A method and device for automatically generating patent intelligence based on a knowledge graph

Abstract

This application belongs to the technical field of patent intelligence analysis, and particularly relates to a method and device for automatically generating patent intelligence based on a knowledge graph. The method includes: Step S1, obtaining an aircraft structure knowledge graph stored in a tree structure; Step S2, determining the position of a structure node to be processed; Step S3, grouping the structure nodes of the structure tree and determining the keyword set of each group; Step S4, corresponding each keyword set to different positions of the patent text to construct a combined retrieval strategy for the patent text; Step S5, obtaining the single retrieval result of each structure node; Step S6, determining the composite retrieval result; Step S7, extracting patent content in the composite retrieval result corresponding to each structure node and its sub-structure nodes according to the patent intelligence analysis template to construct a patent intelligence tree. This application realizes the comprehensive patent intelligence retrieval and generation of the aircraft structure, and improves the coverage and analysis accuracy of patent intelligence.

Description

Technical Field

[0001] This application belongs to the technical field of patent intelligence analysis, and particularly relates to a method and device for automatically generating patent intelligence based on a knowledge graph. Background Art

[0002] Currently, when collecting patent intelligence on aircraft, it is usually necessary to provide keywords and then conduct a patent search. Patent intelligence is extracted from the search results. However, when conducting a patent search, there are defects such as chaotic search conditions, slow search efficiency, and low accuracy of search results. Moreover, during the process of obtaining patent intelligence, the amount of patent data is large, and the correspondence between the obtained patent intelligence and the key technologies of the aircraft is unclear. Summary of the Invention

[0003] To solve the above problems, a first aspect of this application provides a method for automatically generating patent intelligence based on a knowledge graph, which mainly includes:

[0004] Step S1: Obtain an aircraft structure knowledge graph stored in a tree structure;

[0005] Step S2: Determine the position of the initially selected structure node to be processed in the knowledge graph, and use this structure node and all its sub-structure nodes as the structure nodes to be processed;

[0006] Step S3: Divide all the structure nodes between the structure node to be processed and the root node into multiple groups according to the growth order of the tree structure, and construct a keyword set for each group with the names of the structure nodes in each group as keywords;

[0007] Step S4: Map each keyword set to different positions in the patent text to construct a combined retrieval strategy for the patent text;

[0008] Step S5: Use the patent text retrieved by the combined retrieval strategy as the single retrieval result of the structure node to be processed;

[0009] Step S6: Combine the single result of the structure node to be processed with the single retrieval results of all its sub-structure nodes as the composite retrieval result of the initially selected structure node to be processed;

[0010] Step S7: Use the structure node to be processed and all its sub-structure nodes to construct an intelligence tree, extract patent content according to the patent intelligence analysis template in the composite retrieval results corresponding to each structure node and its sub-structure nodes, and attach it to the structure nodes of the intelligence tree.

[0011] Preferably, after step S3, it further includes:

[0012] Correct the keywords based on a pre - constructed word - mapping dictionary, which is used to map aircraft - related words to patent - used words.

[0013] Preferably, step S4 further includes:

[0014] Step S41: Corresponding the keywords of the structure node with the farthest distance from the root node of the structure knowledge graph to the title of the patent text, corresponding the keyword set composed of one or more structure nodes with the second - farthest distance from the root node of the structure knowledge graph to the abstract of the patent text, and corresponding the keyword set composed of one or more structure nodes with the closest distance from the root node of the structure knowledge graph to the description of the patent text;

[0015] Step S42: Connect the retrieval strategies constructed by multiple keyword sets with the relationship "AND" between each other, and connect the keywords within each keyword set with the relationship "OR".

[0016] Preferably, step S6 further includes:

[0017] From bottom to top, for each structure node, remove the retrieval data of all its sub - structure nodes, and use the remaining retrieval data as the corrected single retrieval result of this structure node. The combined corrected single retrieval results of each structure node are used as the composite retrieval result.

[0018] Preferably, step S7 further includes:

[0019] Step S71: For each structure node A, determine the patent catalogs of this structure node A and each sub - structure node B directly connected to this structure node A. Among them, the patent catalog of the sub - structure node B directly connected to this structure node A is the set of the patent catalogs of this sub - structure node B and all its sub - nodes C. The patent catalog of this structure node A is extracted from the corrected single retrieval result of this structure node A. The patent catalog contains multiple patent records, and each patent record includes at least two fields: the patent application date and the patent publication country.

[0020] Step S72: Based on the patent catalogs of the structure node A and each sub - structure node B directly connected to this structure node A, conduct patent technology composition analysis, patent annual application volume distribution analysis, and patent regional deployment distribution analysis.

[0021] The second aspect of this application provides a patent intelligence automatic generation device based on a knowledge graph, mainly including:

[0022] An aircraft structure knowledge graph acquisition module, used to acquire an aircraft structure knowledge graph stored in a tree structure;

[0023] A structural node to be processed determination module, configured to determine the position of the initially selected structural node to be processed in the knowledge graph, and use this structural node and all its sub-structural nodes as the structural nodes to be processed;

[0024] A keyword set acquisition module, configured to divide all the structural nodes between the structural node to be processed and the root node into multiple groups according to the growth order of the tree structure, and construct keyword sets for each group with the names of the structural nodes in each group as keywords;

[0025] A combined retrieval strategy formulation module, configured to correspond each keyword set to different positions of the patent text to construct a combined retrieval strategy for the patent text;

[0026] A single retrieval result collection module, configured to use the patent text retrieved by the combined retrieval strategy as the single retrieval result of the structural node to be processed;

[0027] A composite retrieval result determination module, configured to combine the single result of the structural node to be processed with the single retrieval results of all its sub-structural nodes as the composite retrieval result of the initially selected structural node to be processed;

[0028] An intelligence tree generation module, configured to use the structural node to be processed and all its sub-structural nodes to construct an intelligence tree, extract patent content according to the patent intelligence analysis template from the composite retrieval results corresponding to each structural node and its sub-structural nodes, and mount it on the structural nodes of the intelligence tree.

[0029] Preferably, the apparatus further includes:

[0030] A keyword correction module, configured to correct the keywords based on a pre-constructed word comparison dictionary, and the word comparison dictionary is used to map aircraft words to patent usage words.

[0031] Preferably, the combined retrieval strategy formulation module includes:

[0032] A keyword correspondence unit, configured to correspond the keywords of the structural node farthest from the root node of the structural knowledge graph to the title of the patent text, correspond the keyword set composed of one or more structural nodes with the second farthest distance from the root node of the structural knowledge graph to the abstract of the patent text, and correspond the keyword set composed of one or more structural nodes closest to the root node of the structural knowledge graph to the description of the patent text;

[0033] A retrieval strategy automatic generation unit, configured to connect the retrieval strategies constructed by multiple keyword sets with the relationship "AND" between each other, and connect the keywords within each keyword set with the relationship "OR".

[0034] Preferably, the composite retrieval result determination module includes:

[0035] A duplicate removal processing unit, which, from bottom to top, for each structural node, removes the retrieval data of all its sub-structural nodes, and takes the remaining retrieval data as the corrected single retrieval result of this structural node. The corrected single retrieval results of each structural node are combined as the composite retrieval result.

[0036] Preferably, the intelligence tree generation module includes:

[0037] A patent catalog extraction unit, which, for each structural node A, determines the patent catalog of this structural node A and each sub-structural node B directly connected to this structural node A. Among them, the patent catalog of the sub-structural node B directly connected to this structural node A is the set of the patent catalogs of this sub-structural node B and all its sub-nodes C. The patent catalog of this structural node A is extracted from the corrected single retrieval result of this structural node A. The patent catalog contains multiple patent records, and each patent record includes at least two fields: the patent application date and the patent publication country.

[0038] A data analysis unit, which, based on the patent catalogs of the structural node A and each sub-structural node B directly connected to this structural node A, conducts patent technology composition analysis, patent annual application volume distribution analysis, and patent regional deployment distribution analysis.

[0039] This application realizes all-round patent intelligence retrieval and generation for the aircraft structure, improving the coverage and analysis accuracy of patent intelligence. Brief Description of the Drawings

[0040] Figure 1 is a flowchart of a preferred embodiment of the method for automatically generating patent intelligence based on a knowledge graph in this application.

[0041] Figure 2 is this application Figure 1 A schematic diagram of the aircraft structure knowledge graph structure of the illustrated embodiment.

[0042] Figure 3 is this application Figure 1 A schematic diagram of the analysis result of the patent technology composition of the illustrated embodiment. Detailed Description of the Embodiment

[0043] To make the purpose, technical solution and advantages of the implementation of this application clearer, the technical solution in the implementation mode of this application will be described in more detail below in conjunction with the accompanying drawings in the implementation mode of this application. In the accompanying drawings, the same or similar reference numerals represent the same or similar elements or elements with the same or similar functions from beginning to end. The described implementation mode is a part of the implementation modes of this application, rather than all of the implementation modes. The implementation mode described below by referring to the accompanying drawings is exemplary and is intended to explain this application, and should not be construed as a limitation of this application. Based on the implementation modes in this application, all other implementation modes obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of this application. The implementation mode of this application will be described in detail below in conjunction with the accompanying drawings.

[0044] The first aspect of this application provides a method for automatically generating patent intelligence based on a knowledge graph, as Figure 1 shown, mainly including:

[0045] Step S1, obtain the aircraft structure knowledge graph stored in a tree structure.

[0046] Figure 2 A simplified schematic diagram of the aircraft structure knowledge graph is given. In Figure 2 , the fixed-wing aircraft is the root node of the structure tree, which can also be called the first-level node, the fuselage, wings, and engines are the second-level nodes, the skin, frame, cabin section, etc. are the third-level nodes, 7075 aluminum alloy material, carbon fiber material, truss frame, box frame, torsional stiffness, etc. are the fourth-level nodes, the yield strength parameter, density parameter, etc. are the fifth-level nodes, and also the last-level nodes. In the aircraft structure knowledge graph, this last-level node is the attribute parameter of the upper-level node, that is, the yield strength ≥ 310 MPa is the strength attribute parameter of 7075 aluminum alloy material or carbon fiber material.

[0047] Step S2, determine the position of the initially selected structural node to be processed in the knowledge graph, and use this structural node and all its sub-structural nodes as the structural nodes to be processed.

[0048] The structural node to be processed is usually specified by the user, and one or more can be specified. For example, the skin in Figure 2 is specified as the structural node to be processed. When the system is analyzing, although the user only specifies the skin, all the sub-nodes of the skin need to be analyzed as well. That is, the 7075 aluminum alloy material, carbon fiber material, etc. in Figure 2 are also used as the structural nodes to be processed.

[0049] Step S3: Divide all the structural nodes between the structural node to be processed and the root node into multiple groups according to the growth order of the tree structure, and construct a keyword set for each group with the names of the structural nodes in each group as keywords.

[0050] This step processes each structural node to be processed separately. For example, for the structural node of the skin selected above, all the structural nodes between it and the root node, the fixed-wing aircraft, include the skin, the fuselage, and the fixed-wing aircraft. The above three nodes are divided into multiple groups. For example, the skin is the first group, and in this group, the skin is used as a keyword and is located in the keyword set of the first group. The fuselage and the fixed-wing aircraft are the second group, and in this group, the fuselage and the fixed-wing aircraft are used as keywords and are located in the keyword set of the second group.

[0051] Another example is for the structural node of 7075 aluminum alloy material to be processed. All the structural nodes between it and the root node, the fixed-wing aircraft, include 7075 aluminum alloy, the skin, the fuselage, and the fixed-wing aircraft. The above four nodes are divided into multiple groups. For example, 7075 aluminum alloy is the first group, and in this group, 7075 aluminum alloy is used as a keyword and is located in the keyword set of the first group. The skin and the fuselage are the second group, and in this group, the skin and the fuselage are used as keywords and are located in the keyword set of the second group. The fixed-wing aircraft is the third group, and in this group, the fixed-wing aircraft is used as a keyword and is located in the keyword set of the third group.

[0052] In some alternative embodiments, after step S3, it further includes:

[0053] Correct the keywords based on a pre-constructed word comparison dictionary, which is used to map aircraft terms to patent usage terms.

[0054] Although keyword sets are obtained in step S3, these keywords may not be suitable for patent retrieval, or are too simple and cannot comprehensively and accurately represent the relevant structures. For example Figure 2 the airfoil in [example]. In patent terms, in addition to the airfoil, keywords such as wing configuration can also be used to represent it. Another example is aerodynamic force, which can also be extended to keywords such as flow field pressure, lift, and drag. In this embodiment, a word comparison dictionary is first constructed. The word comparison dictionary contains multiple mapping relationships. Each mapping relationship includes two feature sets that can map to each other. The first feature set uses specialized terms in aircraft design, and the second feature set uses specialized terms in patent writing. The keywords in the two feature sets have a certain overlap. Then, based on the word comparison dictionary, the aircraft terms are corrected to one or more patent usage terms.

[0055] In some alternative embodiments, before correcting the keyword, for each patent usage term mapped to the keyword in the word comparison dictionary, calculate the matching degree between each patent usage term and the structural name of the parent node or child node of the structural node where the keyword is located, and only use the patent usage terms with a matching degree lower than the threshold to correct the keyword.

[0056] As mentioned above, there are multiple patent usage terms mapped to the keyword of aerodynamic force, such as flow field pressure, lift, drag, etc. If in the aircraft structure knowledge graph, the aerodynamic force structural node contains sub-nodes such as lift and drag, through the calculation of this embodiment, it is obvious that the matching degree of the patent usage term "lift" with the lift node, which is the sub-node of the aerodynamic force structural node, is 100%, and the matching degree of the patent usage term "drag" with the drag node, which is the sub-node of the aerodynamic force structural node, is 100%. Therefore, these two terms need to be blocked, and on the basis of the keyword "aerodynamic force", only the patent usage term "flow field pressure" is added as a keyword.

[0057] In this embodiment, the matching degree calculation can be performed using methods such as cosine similarity and Jaccard similarity coefficient. The threshold used in the comparison process can be manually modified, and the value usually ranges from 60% to 90%.

[0058] Step S4: Corresponding each keyword set to different positions in the patent text to construct a combined retrieval strategy for the patent text.

[0059] As mentioned above, the keywords of all structural nodes between the structural node to be processed and the root node are grouped. These keywords may appear in different positions such as the title, abstract, and specification of the patent text. Step S4 is used to perform the matching process between keywords and positions, so as to construct a combined retrieval strategy.

[0060] In some alternative embodiments, step S4 further includes:

[0061] Step S41: Corresponding the keyword of the structural node with the farthest distance from the root node of the structure knowledge graph to the title of the patent text, corresponding the keyword set composed of one or more structural nodes with the second farthest distance from the root node of the structure knowledge graph to the abstract of the patent text, and corresponding the keyword set composed of one or more structural nodes with the closest distance from the root node of the structure knowledge graph to the specification of the patent text;

[0062] Step S42: Connect the retrieval strategies constructed by multiple keyword sets with the relationship "AND" between each other, and connect the keywords within each keyword set with the relationship "OR".

[0063] In step S41, the structural nodes farther from the root node, that is, the structural nodes closer to the node to be processed, the patents related to these nodes are the most needed patents, and the related keywords usually should appear in the title. For example, for the structural node of skin to be processed in the foregoing example, if the patent title contains "skin", such a patent is usually what we need. However, the appearance of "skin" in the title does not indicate that this type of patent is related to an aircraft. Therefore, other keywords are needed for further limitation. For example, in step S41, the keywords of the structural node closest to the root node are corresponding to the position of the specification of the patent text, and the keywords of the nodes located between the root node and the structural node to be processed are corresponding to the abstract position of the patent text. For example, the parent node "fuselage" of "skin" is corresponding to the abstract position, and the parent node "fixed-wing aircraft" of "fuselage" is corresponding to the specification position.

[0064] Then in step S42, the combination of such keywords is connected to form a combined retrieval strategy. For example, a combined retrieval strategy expression is: TI=(skin OR covering layer) AND AB=(fuselage) AND DES=(fixed-wing aircraft). In this expression, TI is the title, AB is the abstract, DES is the specification, OR represents the OR connection relationship, and AND represents the AND connection relationship.

[0065] It can be understood that the above example only gives one combination method of the combined retrieval strategy. In actual use, some keywords can also be corresponding to other positions or combined positions of the patent text. Here, the combined position can be, for example, setting "fuselage" to appear in the title or abstract.

[0066] Step S5: Use the patent text retrieved by the combined retrieval strategy as the single retrieval result of the structural node to be processed.

[0067] As shown in step S2, in addition to the given structural node to be processed, its child nodes are also used as the structural nodes to be processed. Each structural node is retrieved using the corresponding combined retrieval strategy to obtain the retrieval result of the corresponding structural node, and it is marked as the single retrieval result. For example, for the structural node of "skin", 10,000 patent data are retrieved. For its sub-structural node "7075 aluminum alloy", 3,000 patent data are retrieved. For the sub-structural node "carbon fiber", 20,000 patent data are retrieved.

[0068] Step S6: Combine the single result of the structural node to be processed with the single retrieval results of all its sub-structural nodes as the composite retrieval result of the structural node to be processed initially selected.

[0069] In this step, the retrieval results of all structural nodes to be processed are aggregated as the composite retrieval result of the initially selected structural nodes. That is, in the above example, the single retrieval result of the skin, the single retrieval result of 7075 aluminum alloy, and the single retrieval result of carbon fiber together serve as the composite retrieval result of the skin.

[0070] In some alternative embodiments, step S6 further includes:

[0071] From bottom to top, for each structural node, remove the retrieval data of all its sub-structural nodes, and use the remaining retrieval data as the corrected single retrieval result of this structural node. The corrected single retrieval results of each structural node are combined as the composite retrieval result.

[0072] In this embodiment, during the composition of the single retrieval results, duplicate patents need to be removed. For example, if the initially selected structural node to be processed is the skin, there is partial overlap between the 10,000 retrieval results of the skin, the 3,000 retrieval results of 7075 aluminum alloy, and the 20,000 retrieval results of carbon fiber. Therefore, duplicate data needs to be deducted from the 10,000 data of the skin. Another example is that if the initially selected structural node to be processed is the fuselage, refer to Figure 2 , first correct the retrieval results of the skin according to the retrieval results of 7075 aluminum alloy and carbon fiber, correct the retrieval results of the frame based on the retrieval results of the truss structure and the box structure, correct the retrieval results of the cabin section based on the retrieval results of the cockpit and the cargo hold, and finally correct the retrieval results of the fuselage according to the retrieval results of the skin, aluminum alloy, carbon fiber, truss structure, box structure, frame, cockpit, cargo hold, and cabin section.

[0073] Step S7: Use the structural nodes to be processed and all their sub-structural nodes to construct an intelligence tree, extract patent content from the composite retrieval results corresponding to each structural node and its sub-structural nodes according to the patent intelligence analysis template, and mount it on the structural nodes of the intelligence tree.

[0074] With the retrieval results, all structural nodes to be processed can be analyzed.

[0075] For example, for the initially selected fuselage structure node, step S6 gives the composite retrieval results of the fuselage and the composite retrieval results of its three sub-structure nodes, namely, the composite retrieval results of the skin, the frame, and the cabin section. These retrieval results can form the patent intelligence of the fuselage and be attached to the fuselage structure node of the intelligence tree. Similarly, for the initially selected fuselage structure node, for its extended skin sub-structure node, the patent intelligence of the skin can be constructed based on the composite retrieval results of the skin, the composite retrieval results of aluminum alloy and carbon fiber in the sub-structure nodes of the skin, and attached to the skin structure node of the intelligence tree. In this way, the patent intelligence of all structural nodes to be processed can be constructed to form a patent intelligence tree for the initially selected structure to be processed.

[0076] In step S1 of this application, based on the initially selected structural node to be processed, a patent intelligence tree is automatically constructed based on the aircraft structure knowledge graph. The architecture of the patent intelligence tree here is a part of the aircraft structure knowledge graph and is a relatively complete part. For example, for the fuselage, all its sub-nodes are consistent with the aircraft structure knowledge graph. In an alternative implementation, in order to reconstruct a new patent intelligence tree from the aircraft structure knowledge graph, multiple new structure trees with continuous connection relationships can also be initially selected in step S1, so as to directly construct a new patent intelligence tree. For example, select the first-level node fuselage and the second-level nodes skin and frame. At this time, the system will automatically ignore the second-level node cabin section and only use the skin and frame as the second-level nodes of the fuselage. The patent intelligence formed only shows the patent proportion in terms of the skin and frame. Of course, for the skin and frame, the system still treats them as the initially selected structural nodes to be processed and continues to analyze their sub-nodes.

[0077] In some alternative embodiments, step S7 further includes:

[0078] Step S71: For each structural node A, determine the patent catalog of the structural node A and each sub-structural node B directly connected to the structural node A. Among them, the patent catalog of the sub-structural node B directly connected to the structural node A is the set of the patent catalogs of the sub-structural node B and all its sub-nodes C. The patent catalog of the structural node A is extracted from the corrected single retrieval result of the structural node A. The patent catalog contains multiple patent records, and each patent record includes at least two fields: the patent application date and the patent publication country.

[0079] Step S72: Based on the patent catalogs of the structural node A and each sub-structural node B directly connected to the structural node A, conduct patent technology composition analysis, patent annual application volume distribution analysis, and patent regional deployment distribution analysis.

[0080] This embodiment presents the detailed process of generating patent intelligence based on templates. For example, Figure 2 As shown, the structural node A here is the fuselage, the structural node B is the skin, frame, and cabin section, and the structural node C is aluminum alloy and carbon fiber. First, the patent catalog of the skin contains all the retrieved patents of the skin, aluminum alloy, and carbon fiber, that is, the composite retrieval results of the skin, such as 500 items. Similarly, the patent catalog of the frame includes the patents in the composite retrieval results of the frame, such as 600 items, and the patent catalog of the cabin section includes the patents in the composite retrieval results of the cabin section, such as 700 items. The retrieval results of the fuselage include the patents in the corrected single retrieval results of the fuselage, such as 200 items, that is, the number of patents after removing the composite retrieval results of the skin, the composite retrieval results of the frame, and the composite retrieval results of the cabin section from the 2000 patents of the composite retrieval results of the fuselage. This makes the quantity for patent distribution analysis in step S72 more accurate, and the corrected single retrieval results of the fuselage can be used as other patent data in the fuselage excluding the skin, frame, and cabin section, as Figure 3 shown.

[0081] The analysis of other patent application distributions and regional distributions is similar to the above structural composition analysis process, and can be directly obtained using the built-in functions of the excel table based on the application date field and application country field in the patent catalog.

[0082] The second aspect of this application provides a knowledge graph-based automatic patent intelligence generation device corresponding to the above method, mainly including:

[0083] An aircraft structure knowledge graph acquisition module, used to acquire the aircraft structure knowledge graph stored in a tree structure;

[0084] A structural node to be processed determination module, used to determine the position of the initially selected structural node to be processed in the knowledge graph, and use this structural node and all its sub-structural nodes as the structural nodes to be processed;

[0085] A keyword set acquisition module, used to divide all the structural nodes between the structural node to be processed and the root node into multiple groups in the growth order of the tree structure, and construct keyword sets for each group with the names of the structural nodes in each group as keywords;

[0086] A combined retrieval strategy formulation module, used to map each keyword set to different positions in the patent text to construct a combined retrieval strategy for the patent text;

[0087] A single retrieval result collection module, used to use the patent text retrieved by the combined retrieval strategy as the single retrieval result of the structural node to be processed;

[0088] A composite retrieval result determination module, configured to combine the single result of the to-be-processed structural node with the single retrieval results of all its sub-structural nodes as the composite retrieval result of the to-be-processed structural node initially selected;

[0089] An intelligence tree generation module, configured to construct an intelligence tree using the to-be-processed structural node and all its sub-structural nodes, extract patent content from the composite retrieval results corresponding to each structural node and its sub-structural nodes according to a patent intelligence analysis template, and mount it on the structural nodes of the intelligence tree.

[0090] In some alternative embodiments, the apparatus further includes:

[0091] A keyword correction module, configured to correct the keywords based on a pre-constructed word comparison dictionary, where the word comparison dictionary is used to map aircraft words to patent usage words.

[0092] In some alternative embodiments, the combined retrieval strategy formulation module includes:

[0093] A keyword correspondence unit, configured to correspond the keywords of the structural node farthest from the root node of the structural knowledge graph to the title of the patent text, correspond the keyword set composed of one or more structural nodes that are the second farthest from the root node of the structural knowledge graph to the abstract of the patent text, and correspond the keyword set composed of one or more structural nodes closest to the root node of the structural knowledge graph to the specification of the patent text;

[0094] A retrieval strategy automatic generation unit, configured to connect the retrieval strategies constructed by multiple keyword sets to each other using the relationship "AND", and connect the keywords within each keyword set using the relationship "OR".

[0095] In some alternative embodiments, the composite retrieval result determination module includes:

[0096] A duplicate removal processing unit, configured to, from bottom to top, for each structural node, remove the retrieval data of all its sub-structural nodes, and use the remaining retrieval data as the corrected single retrieval result of the structural node. The corrected single retrieval results of each structural node are combined as the composite retrieval result.

[0097] In some alternative embodiments, the intelligence tree generation module includes:

[0098] A patent catalog extraction unit is used to determine, for each structural node A, the patent catalogs of the structural node A and each sub-structural node B directly connected to the structural node A. Among them, the patent catalog of the sub-structural node B directly connected to the structural node A is the set of the patent catalogs of the sub-structural node B and all its sub-nodes C. The patent catalog of the structural node A is extracted from the corrected single search result of the structural node A. The patent catalog contains multiple patent records, and each patent record includes at least two fields: the patent application date and the patent publication country;

[0099] A data analysis unit is used to perform patent technology composition analysis, annual patent application volume distribution analysis, and patent regional deployment distribution analysis based on the patent catalogs of the structural node A and each sub-structural node B directly connected to the structural node A.

[0100] As described above, this is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed in this application should be covered by the protection scope of this application. Therefore, the protection scope of this application should be subject to the protection scope of the claims.

Claims

1. A method for automatically generating patent intelligence based on knowledge graph, characterized in that: include: Step S1, obtaining an aircraft structure knowledge graph stored in a tree structure; Step S2: determine the position of the initially selected structure node to be processed in the knowledge graph, and use the structure node and all its substructure nodes as the structure nodes to be processed; Step S3, dividing all structure nodes between the structure node to be processed and the root node into multiple groups according to the growth order of the tree structure, and constructing a keyword set for each group using the name of each structure node in each group as a keyword; Step S4, mapping each keyword set to different positions in the patent text to construct a combined search strategy for the patent text; Step S5, using the patent text retrieved by the combined search strategy as a single search result of the structure node to be processed; Step S6, combining the single search result of the structure node to be processed with the single search results of all its substructure nodes as the composite search result of the initially selected structure node to be processed; Step S7: construct an intelligence tree using the structure node to be processed and all its substructure nodes, extract patent content from the composite search results corresponding to each structure node and its substructure nodes according to the patent intelligence analysis template, and mount it on the structure node of the intelligence tree.

2. The method for automatically generating patent intelligence based on knowledge graph according to claim 1, characterized in that: Step S3 further includes: The keywords are modified based on a pre-built term comparison dictionary, which is used to map aircraft terms to patent usage terms.

3. The method for automatically generating patent intelligence based on knowledge graph according to claim 1, characterized in that: Step S4 further comprises: Step S41, the keywords of the structure node farthest from the root node of the structure knowledge graph are matched to the title of the patent text, the keyword set consisting of one or more structure nodes with the second farthest distance from the root node of the structure knowledge graph is matched to the abstract of the patent text, and the keyword set consisting of one or more structure nodes closest to the root node of the structure knowledge graph is matched to the description of the patent text; Step S42: connect the search strategies constructed by multiple keyword sets with each other using the relationship “and”, and connect the keywords in each keyword set with the relationship “or”.

4. The method for automatically generating patent intelligence based on knowledge graph according to claim 1, characterized in that: Step S6 further comprises: From the bottom up, for each structure node, the search data of all its child structure nodes are removed, and the remaining search data is used as the modified single search result of the structure node. The modified single search results of each structure node are combined as the composite search result.

5. The method for automatically generating patent intelligence based on knowledge graph according to claim 4, characterized in that: Step S7 further comprises: Step S71: for each structure node A, determine the patent catalog of the structure node A and each sub-structure node B directly connected to the structure node A, wherein the patent catalog of the sub-structure node B directly connected to the structure node A is the set of patent catalogs of the sub-structure node B and all sub-nodes C of the sub-structure node B, and the patent catalog of the structure node A is extracted from the corrected single search result of the structure node A, and the patent catalog contains multiple patent records, and each patent record includes at least two fields: patent application date and patent publication country; Step S72: Based on the patent catalogs of the structure node A and each substructure node B directly connected to the structure node A, conduct patent technology composition analysis, annual patent application volume distribution analysis, and patent regional deployment distribution analysis.

6. A patent information automatic generation device based on knowledge graph, characterized in that: include: An aircraft structure knowledge graph acquisition module is used to acquire an aircraft structure knowledge graph stored in a tree structure; A module for determining a structure node to be processed, used to determine the position of the initially selected structure node to be processed in the knowledge graph, and to take the structure node and all its substructure nodes as the structure nodes to be processed; A keyword set acquisition module is used to divide all structure nodes between the structure node to be processed and the root node into multiple groups according to the growth order of the tree structure, and construct a keyword set for each group using the name of each structure node in each group as a keyword; A combined search strategy formulation module is used to map each keyword set to different positions in the patent text to construct a combined search strategy for the patent text; A single search result collection module, used to use the patent text searched by the combined search strategy as a single search result of the structure node to be processed; A composite search result determination module, used for combining the single result of the structure node to be processed with the single search results of all its substructure nodes as the composite search result of the structure node to be processed that is initially selected; The intelligence tree generation module is used to construct an intelligence tree using the structure node to be processed and all its substructure nodes, extract patent content from the composite search results corresponding to each structure node and its substructure nodes according to the patent intelligence analysis template, and mount it on the structure node of the intelligence tree.

7. The patent information automatic generation device based on knowledge graph according to claim 6, characterized in that: The device also includes: A keyword correction module is used to correct the keywords based on a pre-built word comparison dictionary, wherein the word comparison dictionary is used to map aircraft terms to patent usage terms.

8. The patent information automatic generation device based on knowledge graph according to claim 6, characterized in that: The combined search strategy formulation module includes: A keyword corresponding unit, used to correspond the keywords of the structure node farthest from the root node of the structure knowledge graph to the title of the patent text, correspond the keyword set consisting of one or more structure nodes with the second farthest distance from the root node of the structure knowledge graph to the abstract of the patent text, and correspond the keyword set consisting of one or more structure nodes closest to the root node of the structure knowledge graph to the description of the patent text; The search strategy automatic generation unit is used to connect the search strategies constructed by multiple keyword sets with each other using the relationship "and", and to connect the keywords in each keyword set with the relationship "or".

9. The patent information automatic generation device based on knowledge graph according to claim 6, characterized in that: The composite search result determination module comprises: The deduplication processing unit is used to remove the search data of all substructure nodes of each structure node from the bottom up, and use the remaining search data as the modified single search result of the structure node. The modified single search results of each structure node are combined as the composite search result.

10. The patent information automatic generation device based on knowledge graph according to claim 9, characterized in that: The intelligence tree generation module includes: A patent catalog extraction unit is used to determine, for each structure node A, the patent catalog of the structure node A and each sub-structure node B directly connected to the structure node A, wherein the patent catalog of the sub-structure node B directly connected to the structure node A is the set of patent catalogs of the sub-structure node B and all sub-nodes C of the sub-structure node B, and the patent catalog of the structure node A is extracted from the corrected single search result of the structure node A, wherein the patent catalog contains multiple patent records, and each patent record includes at least two fields: patent application date and patent publication country; The data analysis unit is used to perform patent technology composition analysis, annual patent application volume distribution analysis, and patent regional deployment distribution analysis based on the patent catalog of the structure node A and each substructure node B directly connected to the structure node A.

Images