Digital banquet marketing management method based on multi-dimensional decision algorithm

Abstract

The invention discloses a digital banquet marketing management method based on a multi-dimensional decision algorithm, and relates to the field of digital banquet marketing management. According to the method, multi-dimensional extraction and coding of banquet features are realized through a self-adaptive excitation generation optimization algorithm based on event feature vectors, and the excitation effect is predicted in combination with historical implementation records, so that an optimal excitation scheme template is generated. And furthermore, comprehensive scoring is performed on real-time banquet records through a multi-modal verification algorithm and a confidence fusion algorithm, so that accurate evaluation and automatic reward distribution of the incentive effect are realized, the incentive scheme is more intelligent, personalized and efficient, and the management level and customer participation degree of banquet marketing activities are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of digital banquet marketing management, specifically a digital banquet marketing management method based on a multi-dimensional decision-making algorithm. Background Technology

[0002] With the development of digital marketing and corporate event management, existing digital banquet management systems typically receive banquet reservation information submitted by customers or stores through headquarters or distributor systems, and register and manage the scale, time, location, and basic customer information of the banquet. Some systems combine customers' historical consumption records or past event participation data to generate incentive programs using fixed templates, with incentive methods including coupons, points, or small gifts. However, these systems mostly rely on static rules and manual intervention, lacking intelligent analysis capabilities for the multidimensional characteristics of banquet events. When handling banquets of different sizes, times, and locations, existing systems typically cannot deeply analyze customer behavior, nor can they achieve personalized allocation and dynamic optimization of incentive programs. Furthermore, monitoring the banquet implementation process largely relies on manual uploading or single data records, failing to achieve real-time, comprehensive event data analysis.

[0003] However, existing technologies have significant shortcomings. First, these systems have limited capabilities in extracting features from banquet events, failing to effectively integrate multi-dimensional information such as banquet scale, time distribution, geographical clustering, and customer behavior, resulting in a lack of targeted incentive programs. Second, incentive program generation is mostly based on fixed templates or simple weighted allocation, failing to fully utilize historical banquet implementation records for dynamic optimization, thus making it difficult to guarantee optimal incentive effects and easily wasting incentive resources. Third, existing systems rely heavily on single-modal data or manual verification for the verification and evaluation of real-time banquet records, unable to comprehensively analyze structured data, images, videos, and text logs for multi-modal analysis, leading to problems such as delayed, low-precision, and insufficiently intelligent incentive reward distribution. These shortcomings limit the management efficiency and customer engagement effectiveness of enterprises in banquet marketing activities. Summary of the Invention

[0004] This invention proposes a digital banquet marketing management method based on a multidimensional decision-making algorithm. This method aims to solve the problems of existing digital banquet systems that cannot efficiently and accurately predict incentive effects, lack multidimensional event feature analysis and real-time verification capabilities. Through systematic event feature modeling, algorithm optimization and multimodal verification, it realizes intelligent management of the entire process from banquet demand reception, feature extraction, incentive scheme generation to real-time scoring and reward distribution.

[0005] One digital banquet marketing management method based on a multidimensional decision-making algorithm includes the following steps: S1. Receive banquet demand data from customer terminals, distributors, or terminal stores through the headquarters system, and perform format verification and time-space normalization processing on the received banquet demand data to obtain a standardized structured input dataset. Specifically, in step S1, the headquarters system receives banquet request data submitted by customer terminals, distributors, or terminal stores. This data includes banquet identifiers, customer identifiers, reservation dates and times, banquet locations, and banquet scale, where banquet scale involves the number of tables and the estimated number of attendees. Due to the diverse data sources and formats, the system first performs structured parsing and integrity checks on the received data, including verifying whether fields are complete, data types match, reservation times are within the allowed range, and banquet location coordinates are within the serviceable area. Simultaneously, it identifies outliers and corrects data to eliminate data deviations caused by user input errors or system differences. In terms of time and space, the system uniformly adjusts the time zone for reservation times, normalizes date formats, and divides time periods. It also standardizes geographical coordinates, such as normalizing latitude and longitude or using a gridded representation, to ensure that data from different sources can be compared and analyzed at the same scale. After these processes, the system transforms the raw data into a standardized structured input dataset. This dataset stores various fields in tabular or vector form, providing a consistent data foundation for subsequent feature extraction and vector encoding. It also solves the problems in existing technologies where it is difficult to process multi-source heterogeneous data in a unified manner and where inconsistent data formats make it difficult for algorithms to perform direct calculations.

[0006] S2. Based on the standardized structured input dataset, extract key feature parameters related to banquet size, time distribution, geographical clustering, and customer behavior, and generate event feature vectors through encoding algorithms; Specifically, in step S2, the system extracts key feature parameters related to banquet size, time distribution, geographic clustering, and customer behavior from the standardized structured input dataset obtained in S1. First, for the banquet size field, it calculates the total number of tables, average number of people per table, and total expected number of participants, and performs normalization and standardization processing to eliminate dimensional differences and numerical deviations between banquets of different sizes. Then, it encodes the reservation time, converts the date into a periodic numerical representation, divides the time period, and adds holiday identifiers to reflect the time distribution pattern. For the banquet location, the system analyzes the historical geographical distribution of banquets through clustering algorithms, maps spatial locations to category labels and encodes them, quantifies geographic clustering features, and considers the surrounding historical participation density and regional preferences to capture geographic correlations. Customer behavior features are calculated based on customer identifiers and historical participation records, including participation frequency, average participation size, probability of responding to incentive activities, and participation pattern indicators, and these features are standardized. The system concatenates banquet scale features, time distribution features, geographical clustering features, and customer behavior features in a fixed order to form an event feature vector. This transforms discrete, continuous, and multi-source information into a computable unified vector representation, enabling subsequent algorithms to perform similarity calculations and optimization on event features in a unified dimensional space. This solves the problem of prediction bias caused by feature heterogeneity or lack of unified representation in traditional methods. At the same time, the system incorporates the analysis of historical participation patterns during feature generation to enhance the expressive power of event feature vectors in optimization algorithms.

[0007] S3. Based on the event feature vector and combined with historical banquet implementation records, generate an incentive scheme template through an adaptive incentive generation optimization algorithm; Specifically, in step S3, the system matches and analyzes the event feature vectors generated in step S2 with historical banquet implementation records. First, the historical records are standardized, including cleaning and normalizing the historical event feature vectors and their corresponding incentive response indicators, removing abnormal or missing data to ensure data quality. Then, the matching degree between the current event and historical events is quantified through vector similarity measurement. Several of the most relevant historical records are selected to form a reference set. The system calculates a weighted prediction of the incentive effect based on the incentive response of the historical records and their similarity, achieving effect estimation based on historical data. On this basis, the system performs an adaptive search in the candidate incentive scheme set, controlling the executability of candidate schemes through constraints, such as limiting the total number of incentives, time windows, and resource usage, while dynamically adjusting the parameters of the candidate schemes to maximize the predicted incentive effect. This process organically combines event feature vectors, historical patterns, and constraints to achieve adaptive generation of incentive scheme templates. The entire method solves the problems of traditional static schemes being unable to be adjusted under different event scenarios and having difficulty predicting incentive effects. Technically, it constructs a closed-loop decision-making logic through data standardization, feature matching, similarity weighting, and constraint optimization search mechanisms.

[0008] S4. Send the incentive scheme generated by the algorithm and its corresponding event parameters to the target dealer or terminal store node; Specifically, in step S4, the headquarters system sends the incentive scheme template generated in S3 and its corresponding event feature parameters to the target distributors or terminal store nodes through the system distribution module. To ensure the reliability and consistency of the distribution process, the system standardizes the encoding of the scheme and event parameters and verifies the integrity of the transmission. Simultaneously, it retains the mapping relationship between event identifiers and schemes in the message to prevent misalignment or data loss during parsing by different terminals. The system also records the version and timestamp of the distributed data, enabling terminals to execute incentive activities based on the latest scheme and maintain consistency with headquarters data. The distribution process not only ensures that terminals can parse the scheme and execute according to the template, but also solves the problems of inconsistent execution information across multiple terminals and low matching degree between schemes and event parameters in existing technologies through a unified encoding and mapping mechanism. It also provides a traceable data foundation for subsequent real-time recording and uploading and incentive verification. Technically, the entire process constructs a data consistency guarantee mechanism through encoding, mapping, transmission verification, and version control, achieving a closed-loop connection between the algorithm-generated scheme and actual execution, thereby ensuring that the adaptive incentive algorithm can be implemented in practical applications and maintains logical accuracy.

[0009] The beneficial effects of the invention are: This approach utilizes an adaptive incentive generation optimization algorithm based on event feature vectors to extract and encode banquet features from multiple dimensions. It then combines this with historical implementation records to predict incentive effects, thereby generating an optimal incentive scheme template. Furthermore, a multimodal validation algorithm and a confidence fusion algorithm are used to comprehensively score real-time banquet records, enabling accurate evaluation of incentive effects and automatic reward distribution. This makes the incentive scheme more intelligent, personalized, and efficient, significantly improving the management level and customer engagement of banquet marketing activities. Attached Figure Description

[0010] Figure 1 This is a schematic diagram of a digital banquet marketing management method based on a multidimensional decision-making algorithm, according to an embodiment of the present invention. Detailed Implementation

[0011] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings, but the scope of protection of the present invention is not limited to the following description.

[0012] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention; that is, the described embodiments are only a part of the embodiments of the invention, and not all of them. The components of the embodiments of the invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0013] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention. It should be noted that relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations.

[0014] Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or machine 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 machine. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or machine that includes said element.

[0015] The features and performance of the present invention will be further described in detail below with reference to embodiments.

[0016] Example 1 Among them, such as Figure 1 A digital banquet marketing management method based on a multidimensional decision-making algorithm includes the following steps: S1. Receive banquet demand data from customer terminals, distributors, or terminal stores through the headquarters system, and perform format verification and time-space normalization processing on the received banquet demand data to obtain a standardized structured input dataset. S2. Based on the standardized structured input dataset, extract key feature parameters related to banquet size, time distribution, geographical clustering, and customer behavior, and generate event feature vectors through encoding algorithms; S3. Based on the event feature vector and combined with historical banquet implementation records, generate an incentive scheme template through an adaptive incentive generation optimization algorithm; S4. Send the incentive scheme generated by the algorithm and its corresponding event parameters to the target dealer or terminal store node.

[0017] Furthermore, the banquet demand data includes: banquet identifier, customer identifier, reservation date and time, banquet location, and banquet scale, wherein the banquet scale includes the number of tables and the estimated number of attendees.

[0018] Specifically, the implementation process of the above embodiments is as follows: In S1, the headquarters system receives banquet request data from customer terminals, distributors, or terminal stores via a network interface. This data includes banquet identifiers, customer identifiers, reservation dates and times, banquet locations, and banquet scale (including the number of tables and estimated attendees). The system first parses the received data, mapping the data structures from different sources to a unified internal standardized data model. It also performs type checks and logical validations, such as verifying date validity, time accuracy, and geographical location within the service area. The system further processes the multi-source time fields, including converting them to standard time zones, normalizing dates and times, and dividing them into fixed time intervals for subsequent quantitative analysis of time distribution. Geographic coordinates are mapped to a unified grid or cluster labels, and the system automatically corrects or marks deviations from the service area or abnormal geographical locations to ensure data consistency. For the banquet scale field, the system normalizes the number of tables and estimated attendees, generating a distribution vector representing the number of attendees at each table and the total number of participants, forming a multi-dimensional numerical array.

[0019] In S2, the system uses a standardized structured input dataset as a foundation to extract multidimensional feature parameters of banquet events and generate event feature vectors. For banquet scale, the system calculates the total number of tables, average number of attendees per table, and total expected number of participants, and normalizes and standardizes these values ​​to ensure comparability between banquets of different scales. Time feature processing includes encoding reservation dates into periodic vector representations representing daily, weekly, and monthly cycles, and dividing them into multiple time periods to capture reservation time distribution patterns. Holidays and special dates are also binary-labeled. Geographic feature processing employs clustering analysis algorithms to map banquet locations to category labels and encode them. The labels are optimized and adjusted by referencing historical geographic distribution patterns and surrounding clustering, ensuring that nearby locations or high-frequency participation areas are labeled as belonging to the same category. Customer behavior characteristics are generated by analyzing historical customer records, including participation frequency, average participation size, incentive response patterns, and behavioral trend indicators. The system integrates all extracted features in a fixed order and transforms them into event feature vectors through an encoding algorithm, so that discrete, continuous, and multi-source heterogeneous features can be uniformly represented as computable numerical vectors with fixed vector dimensions and each dimension representing specific feature information. In S3, the system uses event feature vectors as its core, comparing them with historical banquet implementation records. Historical records are cleaned and standardized, including normalizing historical event feature vectors and incentive result data, and removing outlier or missing records. The system calculates the matching degree between the current event and each historical event feature using similarity metrics (such as cosine similarity or Euclidean distance), selecting the highest-matching historical events to form a reference set. The incentive results in the reference set are weighted and combined based on matching degree to form a predictive index for the possible incentive response to the current event. The system searches through a set of candidate incentive schemes, filtering feasible schemes based on constraints, including total incentive limits, execution time windows, resource availability, and logical rules. During the search process, the system dynamically adjusts scheme parameters based on historical event similarity, iteratively optimizing the incentive type, quantity, and allocation strategy in the schemes, ultimately generating an incentive scheme template for the current event. The template includes the scheme structure, applicable conditions, and parameter configuration. In S4, the headquarters system sends the generated incentive scheme template and event feature vectors to target distributors or terminal store nodes via the distribution module. Before distribution, the system performs unified encoding and standardization on the scheme template and feature vectors, including fixing the field order, standardizing data types, and packaging messages to ensure that different terminals can correctly parse them during transmission. The system retains event identifiers and scheme mapping relationships in the messages to ensure that terminals can accurately apply the schemes to the corresponding events. Integrity verification and version control mechanisms are added to the data transmission, including timestamps, checksums, and message sequence numbers, to prevent data loss, duplication, or misparsing. After parsing, the receiving node can directly execute incentive-related operations based on the scheme template, and maintains event mapping logic consistent with the headquarters data during execution. This ensures a continuous and traceable data link between the generated scheme and event features, realizing a complete closed-loop logic from feature extraction and scheme generation to distribution.

[0020] Furthermore, step S2 specifically includes the following sub-steps: S201. Based on the standardized structured input dataset, extract the banquet scale features, time distribution features, geographical clustering features, and customer behavior features respectively; S202. Unify the encoding of banquet scale characteristics, time distribution characteristics, geographical clustering characteristics, and customer behavior characteristics into numerical vectors; S203. Merge the various feature vectors to obtain the event feature vector: ; Among them, the Represents the event feature vector, the Indicating the scale and characteristics of the banquet, the Representing the time distribution characteristics, the Representing geographical clustering features, the This indicates customer behavior characteristics.

[0021] Specifically, the above implementation process is based on a standardized structured input dataset. Multi-dimensional feature extraction and unified encoding are performed on banquet events. First, features are extracted from four dimensions: banquet scale, time distribution, geographic clustering, and customer behavior. Banquet scale features are represented numerically by analyzing the total number of tables, the number of people per table, and the expected number of participants, and then normalizing the data. Time distribution features map reservation dates and times into periodic vectors, including daily, weekly, and monthly cycle information and time period divisions. Holidays and special dates are also identified to quantify time patterns. Geographic clustering features generate category labels based on historical banquet location distribution using clustering algorithms, and weight optimization is performed by combining surrounding density information to ensure consistent representation of adjacent or high-frequency areas. Customer behavior features generate quantifiable indicators by analyzing historical participation records, incentive response patterns, and participation frequency, and are then standardized to form numerical vectors. Subsequently, the system converts these heterogeneous features into numerical vectors through unified encoding, enabling discrete, continuous, and multi-source features to be computed and compared in the same dimensional space. Banquet scale, time, geographic, and customer behavior features are all quantifiable. Finally, the system merges the various encoded vectors in a fixed order to generate a complete event feature vector.

[0022] Furthermore, in step S201, the banquet scale characteristics are obtained by quantifying the number of banquet tables and the expected number of attendees in the standardized structured input dataset; the time distribution characteristics are obtained by periodically encoding the reservation dates and times in the standardized structured input dataset; the geographic clustering characteristics are obtained by performing cluster analysis on the banquet locations in the standardized structured input dataset; and the customer behavior characteristics are obtained by statistically analyzing customer identifiers in the standardized structured input dataset in conjunction with historical banquet implementation records to obtain customer participation frequency, average participation scale, and historical incentive response, and the statistical results are quantified; the historical banquet implementation records are obtained from the historical database of the headquarters system.

[0023] Furthermore, step S3 specifically includes the following sub-steps: S301. Extract the feature vector of the historical banquet and the corresponding execution results of the incentive scheme based on the historical banquet implementation records; S302. Calculate the similarity between the event feature vector of the current event and the feature vector of the historical banquets, and select the top k historical banquet implementation records as the reference set based on the similarity. S303. Weight the historical stimulus responses of the reference set to predict the stimulus effect of the current event and obtain the expected effect index of each optional stimulus scheme under the current event; S304. Select the incentive scheme with the best incentive effect according to the set constraints, and obtain the incentive scheme template by standardizing and normalizing the scheme.

[0024] Specifically, taking the event feature vector generated in step S2 as the core, the system first extracts the corresponding historical banquet feature vector and its incentive scheme execution results from the historical banquet implementation records. The historical records are cleaned and standardized to ensure that the feature dimensions are uniform and the incentive result data is complete and quantifiable. Subsequently, the system calculates the similarity between the event feature vector of the current event and the historical banquet feature vector, usually using cosine similarity or other high-dimensional distance metrics to quantify the matching degree between each historical record and the current event. Based on the calculation results, the system selects the top k with the highest similarity. A reference set is formed from historical banquet records, containing the most relevant historical events and their corresponding incentive scheme execution results. Next, the system weights the historical incentive response data in the reference set, dynamically allocating weights based on the similarity between historical records and the current event. Weighted aggregation yields the predicted performance index for each candidate incentive scheme under the current event, mapping complex historical response patterns to the numerical expectation of the current event. Finally, the system compares and filters the expected performance index within the candidate incentive scheme set based on set constraints, including incentive quantity limits, execution time windows, and logical rules, selecting the optimal incentive scheme. A standardization and normalization process generates an incentive scheme template, which includes the incentive scheme's structural parameters, operable values, and applicable conditions.

[0025] Furthermore, in step S302, the similarity calculation process is specifically represented as follows: ; Among them, the Represents the event feature vector, the Let i represent the feature vector of the i-th historical banquet, the This represents the numerical similarity between the event feature vector and the historical banquet feature vector.

[0026] Furthermore, in step S3, the process of predicting the incentive effect of the current event is specifically represented as follows: ; ; Among them, the This indicates the incentive effect of the current event, the... Indicates the candidate incentive scheme, the Represents a reference set, the It means that the This indicates the incentive effect of historical banquet implementation records. The weighting coefficients representing the historical banquet implementation records, the stated Let represent the feature vector of the j-th historical banquet.

[0027] Furthermore, in step S304, the constraints include: Total incentive quantity constraint, used to ensure that the total number of incentives allocated to each incentive type in the incentive scheme does not exceed the maximum available incentive quantity preset by the system; Single-type incentive quantity constraint, used to constrain the quantity of each incentive type allocated in the scheme to be no less than the preset minimum value and no more than the preset maximum value; The time window constraint is used to ensure that the implementation time of the incentive activity is within the allowed time period of the banquet; Resource constraints are used to ensure that the total amount of resources required for the incentive scheme does not exceed the amount of available resources. The total amount of resources includes budget, inventory, and system capacity. Event applicability constraints are used to restrict incentive schemes to only be applied to banquet event characteristics that are suitable for the scheme.

[0028] Specifically, the selection of incentive schemes must meet multi-dimensional constraints to ensure that the schemes are executable under the current event and that their internal logic is consistent. The total incentive quantity constraint sums the quantities of each incentive type in the candidate incentive schemes and compares the total quantity with the system's preset maximum available incentive quantity. The system automatically eliminates combinations exceeding the limit during scheme generation and selection to ensure that the generated schemes are quantifiable and can be processed and allocated by the system. The single-type incentive quantity constraint sets upper and lower limits for the allocation quantity of each incentive type. The system verifies and adjusts the quantity of each incentive type during candidate scheme construction to ensure that each type is neither lower than the preset minimum value to ensure statistical operability, nor higher than the maximum value to avoid exceeding execution capacity or resource limitations. This constraint is applied to the scheme parameter space through dynamic boundary conditions in the optimization search. The time window constraint maps the allowed time period of the banquet to a time-coded interval. The system matches and verifies the execution time of the incentive activities during incentive scheme generation and selection, eliminating all those exceeding the allowed time. The time-segmented activities ensure the plan is perfectly aligned with the banquet scheduling in terms of time. Resource constraints involve calculating the total resources required for the incentive plan, including budget consumption, inventory usage, and system processing capacity. When generating a plan, the system summarizes the resource consumption of each incentive type and compares it with the current available resources. If a plan exceeds the resource limit, the incentive allocation is automatically adjusted or unsuitable plans are eliminated, thus ensuring the feasibility of resource allocation and the continuity of execution. Event applicability constraints are implemented through matching event feature vectors. The system logically compares the applicable feature range of candidate incentive plans with the feature vector of the current event. Only plans that fully match or conform to the adaptation rules are retained for further optimization and standardization. This mechanism ensures that the incentive plan is operable in the event feature space and consistent with the event logic, thereby completing the screening and generation of incentive plans under multi-dimensional constraints.

[0029] Furthermore, in step S304, the process of selecting the incentive scheme with the best incentive effect is specifically expressed as follows: ; Among them, the The incentive scheme that provides the best incentive effect is described in the following text. Indicates output The largest The Denotes the set of candidate incentive schemes, the Indicates the candidate incentive scheme, the This indicates the motivating effect of a preceding event.

[0030] Specifically, the constraints Represented as: ; Among them, the This represents a set of constraints, including total incentive quantity constraints, single-type incentive quantity constraints, time window constraints, resource limit constraints, and event applicability constraints. This indicates the total number of incentive types, the Indicates the incentive type index, the Indicates incentive type The allocated quantity, the This indicates the maximum allowed total number of incentives for the current event. and These represent the incentive types. Minimum allowed allocation quantity and incentive type The maximum allowed number of allocations, the and These represent the start and end time windows for the banquet's permitted incentive activities, respectively. Indicates incentive type The start time of the activity, the Indicates incentive type The amount of resources required per unit, the This indicates the total amount of resources available for the current event. Indicates incentive type The applicable set of event characteristics. Specifically, each incentive type has a unique identifier to distinguish different incentive types. This identifier is obtained from a predefined list of incentive types in the headquarters or terminal system. The incentive type allocation quantity represents the actual quantity of each incentive type allocated in the current banquet event. Its value is dynamically generated by an adaptive optimization algorithm based on the event feature vector of the current banquet, historical banquet implementation records, and constraints such as resources and quantity. The minimum and maximum allowable allocation quantities for each incentive type are set by the incentive strategy table or historical experience data in the headquarters system to limit the scope of incentive allocation. The start time of the incentive type activity is determined based on the banquet reservation time and the execution rules of the incentive type, obtained from the banquet reservation data and incentive implementation requirements. The amount of resources required for each incentive type, such as budget, inventory, or system processing capacity, is obtained from the resource management table in the headquarters system. The range of banquet characteristics applicable to the incentive type is defined by analyzing historical banquet event data and incentive implementation records, combined with the headquarters strategy table definition, to limit the types of banquets to which the incentive type can be applied. The upper limit of the total incentive quantity and the upper limit of the total resource quantity are set by the headquarters based on the banquet scale, budget, and resource status, obtained from the system resource management table or event planning strategy. The time window for allowing incentive activities is determined by the banquet reservation time and the execution requirements of the incentive type, and is obtained from the banquet reservation data and incentive execution rules. The feature vector of the current banquet event is generated in step S2, extracting key features such as banquet size, time distribution, geographical location, and customer behavior from the standardized structured input dataset, and integrating them through an encoding algorithm. This integrated feature vector serves as input data describing the overall characteristics of the current banquet event and is used to generate an adaptive incentive scheme.

[0031] Furthermore, step S5 is included: receiving real-time banquet records uploaded by distributors or terminal stores through the headquarters system, and comprehensively scoring the parameters of the real-time banquet records through a multimodal verification algorithm combined with a confidence fusion algorithm, and issuing incentive rewards to the corresponding distributors or terminal stores whose comprehensive scores of real-time banquet records exceed a set threshold.

[0032] Furthermore, in step S5, the process of comprehensively scoring the parameters of the real-time banquet record using a multimodal verification algorithm combined with a confidence fusion algorithm specifically includes the following sub-steps: S501. Based on the uploaded real-time banquet records, extract multimodal data including structured data, images, videos, and text logs, and extract feature vectors for each modality. S502. Calculate the corresponding single-modal score and confidence level for the feature vector of each modality; S503. The scores of each modality are weighted and fused with their corresponding confidence levels to obtain a comprehensive score for the real-time banquet record.

[0033] Specifically, in step S5, the headquarters node receives real-time banquet records uploaded by distributors or terminal stores, and performs multimodal feature analysis and fusion processing on the recorded content. First, the system extracts multimodal data such as structured data, images, videos, and text logs from the uploaded real-time banquet records. Each type of modal data is converted into a vector representation using a specialized feature extraction algorithm. The structured data includes quantifiable indicators such as banquet scale, time, and geographical information. Image and video data generate high-dimensional feature vectors using convolutional neural networks or other visual feature extraction algorithms. Text log data generates semantic vectors using natural language processing technology, ensuring that the features of each modal data can be unified. The system represents and processes the data; subsequently, it calculates a single-modal score for the feature vector of each modality and simultaneously generates a confidence index. The single-modal score is quantified by matching with historical records, model predictions, or rules, while the confidence index reflects the reliability and accuracy of the modality data in the overall evaluation. The confidence index can be dynamically adjusted based on data quality, completeness, and model output stability. Finally, the system weights and fuses the scores of each modality with their corresponding confidence indices, and calculates the final score of the real-time banquet record through a fusion algorithm. This comprehensive score, under the combined effect of multimodal information, confidence weights, and historical references, can reflect the overall status of the banquet implementation.

[0034] Furthermore, the multimodal verification algorithm mainly draws on existing multimodal learning techniques. Its basic principle is to vectorize information from different modalities, such as structured data, images, videos, and text logs, to reflect the implementation of the banquet activities from multiple perspectives. Specifically, structured data is normalized to form numerical feature vectors; image feature extraction uses Convolutional Neural Networks (CNNs), a mature algorithm in deep learning for extracting spatial features; video data feature extraction uses Long Short-Term Memory (LSTM) networks, which can capture temporal dynamic features; and text log feature extraction uses pre-trained language models (such as BERT) to extract semantic information. These methods are all existing technologies and are used in this scheme to construct multimodal feature vectors. In the single-modal scoring stage, the feature vectors of each modality are scored using existing regression or classification models. For example, structured data can be scored using linear regression or decision tree models to predict the banquet's execution; image and video features are scored using trained CNN or LSTM models; and text features are scored using text classification or regression models. Simultaneously, a confidence score is calculated for each single-modal score. This confidence score, derived from the quantification of model output probability, variance, or entropy, measures the reliability of the modality score; this process is also based on existing methods. In the confidence score fusion stage, this method employs a weighted fusion strategy to integrate the scores from each modality into a comprehensive score. The basic principle of fusion is multimodal complementarity, meaning that different modalities reflect the banquet implementation from different perspectives, and modalities with higher confidence scores have a larger weight in the comprehensive score. This weighted fusion algorithm itself is an existing multimodal information fusion technology. In this solution, by combining confidence scores, a comprehensive and intelligent evaluation of real-time banquet records is achieved, providing a reliable basis for the accurate distribution of incentive rewards.

[0035] Example 2 Furthermore, as a preferred embodiment of the above-described embodiment one, a digital banquet marketing management system based on a multi-dimensional decision-making algorithm is proposed. This system is implemented based on the digital banquet marketing management method based on a multi-dimensional decision-making algorithm described in embodiment one, and includes a data receiving module, a feature processing module, an incentive scheme generation module, a scheme distribution module, and a real-time recording and scoring module, wherein: The data receiving module is used to receive banquet demand data from customer terminals, distributors, or terminal stores through the headquarters system, and to perform format verification, type checking, and time and space normalization processing on the received data to generate a standardized structured input dataset. The data receiving module further includes: The field parsing unit is used to parse data fields from different sources and map them to the internal standardized data model; The time processing unit is used to perform unified processing on multi-source time fields, including time zone conversion, date and time normalization, and time period encoding. The geoprocessing unit is used to map the banquet location coordinates to a uniform grid or cluster labels and perform anomaly correction. The scale normalization unit is used to numerically normalize the number of banquet tables and the estimated number of people and generate a multidimensional distribution vector; The feature processing module is used to extract multidimensional feature parameters of the banquet event and generate an event feature vector based on the standardized structured input dataset. The feature processing module further includes: The banquet scale feature unit is used to calculate the total number of tables, the average number of people per table, and the total expected number of participants, and then standardizes these parameters. The time distribution feature unit is used to encode the appointment date and time into a periodic vector to represent daily, weekly, and monthly cycles and divide time periods, while also identifying holidays and special dates; Geographic clustering feature units are used to perform cluster analysis on banquet locations, generate category labels, and optimize weights by combining historical distribution patterns and surrounding density information; The customer behavior characteristics unit is used to analyze indicators such as the frequency of participation, average participation size, and incentive response patterns generated from historical customer records, and to standardize them uniformly; The feature integration unit is used to encode and merge the above-mentioned feature vectors in a fixed order to form a unified event feature vector; The incentive scheme generation module is used to generate an incentive scheme template based on the event feature vector and historical banquet implementation records; The incentive scheme generation module also includes: The historical record extraction unit is used to extract the feature vectors of historical banquets and the corresponding incentive execution results, and then perform standardization and cleaning. The similarity calculation unit is used to calculate the degree of matching between the current event feature vector and the historical event feature vector and select the reference set with the highest similarity. The incentive effect prediction unit is used to perform weighted calculations on the historical incentive responses of the reference set and generate the expected effect index of each candidate incentive scheme under the current event. The constraint filtering unit is used to apply total incentive constraints, single-type quantity constraints, time window constraints, resource limit constraints, and event applicability constraints to the candidate scheme set, filter and generate the optimal incentive scheme, and output the incentive scheme template through standardization processing; The scheme delivery module is used to send the incentive scheme template and event feature vector to the target distributor or terminal store node; The scheme distribution module also includes: Encoding and packaging unit, used for unified encoding, standardization, and message packaging; The transmission control unit is used to implement integrity verification, version control, and event identifier mapping; The real-time recording and scoring module is used to receive real-time banquet records uploaded by distributors or terminal stores through the headquarters system, and to perform multimodal feature analysis and comprehensive scoring on the real-time data. The real-time scoring module also includes: The multimodal feature extraction unit is used to extract multimodal feature vectors from structured data, images, videos, and text logs. A single-modal scoring unit is used to calculate the corresponding score and confidence level for each type of modality feature vector; The confidence fusion unit is used to generate a comprehensive score for the real-time banquet record by weighting and fusing the scores of each modality according to their confidence levels.

[0036] Specifically, the implementation principle and process of the above embodiments are as follows: The data receiving and standardization module first receives banquet demand data from customer terminals, distributors, or terminal stores via a network interface. This data includes banquet identifiers, customer identifiers, reservation dates and times, banquet locations, and banquet scale (number of tables and estimated attendees). The data parsing unit parses the input fields, mapping multi-source data to a unified data model and performing type checks and logical validations, including validating the date, time, and geographic coordinates. Missing or abnormal data is marked or removed. Multi-source time fields are processed through time zone conversion, normalization, and time period encoding to generate a computable time series vector. Geographic coordinates are mapped to a unified grid or category labels through cluster analysis. The banquet scale field is normalized to generate a multi-dimensional distribution vector of the number of tables and total attendees, forming a standardized structured input dataset. The feature extraction and encoding module extracts banquet scale features, time distribution features, geographic clustering features, and customer behavior features from the standardized structured input dataset. The banquet scale features are calculated... The system calculates the total number of tables, average number of attendees per table, and total expected number of participants, and normalizes these values ​​to form numerical representations. The time distribution feature encodes the reservation date and time into a periodic vector, including daily, weekly, and monthly cycle information and time period divisions, while also identifying holidays and special dates. Geographic clustering features generate category labels through cluster analysis and optimize them based on historical distribution and surrounding density. Customer behavior features generate quantifiable indicators by analyzing historical customer participation records, incentive response patterns, and participation frequency. All features are uniformly encoded into event feature vectors. Discrete, continuous, and multi-source features are represented in the same dimensional space and merged in a fixed order to obtain a complete event feature vector. The historical comparison and incentive prediction module uses the event feature vector as its core. The historical record extraction unit retrieves historical banquet feature vectors and corresponding incentive scheme execution results from the database, cleans and standardizes the historical records, and the matching unit matches the current event with historical events using a similarity measurement method, selecting the top k most similar events. Historical records form a reference set. The weight calculation unit sums the historical incentive response data in the reference set according to similarity weights to generate a prediction effect index of candidate incentive schemes under the current event. The scheme screening unit screens schemes for feasibility based on constraints including the total number of incentives, the number of single-type incentives, time window, resource limitations, and event applicability. It performs standardization and normalization processing to generate incentive scheme templates, which include scheme structure, applicable conditions, and parameter configurations. The scheme distribution module encodes and standardizes the generated incentive scheme templates and event feature vectors, including fixing the field order, standardizing data types, and packaging messages. It adds integrity verification and version control information and distributes the information to the target dealer or terminal store node. The receiving node parses the message, obtains the scheme template, and maps it to the corresponding event.The real-time recording and verification module receives and uploads real-time banquet records from headquarters. The feature extraction unit extracts multimodal feature vectors from structured data, images, videos, and text logs. The single-modal scoring calculation unit calculates scores and confidence levels for each modality's feature vector. The fusion unit weighted and merges the scores and confidence levels of each modality to obtain a comprehensive score.

[0037] Example 3 Based on the above embodiments one and two, a specific application scenario for a digital banquet marketing management method and system based on a multi-dimensional decision-making algorithm is proposed, and its application process is as follows: Distributor store A001 uploaded a banquet request: Banquet ID E1001, Customer ID C500, Reservation Date 2026-02-15, Reservation Time 18:00, Banquet Location Latitude and Longitude (31.2304, 121.4737), Banquet Size 15 Tables, Estimated Attendance 150 People. The data parsing unit maps the fields to a standardized data model, specifically: Banquet size characteristics Fs = [Total number of tables normalized to 0.3, average number of people per table normalized to 0.3, total number of people normalized to 0.3].

[0038] The time distribution feature Ft = [day of the week code 0.5, daily cycle 18 / 24 = 0.75, monthly cycle 2 / 12 = 0.167, holiday marker 1].

[0039] Geographic clustering feature Fg = [grid number 105 → one-hot vector length 200, surrounding density weight 0.7].

[0040] Customer behavior characteristics Fc = [historical engagement frequency 0.65, average engagement size normalized 0.32, historical incentive response average 0.6].

[0041] The encoding unit merges the feature vectors to generate an event feature vector of length 215.

[0042] The system retrieves the top 100 similar dealer activity records from the historical database, each containing a historical event feature vector and incentive execution results.

[0043] Similarity was calculated, and the top 5 similarities were [0.91, 0.88, 0.84, 0.82, 0.80]. The top k=5 similarities were selected to form a reference set. After calculation, the historical responses corresponding to the excitation scheme s1 were determined. = [0.78, 0.74, 0.7, 0.65, 0.6], weighted calculation ≈0.715; Historical response corresponding to incentive scheme s2 = [0.7, 0.68, 0.65, 0.63, 0.6], weighted calculation ≈ 0.662.

[0044] Define the conditions and constraints, where: Total incentive quantity constraint: The maximum available incentive is 80 units, and the total number of incentives in this plan is 50 ≤ 80.

[0045] Single-type incentive constraints: S1 minimum 10, maximum 30, S2 minimum 10, maximum 30, currently allocated S1=25, S2=25.

[0046] Time window constraint: The banquet is from 18:00 to 22:00, and the incentive execution times of 18:30 and 19:30 are both within the window.

[0047] Resource constraints: Budget 4000 yuan, this plan consumes 3800 yuan.

[0048] Event applicability constraints: The solution is only applicable to corporate client banquets, with a total number of attendees ≥ 50, and the demographic characteristics must match.

[0049] After the screening is completed, 25 templates S1 and 25 templates S2 will be generated. The execution times are 18:30 and 19:30, with a budget of 3800 yuan.

[0050] The sending module packages the template and event feature vectors to generate a message: EventID: E1001 FeatureVector: [0.3,0.3,...,0.6] Incentives: {S1:25, S2:25} ExecutionTime: [18:30, 19:30] Budget: 3800 Send the data to the distributor or retail store A001, attaching a timestamp and verification code. The terminal parses and maps it to the banquet hall E1001, executing the incentive operation. The distributor / retail store uploads real-time banquet records: actual number of tables 15, number of attendees 148, with images, videos, and text logs uploaded simultaneously.

[0051] Multimodal feature extraction: structured vector [0.3, 0.296], image vector 128-dimensional, video vector 128-dimensional, text vector 64-dimensional. Unimodal scoring: structured 0.94, image 0.9, video 0.87, text 0.91; confidence scores 0.9, 0.8, 0.85, 0.7. Weighted fusion yields a comprehensive score ≈ 0.905, exceeding the threshold of 0.8, at which point the system confirms the activation has been completed.

[0052] The above description is merely a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the concept described herein through the above teachings or related technologies or knowledge. Modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

Claims

1. A digital banquet marketing management method based on a multidimensional decision-making algorithm, characterized in that, Includes the following steps: S1. Receive banquet demand data from customer terminals, distributors, or terminal stores through the headquarters system, and perform format verification and time-space normalization processing on the received banquet demand data to obtain a standardized structured input dataset. S2. Based on the standardized structured input dataset, extract key feature parameters related to banquet size, time distribution, geographical clustering, and customer behavior, and generate event feature vectors through encoding algorithms; S3. Based on the event feature vector and combined with historical banquet implementation records, generate an incentive scheme template through an adaptive incentive generation optimization algorithm; S4. Send the incentive scheme generated by the algorithm and its corresponding event parameters to the target dealer or terminal store node.

2. The digital banquet marketing management method based on a multi-dimensional decision-making algorithm as described in claim 1, characterized in that, The banquet demand data includes: banquet identifier, customer identifier, reservation date and time, banquet location, and banquet size, wherein the banquet size includes the number of tables and the estimated number of attendees.

3. The digital banquet marketing management method based on a multi-dimensional decision-making algorithm as described in claim 1, characterized in that, Step S2 specifically includes the following sub-steps: S201. Based on the standardized structured input dataset, extract the banquet scale features, time distribution features, geographical clustering features, and customer behavior features respectively; S202. Unify the encoding of banquet scale characteristics, time distribution characteristics, geographical clustering characteristics, and customer behavior characteristics into numerical vectors; S203. Merge the various feature vectors to obtain the event feature vector: ； Among them, the Represents the event feature vector, the Indicating the scale and characteristics of the banquet, the Representing the time distribution characteristics, the Representing geographical clustering features, the This indicates customer behavior characteristics.

4. The digital banquet marketing management method based on a multi-dimensional decision-making algorithm as described in claim 3, characterized in that, In step S201, the banquet scale characteristics are obtained by quantifying the number of banquet tables and the expected number of attendees in the standardized structured input dataset; the time distribution characteristics are obtained by periodically encoding the reservation dates and times in the standardized structured input dataset; the geographic clustering characteristics are obtained by performing cluster analysis on the banquet locations in the standardized structured input dataset; and the customer behavior characteristics are obtained by statistically analyzing customer identifiers in the standardized structured input dataset in conjunction with historical banquet implementation records to obtain customer participation frequency, average participation scale, and historical incentive response, and the statistical results are quantified. The historical banquet implementation records are obtained from the historical database of the headquarters system.

5. The digital banquet marketing management method based on a multi-dimensional decision-making algorithm as described in claim 1, characterized in that, Step S3 specifically includes the following sub-steps: S301. Extract the feature vector of the historical banquet and the corresponding execution results of the incentive scheme based on the historical banquet implementation records; S302. Calculate the similarity between the event feature vector of the current event and the feature vector of the historical banquets, and select the top k historical banquet implementation records as the reference set based on the similarity. S303. Weight the historical stimulus responses of the reference set to predict the stimulus effect of the current event and obtain the expected effect index of each optional stimulus scheme under the current event; S304. Select the incentive scheme with the best incentive effect according to the set constraints, and obtain the incentive scheme template by standardizing and normalizing the scheme.

6. The digital banquet marketing management method based on a multi-dimensional decision-making algorithm as described in claim 5, characterized in that, In step S302, the similarity calculation process is specifically represented as follows: ； Among them, the Represents the event feature vector, the Let i represent the feature vector of the i-th historical banquet, and the This represents the numerical similarity between the event feature vector and the historical banquet feature vector.

7. The digital banquet marketing management method based on a multi-dimensional decision-making algorithm as described in claim 6, characterized in that, In step S3, the process of predicting the incentive effect of the current event is specifically represented as follows: ； ； Among them, the This indicates the incentive effect of the current event, the... Indicates the candidate incentive scheme, the Represents a reference set, the This indicates the incentive effect of historical banquet implementation records. The weighting coefficients representing the historical banquet implementation records, the stated Let represent the feature vector of the j-th historical banquet.

8. The digital banquet marketing management method based on a multi-dimensional decision-making algorithm as described in claim 5, characterized in that, In step S304, the constraints include: Total incentive quantity constraint, used to ensure that the total number of incentives allocated to each incentive type in the incentive scheme does not exceed the maximum available incentive quantity preset by the system; Single-type incentive quantity constraint, used to constrain the quantity of each incentive type allocated in the scheme to be no less than the preset minimum value and no more than the preset maximum value; The time window constraint is used to ensure that the implementation time of the incentive activity is within the allowed time period of the banquet; Resource constraints are used to ensure that the total amount of resources required for the incentive scheme does not exceed the amount of available resources. The total amount of resources includes budget, inventory, and system capacity. Event applicability constraints are used to restrict incentive schemes to only be applied to banquet event characteristics that are suitable for the scheme.

9. The digital banquet marketing management method based on a multidimensional decision-making algorithm as described in claim 5, characterized in that, In step S304, the process of selecting the incentive scheme with the best incentive effect is specifically expressed as follows: ； Among them, the The incentive scheme that provides the best incentive effect is described in the following text. Indicates that the output is the largest The Denotes the set of candidate incentive schemes, the Indicates the candidate incentive scheme, the This indicates the motivating effect of a preceding event.

10. The digital banquet marketing management method based on a multidimensional decision-making algorithm as described in claim 1, characterized in that, It also includes step S5: receiving real-time banquet records uploaded by distributors or terminal stores through the headquarters system, and comprehensively scoring the parameters of the real-time banquet records through a multimodal verification algorithm combined with a confidence fusion algorithm, and issuing incentive rewards to the corresponding distributors or terminal stores whose comprehensive scores of real-time banquet records exceed the set threshold.

11. The digital banquet marketing management method based on a multi-dimensional decision-making algorithm as described in claim 10, characterized in that, In step S5, the process of comprehensively scoring the parameters of the real-time banquet record using a multimodal verification algorithm combined with a confidence fusion algorithm specifically includes the following sub-steps: S501. Based on the uploaded real-time banquet records, extract multimodal data including structured data, images, videos, and text logs, and extract feature vectors for each modality. S502. Calculate the corresponding single-modal score and confidence level for the feature vector of each modality; S503. The scores of each modality are weighted and fused with their corresponding confidence levels to obtain a comprehensive score for the real-time banquet record.

Images