship table in a relational database.

[0238] According to the preferred implementation of the patent intelligent auxiliary reply method based on the DIKW map of the present invention, it is possible to correct or modify the patent application documents according to the provisions of patent laws and regulations and the requirements of the patent examination opinion notice, so that the patent application documents comply with the provisions of patent laws and regulations and the requirements of the patent examination opinion notice, and assist the patent examination opinion notice respondent to reply to the examination opinion.

[0239] The patent intelligent assisted reply method based on the DIKW graph includes the patent intelligent modification method based on the DIKW graph, wherein the modification requirement of the patent application document to be modified is the examination opinion notice of the patent application document to be modified.

[0240] According to the above functional test data, the above technical solution achieves that the response to the examination opinion complies with the requirements of patent laws and regulations, and the response speed is fast.

[0241] As mentioned above, the present invention also includes a patented intelligent modification device based on the DIKW map.

[0242] Preferably, the patented intelligent auxiliary reply system based on the DIKW graph also includes a load balancing module and a storage and computing integration module.

[0243] As described above, the present invention also includes a patented intelligent auxiliary reply system based on the DIKW graph.

[0244] Furthermore, the server according to the present invention comprises:

[0245] one or more processors;

[0246] a storage device having one or more programs stored thereon;

[0247] When the one or more programs are executed by the one or more processors, the one or more processors implement the patent intelligent modification method based on the DIKW graph described above, or the patent intelligent assisted reply method based on the DIKW graph described above.

[0248] Further, according to the computer-readable medium of the present invention, a computer program is stored thereon, wherein when the program is executed, the patent intelligent modification method based on the DIKW map described above is implemented, or the patent intelligent assisted reply method based on the DIKW map described above is implemented.

[0249] The preferred or specific embodiments of the invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the design concept of the invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning or limited experiments based on the design concept of the invention on the basis of the prior art should be within the scope of the invention and / or the protection scope determined by the claims.

Claims

1. A patent intelligent modification method based on DIKW graph, which can correct or modify patent application documents according to the provisions of patent laws and regulations, so that the patent application documents comply with the provisions of patent laws and regulations, characterized in that: It includes: Collecting three-party data, wherein the three parties are a sender, an intermediary, and a receiver; the three-party data at least includes the patent application document to be modified of the sender, the modification requirements of the patent application document to be modified of the receiver, and the patent data of the intermediary; Importing the three-party data into a patent input relational database, wherein the data stored in the patent input relational database is referred to as relational patent input data; The relational patent input data is converted into a DIKW graph through data initialization, real-time processing and incremental import framework, wherein the DIKW graph is called DIKW graph patent information, and the DIKW graph patent information includes a sender DIKW graph, a receiver DIKW graph P and an intermediate DIKW graph N; The DIKW graph patent information is stored in a patent graph database, wherein the DIKW graph patent information stored in the patent graph database is referred to as patent graph data; The patent graph data is reconstructed through a distributed graph processing framework, wherein the patent graph data after content reconstruction is referred to as patent content reconstruction data, and the content reconstruction includes the following sub-steps: Sub-step S101: According to the entity S that the receiver needs to reconstruct, based on the intermediate party DIKW graph N and the receiver DIKW graph P, find the i-degree node set A of the entity S in the intermediate party DIKW graph = {a1, a2, ..., a n }, where i is initially 1; Sub-step S102: According to the node set A = {a1, a2, ..., a n }, match the node set B = {b1, b2, …, b m }, where b j ∈{A,P}, if B=ε, then i=i+1, and the process returns to the sub-step S101; otherwise, the process proceeds to sub-step S103; Sub-step S103: Each element b in the set B obtained in sub-step S102 j The entity S is passed into a previously trained Embedding layer to obtain two word vectors ξ1 and ξ2, and the cosine similarity is calculated, as shown in formula (4-8): Sub-step S104: input the calculated cosine similarity result into the sigmoid function, convert it into probability, select entities with probability values ​​greater than a set threshold, and replace the original entities to be reconstructed, which is the reconstructed content; All the reconstructed contents of the patent application document to be modified form the modified patent application document; The patent input relational database is established in a MYSQL database system, wherein the MySql includes a Binlog binary log file; the patent graph database is established in a Neo4j graph database system; The data initialization is performed through MySql Connector and Neo4j's APOCJava stored procedure package; the real-time processing and incremental import framework obtains the relational patent input data through the Binlog binary log file, and processes it in real time through the Kafka distributed log system and Neo4j Driver, and imports the real-time processed data into the patent graph database; the distributed graph processing framework includes Spark Graphx; public data is collected through a crawler program; the third-party data is collected through the API interface of the patent application and management program; In the real-time processing and incremental import framework, the Binlog binary log file is a binary log that records all updates to table structures and modifications to table data in the relational patent input data; the Canal Server middleware is used to monitor the Binlog binary log file, and when the relational patent input data changes, the Canal Server sends the data to the Canal Client; the Canal Client includes a KafkaProducer, which is a message producer that can transmit messages to a Kafka message queue, and the Kafka Consumer takes data out of the message queue; an interface provided by the Neo4j is called to write data to the patent graph database; The DIKW graph includes entity relationship triple data, wherein the entity relationship triple data includes a text information table, a phrase part-of-speech table and an entity relationship table; The text information table at least includes text ID, text content, and acquisition time fields; The phrase part-of-speech table includes at least a segmentation ID, a part-of-speech, an entity, a relationship, and the text ID field; the entity relationship table includes at least an entity relationship ID, a relationship ID, and the segmentation ID field; Among them, the text information table forms a one-to-many data constraint relationship with the phrase part-of-speech table through the text ID; the phrase part-of-speech table forms a one-to-many data constraint relationship with the entity relationship table through the word segmentation ID, and the entity relationship table forms a data constraint relationship with the text information table through the text ID in the phrase part-of-speech table; and the field attribute table and logical model of the text information table, phrase part-of-speech table, and entity relationship table are designed, and their physical models are obtained, and SQL statement codes are generated through forward engineering of the Erwin data modeling tool, and the SQL statement codes are executed to generate the text information table, phrase part-of-speech table, and entity relationship table in a relational database.

2. The patent intelligent modification method based on DIKW graph according to claim 1 is characterized in that: It also includes displaying the modified patent application document on a graphical interface in a visual manner, wherein the modified portion is marked with different fonts, colors or annotations.

3. The patent intelligent modification method based on DIKW graph according to claim 2 is characterized in that: Before the content reconstruction, the real-time processing and incremental import framework performs multimodal semantic modeling based on the DIKW graph in the following sub-steps: Sub-step S201: extracting the keywords of the requirement description of the patent application document to be modified by information extraction technology, and recording the keywords in the form of time, space and structure frequency on the data map, wherein the time frequency Tem_f is the time scale of the keyword appearance, the space frequency Spa_f is the coordinate scale of the keyword, and the structure frequency Str_f is the number of relationships of the keyword in the entity relationship model; Sub-step S202: Mark the keywords with a time frequency of 1 as target objects, and calculate the total frequency of the data according to formula (4-1); frequency()=Str_f()*Tem_f()*Spa_f() formula (4-1) Sub-step S203: Calculate the support and confidence of the keyword according to formulas (4-2) and (4-3); and based on the obtained results, clean the discrete data on the data map to delete marginal data and redundant data; where Frequency DG Represents the sum of the frequencies of all nodes on the data graph, r i 、r j refers to two different nodes; Sub-step S204: semantic analysis is performed on the keyword objects recorded on the data graph, taking nouns as entities and verbs as relations to form an information graph; Sub-step S205: Calculate the interaction frequency between nodes on the information graph according to formula (4-4), where the interaction frequency between nodes represents the number of interactions between nodes, E1 and E2 refer to any two computing nodes in the calculation process, and P represents all complete interaction paths between nodes; frequency(E1,E2)=∑ P(E1→E2) 1 Formula (4-4) Sub-step S206: Calculate the comprehensive frequency Total of all non-discrete nodes on the information graph according to formula (4-5) frequency , Frequency IG represents the interaction frequency of entities on the information graph, α and β are the weights of the data graph frequency and the information graph frequency; Total frequency = α Frequency DG * β Frequency IG Formula (4-5) Sub-step S207: Calculate the cohesion of each node in the information graph according to formula (4-6); Different connected entities with the largest entity cohesion are integrated in the form of attributes and operations, and expressed in the form of new nodes on the information map, so as to enhance the cohesion of the patent application document to be modified and improve the abstractness; Degree EI Indicates external interaction, Degree II represents internal interactions; Sub-step S208: updating the frequency recorded on the data graph according to the new nodes and new relationships generated by the information graph; Sub-step S209: Based on the semantic relationship on the information graph, the knowledge graph is assisted to infer new relationships, and the accuracy Cr is calculated according to formula (4-7), where π represents a path between node 1 and node 2, Q(E1→E2) represents all paths between node E1 and node E2, R represents the relationship between the two nodes, Q represents all paths in the circled graph, θ(π) represents the training weight, and when the accuracy exceeds a certain set threshold, the new relationship is considered to be established: Sub-step S210: Optimize the knowledge graph and delete redundant nodes and loops.

4. The patent intelligent modification method based on DIKW graph according to claim 3 is characterized in that: It also includes collecting public data of patents, wherein the public data of patents at least includes the description and / or claims of authorized invention patents, the examination opinion notice of authorized invention patents, and the examination opinion reply statement of authorized invention patents; the public data of patents is used as the patent data of the intermediary.

5. A patent intelligent auxiliary reply method based on DIKW graph, which can correct or modify the patent application documents according to the provisions of patent laws and regulations and the requirements of patent examination opinion notices, so that the patent application documents comply with the provisions of patent laws and regulations and the requirements of patent examination opinion notices, and assist the patent examination opinion notice responder to respond to the examination opinion, characterized in that: The reply method includes the patent intelligent modification method based on the DIKW map as described in any one of claims 1 to 4, wherein the modification requirement of the patent application document to be modified is the examination opinion notice of the patent application document to be modified.

6. A patent intelligent modification device based on DIKW graph, which can correct or modify patent application documents according to the provisions of patent laws and regulations, so that the patent application documents comply with the provisions of patent laws and regulations, characterized in that: It includes: A data collection module is used to collect three-party data, wherein the three parties are a sender, an intermediary, and a receiver; the three-party data at least includes the patent application document to be modified of the sender, the modification requirements of the patent application document to be modified of the receiver, and the patent data of the intermediary; A first relational database system is used to import the three-party data into a patent input relational database, wherein the data stored in the patent input relational database is referred to as relational patent input data; A DIKW graph generation module is used to convert the relational patent input data into a DIKW graph through data initialization, real-time processing and incremental import framework, wherein the DIKW graph is called DIKW graph patent information, and the DIKW graph patent information includes a sender DIKW graph, a receiver DIKW graph P and an intermediate DIKW graph N; A first graph database system is used to store the DIKW graph patent information in a patent graph database, wherein the DIKW graph patent information stored in the patent graph database is referred to as patent graph data; The graph data content reconstruction module reconstructs the content of the patent graph data through a distributed graph processing framework, wherein the patent graph data after content reconstruction is referred to as patent content reconstruction data. The graph data content reconstruction module includes the following submodules: Submodule M101: Entity S reconstructed according to the needs of the receiver, based on the intermediate DIKW The graph N and the receiving party DIKW graph P are used to find the i-degree node set A of the entity S in the intermediate party DIKW graph = {a1, a2, ..., a n }, where i is initially 1; Submodule M102: According to the node set A={a1, a2, …, a n }, match the node set B = {b1, b2, …, b m }, where b j ∈{A,P}, if B=ε, then i=i+1, and return to the submodule M101; otherwise, enter the submodule M103; Submodule M103: Each element b in the set B obtained in substep S102 j The entity S is passed into a previously trained Embedding layer to obtain two word vectors ξ1 and ξ2, and the cosine similarity is calculated, as shown in formula (4-8): Submodule M104: Input the calculated cosine similarity result into the sigmoid function, convert it into probability, select entities with probability values ​​greater than the set threshold, and replace the original entities to be reconstructed, which is the reconstructed content; All the reconstructed contents of the patent application document to be modified form the modified patent application document; The patent input relational database is established in a MYSQL database system, wherein the MySql includes a Binlog binary log file; the patent graph database is established in a Neo4j graph database system; The data initialization is performed through MySql Connector and Neo4j's APOCJava stored procedure package; the real-time processing and incremental import framework obtains the relational patent input data through the Binlog binary log file, and processes it in real time through the Kafka distributed log system and Neo4j Driver, and imports the real-time processed data into the patent graph database; the distributed graph processing framework includes Spark Graphx; public data is collected through a crawler program; the third-party data is collected through the API interface of the patent application and management program; In the real-time processing and incremental import framework, the Binlog binary log file is a binary log that records all updates to table structures and modifications to table data in the relational patent input data; the Canal Server middleware is used to monitor the Binlog binary log file, and when the relational patent input data changes, the Canal Server sends the data to the Canal Client; the Canal Client includes a KafkaProducer, which is a message producer that can transmit messages to a Kafka message queue, and the Kafka Consumer takes data out of the message queue; an interface provided by the Neo4j is called to write data to the patent graph database; The DIKW graph includes entity relationship triple data, wherein the entity relationship triple data includes a text information table, a phrase part-of-speech table and an entity relationship table; The text information table at least includes text ID, text content, and acquisition time fields; The phrase part-of-speech table includes at least a segmentation ID, a part-of-speech, an entity, a relationship, and the text ID field; the entity relationship table includes at least an entity relationship ID, a relationship ID, and the segmentation ID field; Among them, the text information table forms a one-to-many data constraint relationship with the phrase part-of-speech table through the text ID; the phrase part-of-speech table forms a one-to-many data constraint relationship with the entity relationship table through the word segmentation ID, and the entity relationship table forms a data constraint relationship with the text information table through the text ID in the phrase part-of-speech table; and the field attribute table and logical model of the text information table, phrase part-of-speech table, and entity relationship table are designed, and their physical models are obtained, and SQL statement codes are generated through forward engineering of the Erwin data modeling tool, and the SQL statement codes are executed to generate the text information table, phrase part-of-speech table, and entity relationship table in a relational database.

7. The patent intelligent auxiliary reply system based on the DIKW map can correct or modify the patent application documents according to the provisions of patent laws and regulations and the requirements of the patent examination opinion notice, so that the patent application documents comply with the provisions of patent laws and regulations and the requirements of the patent examination opinion notice, and assist the patent examination opinion notice respondent to respond to the examination opinion. Features: The patent intelligent auxiliary reply system based on the DIKW map includes the patent intelligent modification device based on the DIKW map according to claim 6, wherein the modification requirement of the patent application document to be modified is the examination opinion notice of the patent application document to be modified.

8. A server, characterized in that include: one or more processors; a storage device having one or more programs stored thereon; When the one or more programs are executed by the one or more processors, the one or more processors implement the patent intelligent modification method based on the DIKW graph as described in any one of claims 1 to 4, or the patent intelligent assisted reply method based on the DIKW graph as described in claim 5.

Images