Method for optimizing distribution of large-scale competition review works and medium

Abstract

The invention relates to a method for optimizing distribution of large-scale competition review works and a medium. The method comprises the steps that an unweighted undirected graph with experts as nodes and co-review relations as edges is constructed; calculating the maximum value of a Hamilton circle theory based on the number of nodes, and adjusting the number of circles by decreasing progressively to enable the node degree to meet the integer constraint; designing a Hamilton circle construction algorithm, and circularly connecting nodes according to step length to generate a symmetric network structure; expanding each edge into a plurality of heavy edges to enable the node degree to reach a preset value; initializing a sequence containing all works and a review matrix; inserting the works into the matrix according to a network edge relationship, and ensuring that each work is reviewed by an expert and the common review number is strictly matched; and finally, generating a distribution matrix, checking global consistency, and outputting a standardized scheme. According to the method, efficient and balanced distribution of review tasks is realized through Hamilton circle theory optimization and a structured distribution mechanism.

Description

Technical Field

[0001] This invention relates to the field of resource optimization and allocation technology, specifically to a method and medium for optimizing the allocation of works for large-scale competition review. Background Technology

[0002] As innovation competitions continue to expand in scale, the fairness and scientific rigor of the judging process face serious challenges. Currently, large-scale innovation competitions typically employ a two- or three-stage judging process, including online judging in the preliminary round, on-site judging in the semi-final round, and a Q&A session. However, because innovative works often lack clearly defined scoring criteria, significant differences in subjective scoring among different experts make it difficult to guarantee the comparability and objectivity of the judging results. Although increasing the number of judging experts is considered a common strategy to improve fairness, limited funding and resources often result in an insufficient number of experts, further exacerbating the risk of judging errors.

[0003] In existing review schemes, the allocation of works largely relies on random distribution mechanisms. While this ensures a certain degree of objectivity, it has two major flaws: First, the overlap or imbalance between the expert review portfolios makes it difficult to quantify the different experts' scoring preferences (such as optimistic or conservative tendencies) through joint review of works, thus weakening the effectiveness of collective decision-making. Second, traditional methods lack a systematic optimization model, making it impossible to achieve the global optimum of work allocation with limited expert resources, resulting in low review efficiency and difficulty in verifying fairness.

[0004] To address the aforementioned issues, existing technologies attempt to optimize the review process using statistical methods (such as range and standardized scores) and graph theory models, but limitations remain. For example, while solutions based on genetic algorithms can approximate the optimal allocation, they suffer from high computational complexity and rely on massive computing power. Traditional graph theory models lack theoretical support in ensuring the balance and consistency of expert review tasks, making them unsuitable for the practical needs of large-scale competitions. Furthermore, existing methods rarely incorporate multi-objective optimization theory, failing to simultaneously achieve synergistic optimization of "maximizing the number of intersections among experts," "balancing the size of the intersections," and "maximizing the scale of the intersections." Summary of the Invention

[0005] This invention provides a method and medium for optimizing the allocation of judging entries in large-scale competitions. Its purpose is to solve the problems of insufficient fairness and difficulty in ensuring scientific rationality in large-scale innovation competitions due to uneven allocation of judging entries, significant differences in expert scoring, and limited resources.

[0006] To achieve the above objectives, the first aspect of the present invention provides a method for optimizing the allocation of judging entries in a large-scale competition, comprising the following steps:

[0007] Construct an unweighted undirected graph where nodes represent experts and edges represent co-review relationships among experts;

[0008] Based on the number of nodes in the unweighted undirected graph, determine the number of Hamiltonian cycles that can be constructed in the network, and gradually decrease and adjust the number of Hamiltonian cycles until the degree of each node satisfies the preset integer constraint.

[0009] Based on the adjusted number of Hamiltonian cycles, a Hamiltonian cycle construction algorithm is designed to sequentially connect each node with its neighboring nodes at a specified step size to generate a network structure containing multiple Hamiltonian cycles.

[0010] Each edge in the network structure generated by the Hamilton cycle construction algorithm is expanded into multiple heavy edges, so that the degree of each node reaches a preset value;

[0011] Initialize a sequence containing all works to be assigned and generate a review matrix corresponding to the number of experts;

[0012] According to the edge connection relationship of the network structure, the works in the work sequence are inserted into the review matrix in sequence, so that each work is reviewed by a specified number of experts and the co-review constraint is satisfied.

[0013] Generate a collection of expert-reviewed works based on the insertion results, and output an allocation scheme matrix containing the numbers of each expert-reviewed work.

[0014] Furthermore, methods for constructing unweighted and undirected network models include:

[0015] Set an equal workload for each expert, so that each expert reviews a fixed number of different works;

[0016] Construct a weighted undirected graph with all experts as nodes;

[0017] If two experts have jointly reviewed at least one work, then an undirected edge is established between the corresponding nodes;

[0018] The number of works jointly reviewed by each pair of experts is counted, and this number is used as the number of heavy edges in the corresponding edge. The weighted undirected graph is then converted into an unweighted undirected graph, where the existence of an edge only indicates a joint review relationship, and the number of edges reflects the total number of works jointly reviewed.

[0019] Furthermore, the number of Hamiltonian cycles is determined using the following formula:

[0020]

[0021] Where, n h-cycle υ(G) represents the maximum number of Hamiltonian cycles that can be constructed in the network; υ(G) represents the number of nodes in the network.

[0022] Furthermore, methods for gradually decreasing the number of Hamiltonian cycles until the degree of each node satisfies the preset integer constraint include:

[0023] Starting from the initial number of Hamiltonian cycles, gradually reduce the number of Hamiltonian cycles;

[0024] For each Hamiltonian cycle after a reduction, calculate the degree of each node and determine whether it satisfies the condition that the total degree is divisible by a preset integer constraint.

[0025] If the condition is not met, continue to reduce the number of Hamiltonian cycles and repeat the previous step until the number of Hamiltonian cycles that satisfy the divisibility condition is found.

[0026] Furthermore, methods for generating network structures containing multiple Hamiltonian cycles include:

[0027] The initial network structure is constructed based on the number of nodes in the unweighted undirected graph, with the nodes numbered sequentially.

[0028] Iterate through each node and connect its adjacent nodes sequentially according to multiple preset step values;

[0029] For each step value of each node, the corresponding adjacent node number is calculated cyclically using modulo operations;

[0030] Add an edge connecting the current node to its neighboring node in the initial network structure;

[0031] Repeat the above steps until all nodes complete the edge connection through each step value, forming a network structure containing multiple Hamiltonian cycles.

[0032] Furthermore, methods for expanding each edge in the network structure into multiple heavy edges include:

[0033] Calculate the number of multiple edges that need to be expanded for each edge based on the preset total degree and the original degree of each node in the current network structure;

[0034] Traverse each edge in the network structure and add multiple corresponding multiple edges according to the number of multiple edges;

[0035] Verify whether the degree of each node after expansion is equal to the preset total degree. If not, adjust the number of multiple edges and repeat the previous step until the degree of all nodes reaches the preset total degree.

[0036] Furthermore, methods for initializing the sequence of works and generating the review matrix include:

[0037] Generate unique numbers for all works to be assigned in sequence, forming a sequence of works;

[0038] Based on the total number of experts and the number of works each expert needs to review, create a two-dimensional empty matrix. The number of rows in the two-dimensional matrix is ​​equal to the total number of experts, and the number of columns in each row is equal to the number of works to be reviewed.

[0039] The sequence of works is associated with the two-dimensional matrix, such that each matrix element corresponds to a work position to be assigned.

[0040] Furthermore, methods for inserting a sequence of works into the review matrix include:

[0041] Traverse each edge in the network structure and determine the two corresponding experts based on the connection relationship of the edges;

[0042] Extract unassigned works sequentially from the work sequence, and assign a preset number of co-reviewed works to each edge;

[0043] The assigned co-reviewed works are simultaneously inserted into the review matrix rows corresponding to the two experts;

[0044] Check whether the number of experts inserted for each work has reached the specified number. If not, supplement the number of works from the remaining works until the condition is met.

[0045] Repeat the above steps until all works have been assigned and each expert's review matrix row is filled.

[0046] Furthermore, methods for generating a portfolio of expert-reviewed works and outputting an allocation scheme matrix include:

[0047] Traverse each row of the review matrix, extract all non-empty work numbers in that row, and form the review work set of the corresponding expert;

[0048] All the expert review portfolios are organized into an allocation scheme matrix, where each row represents one expert and the elements in the row are arranged in order of the review portfolio numbers.

[0049] Verify that the number of occurrences of each work number in the allocation scheme matrix is ​​equal to the number of designated review experts. If any anomalies are found, mark and correct them.

[0050] The final revised allocation scheme matrix is ​​converted into an output data format.

[0051] To achieve the above objectives, a second aspect of the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of the method for optimizing the allocation of works for large-scale competition judging.

[0052] The beneficial effects of this invention are:

[0053] Compared with existing technologies, this invention provides a method and medium for optimizing the allocation of judging entries in large-scale competitions. Through multi-objective collaborative optimization and theory-driven network modeling, it solves problems such as insufficient fairness, difficulty in verifying scientific rationality, and resource constraints in the allocation of judging entries. Specifically, firstly, an unweighted undirected graph is constructed with experts as nodes and co-judging relationships as edges. The optimal number of circles in the network structure is determined using Hamiltonian cycle programming theory, and adjusted incrementally based on integer constraints to ensure a strictly balanced workload (node ​​degree) for each expert, overcoming the uneven workload defects in traditional random allocation or genetic algorithms. Secondly, a Hamiltonian cycle construction algorithm is designed to generate a network structure containing multiple uniformly distributed Hamiltonian cycles through symmetric connection rules. This ensures that any expert has a co-judging relationship with a fixed number of other experts, and the number of co-judged entries is strictly consistent (e.g., each edge is expanded to 4 multiple edges). This maximizes the number of intersections among experts while simultaneously balancing and maximizing the size of the intersections, effectively quantifying differences in expert scoring preferences. Subsequently, an algorithm for inserting entries precisely allocates them to the review matrix according to network structure constraints. This ensures that each entry is reviewed by a specified number of experts, and that the intersection of the reviewing experts fully conforms to the preset constraints, avoiding the sparse or irregular distribution of review relationships caused by randomness in traditional methods. The final generated allocation scheme matrix combines theoretical optimality with practical feasibility. Without requiring massive computing power, it achieves efficient utilization of review resources, scientific calibration of scoring differences, and global fairness of allocation results through the synergistic effect of Hamiltonian cycle symmetry and integer programming constraints. This provides a verifiable and reproducible standardized allocation scheme for large-scale competitions. Attached Figure Description

[0054] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below.

[0055] Figure 1 This is a flowchart of a method for optimizing the allocation of works for large-scale competition review, as disclosed in an embodiment of the present invention.

[0056] Figure 2 This is a weighted undirected graph disclosed in an embodiment of the present invention.

[0057] Figure 3 This is an unweighted undirected graph disclosed in an embodiment of the present invention.

[0058] Figure 4 This is a schematic diagram of an embodiment of the present invention, using node z1 as an example, to illustrate that each node has 120 degrees.

[0059] Figure 5 These are overall and partial diagrams of an optimal network structure disclosed in an embodiment of the present invention. Detailed Implementation

[0060] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. 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 should fall within the scope of protection of the present invention.

[0061] According to embodiments of the present invention, it should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the following manufacturing method, in some cases the steps shown or described may be performed in a different order than that shown here.

[0062] like Figure 1 As shown, this invention provides a method for optimizing the allocation of judging entries in large-scale competitions, comprising the following steps:

[0063] Step S100: Construct an unweighted undirected graph, where nodes represent experts and edges represent the co-review relationship between experts;

[0064] First, each reviewer is abstracted as a node in a graph. Assuming there are 125 experts, the graph contains 125 nodes, each corresponding to a unique expert. Next, a "co-review relationship" is defined as the association where two experts jointly review at least one work during the review process. If two experts have a co-review relationship, an undirected edge is established between the corresponding nodes. For example, if expert A and expert B jointly reviewed several works, then an edge is connected between node A and node B. Here, "unweighted" means that the edge itself does not directly carry weight information; it only indicates the existence of a co-review relationship.

[0065] Secondly, to ensure a balanced workload for reviewers, the number of works reviewed by each expert is strictly equal (e.g., 120 works per expert). Based on this, an initial weighted undirected graph is constructed (see...). Figure 2 The graph represents the number of works jointly reviewed by each pair of experts, and uses this number as the weight of the corresponding edge (i.e., the number of overlapping edges). For example, if expert A and expert B jointly review 4 works, then the weight of edge AB is 4. In this case, the edges in the graph reflect both the existence of the joint review relationship and the number of works jointly reviewed through their weights.

[0066] Subsequently, the weighted undirected graph is converted into an unweighted undirected graph (see...). Figure 3The transformation rule is as follows: if there is at least one co-reviewed work between two experts (i.e., weight ≥ 1), then the edge is retained, but the weight information is ignored; otherwise, the edge is removed. For example, if expert A and expert B co-review 4 works, then edge AB is still retained as one edge in the unweighted graph, but the weight is no longer distinguished. This transformation simplifies the model complexity and facilitates subsequent graph theory-based structural optimization (such as Hamiltonian cycle construction).

[0067] Ultimately, the unweighted undirected graph visually presents the network topology of the expert group's collaborative review through the edge relationships between nodes.

[0068] Step S200: Based on the number of nodes in the unweighted undirected graph, determine the number of Hamiltonian cycles that can be constructed in the network, and gradually decrease and adjust the number of Hamiltonian cycles until the degree of each node satisfies the preset integer constraint.

[0069] Through theoretical derivation and incremental adjustments, the optimal number of Hamiltonian cycles to satisfy the balance of review tasks is determined, thereby constructing the network structure. The specific implementation process is as follows:

[0070] The maximum number of Hamiltonian cycles that can be constructed by the network is determined by the following formula:

[0071]

[0072] Where, n h-cycle This represents the maximum number of Hamiltonian cycles that can be constructed in the network; υ(G) represents the number of nodes in the network; it can be understood that the formula is defined as follows:

[0073] When the number of nodes in the network υ(G) is even, the maximum number of Hamiltonian cycles that can be constructed is υ(G) / 2; if υ(G) is odd, then it is (υ(G)-1) / 2.

[0074] Taking 125 experts as an example (υ(G)=125), since 125 is an odd number, substituting into the formula yields:

[0075]

[0076] Therefore, a network with 125 nodes can construct a maximum of 62 Hamiltonian cycles.

[0077] In the complete graph (62 Hamiltonian cycles), each node has a degree of 2×n. h-cycle =124. However, with a preset total score of 480 (e.g., each expert needs to review 120 works, and each work is reviewed by 5 experts), the following must be satisfied:

[0078]

[0079] This indicates that the complete graph structure cannot directly satisfy the integer constraint of task load balancing, and the number of Hamiltonian cycles needs to be adjusted.

[0080] Starting with the maximum number of Hamiltonian cycles (62), gradually reduce the number of Hamiltonian cycles. For each cycle reduced, the degree of each node decreases by 2. The adjustment process is as follows:

[0081] Number of cycles 61: The degree of each node = 61 * 2 = 122, the total degree = 125 * 122 = 15,250, which is still not divisible by 480.

[0082] Number of cycles 60: Degree of each node = 60 * 2 = 120, total degree = 125 * 120 = 15,000.

[0083] According to the handshake lemma, the total number of edges is: total degree / 2 = 7500. At this time, the review task (degree) of each expert is 120, which satisfies the multiple relationship with the preset total degree of 480, that is, each edge needs to be expanded into 4 multiple edges.

[0084] The adjusted network consists of 60 Hamiltonian cycles, with each node connected via a symmetric connection rule (see...). Figure 4 It is directly connected to 120 other experts. For example, node z1 is connected to the next 60 nodes through 60 Hamiltonian cycles (step size from 1 to 60), each cycle contributing 2 degrees, for a total of 120 degrees. This structure satisfies: maximized co-review relationships among experts (each expert has an intersection with 120 other experts), strictly uniform distribution of co-review relationships (the number of co-reviewed works is consistent in each intersection), and maximized intersection size by increasing the number of co-reviewed works to 4 through multi-edge expansion.

[0085] The final 60 Hamiltonian circles (see Figure 5 This approach represents the theoretically optimal solution, avoiding the computational complexity of genetic algorithms while ensuring a balanced workload for reviewers through integer programming. In the adjusted network, each expert has a potential co-review relationship with the other 120 experts, providing strict structural constraints for subsequent work allocation and ensuring fairness and efficient resource utilization in the review process.

[0086] Step S300: Based on the adjusted number of Hamiltonian cycles, design a Hamiltonian cycle construction algorithm, and sequentially connect each node with its adjacent nodes at a specified step interval to generate a network structure containing multiple Hamiltonian cycles;

[0087] Based on the optimal number of Hamiltonian cycles (e.g., 60) determined in step S200, the network is constructed using the algorithm (Table 1). The core logic of the algorithm is as follows:

[0088] Input parameters: Number of nodes N = 25 (total number of experts); Step size set J = {1, 2, ..., 60}, each step size corresponds to an independent Hamiltonian cycle.

[0089] Output objective: Generate a network structure containing 60 Hamiltonian cycles, with each node having a degree of 2 * 60 = 120.

[0090] The algorithm generates Hamiltonian cycles by superimposing multiple steps. The specific steps are as follows:

[0091] Create an empty graph G containing 125 nodes, numbered 1, 2, ..., 125.

[0092] For each step size j∈J, perform the following operations:

[0093] Traverse all nodes: For node i (from 1 to 125), calculate the numbers of its neighboring nodes:

[0094] neighbor = (i + j - 1) mod N + 1

[0095] Where i is the current node number, j is the step size, which determines the connection interval between nodes, N is the total number of nodes in the network, i.e. the total number of experts, mod is the modulo operation, used to iteratively process node numbers; and neighbor is the number of the next node connected to the current node i by the step size j.

[0096] For example, when j = 1 and i = 1, the adjacent node is 2; when i = 125, the adjacent node is 1, forming a closed loop. Add an edge connecting node i and the neighbor in the graph.

[0097] Repeat the above process until all edges corresponding to step size j have been added, ultimately generating 60 independent and uniformly distributed Hamiltonian cycles.

[0098] Table 1. Pseudocode of the core part of the Hamliton cycle construction algorithm

[0099]

[0100] Due to the cyclical nature of modular arithmetic, the connection rules for each node are completely consistent. For example, nodes z1 and z... 1+j (Mod. 125) connection, node z2 and z 2+j Connections, and so on (see...) Figure 4 (Partial schematic diagram).

[0101] Each step size j corresponds to an independent Hamiltonian cycle, covering all nodes and with no repeated edges. For example:

[0102] Step size j = 1: Generate a Hamiltonian cycle 1→2→3→...→125→1.

[0103] Step size j = 2: Generate another Hamiltonian cycle 1→3→5→...→2→4→...→1.

[0104] This process continues until the step size j = 60, generating a total of 60 cycles.

[0105] Each node is connected by 60 Hamiltonian cycles, each cycle contributing 2 degrees (in-degree and out-degree), and the total degree is: single node degree = 2 * 60 = 120; which meets the balance constraint adjusted in step S200.

[0106] like Figure 5 As shown, the final network structure exhibits high symmetry, with any two experts forming a co-review relationship through Hamiltonian cycles of different step lengths. For example, expert 1 is connected to expert 2 (step length 1), expert 3 (step length 2), expert 4 (step length 3), etc., and the co-review relationship of each connection is uniformly adjusted to 4 works through subsequent heavy edge expansion (step S400).

[0107] Step S400: Expand each edge in the network structure generated by the Hamilton cycle construction algorithm into multiple heavy edges, so that the degree of each node reaches a preset value;

[0108] In the network structure generated in step S300, each node has an initial degree of 120 (composed of 60 Hamiltonian cycles, each contributing 2 degrees). However, the actual review task requires each expert to review 120 works, and each work is reviewed by 5 experts, resulting in a total of 120 * 5 = 600 reviews. To achieve a balanced resource allocation, the degree of each node needs to be increased from 120 to 480 (i.e., each expert reviews 120 works, and each work corresponds to 4 co-review relationships).

[0109] Therefore, each edge in the network needs to be expanded into multiple multiple edges. Based on the preset total degree of 480 and the original degree of 120, the number of multiple edges that each edge needs to be expanded into is: multiple edge = preset total degree / original degree = 480 / 120 = 4; that is, each single edge is expanded into 4 multiple edges, which means that two experts jointly review 4 works.

[0110] For each edge (u,v) in the network, expand it sequentially. For example, if expert A and expert B are connected in the original network through a Hamiltonian cycle, then the single edge between them needs to be expanded into a 4-fold edge.

[0111] Based on the unweighted and undirected graph, add 3 duplicate edges to each edge (4 in total), with each duplicate edge representing a joint review task. Figure 3The transformation process from an unweighted graph to a weighted graph is demonstrated, and the increase in the number of edges intuitively reflects the increase in the number of works under joint review.

[0112] After expansion, the degree of each node should satisfy: Expanded degree = Original degree * Multiple of overlapping edges = 120 * 4 = 480. If the degree of a node does not reach 480, it is necessary to check whether the edge expansion is missing or duplicated, and readjust accordingly. After each edge is expanded into 4 overlapping edges, the number of works jointly reviewed by any two experts is fixed at 4. For example, if expert A and expert B are connected by 4 overlapping edges, it means that they need to jointly review 4 works to ensure the maximum intersection size.

[0113] Each expert needs to review 120 works. The 4 works to be reviewed for each edge are distributed among different expert combinations. The final total workload is: Total workload = (Total degree * Number of works / Edge) / Number of works reviewed by each expert = 125 * 480 * 1 / 5 = 12,000. This calculation ensures that all works are evenly distributed without duplication or omission.

[0114] If any abnormalities occur after the expansion, such as the degree not meeting the standard, conflicts in the allocation of works under joint review, or damage to network connectivity, corrections will be required.

[0115] Step S500: Initialize a sequence containing all works to be assigned and generate a review matrix corresponding to the number of experts;

[0116] First, generate unique numbers for all entries to be assigned, forming an ordered sequence. Assuming there are 3000 entries in the competition, they will be numbered from 1 to 3000 and stored sequentially in a list or array. This sequence ensures that each entry has a unique identifier, facilitating subsequent assignment and tracking. Based on the total number of experts (e.g., 125) and the number of entries each expert needs to review (e.g., 120), create a two-dimensional empty matrix. The matrix structure is as follows: the number of rows equals the total number of experts (125 rows), with each row corresponding to one expert; the number of columns equals the number of entries each expert needs to review (120 columns), with each column storing one entry number. Each empty space in the matrix will be filled with a specific entry number in subsequent steps.

[0117] The sequence of works is associated with the review matrix to ensure that each work is only inserted into the review list of a specified number of experts (e.g., 5), avoiding duplicate allocation; by pre-allocating an empty matrix, the 120 positions that each expert needs to fill are clearly defined to prevent the workload from overflowing or falling short; by using the row and column indexes of the matrix, experts and their reviewed works are directly associated, which facilitates subsequent querying and verification.

[0118] Step S600: According to the edge connection relationship of the network structure, insert the works in the work sequence into the review matrix in sequence, so that each work is reviewed by a specified number of experts and meets the co-review constraint.

[0119] First, each edge in the multi-edge extended network generated in step S400 is traversed (for example, the edge between expert A and expert B is expanded to 4 multi-edges, indicating that 4 works need to be jointly reviewed). For each edge, unassigned works (such as works X, Y, Z, and W) are extracted sequentially from the work sequence initialized in step S500 and simultaneously inserted into the review matrix rows of both expert A and expert B. This process ensures that the number of works jointly reviewed by each pair of experts is strictly 4, and the distribution of works among experts completely follows the network topology rules. For example, if the edge (A, B) corresponds to 4 works jointly reviewed, then work X will appear in the review lists of both experts A and B, and the same applies to the remaining works. By assigning edges one by one, the joint review relationship is systematically embedded into the review matrix, avoiding unevenness or conflicts caused by random allocation.

[0120] After completing the edge-based allocation, it is verified whether the number of times each work has been reviewed has reached a preset value (e.g., 5 times). If a work is only assigned to 4 experts, a new work is extracted from the remaining unassigned works and added to the qualified expert pair (e.g., selected from unassigned expert pairs). For example, if work M is only assigned to experts C, D, E, and F, then an additional expert G needs to be selected, M is inserted into G's review list, and the co-review relationship of the relevant edges is updated. This process ensures that all works are accurately allocated through dynamic adjustment, and that the workload of experts is strictly balanced (120 works reviewed per expert). The final generated review matrix satisfies the co-review quantity constraint while ensuring the fairness and coverage of the global allocation.

[0121] Step S700: Generate a collection of expert-reviewed works based on the insertion results, and output an allocation scheme matrix containing the numbers of each expert-reviewed work.

[0122] The system iterates through each row of the review matrix (corresponding to each expert), extracting all non-empty work IDs in that row to form an independent expert review work set. For example, expert 1's review work set consists of works 2, 24, 41, ..., 2565, containing 120 unique IDs. Then, the work sets of all 125 experts are integrated row-wise into a two-dimensional allocation scheme matrix, where each row represents one expert, and the elements within each row are arranged according to the review order of the work IDs they were responsible for. Simultaneously, the system verifies whether the frequency of each work ID is within the preset value (e.g., each work is reviewed by 5 experts). If an anomaly is found (e.g., a work is only assigned 4 times), the system will automatically mark it and trigger a supplementary allocation mechanism, selecting an appropriate position from the unassigned expert pairs to complete the allocation.

[0123] The generated allocation scheme matrix needs to be converted into a standardized, outputtable format so that the review system can use it directly.

[0124] In this embodiment, as described in steps S500-S600 above, after determining the network structure, a traversal insertion planning algorithm can be designed to insert works ①, ②, ③... up to the 3000th work into... Figure 5 The pseudocode for the work insertion algorithm in the network structure shown is presented in Table 2:

[0125] Table 2. Pseudocode of the core part of the work insertion algorithm.

[0126]

[0127] The final result is a 125x120 matrix. Each row of the table represents an expert and their portfolio of 120 entries. Below is a table showing the portfolios of the first three experts.

[0128] Table 3 shows the optimal allocation scheme obtained by Hamilton's algorithm (including the review portfolios of 3 experts).

[0129]

[0130]

[0131] According to another aspect of the embodiments of this application, an electronic device is also provided, including a processor and a memory, wherein the processor is configured to implement the steps of the method when executing a computer program stored in the memory.

[0132] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0133] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only exemplary. For example, the division of the units can be a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0134] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0135] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0136] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for optimizing the allocation of judging entries in large-scale competitions, characterized in that, Includes the following steps: Construct an unweighted undirected graph where nodes represent experts and edges represent co-review relationships among experts; Based on the number of nodes in the unweighted undirected graph, determine the number of Hamiltonian cycles that can be constructed in the network, and gradually decrease and adjust the number of Hamiltonian cycles until the degree of each node satisfies the preset integer constraint. Based on the adjusted number of Hamiltonian cycles, a Hamiltonian cycle construction algorithm is designed to sequentially connect each node with its neighboring nodes at a specified step size to generate a network structure containing multiple Hamiltonian cycles. Each edge in the network structure generated by the Hamilton cycle construction algorithm is expanded into multiple heavy edges, so that the degree of each node reaches a preset value; Initialize a sequence containing all works to be assigned and generate a review matrix corresponding to the number of experts; According to the edge connection relationship of the network structure, the works in the work sequence are inserted into the review matrix in sequence, so that each work is reviewed by a specified number of experts and the co-review constraint is satisfied. Generate a collection of expert-reviewed works based on the insertion results, and output an allocation scheme matrix containing the numbers of each expert-reviewed work.

2. The method for optimizing the allocation of judging entries in a large-scale competition as described in claim 1, characterized in that, Methods for constructing unweighted and undirected network models include: Set an equal workload for each expert, so that each expert reviews a fixed number of different works; Construct a weighted undirected graph with all experts as nodes; If two experts have jointly reviewed at least one work, then an undirected edge is established between the corresponding nodes; The number of works jointly reviewed by each pair of experts is counted, and this number is used as the number of heavy edges in the corresponding edge. The weighted undirected graph is then converted into an unweighted undirected graph, where the existence of an edge only indicates a joint review relationship, and the number of edges reflects the total number of works jointly reviewed.

3. The method for optimizing the allocation of judging entries in a large-scale competition as described in claim 1, characterized in that, The number of Hamiltonian cycles is determined by the following formula: Where, n h-cycle υ(G) represents the maximum number of Hamiltonian cycles that can be constructed in the network; υ(G) represents the number of nodes in the network.

4. The method for optimizing the allocation of judging entries in a large-scale competition as described in claim 1, characterized in that, Methods for decreasing the number of Hamiltonian cycles until the degree of each node satisfies a preset integer constraint include: Starting from the initial number of Hamiltonian cycles, gradually reduce the number of Hamiltonian cycles; For each Hamiltonian cycle after a reduction, calculate the degree of each node and determine whether it satisfies the condition that the total degree is divisible by a preset integer constraint. If the condition is not met, continue to reduce the number of Hamiltonian cycles and repeat the previous step until the number of Hamiltonian cycles that satisfy the divisibility condition is found.

5. The method for optimizing the allocation of judging entries in a large-scale competition as described in claim 1, characterized in that, Methods for generating network structures containing multiple Hamiltonian cycles include: The initial network structure is constructed based on the number of nodes in the unweighted undirected graph, with the nodes numbered sequentially. Iterate through each node and connect its adjacent nodes sequentially according to multiple preset step values; For each step value of each node, the corresponding adjacent node number is calculated cyclically using modulo operations; Add an edge connecting the current node to its neighboring node in the initial network structure; Repeat the above steps until all nodes are connected by each step value, forming a network structure containing multiple Hamiltonian cycles.

6. The method for optimizing the allocation of judging entries in a large-scale competition as described in claim 1, characterized in that, Methods for expanding each edge in a network structure into multiple multiple edges include: Calculate the number of multiple edges that need to be expanded for each edge based on the preset total degree and the original degree of each node in the current network structure; Traverse each edge in the network structure and add multiple corresponding multiple edges according to the number of multiple edges; Verify whether the degree of each node after expansion is equal to the preset total degree. If not, adjust the number of multiple edges and repeat the previous step until the degree of all nodes reaches the preset total degree.

7. The method for optimizing the allocation of judging entries in a large-scale competition as described in claim 1, characterized in that, Methods for initializing the sequence of works and generating the review matrix include: Generate unique numbers for all works to be assigned in sequence, forming a sequence of works; Based on the total number of experts and the number of works each expert needs to review, create a two-dimensional empty matrix. The number of rows in the two-dimensional matrix is ​​equal to the total number of experts, and the number of columns in each row is equal to the number of works to be reviewed. The sequence of works is associated with the two-dimensional matrix, such that each matrix element corresponds to a work position to be assigned.

8. The method for optimizing the allocation of judging entries in a large-scale competition as described in claim 7, characterized in that, Methods for inserting a sequence of works into a review matrix include: Traverse each edge in the network structure and determine the two corresponding experts based on the connection relationship of the edges; Extract unassigned works sequentially from the work sequence, and assign a preset number of co-reviewed works to each edge; The assigned co-reviewed works are simultaneously inserted into the review matrix rows corresponding to the two experts; Check whether the number of experts inserted for each work has reached the specified number. If not, supplement the number of works from the remaining works until the condition is met. Repeat the above steps until all works have been assigned and each expert's review matrix row is filled.

9. The method for optimizing the allocation of judging entries in a large-scale competition as described in claim 1, characterized in that, Methods for generating a portfolio of expert-reviewed works and outputting an allocation scheme matrix include: Traverse each row of the review matrix, extract all non-empty work numbers in that row, and form the review work set of the corresponding expert; All the expert review portfolios are organized into an allocation scheme matrix, where each row represents one expert and the elements in the row are arranged in order of the review portfolio numbers. Verify that the number of occurrences of each work number in the allocation scheme matrix is ​​equal to the number of designated review experts. If any anomalies are found, mark and correct them. The final revised allocation scheme matrix is ​​converted into an output data format.

10. A computer-readable storage medium storing a computer program thereon, characterized in that, The computer program, when run by a processor, performs the steps of the method for optimizing the allocation of works for large-scale competition review as described in any one of claims 1-9.

Images