A Big Data-Based Method and System for Processing Personnel Information in Business Administration

Abstract

This invention discloses a method and system for processing personnel information for business administration based on big data, relating to the field of personnel information data processing technology. The method includes the following steps: acquiring personnel information; preprocessing the personnel information to obtain preprocessed text information; processing the preprocessed text information through an embedding layer based on a CNN-BERT model to obtain a word vector matrix; processing the word vector matrix sequentially through a CNN convolutional layer, pooling layer, and fully connected layer to obtain structured data; using an improved ant colony optimization algorithm to perform path optimization for multi-condition combined queries on the structured data to obtain the globally optimal retrieval path; retrieving personnel information that satisfies the multi-condition combined query from the structured data based on the globally optimal retrieval path; and adaptively updating the global pheromone of the improved ant colony optimization algorithm after each round of multi-ant colony collaborative search to accurately locate the target personnel information.

Description

Technical Field

[0001] This invention relates to the field of personnel information data processing technology, and in particular to a method and system for processing personnel information for business administration based on big data. Background Technology

[0002] In recent years, the combination of big data and natural language processing technology has made it possible to deeply mine personnel information, especially by using word vector representation and deep learning models to automatically extract semantic features from text, thereby realizing the structured representation of personnel attributes. However, in the face of complex and ever-changing management scenarios, how to accurately and efficiently extract key information from multi-dimensional text and support flexible and fast multi-condition combination queries remains the main challenge that current technologies face in practical applications.

[0003] Existing methods for text feature extraction often employ single word embedding models or convolutional neural networks, which struggle to simultaneously capture deep semantic relationships and key local features of the text. This results in incomplete and inaccurate structured representations. Furthermore, when implementing multi-condition combined retrieval, traditional query optimization methods often fall into local optima or slow convergence, making it difficult to quickly plan a globally efficient retrieval path in a large database of personnel information. This is especially true when query conditions are complex and data dimensionality is high, leading to a significant decrease in retrieval efficiency. Summary of the Invention

[0004] The technical problem solved by this invention is that existing methods often use a single word embedding model or convolutional neural network when extracting text features, which makes it difficult to capture the deep semantic relationships and local key features of the text at the same time. This results in an insufficiently comprehensive and accurate structured representation. In addition, when implementing multi-condition combined retrieval, traditional query optimization methods often get stuck in local optima or have slow convergence speed, making it difficult to quickly plan a globally efficient retrieval path in a large database of personnel information. Especially when the query conditions are complex and the data dimension is high, the retrieval efficiency drops significantly.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a method for processing personnel information for business administration based on big data, comprising the following steps: Step S1: Obtain personnel information, preprocess the personnel information, and obtain preprocessed text information; Step S2: Based on the CNN-BERT model, the preprocessed text information is processed through the embedding layer to obtain the word vector matrix. The word vector matrix is ​​then processed sequentially through the CNN convolutional layer, pooling layer, and fully connected layer to obtain structured data. Step S3: The improved ant colony optimization algorithm is used to optimize the path of multi-condition combined query for structured data and obtain the globally optimal retrieval path. Based on the globally optimal retrieval path, retrieve personnel information from structured data that meets multiple condition combinations for querying; Step S4: After each round of multi-ant colony collaborative search in the improved ant colony optimization algorithm, the global pheromone of the improved ant colony optimization algorithm is adaptively updated, and the target personnel information set that satisfies the multi-condition combination query is output based on the globally optimal retrieval path.

[0006] As a preferred embodiment of the personnel information processing method for business administration based on big data described in this invention, step S1 involves obtaining personnel information, preprocessing the personnel, and obtaining preprocessed text information. The personnel information includes personnel files, management records, communication texts, and external data; The personnel files include resumes, onboarding forms, and background investigation reports; Management records include performance evaluation reports, multi-source performance feedback, promotion review opinions, and reward and punishment records; The communication texts include work logs, project summaries, performance reports, internal forum posts, and email communication summaries; External data includes job descriptions posted on external recruitment platforms, industry talent assessment reports, and publicly available professional qualification certifications and practice certificates. The personnel information is processed sequentially through text cleaning, word segmentation, and stop word removal to obtain preprocessed text information.

[0007] As a preferred embodiment of the personnel information processing method for business administration based on big data described in this invention, in step S2, based on the CNN-BERT model, the preprocessed text information is processed through the embedding layer to obtain a word vector matrix, and the word vector matrix is ​​processed sequentially through the CNN convolutional layer, pooling layer and fully connected layer to obtain structured data; The structured data includes personnel entity attribute classification labels and their corresponding feature weights; Step S2 includes steps S201 and S202; Step S201: Input the word vector matrix into the CNN-BERT model, and then... convolution kernel, convolution kernel and The convolutional kernels perform sliding convolution operations. After each convolutional kernel traverses the entire word vector matrix, it generates a feature map, expressed as: ; ; in, This represents the feature extracted at position i. This represents a local window from row i to row i+h-1 of the word vector matrix, where h represents the height of the convolution kernel. Indicates weight, Indicates the bias value. Represents the ReLU nonlinear activation function; This means that the convolutional kernel sequentially filters the word vector matrix to form n-h+1 new features. This represents the feature map of n.

[0008] As a preferred embodiment of the personnel information processing method for business administration based on big data described in this invention, in step S202, the feature map is processed and concatenated through a pooling layer to obtain a q feature map, and the q feature map is concatenated through a fully connected layer to obtain an i feature map; The i-feature map is calculated using the Softmax function to obtain the probability distribution and the feature weights of the corresponding classification labels; By summarizing personnel information, classification labels, and corresponding feature weights, structured data is obtained.

[0009] As a preferred embodiment of the personnel information processing method for business administration based on big data described in this invention, step S3 involves using an improved ant colony optimization algorithm to optimize the path of multi-condition combined query of structured data and obtain the globally optimal retrieval path. Based on the globally optimal retrieval path, retrieve personnel information that satisfies multiple condition combination queries from the structured data; The improved ant colony optimization algorithm employs multi-ant colony collaborative search, including at least two independent sub-ant colonies performing path search in parallel, with each sub-ant colony sharing the same global pheromone matrix. Each query condition in the multi-condition combined query is represented as a node to be visited in the improved ant colony optimization algorithm.

[0010] As a preferred embodiment of the personnel information processing method for business administration based on big data described in this invention, in step S301, the initial pheromone matrix distribution is generated using a genetic algorithm, multi-ant colony collaborative path retrieval is performed, and the query starting condition node is marked as visited and set as the current node. Step S302: When there are unvisited query condition nodes, select the next node to be searched according to the transition probability formula and construct a complete search path. The transition probability formula is: ; in, Let represent the probability that an ant moves from node i to node j. This represents the pheromone concentration along path ij at time t. This represents the heuristic information from node i to node j, where α is the pheromone weighting factor and β is the heuristic information weighting factor. This represents the set of nodes that the ant is allowed to visit next; Step S303: Repeat step S302 until all query condition nodes have been accessed, thus constructing a complete search path. Step S304: Calculate the cost of the retrieval path and update the optimal path for this iteration based on the cost; Step S305: When the number of iterations reaches the preset smoothing threshold, perform pheromone smoothing operation; Step S306: Determine whether the improved ant colony optimization algorithm meets the termination condition; If satisfied, output the globally optimal search path; If the condition is not met, return to step S302 for the next iteration; The termination condition is determined to be met if the change in cost of the global optimal retrieval path in K consecutive iterations is less than or equal to a preset convergence threshold.

[0011] As a preferred embodiment of the personnel information processing method for business administration based on big data described in this invention, step S301 includes steps S3011, S3012, S3013 and S3014. Step S3011: Randomly generate a set of retrieval paths for the query condition nodes as the initial population; Step S3012: Decode each retrieval path and calculate the execution cost of each retrieval path as the fitness value using the objective function; Step S3013: Based on the fitness value, a new generation of population is generated through selection, crossover, and mutation operations; Step S3014: Repeat steps S3012 to S3013 until the preset genetic algorithm termination condition is met, and map the obtained global optimal retrieval path to the initial pheromone matrix. The termination condition is determined to be met when the change in the cost of the globally optimal retrieval path, represented by the fitness value, is less than or equal to the preset convergence threshold of the genetic algorithm in consecutive generations, and the genetic algorithm iteration is stopped.

[0012] As a preferred embodiment of the personnel information processing method for business administration based on big data described in this invention, in step S4, after each round of multi-ant colony collaborative search of the improved ant colony optimization algorithm is completed, the global pheromone of the improved ant colony optimization algorithm is adaptively updated. The expression for adaptive global pheromone update is: ; in, This represents the pheromone concentration on path ij after the update. This represents the adaptive pheromone evaporation factor. This represents the total amount of pheromone released by the top a% of ants with the best fitness values ​​on path ij in this iteration. This represents the pheromone concentration on the path from i to j at time t.

[0013] As a preferred embodiment of the personnel information processing method for business administration based on big data described in this invention, the adaptive pheromone evaporation factor ρ increases with the number of iterations, and its expression is: ; in, Indicates the current iteration number. Indicates the total number of iterations; The percentage of ants releasing pheromones, denoted as 'a%', decreases dynamically with the iteration process, expressed as: ; When the number of iterations reaches a preset smoothing threshold, a pheromone smoothing operation is performed. The pheromone smoothing expression is: ; Where δ represents the smoothing coefficient. This represents the preset minimum pheromone concentration. When the number of iterations reaches a preset learning threshold, a learning operator operation is executed among the sub-ant colonies. The expression for the learning operator is: ; in, and represents the pheromone concentrations of sub-ant colonies m and n on path ij, respectively, and γ is the learning operator.

[0014] A personnel information processing system for business administration based on big data, including a preprocessing module, a processing module, an acquisition module, and an update module; The preprocessing module acquires personnel information, preprocesses the personnel information, and obtains preprocessed text information. The processing module, based on the CNN-BERT model, processes the preprocessed text information through the embedding layer to obtain a word vector matrix. The word vector matrix is ​​then processed sequentially through the CNN convolutional layer, pooling layer, and fully connected layer to obtain structured data. The acquisition module uses an improved ant colony optimization algorithm to optimize the path of multi-condition combined queries on structured data and obtain the globally optimal retrieval path; Based on the globally optimal retrieval path, retrieve personnel information from structured data that meets multiple condition combinations for querying; The update module performs adaptive updates to the global pheromone of the improved ant colony optimization algorithm after each round of multi-ant colony collaborative search, and outputs a set of target personnel information that satisfies multiple condition combination queries based on the globally optimal retrieval path.

[0015] The beneficial effects of this invention are as follows: Through the deep synergistic innovation of the CNN-BERT hybrid model and the improved ant colony optimization algorithm, a breakthrough in both deep understanding and ultra-fast retrieval has been achieved simultaneously for the first time in processing massive, multi-source personnel information in business administration. On the one hand, the CNN-BERT model deeply integrates BERT's deep semantic understanding capabilities with CNN's efficient local feature extraction capabilities, enabling it to accurately extract key attributes and latent features of personnel from unstructured text, constructing a structured knowledge base with rich dimensions and accurate representation. This solves the fundamental problems of one-sided feature extraction and superficial semantic understanding in traditional methods, laying a solid foundation for high-quality queries. On the other hand, the introduction of multi-ant colony collaborative search and genetic optimization algorithms... Algorithm initialization and adaptive pheromone update mechanisms together constitute a powerful dynamic path optimization engine. This engine can intelligently plan and quickly converge to the globally optimal retrieval path for complex multi-condition combined queries. Compared with traditional serial queries or basic optimization algorithms, the planning efficiency and execution speed of retrieval paths are improved by orders of magnitude in multi-condition, high-concurrency query scenarios. At the same time, it effectively avoids getting trapped in local optima. It seamlessly connects deep text mining with intelligent query optimization, making the process of accurately locating target personnel information from massive heterogeneous data accurate, fast and resource-efficient. It provides a powerful intelligent tool for real-time decision-making, talent inventory and organizational analysis in business management. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the steps of a big data-based personnel information processing method for business administration, provided as an embodiment of the present invention.

[0017] Figure 2 This is a basic flowchart of a personnel information processing system for business administration based on big data, provided as an embodiment of the present invention. Detailed Implementation

[0018] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0019] Example 1, referring to Figure 1As an embodiment of the present invention, a method for processing personnel information for business administration based on big data is provided, comprising the following steps: Step S1: Obtain personnel information, preprocess the personnel information, and obtain preprocessed text information; Step S2: Based on the CNN-BERT model, the preprocessed text information is processed through the embedding layer to obtain the word vector matrix. The word vector matrix is ​​then processed sequentially through the CNN convolutional layer, pooling layer, and fully connected layer to obtain structured data. Step S3: The improved ant colony optimization algorithm is used to optimize the path of multi-condition combined query for structured data and obtain the globally optimal retrieval path. Based on the globally optimal retrieval path, retrieve personnel information from structured data that meets multiple condition combinations for querying; Step S4: After each round of multi-ant colony collaborative search in the improved ant colony optimization algorithm, the global pheromone of the improved ant colony optimization algorithm is adaptively updated, and the target personnel information set that satisfies the multi-condition combination query is output based on the globally optimal retrieval path.

[0020] In one embodiment, the acquired personnel information is preprocessed by cleaning and word segmentation to form high-quality standardized text, laying the data foundation for subsequent in-depth analysis. Based on the CNN-BERT model, the preprocessed text information is transformed into a word vector matrix with rich semantics through the BERT embedding layer. This matrix is ​​then input into the CNN module, where local features of different granularities are extracted through multi-scale convolutional kernels. After pooling and full-connected layer fusion and dimensionality reduction, high-precision, weighted structured personnel data is finally output. This process improves the ability to accurately capture key attributes and latent features from complex text information. On this basis, when faced with multi-condition combined queries, an improved ant colony optimization algorithm is used for intelligent path planning. The improved ant colony optimization algorithm quickly finds the globally optimal retrieval path in the solution space composed of query conditions through multi-ant colony collaborative search and adaptive pheromone update mechanism, thereby accessing structured data in the most efficient order. This significantly improves the response speed and overall throughput of complex queries and effectively avoids the problem of traditional methods easily getting trapped in local optima. All links in the entire process are closely connected and optimized collaboratively, ultimately achieving a deep understanding and millisecond-level accurate retrieval of massive amounts of business management personnel information.

[0021] Steps S2, S3, and S4 form an organic whole, jointly achieving dual optimization of data processing and efficient retrieval. Step S2 uses the CNN-BERT hybrid model to transform multi-source heterogeneous text into high-quality structured data, which is the cornerstone of subsequent intelligent retrieval. Steps S3 and S4 dynamically optimize the retrieval path of this structured data in multi-condition query scenarios, ensuring that it can still maintain efficient response under complex query conditions. All three are indispensable and together constitute a complete integrated solution for data processing and retrieval.

[0022] Step S1: Obtain personnel information, preprocess the personnel, and obtain preprocessed text information; Personnel information includes personnel files, management records, communication documents, and external data; Personnel files include resumes, onboarding forms, and background check reports; Management records include performance evaluation reports, multi-source performance feedback, promotion review opinions, and reward and punishment records; Communication documents include work logs, project summaries, performance reports, internal forum posts, and email communication summaries; External data includes job descriptions posted on external recruitment platforms, industry talent assessment reports, and publicly available professional qualification certifications and practice certificates. The personnel information is processed sequentially through text cleaning, word segmentation, and stop word removal to obtain preprocessed text information.

[0023] In one embodiment, in step S1, personnel information is acquired, specifically including but not limited to personnel files, management records, communication texts, and external data. Personnel files include resumes, onboarding forms, and background check reports; management records include performance evaluation reports, multi-source performance feedback, promotion review opinions, and reward / punishment records; communication texts include work logs, project summaries, job performance reports, internal forum posts, and email communication summaries; and external data includes job descriptions from recruitment platforms, industry talent assessment reports, and publicly available professional qualification information. Because personnel information is often formatted in a mixed manner and contains noise, it undergoes standardization preprocessing to provide high-quality input for subsequent in-depth analysis. The preprocessing process includes text cleaning, word segmentation, and stop word removal. First, text cleaning uses a regular expression-based method to remove non-textual noise such as URLs, HTML tags, and garbled characters, and unifies the formats of characters, dates, and numbers. Next, word segmentation is performed. Considering the professional nature of business administration texts, the Jieba word segmentation tool is used for precise word segmentation to ensure that professional entities are accurately segmented. The output is standardized preprocessed text information in the form of word sequences. This series of processes effectively removes noise, unifies the format, and highlights key semantic units, laying a reliable data foundation for the subsequent CNN-BERT model to accurately extract structured features.

[0024] Step S2: Based on the CNN-BERT model, the preprocessed text information is processed through the embedding layer to obtain the word vector matrix. The word vector matrix is ​​then processed sequentially through the CNN convolutional layer, pooling layer, and fully connected layer to obtain structured data. Structured data includes personnel entity attribute classification labels and their corresponding feature weights; Step S2 includes steps S201 and S202; Step S201: Input the word vector matrix into the CNN-BERT model, and then... convolution kernel, convolution kernel and The convolutional kernels perform sliding convolution operations. After each convolutional kernel traverses the entire word vector matrix, it generates a feature map, expressed as: ; ; in, This represents the feature extracted at position i. This represents a local window from row i to row i+h-1 of the word vector matrix, where h represents the height of the convolution kernel. Indicates weight, Indicates the bias value. Represents the ReLU nonlinear activation function; This means that the convolutional kernel sequentially filters the word vector matrix to form n-h+1 new features. This represents the feature map of n.

[0025] Step S202: The feature maps are processed and concatenated through pooling layers to obtain the q feature map, and the q feature map is concatenated through fully connected layers to obtain the i feature map; The i-feature map is calculated using the Softmax function to obtain the probability distribution and the feature weights of the corresponding classification labels; By summarizing personnel information, classification labels, and corresponding feature weights, structured data is obtained.

[0026] In one embodiment, the word sequence is input into a CNN-BERT model for processing. First, the BERT embedding layer of the model converts each word into a high-dimensional semantic vector, thus representing the entire text sequence as a word vector matrix containing contextual information. This word vector matrix is ​​input in parallel into three CNN convolutional layers of different scales, using convolutional kernels with heights of 2×D, 3×D, and 4×D, respectively, where D is a fixed dimension of the word vectors, such as 768 dimensions. Sliding convolution operations are performed to adaptively capture key local feature patterns composed of two-word combinations, three-word combinations, and four-word combinations in the text. After each convolutional kernel traverses the matrix, it generates a corresponding feature map. Max pooling is performed on the feature maps output by each convolutional layer to reduce dimensionality and retain the most salient features. Subsequently, the pooling... The multi-scale feature maps are then concatenated to form a comprehensive feature vector. This fused feature vector undergoes further nonlinear transformation and integration through a fully connected layer. Finally, a Softmax classifier calculates the probability distribution of its belonging to a predefined set of personnel entity attribute classification labels. These classification labels include, but are not limited to, professional skills, management experience, compliance risks, educational background, professional qualifications, project experience, and communication skills. This probability value is used as the feature weight corresponding to the attribute. Through this process, unstructured text is transformed into a set of structured data consisting of classification labels and feature weight pairs. This not only achieves accurate quantification of multi-dimensional attributes of personnel but also provides a data foundation for subsequent intelligent retrieval based on multi-condition combinations that can be directly and efficiently calculated and matched.

[0027] Step S3: The improved ant colony optimization algorithm is used to optimize the path of multi-condition combined query for structured data and obtain the globally optimal retrieval path. Based on the globally optimal retrieval path, retrieve personnel information from structured data that meets multiple condition combinations for querying; Among them, the improved ant colony optimization algorithm adopts multi-ant colony collaborative search, including at least two independent sub-ant colonies performing path search in parallel, and each sub-ant colony sharing the same global pheromone matrix; In a multi-condition combined query, each query condition is represented as a node to be visited in the improved ant colony optimization algorithm.

[0028] Step S301: Use a genetic algorithm to generate the distribution of the initial pheromone matrix, perform multi-ant colony collaborative path retrieval, mark the query starting condition node as visited and set it as the current node; Step S302: When there are unvisited query condition nodes, select the next node to be searched according to the transition probability formula and construct a complete search path. The transition probability formula is: ; in, Let represent the probability that an ant moves from node i to node j. This represents the pheromone concentration along path ij at time t. This represents the heuristic information from node i to node j, where α is the pheromone weighting factor and β is the heuristic information weighting factor. This represents the set of nodes that the ant is allowed to visit next; Step S303: Repeat step S302 until all query condition nodes have been accessed, thus constructing a complete search path. Step S304: Calculate the cost of the retrieval path and update the optimal path for this iteration based on the cost; Step S305: When the number of iterations reaches the preset smoothing threshold, perform pheromone smoothing operation; Step S306: Determine whether the improved ant colony optimization algorithm meets the termination condition; If satisfied, output the globally optimal search path; If the condition is not met, return to step S302 for the next iteration; The termination condition is determined to be met if the change in cost of the global optimal retrieval path in K consecutive iterations is less than or equal to the preset convergence threshold.

[0029] Step S301 includes steps S3011, S3012, S3013 and S3014; Step S3011: Randomly generate a set of retrieval paths for the query condition nodes as the initial population; Step S3012: Decode each retrieval path and calculate the execution cost of each retrieval path as the fitness value using the objective function; Step S3013: Based on the fitness value, a new generation of population is generated through selection, crossover, and mutation operations; Step S3014: Repeat steps S3012 to S3013 until the preset genetic algorithm termination condition is met, and map the obtained global optimal retrieval path to the initial pheromone matrix. The termination condition is that when the change in the cost of the globally optimal retrieval path, represented by the fitness value, is less than or equal to the preset convergence threshold of the genetic algorithm over G consecutive generations, the genetic algorithm iteration is stopped.

[0030] In one embodiment, in order to efficiently execute multi-condition combined queries, this solution uses an improved ant colony optimization algorithm to optimize the retrieval path of the structured data generated in step S2. Each retrieval condition, for example, requiring a person to simultaneously meet the professional skills tag weight greater than 0.8 and the management experience tag weight greater than 0.7, is represented as a node to be visited. The goal of the improved ant colony optimization algorithm is to find the optimal order of visiting all these nodes, i.e., the globally optimal retrieval path, in order to minimize the overall retrieval cost. In step S3, the path optimization of multi-condition combined query is not a traditional database index optimization, but a dynamic query order optimization in semantic and weight space for the structured data with feature weights output in step S2. This optimization process relies on the structured feature representation output in step S2, especially the weight information of each attribute, to calculate the filtering cost and transition heuristic information between query conditions. Therefore, step S3 and step S2 are closely coupled in technical logic and together constitute intelligent query planning based on feature weights.

[0031] Furthermore, the adaptive pheromone update in step S4 is crucial for the improved ant colony optimization algorithm in step S3 to quickly converge to the global optimal path. By dynamically adjusting pheromone evaporation, the proportion of elite ants, and cross-population learning strategies, the algorithm ensures a balance between exploration and utilization, avoiding getting trapped in local optima. Step S4 is a core component of the performance optimization of the algorithm in step S3, and the two together achieve efficient solutions for complex query paths.

[0032] The cost of a retrieval path is defined as the estimated total computational cost when performing multi-condition filtering in the order of that path. The core of its calculation formula is to simulate the data volume decay during the sequential query process. Specifically, it is the sum of the sizes of the data subsets to be processed when executing the k-th condition node, where the initial data subset size is the total number of personnel records N. Each time a query condition is applied in the order of the path, that is, after a node is accessed for filtering, the size of the data subset will be reduced according to the filtering rate of that condition and the estimated proportion of personnel records that meet that condition. Therefore, the optimization objective is to find a node access order that minimizes this estimated total amount of data to be processed. This usually means that conditions with strong filtering power, that is, high filtering rate, and can significantly reduce the result set should be executed first. Algorithm initialization and heuristic information: To overcome the low search efficiency caused by insufficient pheromone in the early stages of traditional ant colony algorithms, this step first uses a genetic algorithm to optimize the distribution of the initial pheromone matrix. Specifically, a set of possible search paths for query condition nodes is randomly generated as the initial population. The execution cost of each path is calculated using the aforementioned cost function as the fitness value. Through selection, crossover, and mutation operations, the algorithm iterates and evolves. When the change in the cost of the optimal path over G generations is lower than a preset threshold (G being a preset positive integer convergence window, for example, 10), the threshold is set to a value related to the total data size, for example, 0.001*N, where N is the total number of records. The basis for this setting is that when the improvement in the cost of the optimal path is less than one-thousandth of the total data size, the solution quality is considered to have entered a stable plateau, and the expected benefit of continued iteration is minimal. Therefore, the genetic algorithm iteration is terminated, and this approximately optimal path is mapped to the initial pheromone concentration of the ant colony algorithm. In subsequent ant colony algorithms, heuristic information... The metric defined as the expected gain from the conditional transition at node i to the conditional transition at node j is specifically set as follows: The value is the reciprocal of the estimated screening cost of condition j, which intuitively means prioritizing the condition node with the lower execution cost. The multi-colony collaborative search process begins after initialization. Multiple independent sub-colony groups, such as 2-3 groups, construct search paths in parallel based on a shared global pheromone matrix. However, each sub-colony employs differentiated search strategy parameters. For example, one sub-colony's parameter settings prioritize a higher pheromone following α value, while another prioritizes a higher heuristic information exploration β value. This balances centralized and decentralized exploration, preventing premature convergence. Each ant starts from a predefined initial condition node and, based on the transition probability formula, comprehensively considers the pheromone concentration. With heuristic information Select the next node to be visited until all query condition nodes have been visited once, thus forming a complete retrieval path. Then calculate the total cost of the path and update the optimal path for this iteration. Iteration and convergence: After each round of multi-ant colony collaborative search, the global pheromone is updated according to the adaptive mechanism in step S4. This process is repeated until the algorithm meets the termination condition, that is, the change in the cost of the global optimal search path in K consecutive iterations is less than or equal to the preset convergence threshold. At this time, the output global optimal search path is the most efficient multi-condition query order. The parameter K is the stability judgment window for algorithm convergence, which is a preset positive integer, such as 10. The basis for setting it is that a sufficiently long observation window is needed to confirm the stable trend of the solution and avoid premature termination due to random fluctuations. At the same time, the window should not be too large to prevent invalid iterations. The preset convergence threshold defines a quantitative standard for a sufficiently small change. It can be set as a relative value related to the cost of the current optimal solution, such as 0.1% of the current optimal cost. The basis for setting it is that when the improvement of the solution is negligible and has no meaning for improving actual performance, the calculation can be stopped to improve efficiency. K and the convergence threshold together constitute the termination criterion. When the improvement in the cost of the global optimal path in K consecutive iterations does not exceed the threshold, the algorithm is judged to have converged and the current optimal path is output. Finally, the retrieval is executed. Based on the node order determined by this optimal path, the query conditions corresponding to each node are used as filtering conditions to iteratively filter the structured data output in step S2, namely personnel records and their attribute label weight vectors. For example, for the path condition A is greater than condition C, which is greater than condition B, all records are first filtered using condition A to obtain the intermediate result set S1. Then, condition C is used to filter S1 to obtain S2. Finally, condition B is used to filter S2 to obtain the personnel information that finally meets all conditions. Through this intelligent planning, it is ensured that each filtering operation is performed on the smallest possible dataset, thereby significantly improving the response speed and system throughput of complex combined queries.

[0033] Step S4: After each round of multi-ant colony collaborative search in the improved ant colony optimization algorithm is completed, perform an adaptive update on the global pheromone of the improved ant colony optimization algorithm. The expression for adaptive global pheromone update is: ; in, This represents the pheromone concentration on path ij after the update. This represents the adaptive pheromone evaporation factor. This represents the total amount of pheromone released by the top a% of ants with the best fitness values ​​on path ij in this iteration. This represents the pheromone concentration on the path from i to j at time t.

[0034] The adaptive pheromone evaporation factor ρ increases with the number of iterations, and its expression is: ; in, Indicates the current iteration number. Indicates the total number of iterations; The percentage of ants releasing pheromones, denoted as 'a%', decreases dynamically with the iteration process, expressed as: ; Step S305: When the number of iterations reaches a preset smoothing threshold, perform pheromone smoothing operation. The pheromone smoothing expression is: ; Where δ represents the smoothing coefficient. This represents the preset minimum pheromone concentration. When the number of iterations reaches the preset learning threshold, the learning operator operation between the sub-ant colonies is executed. The expression for the learning operator is: ; in, and represents the pheromone concentrations of sub-ant colonies m and n on path ij, respectively, and γ is the learning operator.

[0035] In one embodiment, after each round of multi-ant colony collaborative search in step S3, the process proceeds to step S4, where the core adaptive global pheromone update is performed to dynamically guide the search direction of the population. This update mechanism is first based on basic rules, where pheromones on all paths undergo a period of evaporation. Subsequently, the ants that performed well in this round are used to enhance the pheromone levels. Specifically, the update formula is as follows: ,in, Let a be the ants with the best fitness values ​​in this iteration, where the fitness value of an ant is defined as the reciprocal of the cost of its path construction, the total amount of pheromone released on the path ij it traverses, and the amount of pheromone released by a single ant k on path ij. The calculation formula is as follows: ,in, Represents the pheromone intensity constant. This represents the total cost of the ant constructing its path this time, and the total pheromone increment on ij. The sum of the amount released by all elite ants on edge ij achieves the inverse relationship between the value and the cost of their respective path construction, meaning that a higher quality solution will leave a stronger positive feedback on its path. To further improve the performance of the improved ant colony optimization algorithm, this scheme introduces a multi-adaptive strategy. Key parameters are not fixed; the pheromone evaporation factor ρ increases adaptively from a lower value during iteration to ensure sufficient initial exploration, and then gradually increases to a higher value to promote rapid convergence in the later stages. Its value is calculated using the following formula: ; Where count is the current iteration number. To predetermine the total number of iterations, the proportion of elite ants releasing pheromones, denoted as a%, is dynamically reduced. The value of a is controlled by a formula, linearly decreasing from 20% at the beginning of the iterations to 10% at the end. The formula is: ; This allows more ants' experience to be shared in the early stages of the search to broaden the search scope, while in the later stages the focus is on the in-depth utilization of a few optimal solutions; In addition, the algorithm includes two advanced operations to maintain robustness: a smoothing operation is performed on the global pheromone when the number of iterations reaches a preset smoothing threshold of approximately one-third of the total number of iterations. This formula represents the in-situ update operation of pheromone concentration; the right side of the equation... To update the original pheromone concentration (old value) of path ij before the update, the left side of the equation is... The updated pheromone concentration (new value); calculated by comparing the current concentration with the preset minimum value. The difference is multiplied by a smoothing coefficient δ, and then subtracted from the old value to adjust the pheromone concentration towards... Dynamically approaching and not falling below To prevent potential paths from being completely abandoned by the algorithm due to excessively low pheromone concentration and to ensure continued exploration potential, the preset smoothing threshold is set based on the fact that in the first 1 / 3 of the improved ant colony optimization algorithm's operation, the pheromones of each path have begun to differentiate but have not yet solidified. Smoothing at this time can effectively prevent potential paths from being prematurely abandoned due to temporarily low pheromone concentration, thereby maintaining the necessary exploration capability. When the number of iterations reaches approximately 2 / 3 of the total number of iterations, which is the preset learning threshold, learning operator operations are performed among the sub-ant colonies. This formula represents the in-situ fusion and update operation of the m-pheromone concentration in the ant colony. The right side of the equation... The original pheromone concentration (old value) of path ij before updating the ant colony m. The pheromone concentration on path ij of the n-th sub-ant colony; weighted... Preserve the pheromone experience and weights of the sub-ant colony m itself. Drawing on the pheromone experience of ant colony n, the fusion result is assigned to the left side. This refers to the updated pheromone concentration of sub-ant colony m, enabling information sharing and collaborative learning among different sub-ant colonies. In this implementation, two different sub-ant colonies, m and n, are randomly selected for bidirectional pheromone experience sharing. The learning coefficient γ is a preset constant, allowing sub-ant colonies with different search strategies, such as focusing on exploration or utilization, to mutually learn from each other's pheromone distribution experience. and The ant colonies belong to different sub-colonies, thus collaboratively accelerating their approach to the global optimum. For ease of implementation, the preset learning threshold is set based on the fact that in the later two-thirds of the algorithm's operation, each sub-colonym has conducted sufficient independent exploration in different regions and accumulated differentiated experiences. At this time, initiating inter-colonym learning can efficiently integrate these scattered valuable information, guiding the entire population to collaboratively accelerate convergence to the global optimum, and preventing individual sub-colonies from stagnating near local optima. Key parameters can be referenced in the following example values: pheromone intensity constant Q can be set to 1.0, smoothing coefficient δ can be set to 0.1, minimum pheromone concentration τ_min can be set to 10% of the initial concentration, learning coefficient γ can be set to 0.2, and smoothing and learning thresholds can be set to 1 / 3 and 2 / 3 of the total number of iterations, respectively. The designed pheromone update and adaptive adjustment mechanism together constitute the learning and decision-making core of the algorithm's brain. It ensures that the search process achieves a dynamic balance between exploring unknown regions and utilizing known optimal solutions, effectively avoiding the problem of traditional methods easily getting trapped in local optima, thereby driving the multi-ant colony system to converge efficiently and stably to the global optimum retrieval path defined in step S3.

[0036] Example 2, refer to Figure 2 In another embodiment of the present invention, which differs from the first embodiment, a personnel information processing system for business administration based on big data is provided, including a preprocessing module, a processing module, an acquisition module, and an update module; The preprocessing module acquires personnel information, preprocesses the personnel information, and obtains preprocessed text information. The processing module, based on the CNN-BERT model, processes the preprocessed text information through the embedding layer to obtain a word vector matrix. The word vector matrix is ​​then processed sequentially through the CNN convolutional layer, pooling layer, and fully connected layer to obtain structured data. The acquisition module uses an improved ant colony optimization algorithm to optimize the path of multi-condition combined queries on structured data and obtain the globally optimal retrieval path; Based on the globally optimal retrieval path, retrieve personnel information from structured data that meets multiple condition combinations for querying; The update module performs adaptive updates to the global pheromone of the improved ant colony optimization algorithm after each round of multi-ant colony collaborative search, and outputs a set of target personnel information that satisfies multiple condition combination queries based on the globally optimal retrieval path.

[0037] In one embodiment, the acquired personnel information is preprocessed by cleaning and word segmentation to form high-quality standardized text, laying the data foundation for subsequent in-depth analysis. Based on the CNN-BERT model, the preprocessed text information is transformed into a word vector matrix with rich semantics through the BERT embedding layer. This matrix is ​​then input into the CNN module, where local features of different granularities are extracted through multi-scale convolutional kernels. After pooling and full-connected layer fusion and dimensionality reduction, high-precision, weighted structured personnel data is finally output. This process improves the ability to accurately capture key attributes and latent features from complex text information. On this basis, when faced with multi-condition combined queries, an improved ant colony optimization algorithm is used for intelligent path planning. The improved ant colony optimization algorithm quickly finds the globally optimal retrieval path in the solution space composed of query conditions through multi-ant colony collaborative search and adaptive pheromone update mechanism, thereby accessing structured data in the most efficient order. This significantly improves the response speed and overall throughput of complex queries and effectively avoids the problem of traditional methods easily getting trapped in local optima. All links in the entire process are closely connected and optimized collaboratively, ultimately achieving a deep understanding and millisecond-level accurate retrieval of massive amounts of business management personnel information.

[0038] This invention, through the deep synergistic innovation of the CNN-BERT hybrid model and the improved ant colony optimization algorithm, achieves a breakthrough in both deep understanding and ultra-fast retrieval for the first time in processing massive, multi-source personnel information in business administration. On one hand, the CNN-BERT model deeply integrates BERT's deep semantic understanding capabilities with CNN's efficient local feature extraction capabilities, enabling it to accurately extract key attributes and latent features of personnel from unstructured text, constructing a rich-dimensional and accurately represented structured knowledge base. This solves the fundamental problems of traditional methods, such as one-sided feature extraction and superficial semantic understanding, laying a solid foundation for high-quality queries. On the other hand, the introduction of multi-ant colony collaborative search and genetic algorithms... Initialization and adaptive pheromone updates together constitute a powerful dynamic path optimization engine. This engine can intelligently plan and quickly converge to the globally optimal retrieval path for complex multi-condition combined queries. Compared with traditional serial queries or basic optimization algorithms, the planning efficiency and execution speed of retrieval paths are improved by orders of magnitude in multi-condition, high-concurrency query scenarios. At the same time, it effectively avoids getting trapped in local optima. It seamlessly connects deep text mining with intelligent query optimization, making the process of accurately locating target personnel information from massive heterogeneous data accurate, fast and resource-efficient. It provides a powerful intelligent tool for real-time decision-making, talent inventory and organizational analysis in business management.

[0039] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0040] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the protection scope of the present invention.

Claims

1. A method for processing personnel information in business administration based on big data, characterized in that, Includes the following steps: Step S1: Obtain personnel information, preprocess the personnel information, and obtain preprocessed text information; Step S2: Based on the CNN-BERT model, the preprocessed text information is processed through the embedding layer to obtain the word vector matrix. The word vector matrix is ​​then processed sequentially through the CNN convolutional layer, pooling layer, and fully connected layer to obtain structured data. Step S3: The improved ant colony optimization algorithm is used to optimize the path of multi-condition combined query for structured data and obtain the globally optimal retrieval path. Based on the globally optimal retrieval path, retrieve personnel information from structured data that meets multiple condition combinations for querying; Step S4: After each round of multi-ant colony collaborative search in the improved ant colony optimization algorithm, the global pheromone of the improved ant colony optimization algorithm is adaptively updated, and the target personnel information set that satisfies the multi-condition combination query is output based on the globally optimal retrieval path.

2. The method for processing personnel information for business administration based on big data as described in claim 1, characterized in that, Step S1: Obtain personnel information, preprocess the personnel, and obtain preprocessed text information; The personnel information includes personnel files, management records, communication texts, and external data; The personnel files include resumes, onboarding forms, and background investigation reports; Management records include performance evaluation reports, multi-source performance feedback, promotion review opinions, and reward and punishment records; The communication texts include work logs, project summaries, performance reports, internal forum posts, and email communication summaries; External data includes job descriptions posted on external recruitment platforms, industry talent assessment reports, and publicly available professional qualification certifications and practice certificates. The personnel information is processed sequentially through text cleaning, word segmentation, and stop word removal to obtain preprocessed text information.

3. The method for processing personnel information for business administration based on big data as described in claim 2, characterized in that, Step S2: Based on the CNN-BERT model, the preprocessed text information is processed through the embedding layer to obtain the word vector matrix. The word vector matrix is ​​then processed sequentially through the CNN convolutional layer, pooling layer, and fully connected layer to obtain structured data. The structured data includes personnel entity attribute classification labels and their corresponding feature weights; Step S2 includes steps S201 and S202; Step S201: Input the word vector matrix into the CNN-BERT model, and then... convolution kernel, convolution kernel and The convolutional kernels perform sliding convolution operations. After each convolutional kernel traverses the entire word vector matrix, it generates a feature map, expressed as: ; ; in, This represents the feature extracted at position i. This represents a local window from row i to row i+h-1 of the word vector matrix, where h represents the height of the convolution kernel. Indicates weight, Indicates the bias value. Represents the ReLU nonlinear activation function; This means that the convolutional kernel sequentially filters the word vector matrix to form n-h+1 new features. This represents the feature map of n.

4. The method for processing personnel information for business administration based on big data as described in claim 3, characterized in that, Step S202: The feature maps are processed and concatenated through pooling layers to obtain a q feature map, and the q feature maps are concatenated through fully connected layers to obtain an i feature map; The i-feature map is calculated using the Softmax function to obtain the probability distribution and the feature weights of the corresponding classification labels; By summarizing personnel information, classification labels, and corresponding feature weights, structured data is obtained.

5. The method for processing personnel information for business administration based on big data as described in claim 4, characterized in that, Step S3: The improved ant colony optimization algorithm is used to optimize the path of multi-condition combined query for structured data and obtain the globally optimal retrieval path. Based on the globally optimal retrieval path, retrieve personnel information that satisfies multiple condition combination queries from the structured data; The improved ant colony optimization algorithm employs multi-ant colony collaborative search, including at least two independent sub-ant colonies performing path search in parallel, with each sub-ant colony sharing the same global pheromone matrix. Each query condition in the multi-condition combined query is represented as a node to be visited in the improved ant colony optimization algorithm.

6. The method for processing personnel information for business administration based on big data as described in claim 5, characterized in that, Step S301: Use a genetic algorithm to generate the distribution of the initial pheromone matrix, perform multi-ant colony collaborative path retrieval, mark the query starting condition node as visited and set it as the current node; Step S302: When there are unvisited query condition nodes, select the next node to be searched according to the transition probability formula and construct a complete search path. The transition probability formula is: ； in, Let represent the probability that an ant moves from node i to node j. This represents the pheromone concentration along path ij at time t. This represents the heuristic information from node i to node j, where α is the pheromone weighting factor and β is the heuristic information weighting factor. This represents the set of nodes that the ant is allowed to visit next; Step S303: Repeat step S302 until all query condition nodes have been accessed, thus constructing a complete search path. Step S304: Calculate the cost of the retrieval path and update the optimal path for this iteration based on the cost; Step S305: When the number of iterations reaches the preset smoothing threshold, perform pheromone smoothing operation; Step S306: Determine whether the improved ant colony optimization algorithm meets the termination condition; If satisfied, output the globally optimal search path; If the condition is not met, return to step S302 for the next iteration; The termination condition is determined to be met if the change in cost of the global optimal retrieval path in K consecutive iterations is less than or equal to a preset convergence threshold.

7. The method for processing personnel information for business administration based on big data as described in claim 6, characterized in that, Step S301 includes steps S3011, S3012, S3013 and S3014; Step S3011: Randomly generate a set of retrieval paths for the query condition nodes as the initial population; Step S3012: Decode each retrieval path and calculate the execution cost of each retrieval path as the fitness value using the objective function; Step S3013: Based on the fitness value, a new generation of population is generated through selection, crossover, and mutation operations; Step S3014: Repeat steps S3012 to S3013 until the preset genetic algorithm termination condition is met, and map the obtained global optimal retrieval path to the initial pheromone matrix. The termination condition is determined to be met when the change in the cost of the globally optimal retrieval path, represented by the fitness value, is less than or equal to the preset convergence threshold of the genetic algorithm in consecutive generations, and the genetic algorithm iteration is stopped.

8. The method for processing personnel information for business administration based on big data as described in claim 7, characterized in that, Step S4: After each round of multi-ant colony collaborative search in the improved ant colony optimization algorithm is completed, perform an adaptive update on the global pheromone of the improved ant colony optimization algorithm. The expression for adaptive global pheromone update is: ; in, This represents the pheromone concentration on path ij after the update. This represents the adaptive pheromone evaporation factor. This represents the total amount of pheromone released by the top a% of ants with the best fitness values ​​on path ij in this iteration. This represents the pheromone concentration on the path from i to j at time t.

9. The method for processing personnel information for business administration based on big data as described in claim 8, characterized in that, The adaptive pheromone volatile factor ρ increases with the number of iterations, and its expression is: ; in, Indicates the current iteration number. Indicates the total number of iterations; The percentage of ants releasing pheromones, denoted as 'a%', decreases dynamically with the iteration process, expressed as: ; When the number of iterations reaches a preset smoothing threshold, a pheromone smoothing operation is performed. The pheromone smoothing expression is: ; Where δ represents the smoothing coefficient. This represents the preset minimum pheromone concentration. When the number of iterations reaches a preset learning threshold, a learning operator operation is executed among the sub-ant colonies. The expression for the learning operator is: ; in, and represents the pheromone concentrations of sub-ant colonies m and n on path ij, respectively, and γ is the learning operator.

10. A personnel information processing system for business administration based on big data, which is applied to the personnel information processing method for business administration based on big data as described in any one of claims 1-9, characterized in that, It includes a preprocessing module, a processing module, an acquisition module, and an update module; The preprocessing module acquires personnel information, preprocesses the personnel information, and obtains preprocessed text information. The processing module, based on the CNN-BERT model, processes the preprocessed text information through the embedding layer to obtain a word vector matrix. The word vector matrix is ​​then processed sequentially through the CNN convolutional layer, pooling layer, and fully connected layer to obtain structured data. The acquisition module uses an improved ant colony optimization algorithm to optimize the path of multi-condition combined queries on structured data and obtain the globally optimal retrieval path; Based on the globally optimal retrieval path, retrieve personnel information from structured data that meets multiple condition combinations for querying; The update module performs adaptive updates to the global pheromone of the improved ant colony optimization algorithm after each round of multi-ant colony collaborative search, and outputs a set of target personnel information that satisfies multiple condition combination queries based on the globally optimal retrieval path.

Images