Client demand mining method and device

Abstract

The invention provides a client demand mining method and device. The client demand mining method comprises the steps of obtaining a client feature vector and a service feature vector; and inputting the customer feature vector and the service feature vector into a self-adaptive ecological balance optimization algorithm, and outputting a recommendation service list of the to-be-processed customer, which comprises the following steps: initializing a population, and based on a preset fitness function, the customer feature vector and the service feature vector, obtaining a recommendation service list of the to-be-processed customer, calculating the fitness score of each candidate service in the current population relative to the to-be-processed customer; and performing iterative optimization on the population by adopting a global search operator, a local development operator and a fine tuning operator, selecting a preset number of candidate services with the highest fitness score from the current population according to a sequence from high to low when the iteration process is stopped, combining the candidate services into a recommendation service list, and outputting the recommendation service list. Therefore, the accuracy and efficiency of customer demand prediction can be improved by utilizing the optimization capability of the adaptive ecological balance optimization algorithm in customer data analysis.

Description

Technical Field

[0001] This application relates to the field of big data analytics, and more particularly to a method and apparatus for mining customer needs. Background Technology

[0002] With the rapid development of information technology and the widespread adoption of the internet, Customer Relationship Management (CRM) systems are playing an increasingly important role in modern enterprises. These systems are not merely storage tools for customer information, but also key platforms for uncovering customer needs, improving customer satisfaction, and achieving precision marketing. To effectively meet the increasingly diverse needs of customers, enterprises typically rely on various machine learning models and optimization algorithms to analyze and predict customer behavior. However, traditional customer needs mining methods are gradually revealing their limitations when faced with large-scale datasets.

[0003] Current customer demand mining and recommendation systems are mainly based on traditional machine learning models, such as decision trees and support vector machines (SVM), or heuristic algorithms, such as genetic algorithms (GA) and particle swarm optimization. While these methods can handle customer data to some extent, they have significant shortcomings in practical applications. First, the local optimum trap is a major weakness of traditional optimization algorithms. These algorithms often get stuck in local optima on complex, high-dimensional datasets, failing to effectively explore the global optimum and resulting in unsatisfactory recommendation performance. Second, low computational efficiency is also a problem that urgently needs to be addressed. Traditional algorithms typically require multiple iterations for global search, which not only increases computational costs but also makes them difficult to adapt to changes in real-time requirements. Finally, many algorithms have relatively fixed search strategies, lacking adaptability and unable to flexibly respond to dynamic changes in customer data, thus affecting the accurate mining of customer needs.

[0004] In summary, traditional customer needs mining methods face numerous challenges in dealing with large-scale datasets and complex customer needs, and there is an urgent need for new optimization algorithms to improve efficiency and accuracy in order to meet the needs of modern enterprises for intelligent customer relationship management. Summary of the Invention

[0005] This application provides a method and apparatus for mining customer needs, in order to solve the technical problems of low accuracy, low computational efficiency and insufficient adaptability of traditional customer needs mining methods in processing large-scale datasets and complex customer needs.

[0006] To solve the above-mentioned technical problems, this application is implemented as follows:

[0007] In a first aspect, embodiments of this application provide a method for mining customer needs, the method comprising:

[0008] Obtain the customer feature vector of the customer to be processed and the service feature vector of the candidate service;

[0009] The customer feature vector and the service feature vector are input into the adaptive ecological balance optimization algorithm to output a recommended service list for the customer to be processed, wherein the recommended service list includes at least one candidate service;

[0010] Specifically, the customer feature vector and the service feature vector are input into the adaptive ecological balance optimization algorithm, and the output recommended service list for the customer to be processed includes:

[0011] Initialize a population containing at least one candidate service;

[0012] Based on the preset fitness function, the customer feature vector, and the service feature vector, calculate the fitness score of each candidate service in the current population relative to the customer to be processed.

[0013] Based on the fitness score, the population is iteratively optimized using a global search operator, a local exploit operator, and a fine-tuning operator. In each iteration, the global search operator, the local exploit operator, and the fine-tuning operator are dynamically allocated and applied to candidate services based on the fitness score to generate a new population. The size of the global search operator, the local exploit operator, and the fine-tuning operator is adaptively adjusted according to the progress of the iteration process.

[0014] When the iteration process stops, select a preset number of candidate services with the highest fitness scores from the current population in descending order, combine them to form the recommended service list and output it.

[0015] Optionally, obtaining the customer feature vector of the customer to be processed and the service feature vector of the candidate service includes:

[0016] Obtain customer data of the customers to be processed; after preprocessing the customer data, extract features from the preprocessed customer data, and concatenate the extracted features into the customer feature vector;

[0017] Obtain service data for candidate services; after preprocessing the service data, extract features from the preprocessed service data, and concatenate the extracted features into the service feature vector.

[0018] Optionally, based on a preset fitness function, the customer feature vector, and the service feature vector, the fitness score of each candidate service relative to the customer to be processed in the current population is calculated, including:

[0019] Calculate the dot product between the service feature vector of the candidate service and the customer feature vector of the customer to be processed;

[0020] Calculate the Euclidean norm of the service feature vector of the candidate service and the customer feature vector of the customer to be processed;

[0021] The fitness score is determined based on the dot product and the Euclidean norm.

[0022] Optionally, based on the fitness score, dynamically assigning and applying the global search operator, local exploitation operator, and fine-tuning operator to candidate services to generate a new population includes:

[0023] Based on the fitness scores, the population is divided into a high fitness score subgroup, a medium fitness score subgroup, and a low fitness score subgroup.

[0024] The global search operator is applied to the low-fitness-score subpopulation, the local exploitation operator is applied to the medium-fitness-score subpopulation, and the fine-tuning operator is applied to the high-fitness-score subpopulation to generate a new population.

[0025] Optionally, adaptively adjusting the sizes of the global search operator, the local development operator, and the fine-tuning operator according to the progress of the iteration process includes:

[0026] As the iterative process progresses over time, the weight of the global search operator decreases, the weight of the fine-tuning operator increases, and the weight of the local development operator increases.

[0027] Optionally, stopping the iteration process includes:

[0028] The current iteration count has reached the preset maximum threshold, or the difference between the current highest fitness score and the previous best fitness score is less than the preset threshold.

[0029] Optionally, when the iteration process stops, after selecting a preset number of candidate services with the highest fitness scores from the current population in descending order and combining them into the recommended service list, the method further includes:

[0030] Obtain feedback data from the customers to be processed regarding the recommended service list;

[0031] Based on the feedback data, the adaptive ecological balance optimization algorithm is optimized to obtain the optimized adaptive ecological balance optimization algorithm;

[0032] Based on the optimized adaptive ecological balance optimization algorithm, the customer demand mining method is then executed.

[0033] Secondly, embodiments of this application provide a customer demand mining device, the device comprising:

[0034] The acquisition module is used to acquire the customer feature vector of the customer to be processed and the service feature vector of the candidate service;

[0035] An execution module is used to input the customer feature vector and the service feature vector into an adaptive ecological balance optimization algorithm and output a recommended service list for the customer to be processed, wherein the recommended service list includes at least one candidate service;

[0036] Specifically, the customer feature vector and the service feature vector are input into the adaptive ecological balance optimization algorithm, and the output recommended service list for the customer to be processed includes:

[0037] Initialize a population containing at least one candidate service;

[0038] Based on the preset fitness function, the customer feature vector, and the service feature vector, calculate the fitness score of each candidate service in the current population relative to the customer to be processed.

[0039] Based on the fitness score, the population is iteratively optimized using a global search operator, a local exploit operator, and a fine-tuning operator. In each iteration, the global search operator, the local exploit operator, and the fine-tuning operator are dynamically allocated and applied to candidate services based on the fitness score to generate a new population. The size of the global search operator, the local exploit operator, and the fine-tuning operator is adaptively adjusted according to the progress of the iteration process.

[0040] When the iteration process stops, select a preset number of candidate services with the highest fitness scores from the current population in descending order, combine them to form the recommended service list and output it.

[0041] Thirdly, embodiments of this application provide an electronic device, including: a processor, a memory, and a program stored in the memory and executable on the processor, wherein when the program is executed by the processor, it implements the steps of a customer demand mining method as described in the first aspect.

[0042] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of a customer demand mining method as described in the first aspect.

[0043] Fifthly, embodiments of this application provide a computer program product, including computer instructions, which, when executed by a processor, implement the steps of a customer demand mining method as described in the first aspect.

[0044] In this embodiment, by obtaining the customer feature vector of the customer to be processed and the service feature vector of the candidate services, and inputting them into an adaptive ecological balance optimization algorithm, this method can effectively generate a personalized list of recommended services. Specifically, the initialized population contains at least one candidate service, and the fitness score of each candidate service is calculated based on a preset fitness function, thereby ensuring that the recommended services are highly matched with customer needs. During the iterative optimization process, global search operators, local development operators, and fine-tuning operators are used for dynamic adjustment, enabling the algorithm to flexibly switch between global and local approaches, avoiding the pitfalls of local optima. Furthermore, as the iteration progresses, the algorithm adaptively adjusts the size of the operators based on the fitness scores, enhancing the flexibility and efficiency of the optimization process. Finally, when the iteration stops, the candidate service with the highest fitness score is selected from the current population to form a list of recommended services and output. This method can significantly improve the accuracy and real-time performance of recommended services, enhance the precision and efficiency of customer demand prediction, and meet the personalized needs of customers. Attached Figure Description

[0045] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0046] Figure 1 A flowchart illustrating a method for mining customer needs as provided in this application embodiment;

[0047] Figure 2 A flowchart illustrating a method for mining customer needs as provided in this application embodiment;

[0048] Figure 3 A flowchart illustrating a method for mining customer needs as provided in this application embodiment;

[0049] Figure 4 A structural block diagram of a customer-required excavation device provided in this application embodiment;

[0050] Figure 5 This is a structural block diagram of a network device provided in an embodiment of this application. Detailed Implementation

[0051] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0052] Figure 1 This illustrates a method for mining customer needs according to an embodiment of this application, such as... Figure 1 As shown, the method includes:

[0053] Step S101: Obtain the customer feature vector of the customer to be processed and the service feature vector of the candidate service;

[0054] Step S102: Input the customer feature vector and service feature vector into the adaptive ecological balance optimization algorithm, and output the recommended service list for the customer to be processed;

[0055] The recommended service list includes at least one candidate service.

[0056] It should be noted that, Figure 1 The illustrated process describes a service recommendation method based on the Adaptive Ecological Balance Optimization (AEBO) algorithm. First, the system obtains the customer feature vector of the client to be processed and the service feature vectors of all candidate services. Then, these feature vectors are input into the AEBO algorithm for processing. This algorithm dynamically optimizes and calculates the matching relationship between customers and services by simulating the adaptive mechanism of population competition and balance in an ecosystem. Finally, it outputs a list of optimal recommended services tailored to the client to be processed, containing one or more optimal candidate services selected by the algorithm.

[0057] Among them, the adaptive ecological balance optimization algorithm simulates the multi-niche co-evolution mechanism in natural ecosystems. It adopts a dynamic adaptive search strategy to balance global search and local development. By optimizing the process of mining customer needs through the interaction between niches, it improves the accuracy and efficiency of matching customer needs.

[0058] Therefore, by introducing an adaptive ecological balance optimization algorithm, the most suitable service can be matched accurately and efficiently for customers from a massive number of candidate services. Its core advantage lies in the algorithm's good adaptability and global optimization capabilities, which can effectively improve the accuracy of service recommendations and user satisfaction.

[0059] like Figure 2As shown, step S102 involves inputting the customer feature vector and service feature vector into the adaptive ecological balance optimization algorithm, and outputting a list of recommended services for the customers to be processed, including:

[0060] Step S201: Initialize a population, which contains at least one candidate service;

[0061] Step S202: Based on the preset fitness function, customer feature vector and service feature vector, calculate the fitness score of each candidate service in the current population relative to the customer to be processed.

[0062] Step S203: Based on the fitness score, the population is iteratively optimized using global search operators, local exploitation operators, and fine-tuning operators. In each iteration, global search operators, local exploitation operators, and fine-tuning operators are dynamically allocated and applied to candidate services based on the fitness score to generate a new population. The size of global search operators, local exploitation operators, and fine-tuning operators is adaptively adjusted according to the progress of the iteration process.

[0063] Step S204: When the iteration process stops, select a preset number of candidate services with the highest fitness scores from the current population in descending order, combine them into a recommended service list and output it.

[0064] It should be noted that, firstly, customer feature vectors (i.e., numerical vectors representing customer attributes, behaviors, or preferences) and service feature vectors (i.e., numerical vectors representing service characteristics, performance, or categories) are obtained for the customer to be processed and for the candidate services. Then, these vectors are input into an adaptive ecological balance optimization algorithm, an optimization technique that simulates the natural ecological balance process, aiming to find the optimal solution through iterative search and adjustment. The algorithm initializes a population containing at least one candidate service (i.e., potentially recommendable service option) and calculates the fitness score of each candidate service relative to the customer to be processed based on a pre-defined fitness function (a mathematical function that evaluates the degree of matching between a candidate service and a customer). (A higher score indicates a better match.) Subsequently, based on these fitness scores, the algorithm iteratively optimizes the population using three operators: a global search operator (for extensively exploring the solution space to avoid local optima), a local development operator (for deep search in local regions), and a fine-tuning operator (for fine-tuning the solution to improve accuracy). During each iteration, the algorithm dynamically allocates and applies these operators. Specifically, it determines which operator to use based on the fitness score of the candidate services (e.g., low-fitness candidates may be preferentially assigned a global search operator to broaden the search scope, while high-fitness candidates may be assigned local development or fine-tuning operators to refine the results). Simultaneously, the size of these operators is adaptively adjusted based on the iteration progress (e.g., the number of iterations or the degree of convergence). For example, the search radius of the global search operator may decrease as iterations progress. The iteration process continues until a stopping condition is met (e.g., reaching the maximum number of iterations or fitness stabilization). Finally, a predetermined number (e.g., top-N) of candidate services are selected from the current population in descending order of fitness score, combined to form a recommended service list, and output.

[0065] For example, in an e-commerce platform, the customer to be processed is a young user who frequently purchases technology products. Their customer feature vector might include [Age: 25, Interest: Electronics, Purchase Frequency: High]. Candidate services include multiple products, such as laptops, smartphones, and books, each with a service feature vector; for example, the vector for a laptop might be [Price: 1000, Category: Technology, Rating: 4.5]. An adaptive ecosystem optimization algorithm initializes a population (containing these candidate products) and calculates the fitness score of each product relative to the customer (e.g., laptops score highly because they match the technology interest). In iterative optimization, the algorithm might apply a global search operator to low-scoring products (such as books) to explore other similar technology products, apply a local development operator to high-scoring products to refine the recommendations (e.g., recommending laptops of a specific brand), and adjust the recommendation weights through fine-tuning operators. Finally, the output recommendation list, such as [Laptops, Smartphones, Smartwatches], is highly matched to the customer's features.

[0066] In summary, through Figure 1 , Figure 2 The method described dynamically and adaptively filters and ranks candidate services to provide highly personalized recommendations for specific customers. This approach combines global exploration and local optimization to ensure the recommendation list is both comprehensive and accurate, making it suitable for scenarios requiring efficient matching of customers and services, such as e-commerce, financial services, or content recommendations. Customer characteristics may include age, purchase history, or interest tags, while service characteristics may include price, category, or user ratings. Through adaptive adjustment, the algorithm balances search efficiency and solution quality, improving the performance of the recommendation system and the user experience.

[0067] In one possible implementation, obtaining the customer feature vector of the customer to be processed and the service feature vector of the candidate service includes: obtaining customer data of the customer to be processed; preprocessing the customer data, extracting features from the preprocessed customer data, and concatenating the extracted features into a customer feature vector; obtaining service data of the candidate service; preprocessing the service data, extracting features from the preprocessed service data, and concatenating the extracted features into a service feature vector.

[0068] It should be noted that this possible implementation describes the method for constructing customer feature vectors and service feature vectors. For a customer to be processed, the system first obtains its raw customer data, then preprocesses the data to remove noise, handle missing values, or standardize it, providing a clean and standardized data foundation for subsequent analysis. After preprocessing, feature extraction is performed on the processed customer data to mine and construct key information that can effectively represent customer attributes and needs. Finally, these extracted features are concatenated in order to form a customer feature vector representing the customer. Similarly, for candidate services, the system obtains its raw service data and performs the same preprocessing procedure to ensure data quality. Then, feature extraction is used to extract core features that describe the service functions and characteristics from the service data, and these features are concatenated to form a service feature vector representing the service.

[0069] The following explanation uses formulas to illustrate how multidimensional customer data is transformed into feature vectors through preprocessing and feature extraction. The feature vector for each customer is:

[0070]

[0071] in, It is the first The feature vector of each customer contains These features represent the customer's basic information, behavioral data, and identity characteristics, respectively.

[0072] Therefore, through systematic data preprocessing and feature extraction, raw, unstructured customer and service data are transformed into well-structured, machine-readable feature vectors. This provides standardized and quantified input data for subsequent adaptive ecological balance optimization algorithms, serving as a crucial prerequisite and guarantee for achieving accurate service recommendations.

[0073] In one possible implementation, the fitness score of each candidate service relative to the client to be processed in the current population is calculated based on a preset fitness function, client feature vector, and service feature vector. This includes: calculating the dot product of the service feature vector of the candidate service and the client feature vector of the client to be processed; calculating the Euclidean norm of the service feature vector of the candidate service and the client feature vector of the client to be processed; and determining the fitness score based on the dot product and the Euclidean norm.

[0074] It should be noted that, for each customer, the adaptive ecological balance optimization algorithm needs to calculate the fitness between the customer's needs and the candidate services. Assume there is a feature vector for a candidate service. The fitness function is typically measured by cosine similarity, which is compared to the customer's feature vector.

[0075]

[0076] in, This represents the dot product of the customer feature vector and the service feature vector; and Let the Euclidean norms of the customer and service feature vectors be represented respectively; Calculation results The range is between [0, 1]. The larger the value, the higher the degree of matching between the service and the customer's needs.

[0077] It's important to note that in calculating the fitness score, the system uses a preset fitness function, combining the customer feature vector of the client to be processed with the service feature vector of each candidate service in the population, and performs quantitative evaluation through two core mathematical operations. First, the system calculates the dot product between the service feature vector of the candidate service and the customer feature vector of the client to be processed; this operation reflects the similarity and alignment of the two vectors in direction. Second, it calculates the Euclidean norm between these two feature vectors, which measures the absolute distance difference between the vectors in space. Finally, the fitness function combines the results of the dot product and the Euclidean norm calculations, transforming them into a final, quantified fitness score. This score directly characterizes the degree of matching between the candidate service and the client to be processed.

[0078] Therefore, by combining two complementary geometric metrics, dot product (directional similarity) and Euclidean norm (absolute distance), this calculation mechanism can comprehensively and accurately evaluate the matching relationship between services and customers from different dimensions, providing a reliable and accurate quality assessment basis for subsequent adaptive ecological balance optimization algorithms.

[0079] In one possible implementation, dynamically allocating and applying global search operators, local exploitation operators, and fine-tuning operators to candidate services based on fitness scores to generate a new population includes: dividing the population into high-fitness-score subgroups, medium-fitness-score subgroups, and low-fitness-score subgroups based on fitness scores; allocating and applying global search operators to the low-fitness-score subgroups, allocating and applying local exploitation operators to the medium-fitness-score subgroups, and allocating and applying fine-tuning operators to the high-fitness-score subgroups to generate a new population.

[0080] It should be noted that during the iterative optimization process of the adaptive ecological balance optimization algorithm, the system dynamically assigns different optimization operators to each candidate service in the current population based on its fitness score to generate a new population. Specifically, the algorithm first divides the entire population into three subgroups based on their fitness scores: a high-fitness-score subgroup, a medium-fitness-score subgroup, and a low-fitness-score subgroup. After the division, the algorithm applies differentiated optimization strategies based on the characteristics and optimization needs of different subgroups: a global search operator is applied to the low-fitness-score subgroup to explore and perturb it significantly, allowing it to escape the current poor solution region and find new potential optimization directions; a local development operator is applied to the medium-fitness-score subgroup to conduct moderate-intensity mining and development in the vicinity of its current solution, hoping to find a better neighboring solution; and a fine-tuning operator is applied to the high-fitness-score subgroup to make subtle adjustments and improve the accuracy of its already relatively good solutions, making its performance even better.

[0081] For example, in an e-commerce movie recommendation scenario: a subgroup with low fitness scores (applying the global search operator): a candidate movie A has a very low fitness score because its "science fiction" tag is very different from the user's preferred "historical documentary". The global search operator will make significant adjustments to it, for example, replacing it with a new candidate movie B of a completely different type (such as "nature documentary") but which may have an implicit connection, in an attempt to explore entirely new recommendation possibilities.

[0082] Fitness Score Subgroup (Applying Local Development Operator): A candidate film C is a "war history film," which has some similarity to the user's preference for "historical documentaries," and scores moderately. The local development operator will explore its vicinity, for example, fine-tuning its weight to make it more biased towards "history" rather than "war," or recommending another film D, which also belongs to the "war history" genre but is directed by a more famous director.

[0083] High-fitness-score subgroup (applying fine-tuning operators): A candidate movie E is a "highly-rated historical documentary," which highly matches user preferences and scores very high. The fine-tuning operators do not change its core recommendation attributes, but rather perform extremely subtle optimizations, such as prioritizing the recommendation of the clearer 4K version, or slightly advancing its ranking in the recommendation list to improve the user experience.

[0084] Overall, this mechanism embodies the core idea of ​​algorithmic intelligence: tailoring optimization strategies based on the current performance (fitness score) of each individual (candidate service). It transforms the traditional single optimization approach into a hierarchical, focused collaborative optimization process, thereby ensuring global exploration capabilities while also considering local development efficiency and the accuracy of the optimal solution, greatly improving the efficiency and effectiveness of the optimization process.

[0085] In one possible implementation, the global search operator acts as a pioneer explorer in the adaptive ecological balance optimization algorithm, used for global search. The mathematical expression of the global search operator includes a balance factor that decreases with the number of iterations, and a random perturbation term. The local development operator acts as a balancer in the adaptive ecological balance optimization algorithm, used for local search. The mathematical expression of the local development operator is related to the neighborhood optimum of the current candidate service and the fitness score of the current candidate service. The fine-tuning operator acts as an elite follower in the adaptive ecological balance optimization algorithm. The mathematical expression of the fine-tuning operator is related to the global optimum, and the perturbation magnitude of the fine-tuning operator is negatively correlated with the fitness score of the global optimum. The global optimum is the candidate service with the highest fitness score after a preset time period of execution of the adaptive ecological balance optimization algorithm.

[0086] In the adaptive ecological balance optimization algorithm, different individuals (pioneer explorers, balancers, and elite followers) match needs through different strategies. The following are the update formulas for each role:

[0087] Pioneers: Pioneers are primarily responsible for global searching. Assuming the current individual's location is... Its update formula is:

[0088] in: The learning factor controls the convergence speed towards the optimal solution; It is the globally optimal solution; This serves as a perturbation factor, controlling the perturbation amplitude and increasing the diversity of the global search. The random perturbation is used to simulate the impact of environmental changes on an individual's location.

[0089] Balancers: Balancers primarily perform fine-grained searches within local regions. Assume the current local optimum is... Its update formula is:

[0090]

[0091] in: This is the step size factor, which controls the accuracy of the local search; The energy factor measures an individual's fitness in a local search. It is the fitness threshold, used to determine whether an individual needs local fine-tuning.

[0092] Elite Followers: Elite followers focus on fine-tuning around the current global optimum, assuming the global optimum is... Its update formula is:

[0093]

[0094] in: To update the factors and control the intensity of fine-tuning; This is a perturbation factor used to prevent overfitting and maintain a certain level of search diversity. It is the globally optimal fitness, and elite followers use this fitness value to guide further optimization.

[0095] It should be noted that in the execution framework of the adaptive ecological balance optimization algorithm, the global search operator, the local development operator, and the fine-tuning operator each play a unique and crucial role.

[0096] The global search operator acts as a pioneer explorer in the algorithm. Its core responsibility is to perform extensive global searches. Its mathematical expression contains a key balancing factor that is designed to decrease with the number of iterations to dynamically regulate the intensity of its exploration. It also includes a random perturbation term to introduce necessary randomness and ensure sufficient exploration of the solution space.

[0097] The local development operator acts as a balancer, focusing on local searches around the existing search area. Its mathematical expression is closely related to two factors: the neighborhood optimal solution of the current candidate service and the fitness score of the candidate service itself. It guides the optimization direction by combining information from nearby excellent individuals and its own performance.

[0098] The fine-tuning operator, acting as an elite follower, is responsible for the final precision enhancement of the discovered high-quality solutions. Its mathematical expression is directly related to the currently found global optimum by the algorithm, and its perturbation amplitude is designed to be negatively correlated with the fitness score of the global optimum. That is, the better the global optimum (the higher the score), the more subtle the adjustment amplitude of the fine-tuning operator, ensuring refined stability of the top solution. It is important to note that the global optimum here is defined as the candidate service with the highest fitness score in the current population after a preset time period of algorithm execution.

[0099] Overall, these three operators, through their ingenious mathematical definitions and division of roles, collectively constitute a collaborative, adaptive, and efficient optimization system. Global search ensures diversity, local exploitation is responsible for uncovering potential regions, and fine-tuning focuses on improving the quality of elite solutions. This design enables the algorithm to achieve an optimal balance between exploration and exploitation, breadth and depth, thereby efficiently and reliably driving the entire population toward the global optimum.

[0100] In one possible implementation, adaptively adjusting the sizes of the global search operator, the local development operator, and the fine-tuning operator according to the progress of the iteration process includes: as the iteration process progresses over time, the weight of the global search operator decreases, the weight of the fine-tuning operator increases, and the weight of the local development operator increases.

[0101] In each iteration, the adaptive ecological balance optimization algorithm dynamically adjusts the ratio of exploration to development. The update formula for the adaptive ecological balance factor is as follows:

[0102]

[0103] in: Indicates the first Ecological balance factors at the next iteration; To adjust the parameters and control the rate of change of the balance factor; To maximize the number of iterations, this factor ensures the algorithm switches phases at appropriate times. Initially, it increases the proportion of global search (exploration phase), while later it enhances local exploration capabilities (development phase).

[0104] It should be noted that in the adaptive ecological balance optimization algorithm, a t+1 This represents the "ecological balance factor," used to dynamically adjust the ratio of global exploration to local development. In the early stages of the algorithm, a... t+1 The smaller the number of iterations, the higher the global exploration weight; as the number of iterations increases, a... t+1 The gradual increase signifies a greater emphasis on development and refinement. Furthermore, the adjustment parameters are not fixed but designed to update dynamically with each iteration (typically a monotonically increasing function, taking positive values). Therefore, this aligns with the "exploration before development" strategy.

[0105] It should be noted that during the iterative process of the adaptive ecological balance optimization algorithm, the algorithm adaptively adjusts the weights or influence of the three core operators—the global search operator, the local development operator, and the fine-tuning operator—according to the progress of the iteration. Specifically, as the iteration progresses over time, the weight of the global search operator gradually decreases, while the weight of the fine-tuning operator increases simultaneously, and the weight of the local development operator also shows an increasing trend. This adjustment is a dynamic, continuous, and adaptive process, closely related to the iteration progress.

[0106] Overall, this adaptive adjustment mechanism is the core of the algorithm's "ecological balance" and "adaptive" characteristics. In the early stages of iteration, the high weight of the global search operator ensures that the algorithm explores the solution space extensively, avoiding premature entrapment in local optima. As iterations deepen, the algorithm gradually shifts its optimization focus to local exploration and fine-tuning. By increasing the weights of these two operators, it enhances the in-depth mining and accuracy improvement of potential high-quality solutions, thereby driving the population to converge quickly to the vicinity of the global optimum. This strategy cleverly simulates the shift from "exploration" to "fine-tuning" in natural evolution, greatly optimizing the algorithm's overall search efficiency and the quality of the final solution.

[0107] In one possible implementation, the iteration process stops when: the current iteration count reaches a preset maximum threshold, or the difference between the current highest fitness score and the previous best fitness score is less than a preset threshold.

[0108] The adaptive ecological balance optimization algorithm will determine the termination condition (such as the maximum number of iterations) during execution. The stability of the optimal solution determines whether to stop the iteration:

[0109]

[0110] in: This is the current iteration number; It is the change in fitness of the optimal solution; This is the minimum threshold for fitness changes. When the change is less than this value, the iteration stops. If the termination condition is met, the algorithm will output the final service recommendation result.

[0111] It should be noted that in the execution flow of the adaptive ecological balance optimization algorithm, the termination of the iteration process is determined by two explicit convergence conditions. First, the current iteration count has reached the algorithm's preset maximum threshold; this is a hard constraint to ensure computational efficiency and prevent the algorithm from falling into an infinite loop. Second, the difference between the highest fitness score generated in the current iteration and the best fitness score recorded in the previous iteration is less than a preset minimum threshold, indicating that the overall quality improvement of the population has stagnated and the algorithm has converged to a relatively stable state. The iterative optimization process stops when either of these conditions is met, or it can be set to stop when both conditions are met.

[0112] Therefore, it can be ensured that the adaptive ecological balance optimization algorithm can output a high-quality and stable list of recommended services within a reasonable time.

[0113] In one possible implementation, when the iteration process stops, a predetermined number of candidate services with the highest fitness scores are selected from the current population in descending order, and combined into a recommended service list. After outputting the list, the method further includes: obtaining feedback data from the customers to be processed regarding the recommended service list; optimizing the adaptive ecological balance optimization algorithm based on the feedback data to obtain the optimized adaptive ecological balance optimization algorithm; and continuing to execute the customer demand mining method based on the optimized adaptive ecological balance optimization algorithm.

[0114] The customer feature vectors optimized using an adaptive ecological balance optimization algorithm were used to generate accurate service recommendations. The calculation process for the recommendation results includes the following steps:

[0115] Matching degree calculation: For each customer Calculate all candidate services The matching degree is used to rank and obtain the optimal recommendation service:

[0116]

[0117] in: It is the first The final recommended service collection for each customer; This indicates selecting the top match. One service.

[0118] Personalized display: Final recommendation results We will personalize the display based on the customer's needs and characteristics to ensure that the recommended services are highly consistent with the customer's needs.

[0119] It's important to note that in the complete service recommendation process, after the iteration stops and the recommended service list is output, the system includes a crucial closed-loop learning and optimization phase. Specifically, the system actively acquires feedback data from customers regarding the recommended service list. This data directly reflects the effectiveness of the recommendations in real-world scenarios. Subsequently, based on this collected feedback data, the system makes targeted adjustments and optimizations to the internal parameters or strategies of the adaptive ecological balance optimization algorithm, resulting in a performance-enhanced adaptive ecological balance optimization algorithm. Finally, based on this optimized adaptive ecological balance optimization algorithm, the system continues to execute subsequent customer demand mining and service recommendation methods, forming a continuous self-evolutionary cycle.

[0120] For example, in a music streaming recommendation scenario: the system initially generates a recommendation list of 10 songs for the user. User feedback data shows that they listened to one jazz song in its entirety and added it to their favorites, but quickly skipped three rock songs. Based on this feedback data, the system analyzes the user's feedback and concludes that they have a strong preference for jazz and a lower interest in rock. Subsequently, it optimizes the adaptive ecological balance algorithm, for example, by adjusting the weight of the "genre" feature in the feature vector or reducing the probability of exploring rock in the global search operator. The optimized adaptive ecological balance algorithm will then be more inclined to recommend jazz and related genres to the same user the next time it performs recommendations, thus providing a more satisfactory recommendation list.

[0121] Overall, by incorporating user feedback as a crucial signal, the algorithm learns from real-world application results, continuously correcting and optimizing itself. This significantly enhances the adaptability and intelligence of the recommendation system, enabling it to better track changes in user preferences and understand deeper needs, thus providing more accurate and personalized service recommendations in the future.

[0122] In summary, in the embodiments of this application (such as...) Figure 3 As shown in the diagram, this method effectively generates a personalized list of recommended services by obtaining the customer feature vector of the customer to be processed and the service feature vector of the candidate services, and inputting them into an adaptive ecological balance optimization algorithm. Specifically, the initial population contains at least one candidate service, and the fitness score of each candidate service is calculated based on a preset fitness function, thereby ensuring that the recommended services are highly matched with customer needs. During the iterative optimization process, global search operators, local development operators, and fine-tuning operators are used for dynamic adjustment, allowing the algorithm to flexibly switch between global and local approaches and avoid the trap of local optima. In addition, as the iteration progresses, the algorithm adaptively adjusts the size of the operators according to the fitness score, enhancing the flexibility and efficiency of the optimization process. Finally, when the iteration stops, the candidate service with the highest fitness score is selected from the current population to form a list of recommended services and output. This method can significantly improve the accuracy and real-time performance of recommended services, improve the accuracy and efficiency of customer demand prediction, and meet the personalized needs of customers.

[0123] Figure 4 This application illustrates a customer-demand-oriented excavation apparatus according to an embodiment of the present application, such as... Figure 4 As shown, the device 40 includes:

[0124] The acquisition module 401 is used to acquire the customer feature vector of the customer to be processed and the service feature vector of the candidate service;

[0125] The execution module 402 is used to input customer feature vectors and service feature vectors into the adaptive ecological balance optimization algorithm and output a recommended service list for the customer to be processed, wherein the recommended service list includes at least one candidate service.

[0126] The process involves inputting customer feature vectors and service feature vectors into an adaptive ecological balance optimization algorithm, which outputs a list of recommended services for the customers to be processed, including:

[0127] Initialize a population containing at least one candidate service;

[0128] Based on the preset fitness function, customer feature vector and service feature vector, calculate the fitness score of each candidate service in the current population relative to the customer to be processed.

[0129] Based on the fitness score, the population is iteratively optimized using global search operators, local exploitation operators, and fine-tuning operators. In each iteration, global search operators, local exploitation operators, and fine-tuning operators are dynamically allocated and applied to candidate services based on the fitness score to generate a new population. The size of global search operators, local exploitation operators, and fine-tuning operators is adaptively adjusted according to the progress of the iteration process.

[0130] When the iteration process stops, select a preset number of candidate services with the highest fitness scores from the current population in descending order, combine them into a recommended service list and output it.

[0131] In one possible implementation, the acquisition module 401 is further configured to acquire customer data of the customer to be processed; after preprocessing the customer data, feature extraction is performed on the preprocessed customer data, and the extracted features are concatenated into a customer feature vector; service data of candidate services is acquired; after preprocessing the service data, feature extraction is performed on the preprocessed service data, and the extracted features are concatenated into a service feature vector.

[0132] In one possible implementation, the execution module 402 is further configured to calculate the dot product of the service feature vector of the candidate service and the customer feature vector of the customer to be processed; calculate the Euclidean norm of the service feature vector of the candidate service and the customer feature vector of the customer to be processed; and determine the fitness score based on the dot product and the Euclidean norm.

[0133] In one possible implementation, the execution module 402 is further configured to divide the population into a high-fitness-score subgroup, a medium-fitness-score subgroup, and a low-fitness-score subgroup based on the fitness scores; assign and apply a global search operator to the low-fitness-score subgroup, assign and apply a local exploitation operator to the medium-fitness-score subgroup, and assign and apply a fine-tuning operator to the high-fitness-score subgroup to generate a new population.

[0134] In one possible implementation, the execution module 402 is further configured to decrease the weight of the global search operator, increase the weight of the fine-tuning operator, and increase the weight of the local development operator as the iteration process progresses over time.

[0135] In one possible implementation, the iteration process stops when: the current iteration count reaches a preset maximum threshold, or the difference between the current highest fitness score and the previous best fitness score is less than a preset threshold.

[0136] In one possible implementation, the global search operator acts as a pioneer explorer in the adaptive ecological balance optimization algorithm, used for global search. The mathematical expression of the global search operator includes a balance factor that decreases with the number of iterations, and a random perturbation term. The local development operator acts as a balancer in the adaptive ecological balance optimization algorithm, used for local search. The mathematical expression of the local development operator is related to the neighborhood optimum of the current candidate service and the fitness score of the current candidate service. The fine-tuning operator acts as an elite follower in the adaptive ecological balance optimization algorithm. The mathematical expression of the fine-tuning operator is related to the global optimum, and the perturbation magnitude of the fine-tuning operator is negatively correlated with the fitness score of the global optimum. The global optimum is the candidate service with the highest fitness score after a preset time period of execution of the adaptive ecological balance optimization algorithm.

[0137] In one possible implementation, the execution module 402 is further configured to, when the iteration process stops, select a preset number of candidate services with the highest fitness scores from the current population in descending order, combine them into a recommended service list, and then obtain feedback data from the customers to be processed regarding the recommended service list; based on the feedback data, optimize the adaptive ecological balance optimization algorithm to obtain the optimized adaptive ecological balance optimization algorithm; and based on the optimized adaptive ecological balance optimization algorithm, continue to execute the customer demand mining method.

[0138] In summary, the customer demand mining method shown in the embodiments of this application has the following advantages:

[0139] Improving the accuracy of customer demand mining: The adaptive ecological balance optimization algorithm can effectively avoid local optima problems by collaboratively searching for individuals in multiple ecological niches, thereby improving the accuracy of customer demand analysis.

[0140] Optimizing computational efficiency: The adaptive search mechanism and energy regulation strategy of the adaptive ecological balance optimization algorithm greatly improve the convergence speed of the algorithm and reduce the consumption of computational resources.

[0141] Enhancing system intelligence: The adaptive ecological balance optimization algorithm can adaptively adjust the search strategy according to the diversity and complexity of customer data, improving the flexibility and accuracy of matching customer needs.

[0142] Furthermore, the customer demand mining method shown in this application embodiment can be applied in a customer data management system based on an Internet information cloud service platform and combined with artificial intelligence algorithms, and has the following technical advantages:

[0143] Adaptive Ecological Balance Optimization Algorithm: By introducing the co-evolutionary mechanism of the ecosystem, it achieves a balance between global search and local optimization, making it particularly suitable for complex customer demand mining and matching scenarios. Compared to traditional optimization algorithms (such as genetic algorithms), the adaptive ecological balance optimization algorithm has advantages in its adaptability, flexibility, and accuracy during the optimization process. It can more accurately meet dynamically changing customer needs and improve algorithm efficiency. The following are the main advantages of the adaptive ecological balance optimization algorithm:

[0144] Dynamic adaptive optimization mechanism: By introducing an adaptive ecological balance mechanism, the ratio of global search and local optimization is dynamically adjusted during the optimization process, so that the algorithm can automatically switch strategies according to the needs of different stages.

[0145] Enhanced global search capabilities: Global search capabilities are enhanced through multiple search paths and niche roles (such as pioneer explorers, balancers, and elite followers), enabling effective exploration within a broader solution space.

[0146] Precise customer demand matching: By simulating the co-evolution mechanism in natural ecosystems, customer needs are dynamically matched with service feature vectors to achieve more accurate demand matching and service recommendations.

[0147] Enhanced flexibility and stability: Supported by the ecological balance mechanism, it can flexibly respond to complex and changing customer needs, and the optimization process is stable, avoiding large fluctuations.

[0148] Autonomous Discovery and Intelligent Conversion of Potential Customers: This application utilizes a self-developed adaptive ecological balance optimization algorithm to autonomously discover potential customers requiring mobile communication services online. By accurately searching for the identity information of potential customers, it automatically matches them with the most suitable service items. This technology enables intelligent and automated management of existing and potential customers within the customer data system, significantly improving the efficiency and accuracy of acquiring new customers. In summary, this application not only enhances the management efficiency of existing customer data but also expands the capabilities of the customer data system through intelligent discovery and conversion mechanisms, making it more user-friendly and automated in actual business operations.

[0149] Furthermore, the technical solution of this application can be widely applied to multiple fields such as customer management, precision marketing, and service recommendation, and has significant advantages, especially in the following practical scenarios:

[0150] Customer needs mining in the telecommunications industry: Through intelligent customer data analysis, the Sloth algorithm can mine and predict the service needs of existing customers in real time, and provide personalized service recommendations based on user behavior, thereby improving customer satisfaction and loyalty.

[0151] Intelligent identification and conversion of potential customers: In the market expansion process of large enterprises, the Sloth algorithm can automatically identify the identity information of potential customers and intelligently match suitable services, driving the conversion of potential customers into existing customers. This technology can significantly improve the conversion efficiency of marketing.

[0152] Cross-platform data integration and analysis: This invention effectively addresses the lack of flexibility and real-time performance in traditional CRM systems when handling large-scale customer data. By leveraging the data collection and processing capabilities of a cloud platform, combined with the adaptive optimization of the sloth algorithm, enterprises can achieve more intelligent cross-platform data analysis and decision support.

[0153] This application provides an electronic device 50, such as... Figure 5 As shown, the electronic device 50 includes a processor 501, a memory 502, and a program stored in the memory 502 and executable on the processor 501. When the program is executed by the processor 501, it implements the steps of a customer demand mining method as shown in the above embodiment.

[0154] This application also provides a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the steps of a customer demand mining method as shown in the above embodiments, achieving the same technical effect. To avoid repetition, it will not be described again here. The computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0155] This application also provides a computer program product, including computer instructions. When executed by a processor, the computer instructions implement the steps of the customer demand mining method shown in the above embodiments and achieve the same technical effect. To avoid repetition, they will not be described again here.

[0156] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0157] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0158] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A method of mining customer needs, characterized by, The method comprises: obtaining a customer feature vector of a to-be-processed customer and a service feature vector of a candidate service; inputting the customer feature vector and the service feature vector into an adaptive ecological balance optimization algorithm to output a recommended service list of the to-be-processed customer, wherein the recommended service list comprises at least one candidate service; wherein inputting the customer feature vector and the service feature vector into the adaptive ecological balance optimization algorithm to output the recommended service list of the to-be-processed customer comprises: initializing a population, the population comprising at least one candidate service; calculating, based on a preset fitness function, the customer feature vector and the service feature vector, a fitness score of each candidate service in the current population relative to the to-be-processed customer; iteratively optimizing the population according to the fitness score by using a global search operator, a local development operator and a fine tuning operator, wherein in each iteration optimization process, the global search operator, the local development operator and the fine tuning operator are dynamically assigned and applied to the candidate service according to the fitness score to generate a new population, and the size of the global search operator, the local development operator and the fine tuning operator is adaptively adjusted according to the progress of the iteration process; when the iteration process stops, selecting, from the current population, a preset number of candidate services with the highest fitness score in descending order to combine into the recommended service list and output.

2. The method of claim 1, wherein, obtaining a customer feature vector of a to-be-processed customer and a service feature vector of a candidate service comprises: obtaining customer data of the to-be-processed customer; after preprocessing the customer data, extracting features from the preprocessed customer data, and splicing the extracted features into the customer feature vector; obtaining service data of the candidate service; after preprocessing the service data, extracting features from the preprocessed service data, and splicing the extracted features into the service feature vector.

3. The method of claim 1, wherein, calculating, based on a preset fitness function, the customer feature vector and the service feature vector, a fitness score of each candidate service in the current population relative to the to-be-processed customer comprises: calculating the dot product of the service feature vector of the candidate service and the customer feature vector of the to-be-processed customer; calculating the Euclidean norm of the service feature vector of the candidate service and the customer feature vector of the to-be-processed customer; determining the fitness score according to the dot product and the Euclidean norm.

4. The method of claim 1, wherein, dynamically assigning and applying the global search operator, the local development operator and the fine tuning operator to the candidate service according to the fitness score to generate a new population comprises: dividing the population into a high fitness score subgroup, a medium fitness score subgroup and a low fitness score subgroup according to the fitness score; assigning and applying the global search operator to the low fitness score subgroup, assigning and applying the local development operator to the medium fitness score subgroup, and assigning and applying the fine tuning operator to the high fitness score subgroup to generate a new population.

5. The method of claim 1, wherein, The self-adaptive adjustment of the sizes of the global search operator, the local development operator and the fine-tuning operator according to the progress of the iteration process comprises: As the iteration process advances in time, the weight of the global search operator decreases, the weight of the fine-tuning operator increases, and the weight of the local development operator increases.

6. The method of claim 1, wherein, The stopping of the iteration process comprises: The current iteration number reaches a preset maximum threshold, or the difference between the current highest fitness score and the last optimal fitness score is less than a preset threshold.

7. The method of claim 1, wherein, When the iteration process stops, the method further comprises: Obtaining feedback data of the recommended service list from the to-be-processed customer; Based on the feedback data, the self-adaptive ecological balance optimization algorithm is optimized to obtain an optimized self-adaptive ecological balance optimization algorithm; Based on the optimized self-adaptive ecological balance optimization algorithm, the customer demand mining method is continued to be executed.

8. A client demand mining apparatus characterized by comprising: The device comprises: An acquisition module for acquiring a customer feature vector of a to-be-processed customer and a service feature vector of a candidate service; An execution module for inputting the customer feature vector and the service feature vector into a self-adaptive ecological balance optimization algorithm to output a recommended service list of the to-be-processed customer, wherein the recommended service list comprises at least one candidate service; Wherein, inputting the customer feature vector and the service feature vector into the self-adaptive ecological balance optimization algorithm to output the recommended service list of the to-be-processed customer comprises: Initializing a population, the population comprising at least one candidate service; Based on a preset fitness function, the customer feature vector and the service feature vector, calculating the fitness score of each candidate service in the current population relative to the to-be-processed customer; According to the fitness score, using a global search operator, a local development operator and a fine-tuning operator to iteratively optimize the population, wherein in each iteration optimization process, according to the fitness score, the global search operator, the local development operator and the fine-tuning operator are dynamically assigned and applied to the candidate service to generate a new population, and the sizes of the global search operator, the local development operator and the fine-tuning operator are self-adaptively adjusted according to the progress of the iteration process; When the iteration process stops, the method further comprises:

9. An electronic device, comprising: Selecting, from the current population, a preset number of candidate services with the highest fitness score in descending order to combine into the recommended service list and output. Comprise:

10. A computer-readable storage medium, characterized in that, A processor, a memory and a program stored on the memory and executable on the processor, the program being executed by the processor to implement the steps of a customer demand mining method according to any one of claims 1 to 7. The computer readable storage medium stores a computer program, and the computer program is executed by the processor to implement the steps of a customer demand mining method according to any one of claims 1 to 7.

11. A computer program product, characterised in that, Computer program comprising computer instructions which, when executed by a processor, implement the steps of a client needs mining method as claimed in any one of claims 1 to 7.

Images