Automatic order dispatching method and system for hotels

Abstract

The invention provides an automatic order dispatching method and system for a hotel, and the method comprises the steps: carrying out the analysis of hotel order contents according to task attributes in hotel orders, extracting key information, including task types, priorities and complexity, constructing task attribute feature vectors, carrying out the clustering of the task attribute feature vectors, and carrying out the clustering of the task attribute feature vectors; dividing the tasks with high similarity into the same category to obtain task sets of different categories; in the hotel order distribution process, the task execution state is monitored in real time, whether the task execution state is abnormal or overtime or not is judged, and if yes, task redistribution is carried out to other available processing personnel in combination with the hotel order task type and the customer service working time; after the hotel order is completed, the task execution data is counted, the average execution time and the completion quality index of each task are calculated, a task execution report is formed, and the hotel order processing flow and the task distribution strategy are continuously improved.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to an automatic order dispatching method and system for a hotel. Background Art

[0002] In the process of supply chain management and order processing, enterprises are faced with the dual challenges of diversified order types and surging order volumes. Traditional manual order processing methods can no longer meet the needs of efficient processing, so many enterprises have begun to introduce automated systems to allocate orders. However, in practical applications, there are still some technical problems that need to be solved in the automated order allocation system. One of the prominent technical contradictions is the conflict between the fine-grained splitting of order tasks and the complex dependencies between tasks. In the order processing process, each order contains multiple tasks, such as order information entry, inventory inspection, payment confirmation, and delivery arrangement. In order to improve the efficiency of order processing, it is necessary to be able to split the order tasks in a fine-grained manner and assign them to different processing personnel or automated modules for parallel processing. There are often complex dependencies between order tasks, which may lead to problems such as data inconsistency, process missing or deadlock in the order processing process. Therefore, the automated order allocation system needs to find a balance between the fine-grained splitting of tasks and the complexity of task dependencies. On the one hand, the system needs to be able to intelligently analyze the dependencies between order tasks, build a task dependency graph, and dynamically adjust the task splitting strategy according to the complexity of the dependencies to avoid difficulties in handling dependencies caused by excessive splitting; on the other hand, the system also needs to split independent tasks as much as possible to fully utilize the advantages of parallel processing and improve the efficiency of order processing. How to find the optimal balance between these two conflicting goals is a key technical challenge facing the automated order distribution system. Summary of the invention

[0003] The present invention provides an automatic order dispatching method for a hotel, which mainly comprises:

[0004] According to the task attributes in the hotel order, the hotel order content is parsed to extract key information, including task type, priority and complexity, and the task attribute feature vector is constructed. The task attribute feature vector is clustered to classify tasks with high similarity into the same category to obtain a set of tasks in different categories.

[0005] Based on each task category of the hotel order, the association rule method is used to mine the dependency relationship between tasks, generate a task dependency graph, determine the execution order and priority of tasks, and dynamically adjust the task allocation strategy according to the task dependency graph and task priority of the hotel order, giving priority to independent tasks and high-priority tasks, while taking into account the parallel processing of tasks;

[0006] During the hotel order allocation process, the task execution status is monitored in real time to determine whether there is an abnormality or timeout in the task execution status. If so, the task is reallocated to other available processing personnel based on the hotel order task type and customer service working hours;

[0007] The task dependency graph of hotel orders is converted into a task execution network. Task nodes contain task attributes, and edges represent dependencies between tasks. Task execution network parameters are dynamically updated. According to the task execution network and task priority, the task allocation strategy is dynamically adjusted to give priority to independent tasks and tasks with high priority, while taking into account the parallel processing of tasks.

[0008] After the hotel order is completed, the task execution data is collected and statistics are collected to calculate the average execution time and completion quality indicators of each task, and a task execution report is formed to continuously improve the hotel order processing process and task allocation strategy.

[0009] The present invention provides an automatic order dispatching system for a hotel, which mainly comprises:

[0010] The order parsing and task classification module is used to obtain hotel orders, parse the hotel order content, extract key information, including task type, priority and complexity, construct task attribute feature vectors based on the key information, cluster the task attribute feature vectors, classify tasks with high similarity into the same category, and obtain task sets of different categories;

[0011] The task dependency mining module is used to mine the dependency between tasks based on each task category of the hotel order using the association rule method, generate a task dependency graph, determine the execution order and priority of the tasks, and dynamically adjust the task allocation strategy according to the task dependency graph and task priority of the hotel order, giving priority to independent tasks and high-priority tasks, while taking into account the parallel processing of tasks;

[0012] The task status monitoring and reallocation module is used to monitor the task execution status in real time during the hotel order allocation process, determine whether the task execution status is abnormal or timed out, and if so, reallocate the task to other available processing personnel based on the hotel order task type and customer service working hours;

[0013] The task execution network management module is used to convert the task dependency graph of hotel orders into a task execution network. The task nodes contain task attributes, and the edges represent the dependencies between tasks. The task execution network parameters are dynamically updated. According to the task execution network and task priority, the task allocation strategy is dynamically adjusted to give priority to independent tasks and high-priority tasks, while taking into account the parallel processing of tasks.

[0014] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0015] The present invention discloses an automatic order dispatching method and system for a hotel. The method realizes dynamic allocation and priority adjustment of tasks by clustering analysis of order task attributes and constructing a task dependency graph. The present invention takes into account the workload, professional ability and shift situation of processing personnel to ensure the balance and efficiency of order distribution. Orders with the same type of problems are preferentially allocated to the same processing personnel to improve processing efficiency. For unprocessed orders, a one-key return to pool method is used for reallocation. The present invention also introduces a task execution network to dynamically update task parameters and optimize task allocation strategies. The order processing process is continuously improved through statistical analysis of task execution data. Finally, according to different types of order status, the task completion status is judged to realize the full process intelligent management of order processing. The present invention significantly improves the efficiency and quality of hotel order processing and provides strong support for hotel management. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 The present invention is a flow chart of an automatic order dispatching method for hotels.

[0017] Figure 2 The present invention is a structural schematic diagram of an automatic order dispatching system for hotels. DETAILED DESCRIPTION

[0018] In order to further understand the content of the present invention, the present invention is described in detail in conjunction with the accompanying drawings and embodiments. The present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It is understood that the specific embodiments described herein are only used to explain the relevant inventions, rather than to limit the invention. It is also necessary to explain that, for ease of description, only the parts related to the invention are shown in the accompanying drawings.

[0019] like Figure 1-Figure 2 In this embodiment, an automatic order dispatching method for a hotel may specifically include:

[0020] S101. According to the task attributes in the hotel order, the hotel order content is parsed to extract key information, including task type, priority and complexity, and a task attribute feature vector is constructed. The task attribute feature vector is clustered to classify tasks with high similarity into the same category to obtain a task set of different categories.

[0021] Acquire hotel order data, perform word segmentation on the order content, and use the TF-IDF algorithm to calculate the word importance score. According to the preset dictionary, map the words with higher importance scores to corresponding numerical values ​​or classification codes to determine the standardized task attribute feature vector. Use the K-means clustering algorithm to cluster the standardized task attribute feature vector, evaluate the clustering effect of different K values ​​through the silhouette coefficient, and select the optimal K value. For the clustering results, calculate the average cosine similarity of samples within each cluster and the cosine similarity between clusters. If the similarity within the cluster is lower than the preset threshold, the cluster is re-divided into sub-clusters; if the similarity between clusters is higher than the preset threshold, these clusters are merged into the same category. Use the optimized clustering results to re-label the original data and generate a training data set with category labels. Use the random forest algorithm to train the task classifier, the input of the task classifier is the attribute feature vector of the new task, and the output is the predicted task category.

[0022] Specifically, we obtain hotel order data, segment the order content, and use the TF-IDF algorithm to calculate the importance of words, and select words with higher importance as features. According to the preset dictionary, we map the extracted keywords to the corresponding numerical values ​​or classification codes to obtain a standardized task attribute feature vector. The dictionary contains mapping rules for key information such as task type, priority, and complexity. For example, "urgent" is mapped to a priority value of 9, and "normal" is mapped to a priority value of 5.

[0023] The K-means clustering algorithm is used to cluster the standardized task attribute feature vectors. The silhouette coefficient is used to evaluate the clustering effect of different K values ​​and select the optimal K value. Each cluster center is calculated through an iterative optimization process, and each task sample is divided into the category corresponding to the closest cluster center. If the Euclidean distance between two cluster centers is less than the preset threshold, the two clusters are merged, and the merged cluster center is the weighted average of the original two cluster centers.

[0024] For the clustering results, the average cosine similarity of samples within each cluster and the cosine similarity between clusters are calculated. The clustering results are optimized and adjusted by setting a similarity threshold. For clusters with internal similarity lower than the threshold, the sub-clustering is re-divided. For clusters with inter-cluster similarity higher than the threshold, they are merged into the same category. The original data is re-labeled using the optimized clustering results to generate a training data set with category labels.

[0025] According to the optimized clustering results, a unique category identifier is assigned to each cluster, and the original task data is associated with the corresponding category identifier and stored. The task classifier is trained using the random forest algorithm, and the optimized clustering results are used as labels for the training data. Through the correspondence between task categories and task attributes, a task classifier is constructed for the rapid classification of subsequent newly added tasks and the dynamic update of task sets. The input of the classifier is the attribute feature vector of the new task, and the output is the predicted task category.

[0026] To process the hotel order data, we first use the Jieba word segmentation tool to segment the order content and obtain a word list. Then we apply the TF-IDF algorithm to calculate the importance score of each word, and select the top 20% of the words as features. The preset dictionary contains mapping rules for task type, priority, and complexity, such as mapping "cleaning" to task type 1, "repair" to task type 2; mapping "urgent" to priority 9, "normal" to priority 5; mapping "simple" to complexity 2, and "complex" to complexity 8. According to these rules, the extracted keywords are converted into numerical values ​​to construct a standardized task attribute feature vector.

[0027] The K-means clustering algorithm is used to cluster the feature vectors. The silhouette coefficient is used to evaluate the clustering effect of K values ​​from 2 to 10, and the K value with the highest silhouette coefficient is selected as the optimal number of clusters. Assuming that the optimal K value is 5, 5 cluster centers are calculated by iteration. In each iteration, the Euclidean distance from each sample to each cluster center is calculated, and the samples are divided into the clusters with the closest distance. If the Euclidean distance between two cluster centers is less than the preset threshold of 0.5, the two clusters are merged, and the new cluster center is the weighted average of the original two cluster centers, and the weight is the proportion of the number of samples in each cluster.

[0028] For the clustering results, the average cosine similarity of samples within each cluster and the cosine similarity between clusters were calculated. The internal similarity threshold was set to 0.8 and the inter-cluster similarity threshold was set to 0.3. For clusters with internal similarity lower than 0.8, a hierarchical clustering algorithm was used for sub-clustering. For cluster pairs with inter-cluster similarity higher than 0.3, they were merged into the same category. The original data was re-labeled using the optimized clustering results to generate a training data set with category labels.

[0029] According to the optimized clustering results, a unique category identifier is assigned to each cluster, such as C1, C2, C3, etc. The original task data is associated with the corresponding category identifier and stored in a relational database. The task classifier is trained using the random forest algorithm, and the optimized clustering results are used as labels for the training data. The number of trees in the random forest is set to 100, the maximum depth is 10, and the minimum number of leaf node samples is 5. The input of the classifier is the attribute feature vector of the new task, and the output is the predicted task category. The classifier performance is evaluated through cross-validation, and the model parameters with the highest F1 score are selected. The task classifier finally constructed can be used to quickly classify new tasks and dynamically update the task set.

[0030] S102. Based on each task category of the hotel order, the dependency relationship between tasks is mined using the association rule method, a task dependency graph is generated, and the execution order and priority of the tasks are determined; according to the task dependency graph and task priority of the hotel order, the task allocation strategy is dynamically adjusted to give priority to independent tasks and high-priority tasks, while taking into account the parallel processing of tasks.

[0031] The task data in the hotel order is obtained, and the task category standard is formulated according to the common task types obtained by the statistical analysis of historical data to classify the tasks; the April algorithm is used to calculate the support and confidence between tasks, and the association rules that meet the minimum support and minimum confidence thresholds are screened out to obtain the dependency set between tasks; a directed acyclic graph is constructed according to the task dependency set, and the execution order of the tasks is determined by a topological sorting algorithm, wherein the weight of the topological sorting result is set as a preset value, the original priority weight is set as a preset value, the weighted score of each task is calculated, and the tasks are sorted according to the score; the task execution sequence is optimized by a dynamic programming algorithm, and the time window parameter is determined according to the average value of the completion time of historical tasks. In each time window, independent tasks and tasks with high priority are preferentially selected for allocation. If the number of independent tasks is insufficient, the task with the least dependency is selected; a task status monitoring mechanism is set, including heartbeat detection and periodic status report, wherein the heartbeat detection is sent once at a preset time, and the status report is generated once at a preset time. When a task execution exception occurs, a dynamic adjustment mechanism is triggered to recalculate the task priority and dependency, and update the task execution sequence and allocation plan.

[0032] Specifically, we obtain the task data in the hotel order, formulate the task category standard based on the common task types obtained from the statistical analysis of historical data, and classify the tasks. We use the April algorithm to calculate the support and confidence between tasks. The specific implementation process includes four steps: generating candidate item sets, pruning, generating frequent item sets, and generating rules. We filter out the association rules that meet the minimum support and minimum confidence thresholds, and obtain the set of dependencies between tasks.

[0033] A directed acyclic graph is constructed based on the set of task dependencies. In the graph, nodes represent tasks, edges represent dependencies, and the weight of the edges is set to the confidence of the association rule. The execution order of tasks is determined by the topological sorting algorithm, while the original priority of the tasks is considered, and the comprehensive score is calculated using the weighted average method. Specifically, the weight of the topological sorting result is set to 0.6, the weight of the original priority is set to 0.4, the weighted score of each task is calculated, and the tasks are sorted according to the score to obtain the task execution sequence that comprehensively considers the dependency and priority.

[0034] The task execution sequence is optimized using a dynamic programming algorithm, and the time window parameters are determined based on the average completion time of historical tasks. In each time window, independent tasks and high-priority tasks are prioritized for allocation. If the number of independent tasks is insufficient, the tasks with the least dependencies are selected. The number of tasks executed simultaneously is controlled by setting the parallelism parameter, which is dynamically adjusted according to the system resource load to obtain a task allocation plan that meets the parallel processing requirements. The task allocation plan is sent to the execution unit, and the initial execution status of each task is recorded.

[0035] For the task allocation plan, set up a task status monitoring mechanism, including heartbeat detection and regular status reports. Heartbeat detection is sent every 30 seconds, and status reports are generated every 5 minutes, including information such as task completion progress and resource occupancy. Real-time tracking of task execution progress and resource occupancy, when task execution exceptions or resource shortages occur, trigger the dynamic adjustment mechanism. Recalculate task priorities and dependencies, update task execution sequences and allocation plans, and achieve real-time optimization of task scheduling. If the difference between the adjusted plan and the original plan exceeds the preset threshold, the task allocation process is re-executed.

[0036] The data processing of hotel order tasks begins with historical data analysis, and common task types such as cleaning, maintenance, and catering services are obtained through statistics. After the task category standards are established, the current order tasks are classified. Subsequently, the Aprior i algorithm is used to calculate the correlation between tasks, setting the minimum support to 0.1 and the minimum confidence to 0.5. The algorithm first generates 1-item set candidates, counts the frequency, and prunes items below the support threshold to generate frequent 1-item sets. Then it iteratively generates 2-item sets, 3-item sets, etc. until no higher-order item sets can be generated. Finally, association rules are generated based on frequent item sets, and rules that meet the confidence threshold are screened out to form a set of task dependencies.

[0037] A directed acyclic graph is constructed based on the dependency set. The nodes represent tasks, the edges represent dependencies, and the edge weights are the confidence of the association rules. The Kahn algorithm is used for topological sorting to determine the preliminary execution order. At the same time, the original priority of the task is considered, and the weighted average method is used to calculate the comprehensive score. Assume that the weight of the topological sorting result is 0.6, the weight of the original priority is 0.4, and the weighted score of each task is calculated. For example, if the topological order of task A is 3 and the priority is 8, its comprehensive score is 30.6+80.4=5. Reorder the tasks according to the scores to obtain the final execution sequence.

[0038] The execution sequence optimization adopts a dynamic programming algorithm. First, the historical task completion time is analyzed, and the average value is calculated to determine the time window parameters. Assuming that the average completion time is 30 minutes, the time window is set to 25 minutes. In each time window, independent tasks and high-priority tasks are given priority. If there are insufficient independent tasks, the task with the least dependencies is selected. The parallelism parameter is dynamically adjusted according to the system CPU utilization. For example, when the CPU utilization is less than 50%, the parallelism is increased by 1, and when it is higher than 80%, the parallelism is reduced by 1. The generated task allocation plan is then sent to the execution unit, and the initial state of each task is recorded.

[0039] The task status monitoring mechanism includes a heartbeat detection every 30 seconds and a status report every 5 minutes. The heartbeat detection uses the TCP keep-alive mechanism. If no response is received for three consecutive times, the task is considered abnormal. The status report contains information such as task completion percentage, CPU usage, and memory usage. When task execution abnormalities or resource shortages are detected, the dynamic adjustment mechanism is triggered. The task priority is recalculated, the dependency graph is updated, and a new execution sequence and allocation plan are generated. If the difference between the new and old plans exceeds 20%, the task allocation process is re-executed to ensure real-time optimization of task scheduling.

[0040] S103. During the hotel order allocation process, the task execution status is monitored in real time to determine whether the task execution status is abnormal or timed out. If so, the task is reallocated to other available processing personnel based on the hotel order task type and customer service working hours.

[0041] Receive the execution status data of the hotel order task; according to the execution status data, use the sliding window algorithm to monitor the task execution progress in real time and obtain the task completion percentage within the window; if the task completion percentage is lower than the preset threshold, determine whether the task has an execution exception or timeout; obtain an abnormal task list through the judgment result; for the abnormal task list, use the decision tree algorithm to classify the abnormal tasks, and the decision tree algorithm uses the information gain ratio as the feature selection standard; according to the classification result, generate a task reallocation priority queue; obtain the information of the currently available processing personnel; use the weighted polling algorithm to calculate the workload of the available processing personnel, and the weight of the weighted polling algorithm is determined according to the historical task completion efficiency of the processing personnel; according to the workload calculation result, select the most suitable new processing personnel for each abnormal task; according to the selection result, use multi-threading technology to realize the parallel processing mechanism; send task allocation requests to multiple new processing personnel through the parallel processing mechanism; receive the response result of the new processing personnel; if the response result is rejection or timeout without response, select the next candidate processing personnel; repeat the task allocation process until all abnormal tasks are reallocated.

[0042] Specifically, the task execution status data of hotel order tasks is obtained, and the sliding window algorithm is used to monitor the task execution progress in real time. The sliding window size is set to 10 minutes, and it slides once every 1 minute to calculate the completion percentage of tasks in the window. The task type completion time threshold obtained based on historical data statistical analysis is used to determine whether execution abnormality or timeout occurs. If the task execution time exceeds the threshold, the abnormal mark is triggered and the abnormal task list is obtained. The task type completion time threshold is determined based on the average completion time of the same type of tasks in the past three months plus two standard deviations. According to the abnormal task list, combined with the task type and customer service working hours, the decision tree algorithm is used to classify the abnormal tasks. In the process of building the decision tree, the information gain ratio is used as the feature selection criterion, the maximum tree depth is set to 5, and the minimum number of leaf node samples is set to 10. Considering factors such as task urgency, customer service professionalism and working hours, a task redistribution priority queue is generated. The task urgency adopts a scoring system of 1-5, which is calculated based on the ratio of the remaining completion time to the expected completion time. The customer service professionalism uses a grade system, which is divided into 5 levels from elementary to advanced, and is evaluated based on historical task processing data. Determine the order of redistribution of each abnormal task, and calculate the workload of the currently available handlers through the task redistribution priority queue using a weighted polling algorithm. The weight is determined based on the handler's historical task completion efficiency. Combined with the handler's professional skill matching, select the most suitable new handler for each abnormal task. Professional skill matching is calculated based on the handler's success rate in completing similar tasks in the past three months. A success rate of more than 90% is scored 5 points, 80-90% is scored 4 points, and so on. After generating the task redistribution plan, consider the handler's current workload and expected completion time, and use a greedy algorithm to optimize the plan to ensure that the newly assigned tasks will not cause the handler to be overloaded. For the optimized task redistribution plan, multi-threading technology is used to implement a parallel processing mechanism, and task allocation requests are sent to multiple new handlers at the same time. Each thread is responsible for the allocation process of a task, and the response timeout is set to 30 seconds. Get the handler's acceptance or rejection feedback, and use a blocking queue to store the feedback results. If there is a rejection or timeout without response, the next candidate handler is automatically selected. Use a fallback mechanism to try up to 3 candidate handlers. If all three assignments fail, the task will be marked as pending manual processing and added to the manual processing queue. The task reallocation process is continuously monitored until all abnormal tasks are reallocated or added to the manual processing queue. The hotel order task execution status monitoring system first obtains real-time data, uses a 10-minute sliding window, slides once every 1 minute, and calculates the completion percentage of the task within the window. The system calculates the average completion time and standard deviation of each type of task by analyzing historical data from the past three months. For example, if the average completion time of a cleaning task is 30 minutes and the standard deviation is 5 minutes, the completion time threshold is set to 40 minutes.When the task execution time exceeds the threshold, the system automatically marks it as abnormal and generates a list of abnormal tasks. Subsequently, the decision tree algorithm classifies abnormal tasks, with the maximum depth of the tree being 5 and the minimum number of leaf node samples being 10. The urgency of the task is scored on a scale of 1-5. If the remaining time is only 20% of the expected time, the urgency score is 5. Customer service professionalism is divided into 5 levels, which are evaluated based on historical processing data. For example, customer service who successfully handles more than 100 similar tasks is rated as level 5. The system generates a task redistribution priority queue based on these factors. Next, a weighted polling algorithm is used to calculate the workload of the processing personnel, and the weight is determined based on the historical task completion efficiency. Professional skill matching is calculated based on the success rate of similar tasks in the past three months. For example, a success rate of 95% is scored as 5 points and 85% is scored as 4 points. After the system generates a preliminary redistribution plan, it optimizes it through a greedy algorithm to ensure that the newly assigned tasks will not cause the workload of any processing personnel to exceed 120% of the current maximum load. Finally, multi-threading technology is used to parallelly process task allocation requests, with each thread responsible for one task and a 30-second response timeout set. In case of rejection or timeout, the system automatically selects the next candidate and tries up to 3 times. After 3 failures, the task is marked as pending manual processing and added to a special queue. The whole process is continuously monitored until all abnormal tasks are processed or enter the manual processing process.

[0043] The order dispatch scope is determined by combining the maximum number of hotel orders that the pre-set processing personnel can handle, the number of orders being followed, the shift status and time, and the type of orders followed up by each person. At the same time, multiple tasks of the same order are deduplicated to obtain a number of non-repeated orders, and are given priority to personnel who are processing fewer orders, to ensure that the number of order types followed up by each person is basically the same, and the order types include new orders and failed orders.

[0044] Obtain the information database of processing personnel, and extract the maximum number of orders that each processing personnel can process, the number of orders currently being followed, the schedule status and time from the database; based on the processing personnel information, use a multidimensional array to store this information, and use bit operations to determine the current availability of the processing personnel; the bit operation method encodes the schedule status, time period and order type into binary bits, and uses bitwise AND operations to determine whether the processing personnel meets the current dispatching conditions; if the processing personnel meets the dispatching conditions, a matching matrix between processing personnel and order types is established, and each element in the matrix represents the processing ability score of the processing personnel for a specific order type; According to the matching matrix, a weighted polling algorithm is used to calculate the order allocation weight of each processing personnel, and the weight calculation formula is weight = (maximum number of orders that can be processed - number of orders currently being followed) * time weight coefficient; the time weight coefficient is calculated using a piecewise linear function, and the remaining working time is divided into three intervals, 0-2 hours, 2-5 hours, and more than 5 hours, and the corresponding coefficients are 0.8, 1.0, and 1.2 respectively; according to the processing personnel weight and non-repetitive order data, a greedy algorithm is used to allocate orders, and the greedy algorithm prioritizes allocating new orders and orders that have failed to be issued to the processing personnel with the highest weight.

[0045] Specifically, the information database of the processing personnel is obtained, and the maximum number of orders that each processing personnel can process, the number of orders currently being followed, the schedule status and time are extracted. The maximum number of orders that can be processed is calculated based on the average processing efficiency of the processing personnel in the past three months and the working hours of the day. For example, if an average of 5 orders are processed per hour and the working hours are 8 hours, the maximum number of orders that can be processed is 40. A multidimensional array is used to store this information, and the current available status of the processing personnel is quickly determined through bit operations. The bit operation method encodes the scheduling status, time period and order type into binary bits. By bitwise AND operation, it is determined whether the processing personnel meets the current dispatching conditions to obtain a preliminary dispatching range list. A matching matrix between processing personnel and order types is established. Each element in the matrix represents the processing ability score of the processing personnel for a specific order type, and the score is calculated based on historical processing data. According to the order information in the order database, a hash table is used to deduplicate multiple tasks of the same order. The key of the hash table is the order number, and the value is a structure array containing the task ID, task type and status. When traversing the order data, if the order number already exists in the hash table, the new task is added to the corresponding array, otherwise a new array is created. After the traversal is completed, the number of non-repeated orders and the task set for each order are obtained. The weighted polling algorithm is used to calculate the order allocation weight of each processing staff. The weight calculation formula is: weight = (maximum number of orders that can be processed - current number of orders) * time weight coefficient. The time weight coefficient is calculated using a piecewise linear function, and the remaining working time is divided into three intervals: 0-2 hours, 2-5 hours, and more than 5 hours. The corresponding coefficients are 0.8, 1.0, and 1.2 respectively. This calculation method ensures that the more remaining working time, the higher the allocation weight. Combined with the non-repeated order data and the weight of the processing staff, the greedy algorithm is used for order allocation. New orders and orders that failed to be issued are preferentially allocated to processing staff with higher weights. The greedy algorithm first sorts the orders by priority, then traverses the order list, and for each order, calculates the matching scores of all available processing staff (matching score = processing staff weight * order type matching degree), and selects the processing staff with the highest score for allocation. At the same time, the number of orders of each type assigned to each processing staff is recorded through a counter. When the number of orders of a certain type reaches the preset threshold, the allocation probability of that type of order is adjusted. The preset threshold is set based on the average distribution ratio of each type of order obtained from historical data analysis. For example, if new orders account for 30%, the adjustment is triggered when the number of new orders exceeds 30% of the total number of allocated orders. Dynamic adjustment ensures the balance of order type distribution. The hotel order allocation system first obtains the processing personnel information from the database and calculates the maximum number of orders that each person can handle. For example, customer service a has processed an average of 6 orders per hour in the past three months and worked 8 hours today, so the maximum number of orders that can be processed is 48. The system uses a 32-bit integer to store the processing personnel status, of which the first 8 bits represent the scheduling status, the middle 16 bits represent the time period, and the last 8 bits represent the type of order that can be processed.Through the bit operation 10110101&11111111, it is quickly determined whether customer service a meets the current order dispatching conditions. Subsequently, the system constructs a 100x10 processing personnel-order type matching matrix, and the matrix element value range is 0-100, which represents the processing capability score. During the order deduplication process, the system uses the order number as the key, and the value is a hash table containing the array of task ID, type and status. If order A contains two tasks, cleaning and maintenance, then its hash table entry is {customer service a:[{id:1,type:"cleaning",status:"pending"},{id:2,type:"maintenance",status:"pending"}]}. When calculating the weight, the system divides the remaining working time into three intervals: 0-2 hours, 2-5 hours, and more than 5 hours, with corresponding coefficients of 0.8, 1.0, and 1.2. If customer service b can handle a maximum of 50 orders, currently processes 15 orders, and has 6 hours of remaining working time, then its weight is (50-15)*1.2=42. During the order allocation process, the system first sorts the orders and then traverses the order list. If order B is a new order, the system calculates the matching scores of all available processing personnel. For example, if customer service a scores 38 and customer service b scores 42, customer service b is selected to handle order B. At the same time, the number of order types handled by each person is recorded. When the number of new orders exceeds 30% of the total allocated orders, the probability of new order allocation is adjusted to ensure a balanced distribution of order types.

[0046] Determine whether any hotel order has the same type of problem. If so, it will be assigned to the same person for processing. If not, it will be assigned to the corresponding type of customer service.

[0047] The TF-IDF algorithm is used to vectorize the order problem description, and the cosine similarity between the current order and the historical order problems is calculated; the type of order problem is determined based on the cosine similarity, and if the cosine similarity exceeds the preset threshold, it is determined to be the same type of problem. The historical order processing records are obtained according to the order problem type identifier, and the customer service matching degree is calculated using a weighted scoring algorithm; in the weighted scoring algorithm, the weight of the number of processing times is 0.3, the weight of customer satisfaction is 0.4, and the weight of the average processing time is 0.3; the customer service matching ranking list is obtained through the weighted scoring algorithm. A multi-factor decision tree is constructed using the customer service expertise table and workload statistics; the characteristics of the multi-factor decision tree include the current workload, expertise matching degree, estimated completion time, and customer service matching degree; the customer service availability is evaluated through the multi-factor decision tree to generate a customer service priority list. Cross-compare the customer service matching ranking list with the customer service priority list, and use the dynamic programming algorithm to calculate the optimal dispatching plan; select the dispatch object according to the preset weight ratio; if the same customer service meets both the highest matching degree and the highest priority, dispatch the order directly to the customer service; otherwise, determine the dispatch object according to the ratio of the matching degree weight to the priority weight; after determining the dispatching result, update the order status and customer service workload data.

[0048] Specifically, the order information in the hotel order database is obtained, the TF-IDF algorithm is used to vectorize the order problem description, and then the cosine similarity is used to calculate the similarity between the current order and the historical order problems. The historical data is analyzed by the ROC curve, and the optimal threshold is selected, such as 0.85. If the similarity exceeds the threshold, it is determined to be the same type of problem, and the order problem type identification is obtained. According to the order problem type identification, the order history processing record is queried, and the customer service list that has handled this type of problem is extracted. The customer service matching degree is calculated using the weighted scoring algorithm, and the weight distribution is that the number of processing times accounts for 30%, the customer satisfaction accounts for 40%, and the average processing time accounts for 30%. The specific calculation formula is, matching degree = 0.3*standardized processing times + 0.4*average satisfaction + 0.3*(1-standardized processing time). Get the customer service matching degree ranking list. Combine the customer service matching degree ranking list with the current customer service work status data obtained by the real-time monitoring system. Use the customer service expertise table and workload statistics to construct a multi-factor decision tree. The characteristics of the decision tree include current workload, expertise matching degree, expected completion time, and customer service matching degree. The depth of the tree is set to 4, and the information gain ratio is used as the feature selection criterion. The customer service availability is comprehensively evaluated through the decision tree, and a customer service priority list is generated. The customer service matching ranking list is cross-compared with the customer service priority list, and the dynamic programming algorithm is used to calculate the optimal dispatching solution. The state dp is defined to represent the optimal solution for assigning the first i orders to the first j customer services. The transfer equation is dp = max(dp, dp + score(i, j)), where score(i, j) is the score assigned to customer service j by order i. The boundary conditions are dp = 0 and dp = -∞. If the same customer service meets both the highest matching degree and the highest priority, the order is directly dispatched to the customer service. Otherwise, the dispatch object is selected according to the preset weight ratio (such as matching degree weight 0.6, priority weight 0.4). The dispatch result is finally determined and the order status and customer service workload data are updated. The hotel order intelligent dispatching system first obtains order information from the database and applies the TF-IDF algorithm to vectorize the problem description. The problem description of order A, "air conditioner does not cool", is processed to obtain the vector [0.4, 0.6, 0.0, 0.8]. The system then calculates the cosine similarity between the vector and the historical order vector. Assuming that the similarity with order B is 0.92, it exceeds the preset threshold of 0.85 and is determined to be a similar problem. Based on historical data, the system extracts a list of customer service staff who have handled such problems. For example, customer service staff A handled 10 times, with an average satisfaction score of 4.5 points (out of 5 points) and an average processing time of 15 minutes; customer service staff B handled 8 times, with an average satisfaction score of 4.8 points and an average processing time of 20 minutes. Applying the weighted scoring algorithm, the matching degree of customer service staff A is 0.31+0.40.9+0.31=0.96, and that of customer service staff B is 0.30.8+0.40.96+0.30.75=0.845. The system then combines the real-time working status to build a decision tree to evaluate customer service availability.Assume that customer service a currently has a workload of 80%, a specialty match of 90%, and an estimated completion time of 30 minutes; customer service b has a workload of 60%, a specialty match of 85%, and an estimated completion time of 45 minutes. The decision tree outputs a priority of 0.88 for customer service a and 0.82 for customer service b. Finally, the system uses a dynamic programming algorithm to consider the match (weight 0.6) and priority (weight 0.4) to calculate the scores of customer service a at 0.60.96+0.40.88=0.928 and customer service b at 0.60.845+0.40.82=0.835. The system finally selects customer service a to process order A, updates the order status to "processing", and adjusts the workload of customer service a to 90%.

[0049] Get the hotel order status when the customer service is off-duty to determine whether there are any unprocessed orders. If so, reallocate the unprocessed orders by returning them to the pool with one click.

[0050] Receive a cron expression with a preset customer service off-duty time, the cron expression is used to set a scheduled task; start an order status check program before the preset customer service off-duty time according to the cron expression, the order status check program obtains the unfinished order list of the current online customer service from the database; obtain the order number, order status, processing progress and remaining estimated processing time in the unfinished order list; use a decision tree algorithm to classify the orders in the unfinished order list to obtain order classification results, the order classification results include three categories: about to be completed, need to be transferred and cannot be completed; for orders that need to be transferred and cannot be completed, start a one-key return to pool mechanism to reset the status of orders that need to be transferred and cannot be completed to a to-be-allocated status; reallocate the orders in the to-be-allocated status according to order priority, customer service skill matching and workload balance; if there is no matching online customer service, add the orders in the to-be-allocated status to the next day's to-be-processed queue.

[0051] Specifically, a cron expression is used to set a scheduled task, and the order status check program is automatically started 15 minutes before the preset customer service off-duty time. The list of unfinished orders of all current online customer service is obtained through database query, including order number, order status, processing progress and remaining estimated processing time. The estimated processing time is calculated based on historical data, using the average processing time of this type of order plus a standard deviation. For example, if the average processing time of a certain type of order is 30 minutes and the standard deviation is 5 minutes, the estimated processing time is 35 minutes. According to the obtained unfinished order list, the decision tree algorithm is used to classify the orders. The features of the decision tree include order processing progress, remaining estimated processing time and order urgency. The processing progress threshold is set to 80%, the remaining time threshold is set to 10 minutes, and the urgency is scored on a scale of 1-5. The depth of the decision tree is set to 3, and the information gain ratio is used as the feature selection criterion. The orders are divided into three categories: "about to be completed", "need to be transferred" and "cannot be completed", and the order classification results are obtained. For orders of "need to be transferred" and "cannot be completed", a one-key return to the pool mechanism is started. Reset the status of these orders to the pending state, and record the processing progress and related information of the original processing customer service. Update the order status and priority in the order pool, and use the weighted scoring method to calculate the order priority. The weight distribution is: 40% for urgency, 30% for remaining processing time, and 30% for customer waiting time. Reorder the orders in the order pool according to the calculation results. Consider factors such as order priority, customer service skill matching, and workload balance to redistribute the returned orders. The customer service skill matching is calculated by cosine similarity, and the skills required for the order and the customer service skills are represented as vectors, and the cosine value of the angle between the two vectors is calculated. The workload balance uses the minimum and maximum normalization algorithm to map the current workload of each customer service to a value between 0 and 1. Considering these three factors comprehensively, the final score is calculated using the weighted summation method, and the online customer service with the highest score is selected for allocation. If there is no suitable online customer service, the order will be added to the next day's pending queue to complete the order redistribution process. The hotel order intelligent return pool system sets a timed task through the cron expression "04521**?", which automatically triggers the order status check at 21:45 every day. The system queries the database to obtain a list of unfinished orders. For example, order A has a progress of 70% and an estimated remaining time of 40 minutes (calculated based on the average processing time of 35 minutes for this type of order plus a standard deviation of 5 minutes). Subsequently, the decision tree algorithm classifies the orders. Taking order A as an example, the processing progress of 70% < 80% threshold, the remaining time of 40 minutes > 10 minutes threshold, and the urgency is 4 (1-5 scoring system). After judgment by the three-layer decision tree, it is classified as "needs to be transferred". The system activates a one-key return to the pool mechanism, resets the status of order A to pending assignment, and records the original customer service processing progress of 70%. The order priority is calculated by weighted scoring, 0.44 (urgency) + 0.30.5 (normalized remaining time) + 0.30.8 (normalized waiting time)

[0052] =1.79. Based on this score, order A is ranked higher in the order pool. The allocation algorithm considers the cosine similarity between the skill vector [0.8, 0.6, 0.4] required for the order and the customer service skill vector (such as customer service 1: [0.9, 0.7, 0.5], customer service 2: [0.7, 0.8, 0.3]), which are 0.996 and 0.986 respectively. At the same time, the customer service workload is calculated (customer service 1: 60%, customer service 2: 40%), which are 1 and 0 respectively after minimum and maximum normalization. Combining the weighted sum of the three factors (if the weights are 0.4, 0.3, and 0.3 respectively), customer service 1 scores 0.878 and customer service 2 scores 0.99. Finally, order A is assigned to customer service 2 with a higher score, completing the intelligent return to the pool and reallocation process.

[0053] S104. Convert the task dependency graph of the hotel order into a task execution network, where the task node contains the task attributes and the edge represents the dependency between tasks. Dynamically update the task execution network parameters, and dynamically adjust the task allocation strategy according to the task execution network and task priority, giving priority to independent tasks and tasks with high priority, while taking into account the parallel processing of tasks.

[0054] Acquire hotel order information carrying task dependencies, the hotel order information is issued by the front-end system; traverse all nodes in the task dependency graph using a depth-first search algorithm based on the hotel order information to obtain a task node sequence; for the task node sequence, convert each task node into an object containing task attributes, the task attributes including task ID, task type, estimated completion time and task priority; construct an initial structure of a task execution network based on the task attribute object, the initial structure including dependencies between task nodes; analyze the topological structure of the task execution network, and use a critical path algorithm to calculate the critical path in the network, The critical path algorithm includes forward traversal to calculate the earliest start time of each node, and backward traversal to calculate the latest start time of each node; based on the calculation results of the critical path algorithm, determine whether the earliest start time and the latest start time of the node are equal. If they are equal, determine that the node is a node on the critical path; topologically sort the task execution network, convert it into a directed acyclic graph, sort it according to the in-degree and priority score of the node, and obtain a preliminary task execution order; based on the task execution order, the maximum flow algorithm is used for task allocation. The maximum flow algorithm combines the node capacity limit and regards each task node as a resource point with the maximum processing capability.

[0055] Specifically, the task dependency graph of the hotel order is obtained, and the depth-first search algorithm is used to traverse all nodes in the graph. The specific implementation process includes using a stack to store the nodes to be visited, pushing the starting node into the stack, and looping to pop the stack, visit the node, and push the unvisited adjacent nodes into the stack until the stack is empty. Each task node is converted into an object containing task attributes, including task ID, task type, estimated completion time, and task priority. The edge is converted into a connection representing the dependency relationship, and the weight of the edge is set to the correlation between the tasks. The correlation calculation is based on the task type and the historical co-occurrence frequency. For example, the correlation between cleaning and inspection tasks is 0.8, and the correlation between cleaning and maintenance tasks is 0.5. Construct the initial structure of the task execution network. According to the topological structure of the task execution network, the critical path algorithm is used to calculate the critical path in the network. The forward traversal calculates the earliest start time of each node, and the backward traversal calculates the latest start time of each node. The nodes with equal start time are the nodes on the critical path. At the same time, a task priority scoring mechanism is adopted, and the calculation formula is: priority score = 0.4* urgency + 0.3* customer importance + 0.3* (1-standardized estimated completion time). Urgency and customer importance are scored on a scale of 1-5, and the estimated completion time is converted to between 0-1 through minimum and maximum standardization. An initial priority score is assigned to each task node. Preliminary task sorting is performed based on the initial priority and network structure. The task execution network is converted into a directed acyclic graph using a topological sorting algorithm, and then sorted according to the node's in-degree and priority score. Nodes with an in-degree of 0 have the highest priority. When the in-degrees are the same, the priority scores are compared. A preliminary task execution order is generated. The task execution status and resource utilization are obtained through a real-time monitoring system, and the dynamic time window method is used to collect task completion and resource occupancy data at fixed time intervals. The exponential moving average algorithm is used to update the parameters of the task execution network, including the actual completion time of the task node and the weight of the edge. The smoothing coefficient of the exponential moving average algorithm is set to 0.2, that is, new data accounts for 20% and historical data accounts for 80%. The update formula is, new value = 0.2 * current observation value + 0.8 * exponential moving average of the previous period. Based on the updated task execution network and task priority, the maximum flow algorithm is used for task allocation. Improvements include considering node capacity limitations and treating each task node as a resource point with maximum processing power. Construct a residual network, use breadth-first search to find augmenting paths, update the flow of the residual network, and repeat until no augmenting path can be found. Consider the tasks in the network as traffic, and the priority of the task as the traffic weight. By maximizing the network traffic, the optimal allocation of tasks is achieved, and independent tasks and high-priority tasks are allocated first. At the same time, resource constraints are considered to ensure that the parallel processing of tasks does not exceed the upper limit of available resources. The hotel order task execution network system first obtains the task dependency graph. For example, order A contains three tasks: cleaning, inspection, and maintenance.The depth-first search algorithm is used to traverse the nodes in the graph, starting with the cleaning task, pushing it into the stack, then accessing the inspection task, and finally the maintenance task. Each task node is converted into an object, such as the attributes of the cleaning task are {ID: 1, type: "cleaning", estimated time: 30 minutes, priority: 0}. The weight of the edge is based on the task relevance, the relevance of cleaning and inspection is 0.8, and the relevance of cleaning and maintenance is 0.5. Next, the critical path algorithm is used to calculate the critical path in the network, and it is found that cleaning-inspection-maintenance constitutes the critical path with a total duration of 90 minutes. The task priority scoring mechanism calculates the priority score of the cleaning task to be 4.2, the inspection task to be 3.8, and the maintenance task to be 4.5. According to the initial priority and network structure, the topological sorting is performed to obtain the execution order: cleaning-inspection-maintenance. The real-time monitoring system collects task execution data every 5 minutes and updates the parameters using the exponential moving average algorithm. Assuming that the actual completion time of the cleaning task is 35 minutes, the updated estimated time is: 0.2*35+0.8*30=31 minutes. Finally, the maximum flow algorithm treats the tasks as traffic for allocation. If there are two employees, the cleaning task is assigned to employee A, and the inspection and maintenance tasks are assigned to employee B, which realizes the parallel processing of tasks while considering task priorities and resource constraints.

[0056] S105. After the hotel order is completed, the task execution data is counted, the average execution time and completion quality indicators of each task are calculated, a task execution report is formed, and the hotel order processing process and task allocation strategy are continuously improved.

[0057] A data collection request carrying task execution log information is received, wherein the data collection request is triggered when a hotel order is completed; the original data such as the start time, end time, executor, completion status, etc. of each task are extracted from the task execution log according to the data collection request; the original data is cleaned and standardized, and outliers and invalid data are eliminated to obtain a valid task execution data set; if the task execution time exceeds 3 times the standard deviation of the historical average time of the task type, it is determined to be an outlier; if the task execution status is "not started" or "cancelled", it is determined to be invalid data; a descriptive statistical method is used to calculate the valid task execution data set to obtain the average execution time, standard deviation, median and quartile of each task type; the correlation between tasks in the task execution data set is analyzed by an association rule mining algorithm to obtain the key nodes that affect the order processing efficiency and quality; the task execution data set, statistical indicators and key node information are integrated to generate a task execution report; a template-based natural language generation technology is used to automatically generate a report text according to the task execution report.

[0058] Specifically, a real-time data collection mechanism is adopted to trigger the data collection program immediately after the hotel order is completed, and the original data such as the start time, end time, executor, and completion status of each task are extracted from the task execution log. Through data cleaning and standardization, outliers and invalid data are eliminated to obtain a valid task execution data set. The outlier judgment standard is set as the execution time exceeds 3 standard deviations of the historical average time of the task type, or the execution time is less than 1 minute. Invalid data includes records with execution status of "not started" or "cancelled". Based on the cleaned data set, the average execution time, standard deviation, median and quartile of each task type are calculated using descriptive statistics. The execution time data of each task type are sorted, the average and median are calculated, and the minimum, maximum and quartile are obtained. The weighted average algorithm is used to calculate the task completion quality index, and the weight factors are distributed as follows: customer evaluation accounts for 50%, internal quality inspection score accounts for 30%, and rework rate accounts for 20%. Generate quantitative indicators of task execution efficiency and quality. Based on the task execution efficiency and quality indicators, combined with various links of the order processing process, the association rule mining algorithm is used to analyze the correlation and influencing factors between tasks. The algorithm parameters are set to a minimum support of 0.1 and a minimum confidence of 0.5. Identify the key nodes and bottlenecks that affect the efficiency and quality of order processing, and generate a list of optimization suggestions for the task execution process. Verify and screen the optimization suggestions, evaluate the potential impact of each suggestion through simulation calculation, and screen out suggestions with low implementation difficulty and significant expected effects. Integrate the task execution data, statistical indicators and screened optimization suggestions into the task execution report. The report text is automatically generated using template-based natural language generation technology. The template contains a fixed report structure and dynamically filled data fields. Use visualization tools to generate charts containing trend charts, distribution charts and heat maps to form a complete task execution report. Store the report data in the data warehouse, set up a monthly regular review mechanism, compare the changes in key indicators for three consecutive months, and automatically generate an improvement effect evaluation report. According to the evaluation results, adjust the key parameters of the order processing process and the weight coefficients of the task allocation strategy to achieve continuous improvement. The hotel order task execution report system triggers data collection immediately after the order is completed. For example, order A contains three tasks: cleaning, inspection and maintenance. The extracted data showed that the cleaning task execution time was 45 minutes, which exceeded the historical average of 30 minutes plus three times the standard deviation of 15 minutes, and was marked as an outlier. The inspection task was 20 minutes and the maintenance task was 60 minutes, both of which were retained. The descriptive statistics calculation results showed that the average time for the cleaning task was 35 minutes, the standard deviation was 5 minutes, and the median was 33 minutes. The task quality index was weighted average, and the cleaning task score was 0.54.5 (customer evaluation) + 0.390 (internal quality inspection) + 0.2*(1-0.05) (rework rate) = 4.04. The association rule mining set the minimum support to 0.1 and the minimum confidence to 0.5, and found that there was a strong correlation between the delay of the cleaning task and the delay of the maintenance task.Generate optimization suggestions to adjust the order of cleaning and maintenance tasks. Through simulation calculations, it is predicted that this suggestion can reduce the average processing time by 10%. Natural language generation technology generates report text based on the preset template "[task type] average execution time is [time], quality score is [score]". Visualization tools draw a heat map of task time distribution, showing that Monday morning tasks take the longest time. After the report is stored in the data warehouse, the system compares key indicators every month, such as the average cleaning task time in June is 2 minutes less than in May. Based on this, the weight of the cleaning task in the task allocation strategy is automatically increased by 5%, achieving continuous process optimization.

[0059] Whether the task is completed is determined based on different types of hotel orders, including the following five situations: for new orders, the order follower is required to confirm the order and the order recipient cancels the order; for cancellation and modification, the order follower is required to click to confirm the process; when processing the supply order, the order follower is required to enter the supply order result; for automatic order failure, the order follower is required to enter the waiting order result; for automatic confirmation failure, the order follower is required to confirm the order or cancel the order, the task is completed, the follower is cleared and the next task comes in and is assigned again.

[0060] The above describes the specific embodiments of the present invention. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art can make various changes or modifications within the scope of the claims, which does not affect the essence of the present invention. In the absence of conflict, the embodiments of the present application and the features in the embodiments can be combined with each other arbitrarily.

Claims

1. An automatic order dispatching method for a hotel, characterized in that: The method comprises: Acquire hotel orders, parse the hotel order content, extract key information, including task type, priority, and complexity, construct task attribute feature vectors based on the key information, cluster the task attribute feature vectors, classify tasks with high similarity into the same category, and obtain task sets of different categories; Based on each task category of the hotel order, the association rule method is used to mine the dependency relationship between tasks, generate a task dependency graph, determine the execution order and priority of tasks, and dynamically adjust the task allocation strategy according to the task dependency graph and task priority of the hotel order, giving priority to independent tasks and high-priority tasks, while taking into account the parallel processing of tasks; During the hotel order allocation process, the task execution status is monitored in real time to determine whether there is an abnormality or timeout in the task execution status. If so, the task is reallocated to other available processing personnel based on the hotel order task type and customer service working hours; The task dependency graph of hotel orders is converted into a task execution network, in which task nodes contain task attributes, edges represent dependencies between tasks, and the task execution network parameters are dynamically updated. According to the task execution network and task priority, the task allocation strategy is dynamically adjusted to give priority to independent tasks and tasks with high priority, while taking into account parallel processing of tasks.

2. The method according to claim 1, characterized in that The hotel order is obtained, the hotel order content is parsed, and key information is extracted, the key information includes task type, priority and complexity, a task attribute feature vector is constructed according to the key information, the task attribute feature vector is clustered, and tasks with high similarity are classified into the same category to obtain task sets of different categories, including: Obtain hotel order data, perform word segmentation on the order content, and use the TF-IDF algorithm to calculate the word importance score; According to a preset dictionary, the words with higher importance scores are mapped to corresponding numerical values ​​or classification codes to determine a standardized task attribute feature vector; The K-means clustering algorithm is used to cluster the standardized task attribute feature vectors, and the clustering effect of different K values ​​is evaluated by the silhouette coefficient to select the optimal K value; For the clustering results, the average cosine similarity of samples within each cluster and the cosine similarity between clusters are calculated; If the similarity within a cluster is lower than the preset threshold, the cluster is re-divided into sub-clusters; If the similarity between clusters is higher than the preset threshold, these clusters are merged into the same category; Use the optimized clustering results to re-label the original data and generate a training data set with category labels; The task classifier is trained using a random forest algorithm, wherein the input of the task classifier is the attribute feature vector of the new task, and the output is the predicted task category.

3. The method according to claim 1, characterized in that: The method uses association rules to mine the dependencies between tasks for each task category based on the hotel order, generates a task dependency graph, and determines the execution order and priority of the tasks; dynamically adjusts the task allocation strategy according to the task dependency graph and task priority of the hotel order, gives priority to independent tasks and tasks with high priority, and takes into account the parallel processing of tasks, including: Obtain task data from hotel orders, formulate task category standards based on common task types derived from historical data statistical analysis, and classify tasks; The Apriori algorithm is used to calculate the support and confidence between tasks, and the association rules that meet the minimum support and minimum confidence thresholds are selected to obtain the dependency set between tasks. Constructing a directed acyclic graph according to the task dependency set, determining the execution order of the tasks by a topological sorting algorithm, wherein the weight of the topological sorting result is set to a preset value, the original priority weight is set to a preset value, calculating a weighted score of each task, and sorting the tasks according to the score; The dynamic programming algorithm is used to optimize the task execution sequence. The time window parameters are determined according to the average completion time of historical tasks. In each time window, independent tasks and tasks with high priority are given priority for allocation. If the number of independent tasks is insufficient, the tasks with the least dependencies are selected. A task status monitoring mechanism is set up, including heartbeat detection and periodic status reports, wherein the heartbeat detection is sent once at a preset time, and the status report is generated once at a preset time. When a task execution exception occurs, a dynamic adjustment mechanism is triggered to recalculate task priorities and dependencies, and update the task execution sequence and allocation plan.

4. The method according to claim 1, characterized in that In the process of hotel order allocation, the task execution status is monitored in real time to determine whether the task execution status is abnormal or timed out. If so, the task is reallocated to other available processing personnel based on the hotel order task type and customer service working hours, including: Obtain the execution status data of the hotel order task, and use the sliding window algorithm to monitor the task execution progress in real time based on the execution status data to obtain the task completion percentage within the window; If the task completion percentage is lower than the preset threshold, determine whether the task has an execution exception or timeout; By judging the results, a list of abnormal tasks is obtained; For the abnormal task list, a decision tree algorithm is used to classify the abnormal tasks, and the decision tree algorithm uses information gain ratio as a feature selection criterion; Based on the classification results, a task redistribution priority queue is generated; Obtaining currently available processing personnel information, and using a weighted polling algorithm to calculate the workload of the available processing personnel, wherein the weight of the weighted polling algorithm is determined according to the historical task completion efficiency of the processing personnel; Select the most suitable new handler for each exception task based on the workload calculation results; For the selection results, multi-threading technology is used to implement parallel processing mechanism; Sending task assignment requests to multiple new processing personnel through the parallel processing mechanism; Receive the response result of the new processing personnel. If the response result is rejection or no response within the time limit, select the next candidate processing personnel; Repeat the task assignment process until all abnormal tasks are reassigned; In the process of hotel order allocation, the task execution status is monitored in real time to determine whether the task execution status is abnormal or timed out. If so, the task is reallocated to other available processing personnel in combination with the hotel order task type and customer service working hours. It also includes: determining the order distribution range in combination with the preset maximum number of hotel orders that the processing personnel can handle, the number of orders being followed, the shift status and time, and the type of orders followed by each person; at the same time, multiple tasks of the same order are deduplicated to obtain the number of non-repeated orders, and are preferentially allocated to personnel who are processing a small number of orders to ensure that the number of order types followed by each person is basically the same, and the order types include new orders and failed orders; determining whether any hotel order has the same type of problem, if so, it is preferentially assigned to the same person for processing, if not, it is preferentially assigned to the corresponding type of customer service; obtaining the hotel order status at the customer service's off-duty time to determine whether there are any unprocessed orders, if so, the unprocessed orders are reallocated through a one-click return to pool method.

5. The method according to claim 4, characterized in that The order dispatching range is determined by combining the preset maximum number of hotel orders that the processing personnel can handle, the number of orders being followed, the schedule status and time, and the type of orders followed by each person; at the same time, multiple tasks of the same order are deduplicated to obtain the number of non-repeated orders, and are preferentially assigned to personnel who are processing fewer orders to ensure that the number of order types followed by each person is basically the same. The order types include new orders, order failures, and include: Obtain the processing personnel information database, and extract the maximum number of orders that each processing personnel can process, the number of orders currently being followed, the schedule status and time from the database; According to the processing personnel information, a multi-dimensional array is used to store this information, and the current available status of the processing personnel is determined through bit operations; The bit operation method encodes the scheduling status, time period and order type into binary bits, and determines whether the processing personnel meet the current order dispatching conditions through bitwise AND operation; If the handler meets the order dispatching conditions, a matching matrix between the handler and the order type is established. Each element in the matrix represents the handler's processing ability score for a specific order type. According to the matching matrix, a weighted round-robin algorithm is used to calculate the order allocation weight of each processing personnel. The weight calculation formula is: weight = (maximum number of processable orders - current number of orders) * time weight coefficient; The time weight coefficient is calculated using a piecewise linear function, dividing the remaining working time into three intervals: 0-2 hours, 2-5 hours, and more than 5 hours, with corresponding coefficients of 0.8, 1.0, and 1.2, respectively; According to the weight of the processing personnel and the non-duplicate order data, the greedy algorithm is used to allocate orders. The greedy algorithm preferentially allocates new orders and orders that have failed to be issued to the processing personnel with the highest weight.

6. The method according to claim 4, Among them, it is characterized in that the judgment of whether any hotel order has the same type of problem, if so, it is assigned to the same person for processing, if not, it is assigned to the corresponding type of customer service, including: The TF-IDF algorithm is used to vectorize the order problem description and calculate the cosine similarity between the current order and the historical order problems; Determine the type of order problem according to the cosine similarity, and if the cosine similarity exceeds a preset threshold, determine that they are the same type of problems; Obtain order history processing records based on order problem type identifier, and use weighted scoring algorithm to calculate customer service matching degree; In the weighted scoring algorithm, the weight of the number of processing times is 0.3, the weight of customer satisfaction is 0.4, and the weight of average processing time is 0.3; Obtain a customer service matching ranking list through a weighted scoring algorithm; Use customer service expertise tables and workload statistics to build a multi-factor decision tree; The characteristics of the multi-factor decision tree include current workload, expertise matching, estimated completion time, and customer service matching; Evaluate customer service availability through a multi-factor decision tree and generate a customer service priority list; Cross-compare the customer service matching ranking list with the customer service priority list, and use the dynamic programming algorithm to calculate the optimal dispatching plan; Select the dispatch object according to the preset weight ratio; If the same customer service meets both the highest matching degree and the highest priority, the order will be directly assigned to the customer service; Otherwise, the dispatch object is determined according to the ratio of the matching weight to the priority weight; After confirming the order dispatch result, update the order status and customer service workload data.

7. The method according to claim 4, Among them, it is characterized in that the hotel order status of the customer service after get off work is obtained, and whether there are any unprocessed orders is determined. If so, the unprocessed orders are reallocated by a one-key return to pool method, including: Receive a cron expression that carries a preset customer service off-duty time, where the cron expression is used to set a scheduled task; According to the cron expression, the order status check program is started before the preset customer service off-duty time, and the order status check program obtains the unfinished order list of the current online customer service from the database; Obtain the order number, order status, processing progress and remaining estimated processing time in the unfinished order list; Using a decision tree algorithm to classify the orders in the unfinished order list to obtain order classification results, the order classification results including three categories: about to be completed, need to be transferred, and cannot be completed; For orders that need to be transferred or cannot be completed, a one-click return to pool mechanism is activated to reset the status of orders that need to be transferred or cannot be completed to the pending allocation status; Reassign the orders in the pending assignment state according to order priority, customer service skill matching and workload balance; If there is no matching online customer service, the order in the pending allocation state will be added to the queue for processing the next day.

8. The method according to claim 1, characterized in that The task dependency graph of the hotel order is converted into a task execution network, where the task node contains the task attribute, the edge represents the dependency relationship between the tasks, the task execution network parameters are dynamically updated, and the task allocation strategy is dynamically adjusted according to the task execution network and the task priority, so as to give priority to the allocation of independent tasks and tasks with high priority, while taking into account the parallel processing of tasks, including: Acquire hotel order information carrying task dependencies, where the hotel order information is issued by a front-end system; According to the hotel order information, a depth-first search algorithm is used to traverse all nodes in the task dependency graph to obtain a task node sequence; For the task node sequence, convert each task node into an object containing task attributes, wherein the task attributes include task ID, task type, estimated completion time and task priority; Construct the initial structure of the task execution network according to the task attribute object. The initial structure includes the dependency relationship between task nodes. Analyze the topological structure of the task execution network and use the critical path algorithm to calculate the critical path in the network. The critical path algorithm includes forward traversal to calculate the earliest start time of each node and backward traversal to calculate the latest start time of each node. According to the calculation results of the critical path algorithm, determine whether the earliest start time and the latest start time of the node are equal. If they are equal, determine that the node is a node on the critical path; Perform topological sorting on the task execution network, convert it into a directed acyclic graph, sort it according to the in-degree and priority score of the nodes, and obtain the preliminary task execution order; Based on the task execution order, the maximum flow algorithm is used for task allocation. The maximum flow algorithm combines the node capacity limit and regards each task node as a resource point with the maximum processing capability.

9. The method according to claim 1, characterized in that: After the hotel order is completed, the task execution data is counted, the average execution time and completion quality indicators of each task are calculated, and a task execution report is formed to continuously improve the hotel order processing process and task allocation strategy, including: Receive a data collection request carrying task execution log information, wherein the data collection request is triggered when a hotel order is completed; Extracting original data such as the start time, end time, executor and completion status of each task from the task execution log according to the data collection request; Cleaning and standardizing the raw data, removing outliers and invalid data, and obtaining a valid task execution data set; If the task execution time exceeds 3 standard deviations of the historical average time of the task type, it is considered an outlier; If the task execution status is "Not Started" or "Cancelled", it is considered invalid data; Descriptive statistics were used to calculate the effective task execution data set to obtain the average execution time, standard deviation, median and quartile of each task type; Analyze the correlation between tasks in the task execution data set by using an association rule mining algorithm to obtain key nodes that affect order processing efficiency and quality; Integrate the task execution data set, statistical indicators and key node information to generate a task execution report; Using template-based natural language generation technology to automatically generate report text according to the task execution report; Determine whether the task is completed according to different types of hotel orders, including the following five situations: For new orders, the order follower needs to confirm the order recipient's cancellation of the order; For cancellation and modification, the order follower needs to click to confirm the processing; For the processing of supply orders, the order follower needs to enter the supply order results; If the automatic order placement fails, the order follow-up person needs to enter the waiting order result; If automatic confirmation fails, the order follower is required to confirm or cancel the order. When the task is completed, the follower is cleared and the next task is assigned.

10. An automatic order dispatching system for a hotel, characterized in that: The system comprises: The order parsing and task classification module is used to obtain hotel orders, parse the hotel order content, extract key information, including task type, priority and complexity, construct task attribute feature vectors based on the key information, cluster the task attribute feature vectors, classify tasks with high similarity into the same category, and obtain task sets of different categories; The task dependency mining module is used to mine the dependency between tasks based on each task category of the hotel order using the association rule method, generate a task dependency graph, determine the execution order and priority of the tasks, and dynamically adjust the task allocation strategy according to the task dependency graph and task priority of the hotel order, giving priority to independent tasks and high-priority tasks, while taking into account the parallel processing of tasks; The task status monitoring and reallocation module is used to monitor the task execution status in real time during the hotel order allocation process, determine whether the task execution status is abnormal or timed out, and if so, reallocate the task to other available processing personnel based on the hotel order task type and customer service working hours; The task execution network management module is used to convert the task dependency graph of hotel orders into a task execution network. The task nodes contain task attributes, and the edges represent the dependencies between tasks. The task execution network parameters are dynamically updated. According to the task execution network and task priority, the task allocation strategy is dynamically adjusted to give priority to independent tasks and high-priority tasks, while taking into account the parallel processing of tasks.

Images