Work distribution method and device based on network model and genetic algorithm, and medium

Abstract

The invention relates to a work distribution method and device based on a network model and a genetic algorithm, and a medium. The method comprises the following steps: weighting an undirected network model; defining a multi-objective optimization function based on the weighted undirected network model; according to the multi-objective optimization function, initializing a genetic algorithm population; calculating an overall connectivity index for the weighted undirected network model corresponding to each individual, and taking the overall connectivity index as an individual fitness value; screening high-fitness individuals according to the fitness values to generate a next-generation population; performing crossover and mutation operation on individuals in the next generation of population; repeating the steps of integral connectivity index calculation, screening, crossover operation and mutation operation until a termination condition is met; and selecting an individual with the highest fitness value from the final population, decoding a gene sequence of the individual to obtain a work distribution scheme, and improving scientificity and objectivity of a review result by optimizing the number, distribution and network connectivity of the works jointly reviewed by experts.

Description

Technical Field

[0001] This invention relates to the field of fair allocation and decision optimization in review systems, specifically to a method, device, and medium for allocating works based on network models and genetic algorithms. Background Technology

[0002] In large-scale innovation competitions, fairness and scientific rigor are the core challenges of the judging process. Traditional judging mechanisms typically distribute entries to experts through random assignment or fixed grouping, but these methods have significant shortcomings when dealing with a large number of entries. For example, the limited overlap or uneven distribution of entries among judges makes it difficult to calibrate scoring criteria, amplifies subjective differences, and ultimately affects the objectivity of the results. Furthermore, limited funding and resources often prevent a significant increase in the number of experts, making it difficult for traditional allocation methods to balance workload and judging quality, easily leading to accumulated errors and a lack of fairness.

[0003] In existing technologies, review schemes often rely on human experience or simple randomization, lacking systematic optimization. For example, small-scale competitions often employ a single review process (such as direct scoring or group review), but in large-scale scenarios, such methods are inefficient and cannot effectively control the co-review relationships among experts. Some studies have attempted to reduce scoring bias through statistical methods (such as standard score calculation), but have failed to fundamentally solve the structural problems of expert allocation. Furthermore, while traditional network models can describe expert relationships, they lack dynamic optimization capabilities and struggle to meet the complex requirements of multiple objectives (such as maximizing the number of intersections and homogenizing the intersection distribution).

[0004] In recent years, intelligent optimization algorithms (such as genetic algorithms) have shown potential in resource allocation problems, but their application in review systems still has limitations. Existing solutions often ignore the connectivity and weight distribution of the expert review network, leading to local optima or deviations from actual needs in the allocation results. For example, simple genetic algorithms may struggle to optimize information sharing among experts while ensuring a balanced workload due to unreasonable encoding methods or imperfect fitness function design. Summary of the Invention

[0005] This invention provides a method, device, and medium for allocating works based on network models and genetic algorithms. Its purpose is to solve the problems of inconsistent scoring standards, significant subjective differences, and insufficient fairness in judging caused by the uneven distribution of works among judges in large-scale innovation competitions.

[0006] To achieve the above objectives, the first aspect of this invention provides a method for allocating works based on a network model and a genetic algorithm, comprising the following steps: Construct a weighted undirected network model with experts as nodes; A multi-objective optimization function is defined based on the weighted undirected network model. Based on the multi-objective optimization function, the genetic algorithm population is initialized, and each individual represents a work allocation scheme through an encoding method; For each individual, the overall connectivity index is calculated using the weighted undirected network model, and this overall connectivity index is used as the individual fitness value. The next generation of the population is generated by selecting individuals with high fitness values ​​based on these fitness values. The individuals in the next generation population are crossovered, and some gene sequences are randomly exchanged to generate new individuals; The new individual is subjected to a mutation operation, and the gene sequence is randomly adjusted to modify the expert set for work allocation; Repeat the steps of calculating the overall connectivity index, filtering, crossover operation, and mutation operation until the termination condition is met; The individual with the highest fitness value is selected from the final population, and its gene sequence is decoded to obtain the work allocation scheme.

[0007] Furthermore, the method for constructing a weighted undirected network model with experts as nodes includes: Each reviewer is treated as an independent node in the network; If any two reviewers jointly review at least one work, then an undirected edge is established between the independent nodes corresponding to the two reviewers. The weight of the undirected edge is defined as the number of works jointly reviewed by two review experts, where the number of jointly reviewed works is the size of the intersection of the two review experts' review sets.

[0008] Furthermore, the method for defining the multi-objective optimization function includes: The reciprocal of the weight of each edge in the network is defined as the length of the edge. By calculating the sum of the shortest path distances between all pairs of nodes in the network, an overall network connectivity index is constructed. The goal is to minimize the overall network connectivity index and maximize the sum of edge weights. By constraining the network connectivity and the uniform distribution of the shortest path distance, a uniform edge weight distribution is simultaneously achieved; The number of connections is maximized by increasing the number of works reviewed by different experts.

[0009] Furthermore, the method for calculating the overall connectivity index includes: The length of an edge is defined as the reciprocal of its weight. For two adjacent nodes, the distance between the two nodes is directly taken as the length of the edge. For two non-adjacent nodes, calculate their shortest path distance; If two nodes belong to different connected components, their distance is defined as infinity; The sum of the shortest path distances for all pairs of nodes in the network is used as the overall connectivity index.

[0010] Furthermore, for two non-adjacent nodes, a modified Dijkstra's algorithm is used to calculate their shortest path distance, specifically including: Search for a path connecting two nodes, where the path is formed by connecting several edges in sequence; Use the sum of the lengths of all edges in the path as the initial distance of the path; Based on the initial distance, the number of intermediate nodes along the path is added to obtain the final shortest path distance.

[0011] Furthermore, the method for selecting high-fitness individuals to generate the next generation population based on the fitness value includes: Sort the individuals in the population according to their fitness values ​​from highest to lowest; Using a roulette wheel selection method, with the fitness value as the weight of the selection probability, individuals with the same number of individuals as the population are selected from the current population. The selected individuals will be used as the next generation of the population.

[0012] Furthermore, the crossover operation is a single-point crossover, where a crossover point is randomly selected and partial gene sequences of two individuals are exchanged to generate a new individual.

[0013] Furthermore, the termination condition includes reaching a preset genetic generation number or fitness value without improvement for multiple consecutive generations.

[0014] To achieve the above objectives, a second aspect of the present invention provides an electronic device including a memory and a processor, the memory being used to store a program that supports the processor in executing the work allocation method based on a network model and a genetic algorithm, and the processor being configured to execute the program stored in the memory.

[0015] To achieve the above objectives, a third aspect of the present invention provides a computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, performs the steps of the work allocation method based on a network model and a genetic algorithm.

[0016] The beneficial effects of this invention are: Compared with existing technologies, this invention provides a method, device, and medium for allocating works based on network models and genetic algorithms. By integrating network modeling and genetic algorithms, it effectively solves the key problem caused by the uneven allocation of works among review experts in large-scale innovation competitions. Specifically, firstly, a weighted undirected network model with experts as nodes is constructed, using the number of works jointly reviewed by experts as edge weights to intuitively quantify the strength of the association between experts. By defining a multi-objective optimization function (maximizing the number of expert intersections, homogenizing the intersection distribution, and improving the overall network connectivity), the complex review allocation problem is transformed into a mathematical optimization objective. Subsequently, utilizing the global search capability of the genetic algorithm, each allocation scheme is encoded as an individual gene. Through population initialization, iterative selection of high-fitness individuals (based on network connectivity indicators), and crossover and mutation operations, the optimal solution space is dynamically explored.

[0017] During the optimization process, the modified Dijkstra algorithm ensures that the calculation of network connectivity indicators not only focuses on edge weights but also considers path length and the number of nodes, avoiding information silos and guaranteeing the comparability of scores among experts. The crossover operation breaks local optima by exchanging gene sequences, promoting the integration of different allocation schemes; the mutation operation fine-tunes the expert set, balancing the workload of individual experts and optimizing the distribution of co-review relationships. The final allocation scheme significantly reduces the standard deviation, ensuring that the scoring criteria tend to be consistent during cross-calibration and greatly mitigating the impact of subjective bias. Compared with traditional methods, this invention achieves the scientific allocation of review resources through structured network analysis and intelligent optimization, fundamentally improving the fairness, objectivity, and systematic nature of the review results. Attached Figure Description

[0018] 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.

[0019] Figure 1 This is a simple allocation scheme diagram disclosed in an embodiment of the present invention.

[0020] Figure 2 This is a flowchart of a work allocation method based on a network model and genetic algorithm disclosed in an embodiment of the present invention.

[0021] Figure 3 This is a weighted undirected network graph corresponding to a simple allocation scheme disclosed in an embodiment of the present invention.

[0022] Figure 4 This is a simple example diagram illustrating an optimization direction disclosed in an embodiment of the present invention.

[0023] Figure 5 This is an example diagram of an optimization to achieve the optimal structure disclosed in an embodiment of the present invention.

[0024] Figure 6 This is a size distribution diagram of the review portfolios of various experts disclosed in an embodiment of the present invention.

[0025] Figure 7 This is a box diagram showing the size distribution of the review portfolios for each expert, as disclosed in an embodiment of the present invention.

[0026] Figure 8 This is a box plot showing the intersection size distribution of various expert review portfolios disclosed in an embodiment of the present invention.

[0027] Figure 9 This is a scatter plot showing the distribution of the number of intersections between the portfolios of two experts in terms of specific relationships, as disclosed in an embodiment of the present invention.

[0028] Figure 10 This is a vertical histogram showing the distribution of the number of intersections between the portfolios of two experts in terms of specific relationships, as disclosed in an embodiment of the present invention. Detailed Implementation

[0029] 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.

[0030] 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.

[0031] Before describing the method of this invention, it should be noted that in the judging of large-scale innovation competitions, the traditional work allocation scheme usually adopts a simple fixed group model. The following example, using a scenario with 3,000 participating teams (corresponding to 3,000 works) and 125 judging experts, illustrates the specific implementation method and limitations of the existing technology: use This refers to the 3,000 innovative works submitted by 3,000 participating teams, with each work... Each corresponds to a set of experts under review, using This refers to the 3,000 different expert panel members.

[0032] use This indicates 125 expert reviewers, each expert Each corresponds to a portfolio for judging, using This refers to the 125 different collections of works submitted for judging.

[0033] Assuming each expert is assigned the same workload, the average number of works each expert must review, i.e., the average size of the review portfolio for all experts, is: (1) in, This represents the average number of works reviewed by each expert. Indicates the first Each expert's portfolio is reviewed by a specific group of works. Assuming the competition organizing committee assigns the same workload to each expert, then: (2) This means that each expert needs to review 120 different works.

[0034] The simplest allocation scheme is to use the first 5 experts The judging panel's portfolio consists of the first 120 works. (3) And so on, the next 5 experts The portfolio for the review was set as Until the last 5 experts The portfolio for the review was set as .

[0035] The expert allocation scheme according to this rule is as follows: Figure 1 As shown, the graph arranges all works on the horizontal axis, and the ellipses on the vertical axis represent a single task reviewed by a particular expert. Taking work 1 as an example, the five ellipses along the vertical axis correspond to the following experts: This constitutes the set of experts who reviewed Work 1: (4) At this point, the simplest allocation scheme minimizes the intersection of the portfolios reviewed by different experts. (Based on the experts...) Taking the portfolio reviewed by Expert 1 as an example... The portfolios reviewed by Expert 2, Expert 3, Expert 4, and Expert 5 are exactly the same, as detailed in Formula 1. However, the portfolio reviewed by Expert 1... The intersection of the portfolios reviewed by all other experts is an empty set, that is... (5) in, The symbol for an empty set indicates a work that was not jointly reviewed by two experts.

[0036] In summary, the review sets of experts 1-5 completely overlap (Formulas 3-4) but have no overlap with subsequent experts (Formula 5). This makes it difficult to eliminate subjective biases in expert scoring (such as optimistic / conservative tendencies) through score comparison and fails to construct a globally interconnected network structure. Only a few experts have co-review relationships, while most experts have no overlap; some edge weights (number of co-reviewed works) are extremely high (e.g., 120 between experts 1-5), while other edge weights are 0.

[0037] To address the shortcomings of the existing technology mentioned above, such as Figure 2 As shown, this invention provides a method for allocating works based on a network model and a genetic algorithm, comprising the following steps: Step S100: Construct a weighted undirected network model with experts as nodes; Step S200: Define a multi-objective optimization function based on the weighted undirected network model; Step S300: According to the multi-objective optimization function, initialize the genetic algorithm population, and each individual represents a work allocation scheme through an encoding method; Step S400: Calculate the overall connectivity index for the weighted undirected network model corresponding to each individual, and use the overall connectivity index as the individual fitness value; Step S500: Select individuals with high fitness based on the fitness values ​​to generate the next generation population; Step S600: Perform a crossover operation on the individuals in the next generation population, randomly exchanging some gene sequences to generate new individuals; Step S700: Perform a mutation operation on the new individual, randomly adjust the gene sequence to modify the expert set for work allocation; Step S800: Repeat the steps of calculating the overall connectivity index, filtering, crossover operation and mutation operation until the termination condition is met; Step S900: Select the individual with the highest fitness value from the final population and decode its gene sequence to obtain the work allocation scheme. In this embodiment, as described in step S100 above, each expert in the set of review experts is treated as an independent node in the network. Assume there are a total of... Experts (e.g.) Then the set of network nodes It can be represented as: (6) in, Indicates the first Experts Each node corresponds to a collection of works for review. This indicates all the works assigned for review by the expert. Go through all experts' views To judge the portfolio of the review and Is the intersection of these two sets non-empty? If Then at node and Establish an undirected edge between them This indicates that the two experts have a joint review relationship. The final edge set is generated. ,satisfy: (7) For each undirected edge Calculate its weight The size of the intersection of the portfolios reviewed by the two experts, i.e.: (8) in, For the image The middle without direction Weight on, It represents the number of elements in a set.

[0038] This weight directly reflects the number of works jointly reviewed by experts; the greater the intersection, the higher the edge weight. For example, if experts... and Five works were jointly reviewed, and then... The weight is 5.

[0039] Comprehensive Node Set Edge set and weight (Formulas 6, 7, and 8) Construct a weighted undirected graph ,in, This is the weight matrix. This weighted undirected graph corresponds to a certain work allocation scheme. Figure 1 The simplest allocation scheme in the graph is transformed into a weighted undirected graph, such as Figure 3 As shown.

[0040] analyze Figure 3 The networks in the text include: (9) (10) in, Representation diagram The number of nodes in the set; Representation diagram The number of edge sets; The network contains 125 connected components, each corresponding to a group of experts who review the same portfolios (e.g., ...). Figure 3 For example, an expert is only connected to an existing edge (with a weight of 120), and has no edge connections to the other 120 experts.

[0041] In this embodiment, as described in step S200 above, a multi-objective optimization function is defined based on the weighted undirected network model, and its specific implementation is as follows: First, the reciprocal of the weight of each edge in the network is defined as the edge length. The edge weight represents the number of works jointly reviewed by two experts (i.e., the intersection of the reviewed works sets); a larger weight indicates richer shared review information among the experts. By taking the reciprocal, the objective of maximizing the number of jointly reviewed works is transformed into minimizing the edge length. The specific formula is as follows: (11) in, For the image Mid-node experts and The length of the edge between them; weight The larger the number of works reviewed, the smaller the edge length. This design allows the optimization objective to indirectly maximize the edge weight by minimizing the total edge length. Secondly, an overall network connectivity metric is constructed. This metric is obtained by calculating the sum of the shortest path distances between all pairs of nodes in the network. For adjacent nodes, the distance is directly calculated using the edge length (the reciprocal of the weight); for non-adjacent nodes, a modified Dijkstra's algorithm is used to calculate the shortest path distance. Specifically, the formula for calculating the shortest path distance between non-adjacent nodes is: (12) in, Representing paths in a network graph The total length of the path consists of the lengths of all edges on the path and a penalty term for the number of nodes. , , Indicates an undirected edge; Representing a path The number of intermediate nodes traversed (i.e., the path hop count, which is equal to the number of edges); , , Represents an undirected edge The weighting of paths. This design encourages path simplicity and connectivity.

[0042] The modified Dijkstra's shortest distance algorithm is used to ensure network connectivity by optimizing the overall shortest paths, while increasing the weights while maintaining average edge weights. Therefore, the optimization model is established as follows: (13) in, and For the image Subgraph; Representation diagram The middle without direction The weight, is an expert in the path and the starting point of the path The edge weight between them represents the number of works jointly reviewed by both parties.

[0043] This indicator includes four constraints: Constraint ① indicates that the distance between two distinct nodes in the network must be calculated.

[0044] Constraint ② indicates that the distance between two adjacent nodes is the reciprocal of their edge weights.

[0045] Constraint ③ means that the distance between two non-adjacent nodes is the sum of the reciprocals of the weights of all edges on its Dijkstra's shortest path, plus the number of nodes passed through.

[0046] Constraint ④ indicates that if two nodes belong to two connected components in the graph, then the distance between them is positive infinity.

[0047] Using the optimization model described above, the optimal review task allocation scheme can be generated by optimizing the network structure.

[0048] Furthermore, by constraining the uniform distribution of shortest path distances, the weight distribution of each edge in the network is ensured to tend towards uniformity. This mechanism corresponds to the requirement of "uniform edge weight distribution" in the optimization objective, avoiding situations where some experts have excessive overlap while others have no overlap. For example, in Figure 4 In the simple example shown (4 experts reviewing 6 works), the initial allocation scheme is as follows ( Figure 4 This resulted in Expert 1 and Expert 2 having completely identical portfolios (intersection of 3), while their intersection with other experts was 0, indicating poor network connectivity and uneven edge weight distribution. After optimization ( Figure 5 Each expert has one jointly reviewed work (with the same intersection size), the network is fully connected and the edge weights are uniform, which greatly improves the comparability of scores.

[0049] Finally, the number of connections is maximized by increasing the number of works co-reviewed by different experts. This means ensuring that as many experts as possible have connections (co-reviewing at least one work), thereby expanding the information coverage. This goal, combined with network connectivity metrics, ensures that expert scoring data can be cross-calibrated through co-reviewed works, reducing the impact of subjective differences on fairness. For example, in the optimized... Figure 5 In this process, all experts are connected (to review one work together), and the network is fully connected, which fully satisfies the requirements of "maximizing the number of connections" and "maximizing information".

[0050] In summary, step S200 transforms the multi-objective optimization problem into a mathematically solvable model by defining an overall network connectivity index, and by combining graph theory and statistical methods, provides a clear optimization direction and evaluation criteria for subsequent genetic algorithms.

[0051] In this embodiment, as described in step S300 above, the initialization process is as follows: First, define the number of individuals in the population as... (For example, set to 200), each individual represents a review task allocation scheme, describing how 3000 works are distributed among 125 experts, ensuring that each work is reviewed by 5 experts. Each individual's genetic code is in matrix form (e.g., a 5×3000 matrix), where each column corresponds to a set of experts for a given work, and each column contains 5 expert IDs (integers ranging from 1 to 125). The range of gene values ​​is determined by combinatorial calculations, i.e., selecting 5 experts from the 125 experts for each work: (14) It can be seen that the range of values ​​for each gene in any individual is... The value can be an integer.

[0052] During initialization, it is necessary to ensure that the five experts for each work are unique and meet the actual review requirements. For example, the "simplest allocation scheme" mentioned in the appendix sets the review set of the first five experts to be the same (as shown in Formula 3), but the initial population should avoid such extreme cases and instead generate diverse individuals through random allocation. Specifically, the five expert numbers for each work are randomly sampled (sampling without replacement) to form an initial gene sequence, thereby constructing multiple individuals with different allocation schemes. This process must meet the following constraints: the number of works reviewed by each expert should be as balanced as possible (theoretically, the average is 120 works), and the review set of experts for all works should be complete and without missing parts.

[0053] In this embodiment, as described in step S400 above, the overall connectivity index of the weighted undirected network model corresponding to each individual (i.e., a work allocation scheme) is calculated and used as a fitness value to evaluate the merits of the allocation scheme. The specific process is as follows: According to formula (11), the length of each edge in the network is defined as the reciprocal of the number of works reviewed by two experts. This design transforms "maximizing the number of reviews" into "minimizing the edge length", thereby encouraging experts to share more works during optimization.

[0054] The distance between two nodes is directly calculated using the length of the edge (i.e., the reciprocal of the weight); a modified Dijkstra algorithm is used to calculate the shortest path. According to formula (12), the path distance is the sum of the lengths of all edges on the path plus the number of intermediate nodes along the way; this mechanism encourages path simplicity and avoids redundant connections by adding a penalty term for the number of intermediate nodes. The distance is set to infinity (Formula 13), which forces the entire network to be connected.

[0055] The sum of the shortest path distances between all pairs of nodes in the network is used as the overall connectivity index. A smaller index indicates better network connectivity, a larger number of collaboratively reviewed works, and a more even distribution. By minimizing this index, the algorithm simultaneously achieves the dual objectives of "maximizing the sum of edge weights" and "uniforming the distribution of edge weights." The overall connectivity index is then used as the fitness value of an individual. Since genetic algorithms typically optimize by maximizing fitness, the index needs to be converted by taking its reciprocal or a negative value.

[0056] In this embodiment, as described in step S500 above, firstly, all individuals in the current population are sorted from high to low according to their fitness values. Individuals with higher fitness values ​​(corresponding to lower network connectivity indices) represent those with better allocation schemes. Next, a Roulette Wheel Selection method is used for selection, where the fitness value of each individual is used as the weight of the selection probability; individuals with higher fitness have a greater probability of being selected. For example, if the fitness value of individual A is twice that of individual B, then the probability of individual A being selected is also twice that of individual B. Through this method, individuals of the same number (e.g., 200) as the original population are selected from the current population to form the next generation. This mechanism preserves the genetic information of high-quality individuals while maintaining population diversity through probabilistic selection, thus preventing the algorithm from getting trapped in local optima.

[0057] In this embodiment, as described in step S600 above, a new individual is generated through a crossover operation. Specifically, a single-point crossover mechanism is adopted, where a random position in the gene sequence is selected as the crossover point, and partial gene sequences of two parent individuals are exchanged. The crossover probability is set to 80% to ensure that most individuals participate in gene mixing, thereby introducing diversity and exploring a better solution space.

[0058] In this embodiment, as described in step S700 above, a mutation operation is performed on the newly generated individuals, randomly adjusting a portion of the expert IDs in the gene sequence at a mutation rate of 1%. For example, a set of experts who reviewed a work is randomly selected, and one expert is replaced with the ID of another expert who did not participate. This mechanism breaks population convergence through local random perturbation, preventing the algorithm from getting trapped in local optima. In some embodiments, setting the mutation rate too low can cause the algorithm to easily get trapped in local optima; setting the mutation rate too high can cause the algorithm to optimize too slowly. The newly generated individuals replace some individuals in the original population to form the next generation population. Typically, individuals with lower fitness are replaced.

[0059] In this embodiment, as described in step S800 above, fitness calculation, screening, crossover, and mutation operations are repeatedly performed until termination conditions are met. Termination conditions include reaching a preset number of generations (e.g., 100 generations) or the fitness value not significantly improving for multiple consecutive generations. Through repeated optimization, the population gradually eliminates individuals with low fitness, retains and enhances high-quality genes, and ultimately approaches the global optimum.

[0060] In this embodiment, as described in step S900 above, the individual with the highest fitness value is selected from the final population, and its gene sequence is decoded into a specific work allocation scheme. For example, each column of the 5×3000 matrix is ​​parsed into the IDs of the 5 review experts for a work, generating a complete allocation table. This scheme ensures that each work is reviewed by 5 experts, and that the co-review relationship among experts is maximized and evenly distributed. The final result is output in an Excel file (e.g., "Genetic Algorithm Results - 3000×5 Matrix.xlsx"), which is directly applied to the actual review process, taking into account fairness, scientific rigor, and operability.

[0061] To further illustrate the method of the present invention, the following results analysis will be performed: The population size was set to 200, the number of generations was set to 100, and parameters of crossover probability of 0.8 and mutation probability of 0.01 were used. Experts were used as nodes, and the intersection size of the works reviewed by different experts was used as the basis for connecting edges. The optimal allocation scheme was optimized by using the optimization model of formula (12) and the fitness function of formula (14).

[0062] A personal computer equipped with an Intel Core i9 processor (model: i9-12900HX, clock speed: 2.3GHz, 16 cores) was used as the hardware platform for running the genetic algorithm. Windows 11 operating system and Matlab programming language version 2022b were used to implement and run the genetic algorithm.

[0063] The final optimal review scheme can be represented by a 5*3000 matrix, where each column of the matrix contains 5 elements representing the 5 expert sets for a single work. All review schemes can be found in "Genetic Algorithm Results - Expert Sets of Reviewed Works.xlsx" and "Genetic Algorithm Results - 3000×5 Matrix.xlsx". The allocation scheme for the first 30 works is shown here, as shown in Table 1.

[0064] Table 1. Optimal allocation schemes obtained by genetic algorithm

[0065] To improve the efficiency of the genetic algorithm in finding the optimal solution and to relax the conditions to better reflect real-world scenarios, this embodiment does not limit the number of expert review submissions when running the genetic algorithm. The distribution of the number of expert review submissions after the final solution is as follows: Figure 6 , Figure 7 As shown.

[0066] Depend on Figure 6 , Figure 7 It can be observed that the number of expert reviews for each item in the optimal allocation scheme of the genetic algorithm is very close to the theoretical average of 120. The calculated mean number of reviews for each item is 120, with a standard deviation of 9.4528.

[0067] The following analysis examines the number of works reviewed by each expert.

[0068] Figure 8 The box plot shows the intersection size of the various expert review sets. As can be seen from the box plot, the intersection of most expert review sets is 4.

[0069] Figure 9 The length of the horizontal axis and Figure 10 The vertical axis length is 7750, because randomly selecting 2 experts from 125 experts results in... Seed selection method.

[0070] Figure 9 and Figure 10 This also shows that the intersection of most expert review submissions is 4. The dense color distribution in these two graphs clearly shows that the intersection between most experts is less than or equal to 4.

[0071] In summary, this invention first uses network modeling to represent the expert review process, with experts as nodes, collaborative review relationships between experts as edges, and the number of works reviewed as edge weights. Second, a modified Dijkstra's algorithm is employed, which ensures the shortest distance metric of the network, thus guaranteeing network connectivity. Then, while ensuring a uniform weight distribution within the network, the weight values ​​are maximized to meet the optimization objective. Finally, a genetic algorithm is used to optimize the shortest distance metric of the network, resulting in the optimal work allocation scheme.

[0072] 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.

[0073] 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.

[0074] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0075] 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.

[0076] 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.

[0077] 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 allocating works based on a network model and a genetic algorithm, characterized in that, The steps include: Construct a weighted undirected network model with experts as nodes; A multi-objective optimization function is defined based on the weighted undirected network model. Based on the multi-objective optimization function, the genetic algorithm population is initialized, and each individual represents a work allocation scheme through an encoding method; For each individual, the overall connectivity index is calculated using the weighted undirected network model, and this overall connectivity index is used as the individual fitness value. The next generation of the population is generated by selecting individuals with high fitness values ​​based on these fitness values. The individuals in the next generation population are crossovered, and some gene sequences are randomly exchanged to generate new individuals; The new individual is subjected to a mutation operation, and the gene sequence is randomly adjusted to modify the expert set for work allocation; Repeat the steps of calculating the overall connectivity index, filtering, crossover operation, and mutation operation until the termination condition is met; The individual with the highest fitness value is selected from the final population, and its gene sequence is decoded to obtain the work allocation scheme.

2. The work allocation method based on network model and genetic algorithm as described in claim 1, characterized in that, The method for constructing a weighted undirected network model with experts as nodes includes: Each reviewer is treated as an independent node in the network; If any two reviewers jointly review at least one work, then an undirected edge is established between the independent nodes corresponding to the two reviewers. The weight of the undirected edge is defined as the number of works jointly reviewed by two review experts, where the number of jointly reviewed works is the size of the intersection of the two review experts' review sets.

3. The work allocation method based on network model and genetic algorithm as described in claim 1, characterized in that, The method for defining the multi-objective optimization function includes: The reciprocal of the weight of each edge in the network is defined as the length of the edge. By calculating the sum of the shortest path distances between all pairs of nodes in the network, an overall network connectivity index is constructed. The goal is to minimize the overall network connectivity index and maximize the sum of edge weights. By constraining the network connectivity and the uniform distribution of the shortest path distance, a uniform edge weight distribution is simultaneously achieved; The number of connections is maximized by increasing the number of works reviewed by different experts.

4. The work allocation method based on network model and genetic algorithm as described in claim 1, characterized in that, The calculation method for the overall connectivity index includes: The length of an edge is defined as the reciprocal of its weight. For two adjacent nodes, the distance between the two nodes is directly taken as the length of the edge. For two non-adjacent nodes, calculate their shortest path distance; If two nodes belong to different connected components, their distance is defined as infinity; The sum of the shortest path distances for all pairs of nodes in the network is used as the overall connectivity index.

5. The work allocation method based on network model and genetic algorithm as described in claim 4, characterized in that, For two non-adjacent nodes, the modified Dijkstra algorithm is used to calculate their shortest path distance, specifically including: Search for a path connecting two nodes, where the path is formed by connecting several edges in sequence; Use the sum of the lengths of all edges in the path as the initial distance of the path; Based on the initial distance, the number of intermediate nodes along the path is added to obtain the final shortest path distance.

6. The work allocation method based on network model and genetic algorithm as described in claim 1, characterized in that, The method for selecting high-fitness individuals to generate the next generation population based on the fitness value includes: Sort the individuals in the population according to their fitness values ​​from highest to lowest; Using a roulette wheel selection method, with the fitness value as the weight of the selection probability, individuals with the same number of individuals as the population are selected from the current population. The selected individuals will be used as the next generation of the population.

7. The work allocation method based on network model and genetic algorithm as described in claim 1, characterized in that, The crossover operation is a single-point crossover, in which a crossover point is randomly selected and partial gene sequences of two individuals are exchanged to generate a new individual.

8. The work allocation method based on network model and genetic algorithm as described in claim 1, characterized in that, The termination conditions include reaching a preset genetic generation number or fitness value without improvement for multiple consecutive generations.

9. An electronic device, comprising a memory and a processor, characterized in that, The memory is used to store programs that support the processor in executing the work allocation method based on any one of claims 1-8, and the processor is configured to execute the programs stored in the memory.

10. A computer-readable storage medium storing a computer program thereon, characterized in that, The computer program, when run by a processor, executes the steps of the work allocation method based on network model and genetic algorithm as described in any one of claims 1-8.

Images