Popularization task intelligent distribution method and system based on label matching
By building a task and executor tag library and using a multi-dimensional deep learning model to intelligently allocate promotion tasks, the problems of low task allocation efficiency, poor matching degree and insufficient monitoring in the existing technology are solved, efficient marketing delivery and resource optimization are achieved, and the effect of private domain promotion and customer retention are improved.
Patent Information
- Application Number
- CN202510359977.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-25
- Publication Date
- 2025-08-12
AI Technical Summary
The existing promotion task management methods have problems such as inefficient task release and allocation, difficulty in tracking promotion effect, chaotic KOC/KOS management, and unreasonable resource allocation, and lack of intelligent task allocation and monitoring mechanisms.
By building a task tag rules and executor tag library, using a multivariate deep learning model to match tasks and executors, combining Word2Vec, BERT, LightGBM and DeepFM models for task allocation optimization, and using incremental learning and reinforcement learning methods to update the model, setting up incentive mechanisms and settlement mechanisms.
It improves the timeliness of task allocation, improves marketing delivery efficiency and task matching accuracy, reduces operational labor costs, optimizes resource allocation, improves the efficiency and effectiveness of private domain promotion, and enhances customer retention rate.
Smart Images
Figure CN120471318A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of media promotion operations, and in particular to a method and system for intelligently allocating promotion tasks based on tag matching. Background Art
[0002] In the current environment of widespread private domain traffic and self-media operations, businesses need to promote and disseminate their products and services through KOCs (Key Opinion Consumers) and KOSs (Key Opinion Sales). However, existing promotion task management methods have the following problems and shortcomings:
[0003] (1) Inefficient task release and allocation: There is a lack of a unified task release and management platform; manual task allocation is prone to deviations and errors, making it impossible to achieve accurate matching and intelligent scheduling.
[0004] (2) It is difficult to track the promotion effect: data collection is scattered and incomplete; there is a lack of real-time monitoring and early warning mechanisms; and there is no unified evaluation standard for promotion effects.
[0005] (3) KOC / KOS management is chaotic: there is a lack of systematic talent pool management, making it difficult to evaluate the ability level of executors; there is a lack of appropriate incentive and credit evaluation mechanisms.
[0006] (4) Irrational resource allocation: low efficiency in the use of promotion budget; suboptimal task allocation; and lack of data-supported decision-making mechanisms. Summary of the Invention
[0007] In view of the above-mentioned deficiencies in current technology, the present invention provides a method for intelligent allocation of promotion tasks based on tag matching. Through intelligent task allocation, it can improve the timeliness of task allocation, improve marketing delivery efficiency, reduce operating labor costs, and improve the efficiency and effectiveness of private domain promotion.
[0008] To achieve the above objectives, the embodiments of the present invention adopt the following technical solutions:
[0009] A method for intelligently allocating promotion tasks based on tag matching, comprising the following steps:
[0010] Obtain promotion task information and executor information;
[0011] Based on the promotion task information, classify and standardize the promotion tasks and build task labeling rules;
[0012] Build an executor tag library based on the executor information;
[0013] Based on the task label rules and executor label library, a multi-dimensional deep learning model is used to match promotion tasks and executors to obtain task allocation results.
[0014] According to one aspect of the present invention, the promotion task information includes: task objectives, task requirements, task budget, task cycle, and task acceptance criteria.
[0015] According to one aspect of the present invention, the classification and standardization of promotion tasks includes:
[0016] Identify the type of task;
[0017] Prioritize tasks;
[0018] Divide the difficulty of the task;
[0019] Set the matching conditions for the task;
[0020] Split and optimize tasks;
[0021] Assess the resource requirements of the task;
[0022] Assess the risk of the task.
[0023] According to one aspect of the present invention, the task tags include: task type tags, product type tags, target audience tags, and professional requirement tags; the executor tags include: basic tags, capability tags, industry tags, and credit tags.
[0024] According to one aspect of the present invention, the method of matching promotion tasks and executors based on task label rules and executor label libraries using a multivariate deep learning model to obtain task assignment results includes:
[0025] Based on the Word2Vec model, the task label rules are converted into task vectors;
[0026] Extract semantic features of task vectors based on the BERT model;
[0027] Based on multimodal processing, semantic features are fused and enhanced according to the correlation and important features of task vectors to obtain optimized features;
[0028] Based on the LightGBM model, the promotion tasks and executors are matched and sorted according to the optimization features and executor tag library to obtain the task allocation results;
[0029] Based on the DeepFM model, the matching scores between promotion tasks and executors are calculated, and the task allocation results are adjusted and optimized according to the matching scores.
[0030] According to one aspect of the present invention, the method for intelligently allocating promotion tasks based on tag matching further includes:
[0031] Update and optimize the multivariate deep learning model based on incremental learning methods;
[0032] Evaluate the models and select the better model based on reinforcement learning methods.
[0033] According to one aspect of the present invention, the method for intelligently allocating promotion tasks based on tag matching further includes:
[0034] Obtain the actual acceptance rate and completion quality of promotion tasks and evaluate the accuracy of task allocation;
[0035] Optimize task tag rules and executor tag library based on the accuracy of task assignment.
[0036] According to one aspect of the present invention, the method for intelligently allocating promotion tasks based on tag matching further includes:
[0037] Track and record the execution progress, interaction and conversion data of promotion tasks;
[0038] Identify abnormal situations in promotion tasks and conduct early warning processing;
[0039] The completion quality, interactive effect and conversion rate of promotion tasks are evaluated based on machine learning methods.
[0040] According to one aspect of the present invention, the method for intelligently allocating promotion tasks based on tag matching further includes:
[0041] Generate a settlement plan for the executor based on the preset settlement mechanism according to the completion quality, interaction effect and conversion rate of the promotion task;
[0042] Based on the completion quality, interactive effect and conversion rate of the promotion tasks, an incentive strategy is generated for the executors based on the preset incentive mechanism.
[0043] A system for intelligently allocating promotional tasks based on tag matching, based on the above-mentioned method for intelligently allocating promotional tasks based on tag matching, comprises:
[0044] Information acquisition module, used to obtain promotion task information and executor information;
[0045] The task processing module is used to classify and standardize promotion tasks based on promotion task information and build task labeling rules;
[0046] The executor module is used to build an executor tag library based on the executor information;
[0047] The matching module is used to match promotion tasks and executors based on task label rules and executor label library, using a multi-dimensional deep learning model to obtain task allocation results.
[0048] Advantages of the present invention:
[0049] The present invention provides a method for intelligent allocation of promotion tasks based on tag matching. Through intelligent task allocation, it can greatly improve the timeliness of task allocation, improve marketing delivery efficiency, improve task matching accuracy, increase task acceptance rate and task completion rate, reduce operating manpower costs, optimize resource allocation effects, improve the efficiency and effect of private domain promotion, improve conversion efficiency and customer retention rate, and create greater economic value; it can effectively solve the problems of low efficiency, poor matching, insufficient monitoring, etc. in private domain promotion task management, and has significant practical value and promotion value. BRIEF DESCRIPTION OF THE DRAWINGS
[0050] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0051] Figure 1 This is a flow chart of a method for intelligently allocating promotion tasks based on tag matching according to the first embodiment of the present invention;
[0052] Figure 2 This is a flowchart of a method for intelligently allocating promotion tasks based on tag matching as described in Example 2 of the present invention. DETAILED DESCRIPTION
[0053] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.
[0054] Example 1
[0055] like Figure 1 As shown, a method for intelligently allocating promotion tasks based on label matching includes the following steps:
[0056] S1: Get promotion task information and executor information.
[0057] Specifically, the promotion task information includes: task objectives, task requirements, task budget, task cycle, and task acceptance criteria. When collecting task information, it is necessary to collect basic task information, determine task objectives and requirements, set task budget and cycle, and define acceptance criteria.
[0058] In this method, performers are individuals such as KOCs (Key Opinion Consumers) and KOSs (Key Opinion Sales) who carry out promotional tasks. When collecting performer information, it's necessary to obtain information such as their active platforms, number of followers, follower type, follower distribution, their promotional capabilities, and past task completion history.
[0059] S2: Based on the promotion task information, classify and standardize the promotion tasks and build task labeling rules.
[0060] Specifically, the classification and standardization of promotion tasks includes:
[0061] (1) Identify the type of task. Task types generally include seeding, advertising promotion, live streaming, etc.
[0062] (2) Determine the priority of the task. Set the priority according to the situation of different tasks. The priority of different tasks will be different. For example, some tasks promote products that are about to be launched and the time is tight, so the task priority will be high.
[0063] (3) Divide the difficulty of tasks. Set the difficulty of tasks according to the situation of different tasks. The difficulty of different tasks will be different. For example, the difficulty of recommending new products will be higher than that of recommending products that are already popular among the public.
[0064] (4) Set the matching conditions for tasks. Different tasks require executors that match the conditions.
[0065] (5) Split and optimize tasks. Some tasks can be split and assigned to different executors in different steps; some tasks can also be merged. By optimizing tasks, the success rate of task execution can be improved.
[0066] (6) Evaluate the resource requirements of the task. Different tasks require corresponding resources.
[0067] (7) Assess the risks of the task. When executing a promotion task, there are certain risks, which need to be assessed in advance so that they can be handled in a timely manner.
[0068] In addition, this step may also include: setting control parameters of the task, etc.
[0069] The task tags include:
[0070] (1) Task type labels, including: grass-planting tasks, live broadcast tasks, product promotion tasks, community operation tasks, etc.
[0071] (2) Product type label, including: specific industry, category, price range and other requirements.
[0072] (3) Target audience labels, including: gender, age, spending power and other profile requirements.
[0073] (4) Professional requirements label, including: skills, experience level and other qualification requirements required by the executor.
[0074] This step classifies and standardizes the information of promotion tasks to obtain the task type, priority, difficulty, matching conditions, resource requirements, task tags, etc., and construct task tag rules.
[0075] S3: Build an executor tag library based on the executor information.
[0076] Specifically, the executor tag includes:
[0077] (1) Basic tags, including: number of fans, age distribution of fans, occupation type of fans, geographical distribution of fans, content areas, active platforms, fan interactions, etc.
[0078] (2) Ability labels, including: grass-growing ability, live streaming and selling ability, community operation ability, etc.
[0079] (3) Industry labels, including: beauty, clothing, food, mother and baby, games, film and television commentary, comedy, life, animation, technology, handicrafts, digital appliances and other vertical field experiences.
[0080] (4) Credit labels, including: task completion rate, punctuality rate, praise rate, default rate, etc.
[0081] This step processes the performer information to obtain various basic information and capabilities of the performer, thereby building a performer tag library.
[0082] S4: Based on the task label rules and executor label library, a multi-dimensional deep learning model is used to match promotion tasks and executors to obtain task assignment results.
[0083] Step S4 includes:
[0084] (1) Based on the Word2Vec model, the task label rules are converted into task vectors.
[0085] Word2Vec is a technology used to map words into a continuous vector space. Word2Vec generates word embeddings by analyzing large amounts of text data, capturing the semantic relationships between words. Word2Vec is based on the fundamental idea that if two words appear close together in many contexts, their semantics and usage are likely similar. Through this approach, Word2Vec can generate high-dimensional vectors (typically 100 to 300 dimensions) while preserving the distance relationships between words.
[0086] The Word2Vec model is used to convert the task labeling rules constructed in step S2 into semantic vectors for subsequent processing.
[0087] (2) Based on the BERT model, the semantic features of the task vector are extracted.
[0088] The BERT (Bidirectional Encoder Representations from Transformers) model is a Transformer-based autoencoding language model. The BERT model maps input tokens (text tags), text fragments, and positions, then adds these mappings together to form the input layer. Next, a multi-layer bidirectional Transformer encoder is used to process the input sequence. This bidirectional structure enables BERT to simultaneously consider the left and right context of the input text, thereby more accurately understanding the meaning of the language.
[0089] Using the BERT model, the semantic features of the task vector in the previous step can be extracted.
[0090] (3) Based on multimodal processing, semantic features are fused and enhanced according to the correlation and important features of task vectors to obtain optimized features.
[0091] Specifically, multimodal processing can use convolutional neural networks to extract local correlation features between labels, adopt long short-term memory networks to process the historical task sequences of executors, combine attention mechanisms to highlight important features, use Transformer models to process long sequence dependencies, perform feature fusion and enhancement on semantic features, and finally obtain optimized features.
[0092] (4) Based on the LightGBM model, the promotion tasks and executors are matched and sorted according to the optimization features and executor tag library to obtain the task allocation results.
[0093] LightGBM (Light Gradient Boosting Machine) is a machine learning model based on gradient boosting decision trees. It uses a collection of decision trees for learning and iteratively builds a series of weak learners (usually decision trees) to improve the overall performance of the model. Each step builds a new tree to reduce the error of the previous stage. Through cumulative reduction, the prediction error continues to decrease, ultimately resulting in an accurate prediction model.
[0094] Specifically, models such as Word2Vec can be used to convert the actor tag library into actor vector features. Training samples can be constructed based on the task's optimized features and actor vector features. Each sample can be represented as a feature vector that contains the task's optimized features, actor vector features, and a predicted matching score. The LightGBM model is pre-trained using these pre-processed training samples.
[0095] During actual prediction, new data is input into the trained LightGBM model for prediction. The model will output the matching score for each task-executor, and sort the promoted tasks and executors according to the score.
[0096] Preferably, step S4 may further include:
[0097] (5) Based on the DeepFM model, the matching scores between promotion tasks and executors are calculated, and the task allocation results are adjusted and optimized according to the matching scores.
[0098] The DeepFM model is a hybrid model that combines a deep learning model (DNN) and a factorization machine (FM) model. It can simultaneously learn low-order explicit feature combinations and high-order implicit feature combinations without manual feature engineering. Therefore, it is often used in recommendation systems or advertising systems.
[0099] Preferably, the method further comprises:
[0100] S5: Update and optimize the multivariate deep learning model based on incremental learning methods.
[0101] In actual applications, the task label rules and executor label library can be updated regularly according to actual conditions. On the basis of the original matching model, only the changes caused by the new data are updated and optimized without rebuilding the matching model.
[0102] S6: Evaluate the model and select the better model based on reinforcement learning method.
[0103] In practical applications, when updating and optimizing the matching model, reinforcement learning methods can be used to construct the state space, action space and optimization evaluation function, and a better matching model can be selected through strategy selection iteration.
[0104] The beneficial effects of this embodiment are: through intelligent task allocation, this method can greatly improve the timeliness of task allocation, improve marketing delivery efficiency, improve task matching accuracy, increase task acceptance rate and task completion rate, reduce operating labor costs, optimize resource allocation effects, improve the efficiency and effectiveness of private domain promotion, improve conversion efficiency and customer retention rate, and create greater economic value. This method can effectively solve the problems of inefficiency, poor matching, and insufficient monitoring in private domain promotion task management, and has significant practical value and promotion value. This method has strong scalability and adaptability, can be customized according to the needs of different industries and enterprises, and has good market application prospects.
[0105] Example 2
[0106] like Figure 2 As shown, a method for intelligently allocating promotion tasks based on label matching includes the following steps:
[0107] S1: Get promotion task information and executor information.
[0108] Specifically, the promotion task information includes: task objectives, task requirements, task budget, task cycle, and task acceptance criteria. When collecting task information, it is necessary to collect basic task information, determine task objectives and requirements, set task budget and cycle, and define acceptance criteria.
[0109] In this method, performers are individuals such as KOCs (Key Opinion Consumers) and KOSs (Key Opinion Sales) who carry out promotional tasks. When collecting performer information, it's necessary to obtain information such as their active platforms, number of followers, follower type, follower distribution, their promotional capabilities, and past task completion history.
[0110] S2: Based on the promotion task information, classify and standardize the promotion tasks and build task labeling rules.
[0111] Specifically, the classification and standardization of promotion tasks includes:
[0112] (1) Identify the type of task. Task types generally include seeding, advertising promotion, live streaming, etc.
[0113] (2) Determine the priority of the task. Set the priority according to the situation of different tasks. The priority of different tasks will be different. For example, some tasks promote products that are about to be launched and the time is tight, so the task priority will be high.
[0114] (3) Divide the difficulty of tasks. Set the difficulty of tasks according to the situation of different tasks. The difficulty of different tasks will be different. For example, the difficulty of recommending new products will be higher than that of recommending products that are already popular among the public.
[0115] (4) Set the matching conditions for tasks. Different tasks require executors that match the conditions.
[0116] (5) Split and optimize tasks. Some tasks can be split and assigned to different executors in different steps; some tasks can also be merged. By optimizing tasks, the success rate of task execution can be improved.
[0117] (6) Evaluate the resource requirements of the task. Different tasks require corresponding resources.
[0118] (7) Assess the risks of the task. When executing a promotion task, there are certain risks, which need to be assessed in advance so that they can be handled in a timely manner.
[0119] In addition, this step may also include: setting control parameters of the task, etc.
[0120] The task tags include:
[0121] (1) Task type labels, including: grass-planting tasks, live broadcast tasks, product promotion tasks, community operation tasks, etc.
[0122] (2) Product type label, including: specific industry, category, price range and other requirements.
[0123] (3) Target audience labels, including: gender, age, spending power and other profile requirements.
[0124] (4) Professional requirements label, including: skills, experience level and other qualification requirements required by the executor.
[0125] This step classifies and standardizes the information of promotion tasks to obtain the task type, priority, difficulty, matching conditions, resource requirements, task tags, etc., and construct task tag rules.
[0126] S3: Build an executor tag library based on the executor information.
[0127] Specifically, the executor tag includes:
[0128] (1) Basic tags, including: number of fans, age distribution of fans, occupation type of fans, geographical distribution of fans, content areas, active platforms, fan interactions, etc.
[0129] (2) Ability labels, including: grass-growing ability, live streaming and selling ability, community operation ability, etc.
[0130] (3) Industry labels, including: beauty, clothing, food, mother and baby, games, film and television commentary, comedy, life, animation, technology, handicrafts, digital appliances and other vertical field experiences.
[0131] (4) Credit labels, including: task completion rate, punctuality rate, praise rate, default rate, etc.
[0132] This step processes the performer information to obtain various basic information and capabilities of the performer, thereby building a performer tag library.
[0133] S4: Based on the task label rules and executor label library, a multi-dimensional deep learning model is used to match promotion tasks and executors to obtain task assignment results.
[0134] Step S4 includes:
[0135] (1) Based on the Word2Vec model, the task label rules are converted into task vectors.
[0136] Word2Vec is a technology used to map words into a continuous vector space. Word2Vec generates word embeddings by analyzing large amounts of text data, capturing the semantic relationships between words. Word2Vec is based on the fundamental idea that if two words appear close together in many contexts, their semantics and usage are likely similar. Through this approach, Word2Vec can generate high-dimensional vectors (typically 100 to 300 dimensions) while preserving the distance relationships between words.
[0137] The Word2Vec model is used to convert the task labeling rules constructed in step S2 into semantic vectors for subsequent processing.
[0138] (2) Based on the BERT model, the semantic features of the task vector are extracted.
[0139] The BERT (Bidirectional Encoder Representations from Transformers) model is a Transformer-based autoencoding language model. The BERT model maps input tokens (text tags), text fragments, and positions, then adds these mappings together to form the input layer. Next, a multi-layer bidirectional Transformer encoder is used to process the input sequence. This bidirectional structure enables BERT to simultaneously consider the left and right context of the input text, thereby more accurately understanding the meaning of the language.
[0140] Using the BERT model, the semantic features of the task vector in the previous step can be extracted.
[0141] (3) Based on multimodal processing, semantic features are fused and enhanced according to the correlation and important features of task vectors to obtain optimized features.
[0142] Specifically, multimodal processing can use convolutional neural networks to extract local correlation features between labels, adopt long short-term memory networks to process the historical task sequences of executors, combine attention mechanisms to highlight important features, use Transformer models to process long sequence dependencies, perform feature fusion and enhancement on semantic features, and finally obtain optimized features.
[0143] (4) Based on the LightGBM model, the promotion tasks and executors are matched and sorted according to the optimization features and executor tag library to obtain the task allocation results.
[0144] LightGBM (Light Gradient Boosting Machine) is a machine learning model based on gradient boosting decision trees. It uses a collection of decision trees for learning and iteratively builds a series of weak learners (usually decision trees) to improve the overall performance of the model. Each step builds a new tree to reduce the error of the previous stage. Through cumulative reduction, the prediction error continues to decrease, ultimately resulting in an accurate prediction model.
[0145] Specifically, models such as Word2Vec can be used to convert the actor tag library into actor vector features. Training samples can be constructed based on the task's optimized features and actor vector features. Each sample can be represented as a feature vector that contains the task's optimized features, actor vector features, and a predicted matching score. The LightGBM model is pre-trained using these pre-processed training samples.
[0146] During actual prediction, new data is input into the trained LightGBM model for prediction. The model will output the matching score for each task-executor, and sort the promoted tasks and executors according to the score.
[0147] (5) Based on the DeepFM model, the matching scores between promotion tasks and executors are calculated, and the task allocation results are adjusted and optimized according to the matching scores.
[0148] The DeepFM model is a hybrid model that combines a deep learning model (DNN) and a factorization machine (FM) model. It can simultaneously learn low-order explicit feature combinations and high-order implicit feature combinations without manual feature engineering. Therefore, it is often used in recommendation systems or advertising systems.
[0149] S5: Obtain the actual acceptance rate and completion quality of promotion tasks to evaluate the accuracy of task assignment.
[0150] You can set scoring rules based on the actual acceptance rate and completion quality of the promotion tasks and calculate the accuracy score of the task assignment.
[0151] S6: Optimize the task labeling rules and executor labeling library based on the accuracy of task assignment.
[0152] Based on the accuracy of task assignment, evaluate the contribution of each label to the matching accuracy, update the task labels and executor labels, regularly clean up inefficient labels, and add new labels.
[0153] Preferably, the method further comprises:
[0154] S7: Track and record the execution progress, interaction and conversion data of promotion tasks.
[0155] After the promotion task is assigned to the executor for execution, tracking records are required to timely understand the progress of task execution, fan interaction, conversion rate, etc.
[0156] S8: Identify abnormal situations in promotion tasks and conduct early warning processing.
[0157] During the execution of the task, abnormal situations may occur, such as errors in the advertising copy and public opinion crises of the executors. Timely warnings and processing are needed to reduce losses.
[0158] S9: Evaluate the completion quality, interaction effect and conversion rate of promotion tasks based on machine learning methods.
[0159] For example, models such as logistic regression, random forest, and XGBoost can be used to evaluate task completion quality, interaction effects, conversion rates, etc.
[0160] S10: Generate a settlement plan for the executor based on the preset settlement mechanism, according to the completion quality, interactive effect and conversion rate of the promotion task.
[0161] Specifically, the settlement mechanism may include calculating basic remuneration, evaluating performance rewards, adjusting execution credit, etc., and finally generating a settlement plan to settle task remuneration for the executor.
[0162] S11: Generate incentive strategies for executors based on the preset incentive mechanism, according to the completion quality, interactive effect and conversion rate of the promotion tasks.
[0163] In order to motivate executors to better complete promotion tasks, you can also design incentive mechanisms, set reward strategies, credit scores, task allocation priorities, give gifts, etc., to generate incentive strategies for executors.
[0164] The beneficial effects of this embodiment are: this method can improve the efficiency of marketing delivery through intelligent task allocation, improve the conversion efficiency and customer retention rate of private domain promotion; use effect tracking and early warning processing to improve customer service satisfaction and reduce the complaint rate; by establishing a settlement mechanism and incentive mechanism, it can greatly improve the quality of the executors' task completion.
[0165] Example 3
[0166] A system for intelligently allocating promotional tasks based on tag matching, based on the method for intelligently allocating promotional tasks based on tag matching as described in Embodiment 1 or 2, includes:
[0167] Information acquisition module, used to obtain promotion task information and executor information;
[0168] The task processing module is used to classify and standardize promotion tasks based on promotion task information and build task labeling rules;
[0169] The executor module is used to build an executor tag library based on the executor information;
[0170] The matching module is used to match promotion tasks and executors based on task label rules and executor label library, using a multi-dimensional deep learning model to obtain task allocation results.
[0171] Preferably, the system further includes:
[0172] Incremental update module, used to update and optimize the multivariate deep learning model based on incremental learning method;
[0173] The AB update module is used to evaluate the model using the A / B testing method and select the better model.
[0174] Preferably, the system further includes:
[0175] The accuracy evaluation module is used to obtain the actual acceptance rate and completion quality of promotion tasks and evaluate the accuracy of task assignment;
[0176] The label optimization module is used to optimize the task label rules and executor label library based on the accuracy of task assignment.
[0177] Preferably, the system further includes:
[0178] Tracking module, used to track and record the execution progress, interaction and conversion data of promotion tasks;
[0179] The early warning module is used to identify abnormal situations in promotion tasks and perform early warning processing;
[0180] The task evaluation module is used to evaluate the completion quality, interaction effect and conversion rate of promotion tasks based on machine learning methods.
[0181] Preferably, the system further includes:
[0182] The settlement module is used to generate a settlement plan for the executor based on the completion quality, interaction effect and conversion rate of the promotion task and the preset settlement mechanism;
[0183] The incentive module is used to generate incentive strategies for executors based on the preset incentive mechanism according to the completion quality, interaction effect and conversion rate of the promotion tasks.
[0184] Example 4
[0185] A computer program product includes a computer program, which, when executed, implements the steps of the method for intelligently allocating promotion tasks based on tag matching as described in embodiment one or two.
[0186] Example 5
[0187] A readable storage medium stores a computer program as described in Example 4, and when the computer program is executed, the steps of the method for intelligently allocating promotion tasks based on label matching as described in Example 1 or 2 are implemented.
[0188] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.
Claims
1. A method for intelligently allocating promotion tasks based on label matching, characterized in that: The following steps are involved: Obtain promotion task information and executor information; Based on the promotion task information, classify and standardize the promotion tasks and build task labeling rules; Build an executor tag library based on the executor information; Based on the task label rules and executor label library, a multi-dimensional deep learning model is used to match promotion tasks and executors to obtain task allocation results.
2. The method for intelligently allocating promotion tasks based on tag matching according to claim 1 is characterized in that: The promotion task information includes: task objectives, task requirements, task budget, task cycle, and task acceptance criteria.
3. The method for intelligently allocating promotion tasks based on tag matching according to claim 1 is characterized in that: The classification and standardization of promotion tasks include: Identify the type of task; Prioritize tasks; Divide the difficulty of the task; Set the matching conditions for the task; Split and optimize tasks; Assess the resource requirements of the task; Assess the risk of the task.
4. The method for intelligently allocating promotion tasks based on tag matching according to claim 1 is characterized in that: The task tags include: task type tag, product type tag, target audience tag, and professional requirement tag; the executor tags include: basic tag, capability tag, industry tag, and credit tag.
5. The method for intelligently allocating promotion tasks based on tag matching according to claim 1 is characterized in that: Based on the task label rules and the executor label library, the multi-dimensional deep learning model is used to match the promotion tasks and executors, and the task allocation results obtained include: Based on the Word2Vec model, the task label rules are converted into task vectors; Extract semantic features of task vectors based on the BERT model; Based on multimodal processing, semantic features are fused and enhanced according to the correlation and important features of task vectors to obtain optimized features; Based on the LightGBM model, the promotion tasks and executors are matched and sorted according to the optimization features and executor tag library to obtain the task allocation results; Based on the DeepFM model, the matching scores between promotion tasks and executors are calculated, and the task allocation results are adjusted and optimized according to the matching scores.
6. The method for intelligently allocating promotion tasks based on tag matching according to claim 5 is characterized in that: The method for intelligently allocating promotion tasks based on tag matching also includes: Update and optimize the multivariate deep learning model based on incremental learning methods; Evaluate the models and select the better model based on reinforcement learning methods.
7. The method for intelligently allocating promotion tasks based on tag matching according to claim 1 is characterized in that: The method for intelligently allocating promotion tasks based on tag matching also includes: Obtain the actual acceptance rate and completion quality of promotion tasks and evaluate the accuracy of task allocation; Optimize task tag rules and executor tag library based on the accuracy of task assignment.
8. The method for intelligently allocating promotion tasks based on tag matching according to claim 1 is characterized in that: The method for intelligently allocating promotion tasks based on tag matching also includes: Track and record the execution progress, interaction and conversion data of promotion tasks; Identify abnormal situations in promotion tasks and conduct early warning processing; The completion quality, interactive effect and conversion rate of promotion tasks are evaluated based on machine learning methods.
9. The method for intelligently allocating promotion tasks based on tag matching according to claim 8 is characterized in that: The method for intelligently allocating promotion tasks based on tag matching also includes: Generate a settlement plan for the executor based on the preset settlement mechanism according to the completion quality, interaction effect and conversion rate of the promotion task; Based on the completion quality, interactive effect and conversion rate of the promotion tasks, an incentive strategy is generated for the executors based on the preset incentive mechanism.
10. A promotion task intelligent allocation system based on tag matching, characterized in that: The method for intelligently allocating promotion tasks based on label matching according to any one of claims 1 to 9 comprises: Information acquisition module, used to obtain promotion task information and executor information; The task processing module is used to classify and standardize promotion tasks based on promotion task information and build task labeling rules; The executor module is used to build an executor tag library based on the executor information; The matching module is used to match promotion tasks and executors based on task label rules and executor label library, using a multi-dimensional deep learning model to obtain task allocation results.