AI intelligent patent retrieval and analysis device
By introducing anti-collision and dustproof components into the intelligent analysis device, and combining knowledge graphs and QUBO models, the stability and heat dissipation issues of the device in industrial environments were solved, enabling efficient creative evaluation.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-19
- Publication Date
- 2026-04-03
AI Technical Summary
Existing intelligent analysis devices are susceptible to damage from impacts in industrial environments, have low heat dissipation efficiency, are inconvenient to maintain, and lack objectivity and consistency in patent inventiveness assessment.
Physical protection capabilities are enhanced by employing anti-collision and dustproof components, and a retrieval and analysis system is integrated to conduct non-obviousness assessment using knowledge graphs and QUBO models.
To ensure stable operation of the device in complex environments, improve heat dissipation efficiency, facilitate maintenance, provide quantitative and creative evaluation results, and enhance the accuracy and consistency of evaluations.
Smart Images
Figure CN121785435A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of interdisciplinary technology of artificial intelligence and patent information processing, specifically to an AI-powered intelligent patent retrieval and analysis device. Background Technology
[0002] With the continuous development of artificial intelligence technology, it has become a trend to deploy dedicated intelligent analysis devices in specific fields. However, these analysis devices are usually equipped with sophisticated processors and storage units and run on general-purpose server or workstation hardware.
[0003] These general-purpose hardware devices are primarily designed for stable operation in standard data center or office environments, lacking effective buffering and protection against physical shocks and vibrations. When deployed in industrial environments or needing to be moved, accidental collisions can loosen or damage internal components, affecting the system's normal operation. Furthermore, the cooling systems of these general-purpose hardware devices often employ standard air-cooling designs. In dusty environments, airborne dust is easily drawn in and accumulates on the heatsinks and fans, reducing cooling efficiency and leading to overheating, performance degradation, or even system crashes. Moreover, their structure typically does not consider quick and easy cleaning and maintenance.
[0004] In the field of patent analysis, assessing the non-obviousness (or inventiveness) of an invention is the core and most challenging aspect. Currently, this assessment relies primarily on manual searches and subjective judgments by patent examiners or field experts. This process is not only extremely time-consuming and labor-intensive, but the consistency and objectivity of the assessment results are also significantly influenced by the examiner's personal knowledge and experience. Existing auxiliary search systems mainly focus on keyword-based literature retrieval, citation network analysis, or basic semantic similarity matching. These methods can find prior art literature that is literally or thematically related, but they struggle to make in-depth obviousness inferences about combinations of technical features from different technical fields. They lack a model that can simulate the complex relationships between technical concepts and quantify the difficulty of combination, and they cannot learn and evolve from historical analysis. Their output is usually a list of related documents, and the final inventiveness judgment still requires manual completion, failing to provide an objective, repeatable, and quantitative assessment result. Summary of the Invention
[0005] To address the shortcomings of existing technologies, this invention provides an AI-powered intelligent patent retrieval and analysis device that solves the problem that when the device is subjected to a collision, the anti-collision pad alone has limited buffering effect, which can easily cause the internal system of the analysis device to be damaged due to excessive impact force.
[0006] To achieve the above objectives, the present invention provides the following technical solution: an AI-powered intelligent patent retrieval and analysis device, comprising: an analysis device body, anti-collision components on both sides of the analysis device body, the anti-collision components being used to reduce the impact on the analysis device body and prevent it from making hard contact with external forces, a dustproof component on the side wall of the analysis device body, a connecting component on the side wall of the dustproof component, and a retrieval and analysis system integrated inside the analysis device body;
[0007] The anti-collision assembly includes a fixed plate, the sidewalls of which are disposed on both sides of the analytical device body. Anti-collision pads are fixedly connected to the sidewalls of the fixed plate. A pair of fixed frames are fixedly connected to the sidewalls of the analytical device body. Dampers are fixedly installed on the analytical device body and the sidewalls of the fixed plate. A fixed rod is fixedly connected inside the fixed frame. A slider is slidably connected to the sidewall of the fixed rod. The sidewall of the slider is slidably connected inside the fixed frame. A fixed block is fixedly connected to the sidewalls of the fixed plate and the slider. A rotating rod is rotatably connected inside the fixed block.
[0008] Preferably, the anti-collision component further includes a spring, which is sleeved on the side wall of the fixed rod, with one end of the spring fixedly connected inside the fixed frame and the other end of the spring fixedly connected to the side wall of the slider.
[0009] Preferably, the dustproof component includes a heat dissipation frame, the sidewall of which is slidably connected inside the body of the analytical device, and recesses are provided at the upper and lower ends of the heat dissipation frame.
[0010] Preferably, the connecting component includes a mounting frame, the sidewall of which is fixedly connected to the sidewall of the analytical device body, a first neodymium magnet is fixedly connected inside the mounting frame, a connecting frame is fixedly connected to the sidewall of the heat dissipation frame, a second neodymium magnet is fixedly connected inside the connecting frame, the sidewall of the connecting frame is slidably connected inside the mounting frame, and the first neodymium magnet and the second neodymium magnet are attracted to each other.
[0011] To address the aforementioned technical problems, this invention provides an AI-powered intelligent patent retrieval and analysis device, which integrates a retrieval and analysis system.
[0012] The retrieval and analysis system includes: a knowledge graph construction module, a QUBO model construction module, a non-obviousness measurement module, and a weight inverse enhancement module.
[0013] The knowledge graph construction module is used to decompose patent documents into multiple technical concept primitives and construct a knowledge hypergraph containing multiple nodes and dynamic connection weights.
[0014] The QUBO model construction module is used to parse the target invention to be analyzed into a set of target technical concept primitives, and combine the dynamic connection weights of the knowledge hypergraph to construct a quadratic unconstrained binary optimization model.
[0015] The non-obviousness measurement module is used to solve the quadratic unconstrained binary optimization model using the quantum annealing algorithm, obtain the ground state solution representing the most obvious combination of technologies, and calculate the energy difference between the target invention and the ground state solution to obtain the non-obviousness index.
[0016] The weight inverse enhancement module is used to update the dynamic connection weights in the knowledge hypergraph based on the ground state solution obtained by the non-obviousness measurement module.
[0017] The knowledge graph construction module, QUBO model construction module, non-obviousness measurement module, and weight inverse enhancement module work together to provide a quantitative analysis result for the assessment of the inventiveness level of a patent by outputting the non-obviousness index.
[0018] Preferably, when constructing the knowledge hypergraph, the knowledge graph construction module calculates the initial connection weight between any two technical concept primitives using an initial weight formula;
[0019] The initial weighting formula integrates the co-occurrence frequency and semantic similarity of the two technical concept primitives in the literature corpus, and is used to quantify the initial obviousness between the two technical concept primitives from two dimensions: structural relevance and content relevance.
[0020] The initial weight formula is:
[0021] J ij (0)=-(w c ·C norm (i,j)+w s Sim(v) i ,v j ));
[0022] Among them, J ij (0) represents the initial connection weights, w c With w s C is the preset weighting coefficient. norm (i,j) represents the normalized co-occurrence frequency, Sim(v i ,v j ) as a technological concept primitive v i With v j Semantic similarity.
[0023] Preferably, the quadratic unconstrained binary optimization model constructed by the QUBO model construction module has a Hamiltonian that includes a cost term and a constraint term.
[0024] The cost item is composed of the dynamic connection weights output by the knowledge graph construction module;
[0025] The constraint terms are used to incentivize or penalize the operation of adopting or not adopting the technical concept primitives in the target technical concept primitive set.
[0026] Its Hamiltonian formula is:
[0027] H = ∑ i<j J ij (t)·q i q j +∑ i P i ·q i ;
[0028] Where H is the system Hamiltonian, J ij (t) represents the dynamic connection weights, q i P is a binary variable representing whether or not a technological concept primitive is adopted. i This is the incentive or penalty coefficient.
[0029] In one specific embodiment, the non-obviousness measurement module specifically includes:
[0030] The system includes a ground state energy calculation unit, a target energy calculation unit, and an exponent generation unit.
[0031] The ground state energy calculation unit is used to determine the energy value E of the ground state solution based on the solution result of the quantum annealing algorithm. ground ;
[0032] The target energy calculation unit is used to substitute the target technology concept primitive set into the quadratic unconstrained binary optimization model to determine the energy value E of the target invention. inv ;
[0033] The index generation unit is used to calculate the difference between the energy value of the target invention and the energy value of the ground state solution, so as to generate the non-obviousness index ISI.
[0034] The calculation formula is as follows:
[0035] ISI = E inv -E ground ;
[0036] Preferably, the specific working process of the weight inverse enhancement module includes: performing automated cross-validation on the ground state solution in the existing technology literature library to confirm whether the technology combination represented by the ground state solution exists in reality; and calculating a weight update increment based on the verification result to reduce the dynamic connection weight value corresponding to the technology combination that has been verified as obvious.
[0037] Its dynamic connection weight update formula is:
[0038] J ij (t+1)=J ij (t)+ΔJ ij ;
[0039] Among them, J ij (t+1) represents the updated dynamic connection weights, ΔJ ij Update the weight increment.
[0040] Furthermore, when the weighted inverse enhancement module performs the automated cross-validation, it includes: performing a strong validation step of single-source semantic fingerprint matching; and after the strong validation step fails, performing a weak validation step of the minimum document set coverage algorithm.
[0041] Furthermore, the weight inverse enhancement module calculates the weight update increment ΔJ. ij The calculation factors used include: a preset global learning rate α, a verification strength factor β that distinguishes between strong and weak verification results, and an energy value E related to the ground state solution. ground The associated energy confidence factor γ, and a factor related to the publication time T of the literature used for validation. pub The associated time decay factor δ. Its calculation formula is:
[0042] ΔJ ij =-(α·β·γ·δ).
[0043] This invention provides an AI-powered intelligent patent retrieval and analysis device. It has the following beneficial effects:
[0044] 1. By setting up an anti-collision component, the present invention uses a spring and a damper in combination. When the analysis device body is subjected to external impact, the spring first absorbs the impact energy, and then the damper dissipates the energy, avoiding hard collisions and continuous vibrations, improving the overall physical protection capability of the device, ensuring the stable operation of the internal retrieval and analysis system, and enhancing the applicability and reliability of the device in complex environments.
[0045] 2. By setting up dustproof components and connecting components, the present invention utilizes the magnetic adsorption effect of the first neodymium magnet and the second neodymium magnet to achieve a detachable connection between the heat dissipation frame and the main body of the analysis device. This allows users to easily disassemble the heat dissipation frame for dust cleaning without tools, ensuring good heat dissipation performance inside the device, preventing the performance of the retrieval and analysis system from deteriorating due to heat accumulation, and improving the ease of maintenance and long-term stability of the device.
[0046] 3. This invention integrates a retrieval and analysis system and incorporates a weighted inverse enhancement module. This transforms the assessment of non-obviousness from static comparison to dynamic optimization. It not only outputs a quantified non-obviousness index, providing an objective measure of inventiveness, but also verifies the ground-state solution and feeds back to correct the dynamic connection weights of the knowledge hypergraph through the weighted inverse enhancement module. This enables the system to learn and evolve on its own. As the number of processed cases increases, the system's accuracy in judging the difficulty of technology combinations continuously improves, thus providing a continuously optimized and more reliable decision-making basis for patent analysis. Attached Figure Description
[0047] Figure 1 This is a perspective view of the present invention;
[0048] Figure 2 This is a schematic diagram of the anti-collision component of the present invention;
[0049] Figure 3 This is a schematic diagram of the interior of the anti-collision plate of the present invention;
[0050] Figure 4 This is a three-dimensional view of the back of the present invention;
[0051] Figure 5 This is a schematic diagram of the heat sink frame structure of the present invention;
[0052] Figure 6 For the present invention Figure 5 Enlarged structural diagram at point A;
[0053] Figure 7 This is a schematic diagram of the functional modules of a retrieval and analysis system according to an embodiment of the present invention;
[0054] Figure 8 This is a schematic diagram of the overall workflow of a retrieval and analysis system according to an embodiment of the present invention.
[0055] The components include: 1. the main body of the analytical device; 2. the fixing plate; 3. the anti-collision pad; 4. the fixing frame; 5. the damper; 6. the fixing rod; 7. the slider; 8. the spring; 9. the fixing block; 10. the rotating rod; 11. the heat dissipation frame; 12. the mounting frame; 13. the first neodymium magnet; 14. the connecting frame; and 15. the second neodymium magnet. Detailed Implementation
[0056] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0057] Please see the appendix Figure 1 - Appendix Figure 3 This invention provides an AI-powered intelligent patent retrieval and analysis device, characterized in that it includes:
[0058] The analysis device body 1 has anti-collision components on both sides. The anti-collision components are used to reduce the impact on the analysis device body 1 and prevent it from making hard contact with external forces. The side wall of the analysis device body 1 is equipped with a dustproof component, and the side wall of the dustproof component is equipped with a connecting component. The analysis device body 1 integrates a retrieval and analysis system.
[0059] The anti-collision assembly includes a fixed plate 2, the side walls of the fixed plate 2 are set on both sides of the analyzer body 1, the side walls of the fixed plate 2 are fixedly connected to anti-collision pads 3, the side walls of the analyzer body 1 are fixedly connected to a pair of fixed frames 4, the analyzer body 1 and the side walls of the fixed plate 2 are fixedly installed with dampers 5, the inside of the fixed frame 4 is fixedly connected to a fixed rod 6, the side wall of the fixed rod 6 is slidably connected to a slider 7, the side wall of the slider 7 is slidably connected inside the fixed frame 4, the side walls of the fixed plate 2 and the slider 7 are fixedly connected to a fixed block 9, and the inside of the fixed block 9 is rotatably connected to a rotating rod 10.
[0060] The analysis device body 1 is the core carrier of the entire device, used to integrate various systems and components to realize AI intelligent patent retrieval and analysis functions. The anti-collision components set on both sides of the analysis device body 1 are used to reduce the impact on the analysis device body 1 and prevent it from making hard contact with external forces, thus protecting the analysis device body 1 and preventing damage to the internal system due to collisions. The dustproof components set on the side wall of the analysis device body 1 can prevent dust from entering the analysis device body 1, keep the interior clean, and ensure the normal operation of the system. The connecting components set on the side wall of the dustproof components are used to connect and fix the dustproof components to the analysis device body 1. The retrieval and analysis system integrated inside the analysis device body 1 is the key to realizing AI intelligent patent retrieval and analysis, and can perform operations such as retrieval and analysis of patent information.
[0061] The sidewalls of the fixed plate 2 are set on both sides of the body 1 of the analytical device, serving to connect and support components such as the anti-collision pads 3. The anti-collision pads 3, which are fixedly connected to the sidewalls of the fixed plate 2, can absorb part of the impact force when a collision occurs through their own buffering characteristics, reducing the impact on the body 1 of the analytical device. A pair of fixed frames 4, which are fixedly connected to the sidewalls of the body 1 of the analytical device, provide installation space and guidance for components such as the fixed rod 6 and the slider 7. The damper 5, which is fixedly installed on the sidewalls of the fixed plate 2 and the body 1 of the analytical device, can provide damping force when a collision occurs, slowing down the impact speed of the collision and further reducing the damage caused by the collision. The fixed rod 6, which is fixedly connected inside the fixed frame 4, provides a track for the sliding of the slider 7, ensuring the stability of the movement of the slider 7. The slider 7, which is slidably connected to the sidewalls of the fixed rod 6, can slide on the fixed rod 6, and achieves a buffering effect in conjunction with components such as the spring 8. The fixed block 9, which is fixedly connected to the sidewalls of the fixed plate 2 and the slider 7, is used to connect the fixed plate 2 and the slider 7 and provide an installation position for the rotating rod 10.
[0062] Please see the appendix Figure 1 - Appendix Figure 3 The anti-collision component also includes a spring 8, which is sleeved on the side wall of the fixed rod 6. One end of the spring 8 is fixedly connected to the inside of the fixed frame 4, and the other end of the spring 8 is fixedly connected to the side wall of the slider 7.
[0063] Spring 8 is sleeved on the side wall of fixed rod 6. One end of spring 8 is fixedly connected to the inside of fixed frame 4, and the other end of spring 8 is fixedly connected to the side wall of slider 7. When a collision occurs, slider 7 slides on fixed rod 6, which will compress or stretch spring 8. Spring 8 uses its own elastic deformation to absorb the energy generated by the collision, thereby further reducing the impact of the collision on the body of the analysis device 1 and enhancing the buffering effect of the anti-collision component.
[0064] Please see the appendix Figure 5 The dustproof component includes a heat sink frame 11, the sidewall of which is slidably connected to the inside of the analyzer body 1, and recesses are provided at the upper and lower ends of the heat sink frame 11.
[0065] The heat dissipation frame 11 is slidably connected to the inside of the analyzer body 1. The heat dissipation frame 11 has recessed holes at both the top and bottom. The heat dissipation frame 11 can dissipate heat from the inside of the analyzer body 1, while preventing dust from entering the inside of the analyzer body 1, ensuring that the internal system operates in a clean and suitable temperature environment. The recessed holes at both the top and bottom are used by workers to remove and place the frame during maintenance.
[0066] Please see the appendix Figure 5 - Appendix Figure 6The connecting assembly includes a mounting frame 12, the side wall of which is fixedly connected to the side wall of the analyzer body 1. A first neodymium magnet 13 is fixedly connected inside the mounting frame 12. A connecting frame 14 is fixedly connected to the side wall of the heat dissipation frame 11. A second neodymium magnet 15 is fixedly connected inside the connecting frame 14. The side wall of the connecting frame 14 is slidably connected inside the mounting frame 12. The first neodymium magnet 13 and the second neodymium magnet 15 are attracted to each other.
[0067] The mounting frame 12 is fixedly connected to the side wall of the analyzer body 1, providing a mounting carrier for the first neodymium magnet 13. The first neodymium magnet 13, fixedly connected inside the mounting frame 12, achieves connection and fixation between the connecting frame 14 and the mounting frame 12 through the attraction of the second neodymium magnet 15. The connecting frame 14, fixedly connected to the side wall of the heat dissipation frame 11, is used to connect the heat dissipation frame 11 and the second neodymium magnet 15, and cooperates with the mounting frame 12 to achieve connection. The second neodymium magnet 15, fixedly connected inside the connecting frame 14, attracts the first neodymium magnet 13, thereby fixing the heat dissipation frame 11 to the analyzer body 1, and also facilitating the disassembly and installation of the heat dissipation frame 11, and making it easy to clean or maintain the heat dissipation frame 11.
[0068] See attached document Figure 7 , Figure 7 This is a schematic diagram of the functional modules of a retrieval and analysis system according to an embodiment of the present invention.
[0069] The retrieval and analysis system is integrated inside the analysis device body 1, and its functions are executed through the processor and memory inside the device body 1.
[0070] The retrieval and analysis system provided by the present invention may include: a knowledge graph construction module 20, a QUBO model construction module 21, a non-obviousness measurement module 22, and a weight inverse enhancement module 23.
[0071] In one specific embodiment, the knowledge graph construction module 20 is used to decompose the collection of patent documents stored in the database into multiple technical concept primitives, and construct a knowledge hypergraph containing multiple nodes and dynamic connection weights based on the association between the technical concept primitives.
[0072] The specific working process of the knowledge graph construction module 20 is as follows: First, the document text is processed through a natural language processing model to extract standardized technical feature phrases. Each phrase is defined as a technical concept primitive v. i Secondly, each unique technological concept primitive v i As a node in a knowledge hypergraph; finally, calculate any two nodes v i and v j The initial connection weights J between them ij (0).
[0073] Initial connection weight J ij The formula for calculating (0) is as follows:
[0074] J ij (0)=-(w c ·C norm (i,j)+w s Sim(v) i ,v j ));
[0075] Among them, J ij (0) represents node v i With v j Initial connection weights between; w c With w s C is a pre-set weighting coefficient used to balance the importance of different information sources. norm (i,j) represents the technological concept primitive v. i With v j The co-occurrence frequency in the literature collection after normalization; Sim(v i ,v j () is a technique for extracting conceptual primitives v through language models. i With v j The semantic similarity value is calculated after the word embedding vectors are obtained.
[0076] The QUBO model building module 21 is connected to the knowledge graph building module 20. Upon receiving a target invention to be analyzed, it parses it into a set of target technical concept primitives and, combined with the dynamic connection weights output by the knowledge graph building module 20, constructs a quadratic unconstrained binary optimization (QUBO) model. This module first decomposes the target invention into a set S of target technical concept primitives. inv Then, for each technical concept primitive v in the system i Define a binary variable q i ∈{0,1}; Finally, construct a Hamiltonian H to describe the total energy of the technology combination system.
[0077] The formula for Hamiltonian H is as follows:
[0078] H = ∑ i<j J ij (t)·q i q j +∑ i P i ·q i ;
[0079] Where H is the system Hamiltonian; J ij (t) represents node v at the t-th iteration. i and vj Dynamic connection weights between; q i To correspond to the basic technical concept v i A binary variable; P i As an incentive or penalty coefficient, when the technological concept primitive v i Belonging to the target technology concept primitive set S inv At that time, P i Take a preset negative value if it does not belong to set S. inv At that time, P i Take a preset positive value or zero.
[0080] The non-obviousness measurement module 22 is connected to the QUBO model building module 21 and is used to receive and solve a quadratic unconstrained binary optimization model. This module uses a quantum annealing algorithm solver to find the combination of variables that minimizes the Hamiltonian H; this combination is the ground state solution. This module includes a ground state energy calculation unit, a target energy calculation unit, and an exponent generation unit.
[0081] The ground-state energy calculation unit is used to determine the energy value E of the ground-state solution based on the solution results of the quantum annealing algorithm. ground The target energy calculation unit is used to process the target technology concept primitive set S. inv All corresponding binary variables q i Set the value to 1, and the other variables to 0. Substitute these values into the formula for the Hamiltonian H to calculate the energy value E of the target invention. inv The exponent generation unit is used to calculate E. inv With E ground The difference between them is used to generate the Non-Obviousness Index (ISI), which is calculated using the following formula:
[0082] ISI = E inv -E ground ;
[0083] The weight inverse enhancement module 23 is connected to the non-obviousness measurement module 22 and the knowledge graph construction module 20. This module is used to update the dynamic connection weights in the knowledge graph construction module 20 after the non-obviousness measurement module 22 obtains the ground-state solution. This module first performs automated cross-validation on the ground-state solution. The specific validation steps include: performing a strong validation step using single-source semantic fingerprint matching; if the strong validation step fails, then performing a weak validation step using the minimum document set coverage algorithm.
[0084] After successful cross-validation, the weight inverse enhancement module 23 calculates a weight update increment ΔJ for each pair of technical concept primitives constituting the ground state solution. ij The formula for calculating this increment is as follows:
[0085] ΔJ ij=-(α·β·γ·δ);
[0086] Where α is the preset global learning rate constant; β is the validation strength factor, which takes different values depending on the success of strong or weak validation; and γ is the energy value E of the ground state solution. ground The associated energy confidence factor is calculated using the formula γ = 1 / (1 + exp(k·E)). ground ), where k is a preset positive coefficient; δ is a coefficient related to the publication time T of the literature used for verification. pub The associated time decay factor is calculated using the formula δ=exp(-λ·(T)). current -T pub ), where λ is the preset time decay constant, T current This is the current time.
[0087] Finally, the weight inverse enhancement module 23 updates the calculated weight increment ΔJ through an asynchronous batch processing mechanism. ij The dynamic connection weights applied to knowledge hypergraphs are updated using the formula J. ij (t+1)=J ij (t)+ΔJ ij J ij (t+1) represents the updated dynamic connection weights.
[0088] See attached document Figure 8 , Figure 8 This is a schematic diagram of the overall workflow of a retrieval and analysis system according to an embodiment of the present invention.
[0089] The following section will use an invention to be analyzed as an example to illustrate the collaborative workflow of the various modules within the system.
[0090] First, during the task receiving and model building phase, the analysis device body 1 receives a technical document of the target invention to be analyzed through its input interface.
[0091] The QUBO model building module 21 is activated to process the technical document. Through its internal natural language processing function, it decomposes the target invention into a specific set of target technical concept primitives S. inv .
[0092] Subsequently, the QUBO model building module 21 sends a request to the knowledge graph building module 20 to obtain the target technology concept primitive set S. inv All dynamic connection weights J corresponding to other technical concept primitives in the related technical field. ij (t).
[0093] The knowledge graph construction module 20 retrieves and returns these weight data from its maintained knowledge hypergraph. Based on these weight data and preset incentive and penalty coefficients, the QUBO model construction module 21 constructs a complete quadratic unconstrained binary optimization model representing this analysis task and outputs the model in matrix form.
[0094] Next, the quantum annealing solution and exponential output stage begins. The non-obviousness measurement module 22 receives the matrix of the quadratic unconstrained binary optimization model from the QUBO model construction module 21. This module loads this matrix into the quantum annealing algorithm solver integrated within the device. The solver performs the annealing process to obtain the ground-state solution that minimizes the system's Hamiltonian H, i.e., the configuration of a set of binary variables.
[0095] The ground state energy calculation unit inside the non-obviousness measurement module 22 calculates the energy value E of the ground state solution based on this ground state solution. ground Meanwhile, the target energy calculation unit inside the module will represent the set of technical concept primitives S of the target invention. inv Substituting the corresponding binary variable configurations into the same Hamiltonian H formula, the energy value E of the target invention is calculated. inv .
[0096] Finally, the exponential generating unit calculates E. inv With E ground The difference is used to generate the final non-obviousness index (ISI), which is then presented through the output interface of the analysis device body 1.
[0097] While the non-obviousness measurement module 22 outputs its results, the system enters the background asynchronous verification and graph enhancement phase. The non-obviousness measurement module 22 sends the obtained ground-state solution (i.e., a combination of a set of technical concept primitives) to the weighted inverse enhancement module 23 via an internal message queue. This asynchronous communication method ensures that the user's response in the foreground and the system's learning process in the background do not block each other.
[0098] Upon receiving the ground-state solution, the weighted inverse enhancement module 23 initiates an automated cross-validation process. In the strong validation step, the module encodes all technical concept primitives in the ground-state solution into a high-dimensional semantic vector and performs a nearest neighbor search in a pre-established index containing vectors from all existing technical documents. If a vector from a single document is found whose distance to the ground-state solution vector is less than a preset threshold, the strong validation is successful.
[0099] If strong verification fails, the system automatically switches to weak verification and starts a set coverage algorithm to find the minimum number of document combinations in the literature database that can completely cover all technical concept primitives in the ground state solution.
[0100] After successful verification, the weight inverse enhancement module 23 calculates the weight update increment ΔJ for each pair of technical concept primitives in the ground state solution according to the preset formula. ij .
[0101] These increments are collected into a batch processing task. When triggered by a system-preset time window or task volume threshold, the weight inverse enhancement module 23 applies all weight update increments of this batch to the knowledge hypergraph maintained by the knowledge graph construction module 20 in a database transaction manner, completing the adjustment of the dynamic connection weights J. ij Update (t).
[0102] Through this process, the system completes an iterative optimization of its internal knowledge base while simultaneously performing a non-obviousness assessment.
Claims
1. An AI-powered intelligent patent retrieval and analysis device, characterized in that, include: The analysis device body (1) is provided with anti-collision components on both sides of the analysis device body (1). The anti-collision components are used to reduce the impact on the analysis device body (1) and prevent it from making hard contact with external forces. The analysis device body (1) is provided with dustproof components on the side wall. The dustproof components are provided with connecting components on the side wall. The analysis device body (1) integrates a retrieval and analysis system inside. The anti-collision assembly includes a fixed plate (2), the sidewalls of which are disposed on both sides of the body of the analytical device (1), an anti-collision pad (3) is fixedly connected to the sidewalls of the fixed plate (2), a pair of fixed frames (4) are fixedly connected to the sidewalls of the body of the analytical device (1), a damper (5) is fixedly installed on the body of the analytical device (1) and the sidewalls of the fixed plate (2), a fixed rod (6) is fixedly connected inside the fixed frame (4), a slider (7) is slidably connected to the sidewalls of the fixed rod (6), the sidewalls of the slider (7) are slidably connected inside the fixed frame (4), a fixed block (9) is fixedly connected to the sidewalls of the fixed plate (2) and the slider (7), and a rotating rod (10) is rotatably connected inside the fixed block (9).
2. The AI-powered intelligent patent retrieval and analysis device according to claim 1, characterized in that, The anti-collision assembly also includes a spring (8), which is sleeved on the side wall of the fixed rod (6). One end of the spring (8) is fixedly connected to the inside of the fixed frame (4), and the other end of the spring (8) is fixedly connected to the side wall of the slider (7).
3. The AI-powered intelligent patent retrieval and analysis device according to claim 2, characterized in that, The dustproof component includes a heat dissipation frame (11), the side wall of which is slidably connected to the inside of the analysis device body (1), and the heat dissipation frame (11) has recessed holes at its upper and lower ends.
4. The AI-powered intelligent patent retrieval and analysis device according to claim 3, characterized in that, The connecting assembly includes a mounting frame (12), the sidewall of which is fixedly connected to the sidewall of the analytical device body (1), a first neodymium magnet (13) is fixedly connected inside the mounting frame (12), a connecting frame (14) is fixedly connected to the sidewall of the heat dissipation frame (11), a second neodymium magnet (15) is fixedly connected inside the connecting frame (14), and the sidewall of the connecting frame (14) is slidably connected inside the mounting frame (12), wherein the first neodymium magnet (13) and the second neodymium magnet (15) are attracted to each other.
5. The AI-powered intelligent patent retrieval and analysis device according to claim 1, characterized in that, The retrieval and analysis system includes: A knowledge graph construction module is used to decompose patent documents into multiple technical concept primitives and construct a knowledge hypergraph containing multiple nodes and dynamic connection weights. The QUBO model building module is used to parse the target invention to be analyzed into a set of target technical concept primitives, and combine the dynamic connection weights of the knowledge hypergraph to build a quadratic unconstrained binary optimization model. The non-obviousness measurement module is used to solve the quadratic unconstrained binary optimization model through quantum annealing algorithm, obtain the ground state solution representing the most obvious combination of technologies, and calculate the energy difference between the target invention and the ground state solution to obtain the non-obviousness index. The weight inverse enhancement module is used to update the dynamic connection weights in the knowledge hypergraph based on the ground state solution obtained by the non-obviousness measurement module. The knowledge graph construction module, QUBO model construction module, non-obviousness measurement module, and weight inverse enhancement module work together to provide a quantitative analysis result for the assessment of the inventiveness level of a patent by outputting the non-obviousness index.
6. The AI-powered intelligent patent retrieval and analysis device according to claim 5, characterized in that, When constructing the knowledge hypergraph, the knowledge graph construction module calculates the initial connection weight between any two technical concept primitives using an initial weight formula. The initial weighting formula integrates the co-occurrence frequency and semantic similarity of the two technical concept primitives in the literature corpus, and is used to quantify the initial obviousness between the two technical concept primitives from two dimensions: structural relevance and content relevance.
7. The AI-powered intelligent patent retrieval and analysis device according to claim 5, characterized in that, The quadratic unconstrained binary optimization model constructed by the QUBO model construction module has a Hamiltonian that includes a cost term and a constraint term. The cost item is composed of the dynamic connection weights output by the knowledge graph construction module; The constraint terms are used to incentivize or penalize the operation of adopting or not adopting the technical concept primitives in the target technical concept primitive set.
8. The AI-powered intelligent patent retrieval and analysis device according to claim 5, characterized in that, The non-obviousness measurement module specifically includes: The ground state energy calculation unit is used to determine the energy value of the ground state solution based on the solution result of the quantum annealing algorithm. The target energy calculation unit is used to substitute the target technology concept primitive set into the quadratic unconstrained binary optimization model to determine the energy value of the target invention. An index generation unit is used to calculate the difference between the energy value of the target invention and the energy value of the ground state solution to generate the non-obviousness index.
9. The AI-powered intelligent patent retrieval and analysis device according to claim 5, characterized in that, The specific working process of the weighted inverse enhancement module includes: Automated cross-validation of the ground state solution is performed in an existing technical literature database to confirm whether the combination of technologies represented by the ground state solution exists in reality; Calculate a weight update increment based on the verification results, and reduce the dynamic connection weight values corresponding to the technology combinations that have been verified as obvious.
10. The AI-powered intelligent patent retrieval and analysis device according to claim 9, characterized in that, When the weight inverse enhancement module performs the automated cross-validation, it includes: Perform a strong verification step involving single-source semantic fingerprint matching; After the strong verification step fails, the weak verification step of the minimum document set coverage algorithm is executed.