Method and system for optimizing customer service based on generative artificial intelligence

Abstract

The application discloses a customer service optimization method and system based on generative artificial intelligence, and relates to the technical field of artificial intelligence. The method comprises the following steps: reading the attributes of a retained parcel and the retention duration, combining the site load and the historical average parcel pickup duration of a user to set the sending frequency of a reminder message and configure the sending time zone; retrieving a sample data set from a parcel operation database based on a user portrait and a preset parcel pickup correlation factor; training a generative artificial intelligence model with the sample data to build a parcel pickup probability predictor; and in the message sending time zone, determining the optimal time node corresponding to the maximum predicted parcel pickup probability by using the predictor to optimize, and sending a reminder message to the user. The technical effects of accurate reminding, improved processing efficiency and improved user experience are achieved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a customer service optimization method and system based on generative artificial intelligence. Background Technology

[0002] In the postal customer service sector, SMS alerts are a common method for notifying users about delayed parcels. Traditionally, SMS alerts are triggered at preset fixed times to inform users of their parcel's status. While this method serves a notification purpose to some extent, it has significant limitations and struggles to meet the increasingly complex parcel management and personalized user needs.

[0003] Current technologies for sending SMS alerts for overdue packages cannot adjust in real-time and accurately based on numerous dynamic factors such as the package's actual status, package attributes, site load, and user behavior. Specifically, regardless of the package's attributes, the current site load, or the user's past package pickup habits, alert SMS messages are sent at a uniform, fixed time. This often results in insufficient precision in the timing of alerts. In many cases, SMS messages may be sent when the user is unable to pick up the package, or factors such as site processing capacity and package priority may not be fully considered, thus failing to provide effective alerts, impacting package processing efficiency, and reducing user experience. Summary of the Invention

[0004] This invention provides a customer service optimization method and system based on generative artificial intelligence to solve the technical problems of insufficient accuracy in reminder timing, which affects processing efficiency and user experience in the prior art, and achieves the technical effects of accurate reminders, improved processing efficiency, and improved user experience.

[0005] In a first aspect, the present invention provides a customer service optimization method based on generative artificial intelligence, wherein the customer service optimization method based on generative artificial intelligence includes:

[0006] Read the package attributes and dwell time of the delayed package, and set the real-time sending frequency of reminder SMS messages based on the site load and the user's historical average package pickup time, and configure the SMS sending time zone.

[0007] Using user profiles as comparison constraints, information is retrieved from the package operation database according to preset pickup association factors to obtain a sample dataset.

[0008] Using the sample dataset as training data, a generative artificial intelligence model is trained to construct a pickup probability predictor.

[0009] Using the pickup probability predictor, the optimal time node is determined by optimizing the SMS sending time within the SMS sending time zone, and a reminder SMS is sent to the user at the optimal time node, wherein the optimal time node is the time node corresponding to the maximum predicted pickup probability.

[0010] In one feasible implementation, the package attributes include package volume, item value, and storage environment constraints.

[0011] In one feasible implementation, setting the real-time sending frequency of reminder SMS messages and configuring the SMS sending time zone includes:

[0012] Read the current number of packages at the site, and set the ratio of the current number of packages at the site to a preset quantity threshold as the site load.

[0013] The urgency of package retrieval is assessed based on factors such as package volume, item value, storage environment constraints, dwell time, site load, and the user's historical average retrieval time to determine the urgency level.

[0014] The preset SMS sending frequency is compensated and rounded according to the urgency of the pickup to obtain the real-time sending frequency, and the SMS sending time zone is configured based on the real-time sending frequency.

[0015] In one feasible implementation, the urgency of package pickup is assessed based on factors such as package volume, item value, storage environment constraints, dwell time, site load, and the user's historical average pickup time, including:

[0016] Using the coefficient of variation method and historical parcel operation data, weights are assigned to the parcel volume, item value, storage environment constraints, dwell time, site load, and user's historical average parcel pickup time. An urgency assessment function is constructed based on the weight assignment results, wherein the urgency of pickup is positively correlated with parcel volume, item value, storage environment constraints, dwell time, site load, and user's historical average parcel pickup time.

[0017] After dimensionless processing of the package volume, item value, storage environment constraints, dwell time, site load, and user's historical average pickup time, the urgency of pickup is evaluated and determined using the urgency assessment function.

[0018] In one feasible implementation, user profiles are used as comparison constraints, and information is retrieved from the parcel operation database according to preset pickup association factors to obtain a sample dataset, including:

[0019] Obtain a user profile, which includes basic attribute features and sending and receiving behavior features.

[0020] Configure preset pickup association factors, wherein the preset pickup association factors include at least time nodes, weather conditions, and the number of packages to be picked up at the station.

[0021] Using the user profile as a comparison constraint, information is retrieved from the parcel operation database according to the preset pickup association factors to obtain a sample association factor set. The pickup ratio of different sample association factors within a preset time window is statistically analyzed and set as the sample pickup probability to obtain the sample pickup probability set.

[0022] The sample dataset is obtained by mapping and combining the sample association factor set and the sample selection probability set.

[0023] In one feasible implementation, a generative artificial intelligence model is trained using the sample dataset as training data to construct a pickup probability predictor, including:

[0024] The sample dataset is divided into K equal parts, and selected K times with replacement to construct the first training set. The selection is iteratively repeated K times to obtain K training sets, where K is an integer greater than 5.

[0025] Using the K training sets, generative artificial intelligence models are trained separately until the models converge, resulting in K probability prediction branches for item pickup. These branches are then combined to obtain the item pickup probability predictor.

[0026] In one feasible implementation, the pickup probability predictor is used to optimize the SMS sending time within the SMS sending time zone to determine the optimal time node, including:

[0027] The SMS sending time zone is divided according to a preset time window to determine multiple candidate sending times.

[0028] Based on preset pickup association factors, the association information of the multiple candidate sending times within a future preset time window is collected to obtain multiple real-time association factors.

[0029] Using the pickup probability predictor, the pickup probability is predicted for each of the multiple real-time correlation factors, and multiple predicted pickup probabilities are output. The candidate sending time corresponding to the highest predicted pickup probability is set as the optimal time node.

[0030] In one feasible implementation, the pickup probability predictor is used to predict the pickup probability of the plurality of real-time correlation factors, including:

[0031] The number of branches selected, P, for predicting the probability of pickup are configured according to the urgency of pickup.

[0032] The first real-time correlation factor is randomly selected from the plurality of real-time correlation factors.

[0033] Within the K probability prediction branches of the pickup probability predictor, P pickup probability prediction branches are randomly selected, and the pickup probability is predicted for the first real-time correlation factor respectively to obtain P predicted pickup probabilities. The average value is used to calculate the first predicted pickup probability, and multiple predicted pickup probabilities are obtained by sequential analysis.

[0034] In one feasible implementation, the number P of selection branches for the pickup probability prediction is configured according to the urgency of pickup, including:

[0035] The ratio of the urgency level of the pickup to the preset urgency level is set as the branch adjustment coefficient.

[0036] The product of the branch adjustment coefficient and the initial number of branches is rounded down to obtain the selection number P, where the initial number of branches is 2.

[0037] Secondly, the present invention also provides a customer service optimization system based on generative artificial intelligence, wherein the customer service optimization system based on generative artificial intelligence includes:

[0038] The SMS configuration module is used to read the package attributes and dwell time of the delayed packages, and set the real-time sending frequency of reminder SMS messages based on the site load and the user's historical average package pickup time, and configure the SMS sending time zone.

[0039] The information retrieval module is used to retrieve sample datasets from the parcel operation database by comparing user profiles and according to preset pickup association factors.

[0040] A training module is used to train a generative artificial intelligence model using the sample dataset as training data, and to build a probability predictor for item pickup.

[0041] The prediction and sending module is used to use the pickup probability predictor to optimize the SMS sending time within the SMS sending time zone to determine the optimal time node, and send a reminder SMS to the user at the optimal time node, wherein the optimal time node is the time node corresponding to the maximum predicted pickup probability.

[0042] This invention discloses a customer service optimization method and system based on generative artificial intelligence, comprising: acquiring attribute information and dwell time of pending packages, and dynamically calculating the sending frequency of reminder SMS messages based on the current site's operating load and the user's historical average pickup time, and defining the time period range for SMS sending accordingly; retrieving information from the package operation database according to set pickup-related factors based on user profile data, and extracting a sample dataset for model training; training the generative artificial intelligence model using the sample dataset to construct a pickup probability prediction model for specific users; performing probability optimization analysis within a preset SMS sending time period using the prediction model, and selecting the time point with the highest pickup probability as the optimal time node for SMS sending; sending reminder SMS messages to target users at the optimal time node to improve the actual pickup response rate after SMS delivery. The customer service optimization method and system based on generative artificial intelligence disclosed in this invention solves the technical problems of insufficient accuracy in reminder timing, affecting processing efficiency and user experience, and achieves the technical effects of accurate reminders, improved processing efficiency, and improved user experience. Attached Figure Description

[0043] Figure 1 This is a flowchart illustrating the customer service optimization method based on generative artificial intelligence of the present invention.

[0044] Figure 2 This is a schematic diagram of the customer service optimization system based on generative artificial intelligence according to the present invention.

[0045] Figure labeling: SMS configuration module 11, information retrieval module 12, generation and training module 13, prediction and sending module 14. Detailed Implementation

[0046] The above technical solutions will now be described in detail with reference to the accompanying drawings and specific embodiments to provide a better understanding of them. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. It should be understood that the present invention is not limited to the exemplary embodiments used only to explain the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention. Furthermore, it should be noted that, for ease of description, only the parts related to the present invention are shown in the drawings, not all of them.

[0047] Example 1, as Figure 1 This is a flowchart illustrating the customer service optimization method based on generative artificial intelligence according to the present invention, wherein the customer service optimization method based on generative artificial intelligence includes:

[0048] S100: Reads the package attributes and dwell time of the delayed package, sets the real-time sending frequency of reminder SMS messages based on the site load and the user's historical average pickup time, and configures the SMS sending time zone.

[0049] Specifically, delayed parcels refer to parcels that, during the logistics process, fail to be delivered to the recipient in a timely manner due to various reasons (such as delivery delays, incomplete addresses, or the recipient's failure to pick up the parcel in a timely manner).

[0050] Specifically, the system reads the package attributes and dwell time of delayed parcels. Package attributes include package volume, item value, and storage environment constraints, which comprehensively reflect the characteristics and storage requirements of the parcel. Dwell time refers to the length of time a parcel remains at the station. Additionally, the system considers station load (the relationship between the station's current parcel capacity and processing ability) and the user's historical average parcel pickup time (the average time interval reflected by the user's past pickup behavior).

[0051] By combining these factors, you can set the real-time sending frequency of reminder SMS messages (the number of times an SMS message is sent within a certain period of time) and configure the SMS sending time zone (the time range or time period for sending SMS messages). For example, assuming a station currently has a large number of packages, a heavy load, and a relatively long average pickup time for users, you can set a lower real-time sending frequency and configure the SMS sending time zone to be during the station's working hours and periods when user activity is more frequent, such as 8 pm to 10 pm on weekdays. This ensures that users receive reminders at their convenience, while also avoiding excessive SMS sending during busy periods that could interfere with station staff.

[0052] In some embodiments, the package attributes include package volume, item value, and storage environment constraints.

[0053] Optional package attributes include package volume, item value, and storage environment constraints. Package volume affects the choice of storage space and transportation method; item value relates to the importance of the package and potential insurance needs; and storage environment constraints involve the package's specific requirements for storage conditions, such as whether moisture protection and shock protection are required.

[0054] In some embodiments, setting the real-time sending frequency of reminder SMS messages and configuring the SMS sending time zone includes:

[0055] The current number of packages at the station is read, and the ratio of the current number of packages at the station to a preset quantity threshold is set as the station load. The urgency of package retrieval is assessed based on the package volume, item value, storage environment constraints, dwell time, station load, and the user's historical average retrieval time, and the urgency of retrieval is determined. The preset SMS sending frequency is compensated and rounded according to the urgency of retrieval to obtain the real-time sending frequency, and the SMS sending time zone is configured based on the real-time sending frequency.

[0056] Specifically, "package urgency" refers to a comprehensive quantitative indicator that quantifies the urgency of a package needing to be picked up, based on multiple dimensions of parameters including package attributes (including package volume, item value, and storage environment constraints), package dwell time, site load, and the user's historical average pickup time. "Site load" refers to the ratio of the current number of packages at the site to a preset package quantity threshold, reflecting the real-time status of the site's storage pressure and processing capacity. "SMS sending time zone" refers to the time interval within a day used to allocate SMS reminder sending times, the width of which is dynamically determined by the SMS sending frequency.

[0057] Specifically, the system first reads the current number of packages at the site and then compares this number with a preset package quantity threshold to obtain the current site load value. For example, if the current number of packages is 300 and the threshold is 500, the site load is 0.6. Simultaneously, the system reads information such as the target package's volume, item value, storage environment constraints, dwell time, and the corresponding user's historical average pickup time. Based on these parameters, a weighted or model algorithm is used to assess the urgency of package pickup. For example, packages with larger volume, higher value, or special storage requirements have higher urgency for the same dwell time; the higher the site load, the larger the urgency weighting coefficient.

[0058] Furthermore, the urgency of package pickup is used as a correction factor to compensate for and round the preset SMS sending frequency, resulting in the final real-time SMS sending frequency. For example, if the preset frequency is once a day, and the urgency compensation coefficient is 1.5, the final frequency is rounded up to twice a day. Finally, based on the real-time sending frequency, the 24 hours of a day are divided into a corresponding number of equal-width time zones. For example, if the sending frequency is twice a day, then every 12 hours constitutes an SMS sending time zone, with SMS reminders sent once in the morning and once in the afternoon.

[0059] Through the above process, the dynamic adaptive adjustment of the delayed package notification strategy is achieved. Based on the actual attributes of the package, site load, and user behavior, the frequency and timing of SMS notifications can be optimized in real time, effectively improving the timeliness of package pickup and reducing the risk of long-term package delays. Simultaneously, appropriately configuring the SMS sending time zone avoids users receiving invalid notifications during inactive periods, improving user experience and site operational efficiency.

[0060] In some implementations, the urgency of package pickup is assessed based on factors such as package volume, item value, storage environment constraints, dwell time, site load, and the user's historical average pickup time, including:

[0061] Using the coefficient of variation method and historical parcel operation data, weights are assigned to parcel volume, item value, storage environment constraints, dwell time, site load, and the user's historical average pickup time. An urgency assessment function is constructed based on the weight assignment results, where pickup urgency is positively correlated with parcel volume, item value, storage environment constraints, dwell time, site load, and the user's historical average pickup time. After dimensionless processing of parcel volume, item value, storage environment constraints, dwell time, site load, and the user's historical average pickup time, the urgency of pickup is assessed and determined using the urgency assessment function.

[0062] Specifically, the coefficient of variation method is first used, combined with historical parcel operation data, to assign weights to six factors: parcel volume, item value, storage environment constraints, dwell time, site load, and average historical pickup time for users. That is, based on historical data, the coefficient of variation for each factor is calculated; the larger the coefficient of variation, the greater the weight, because this indicates that the factor changes more significantly under different conditions and has a more critical impact on the urgency of pickup. For example, if the coefficient of variation for parcel volume is 0.3, item value is 0.5, and storage environment constraints are 0.6, then storage environment constraints have the highest weight.

[0063] Furthermore, an urgency assessment function is constructed based on the aforementioned weights. This function can be in the form of the sum of the products of each factor and its weight.

[0064] Specifically, the six indicators mentioned above are dimensionless, for example, by using the min-max normalization method to scale the values ​​of each indicator to the range of [0, 1], and then substituted into the urgency assessment function to calculate the urgency of picking up the package. The higher the urgency value, the more urgent the need for pickup.

[0065] The above process allows for precise quantification of the urgency of each package. This enables differentiated handling based on the urgency level when setting the real-time sending frequency and configuring the SMS sending time zone. For high-urgency packages, the sending frequency can be increased, and reminder SMS messages can be sent preferentially during time periods that better align with users' pickup habits (i.e., the SMS sending time zone). This more effectively encourages users to pick up their packages promptly, reduces package dwell time, improves package processing efficiency, and ultimately enhances the user experience.

[0066] S200: Using user profiles as comparison constraints, information is retrieved from the package operation database according to preset pickup association factors to obtain a sample dataset.

[0067] Specifically, a user profile is a model constructed based on a user's basic attribute characteristics (such as age, gender, occupation, etc.) and delivery behavior characteristics (such as pickup frequency, pickup time preference, etc.). It is used to comprehensively describe user characteristics and can reflect users' personalized needs and behavioral patterns. In the information retrieval process, the user profile is used as a condition constraint (i.e., comparison constraint) to ensure that the retrieved information matches the user characteristics, thereby improving the accuracy and relevance of the retrieval.

[0068] Specifically, the preset pickup association factors are various factors related to user pickup behavior, such as time of day, weather conditions, and the number of packages waiting to be picked up at the station. These factors may affect the likelihood of a user picking up their package. The package operation database stores a large amount of data related to package operations, including basic package information, user drop-off and pick-up records, and station operational status, serving as the fundamental data source for information retrieval and data analysis. The sample dataset is a collection of data retrieved and filtered from the package operation database that matches user profiles and preset pickup association factors, used for subsequent model training.

[0069] This step allows for the precise filtering of data from massive parcel operation data that matches specific user profiles and preset pickup correlation factors, forming a high-quality sample dataset. This provides a rich and accurate data foundation for subsequent training of generative artificial intelligence models, enabling the models to better learn the patterns and characteristics of user pickup behavior.

[0070] In some embodiments, using user profiles as comparison constraints, information is retrieved from the package operation database according to preset pickup association factors to obtain a sample dataset, including:

[0071] A user profile is obtained, comprising basic attribute features and parcel collection and delivery behavior features; preset parcel collection association factors are configured, comprising at least time nodes, weather conditions, and the number of parcels to be collected at the station; using the user profile as a comparison constraint, information is retrieved from the parcel operation database according to the preset parcel collection association factors to obtain a sample association factor set, and the parcel collection ratio of different sample association factors within a preset time window is statistically analyzed and set as the sample parcel collection probability to obtain a sample parcel collection probability set; the sample association factor set and the sample parcel collection probability set are mapped and combined to obtain the sample dataset.

[0072] Specifically, basic attribute characteristics refer to a user's basic information such as age, gender, occupation, and residential area. Parcel collection and delivery behavior characteristics refer to a user's behavioral patterns during the parcel collection and delivery process, such as pickup frequency, frequently selected pickup time periods, and preferred pickup locations. For example, parcel collection and delivery behavior characteristics include:

[0073] Table 1. Exemplary characteristics of mail receiving and sending behavior

[0074] Feature Name Meaning Explanation Average pickup time Average time from when the package arrives at the station to when the user actually picks it up Pickup time preference Common pickup times for users (such as weekday evenings, weekend days, etc.) Pickup frequency Number of packages picked up by the user per week or month Package type preference The types of goods that users frequently receive (such as fresh produce, electronics, books, etc.) Historical retention rate The proportion of user packages exceeding the set retention threshold Frequently used sites User's preferred pickup locations Complaint / Feedback Records Do users frequently complain about package delays, service issues, etc.?

[0075] Specifically, time points refer to time factors related to a user's package pickup behavior, including specific hours, dates, and days of the week, such as weekday evenings or weekend afternoons. Weather conditions refer to the weather conditions in the user's area during the pickup period, such as sunny, rainy, or snowy weather; different weather conditions may affect a user's willingness to go out and pick up their package. The number of packages waiting to be picked up at a particular station refers to the number of packages waiting for users to pick up at a specific station. This value may affect the convenience and willingness of users to pick up their packages; for example, too many packages at a station may cause users to spend more time searching for their packages.

[0076] Specifically, the sample association factor set is a data set retrieved from the database that is related to the preset pickup association factors. Each sample association factor represents various influencing factors in a specific pickup scenario.

[0077] Specifically, first, user profiles containing basic attribute characteristics and sending / receiving behavior characteristics are obtained to comprehensively understand user situations. Next, preset pickup-related factors are configured, including at least time points, weather conditions, and the number of packages waiting to be picked up at the site. Then, using the user profiles as comparison constraints, information is retrieved from the package operation database. Database query technology is used to match the user profiles with user records in the database, filtering out data that meets the conditions to form a sample set of related factors. Finally, within a preset time window, the pickup ratio for different sample related factors is statistically analyzed to obtain a sample pickup probability set.

[0078] For example, in the past month, for a certain user, under the conditions of weekday evenings between 8 PM and 10 PM, sunny weather, and fewer than 50 packages waiting to be picked up at the pickup station, the number of successful pickups was 8, and the number of packages meeting the conditions was 10. The corresponding sample pickup probability is 80%. Finally, the sample association factor set and the sample pickup probability set are mapped and combined to obtain the sample dataset. For example, the above sample association factors (weekday evenings between 8 PM and 10 PM, sunny weather, fewer than 50 packages waiting to be picked up at the pickup station) are combined with the corresponding sample pickup probability of 80% to form a data record; multiple such records constitute the sample dataset.

[0079] Through the above process, based on user profiles and multi-dimensional pickup-related factors, we can accurately mine and quantify the probability of users' actual pickup behavior in different scenarios, forming a personalized, data-driven behavioral analysis sample dataset. This provides support for subsequent applications such as pickup behavior prediction, intelligent reminder strategy formulation, and user segmentation operations, and helps improve package circulation efficiency and user service experience.

[0080] S300: Using the sample dataset as training data, train a generative artificial intelligence model to construct a pickup probability predictor.

[0081] In some embodiments, using the sample dataset as training data, a generative artificial intelligence model is trained to construct a pickup probability predictor, including:

[0082] The sample dataset is divided into K equal parts, and selected K times with replacement to construct the first training set. The first training set is then selected K times iteratively to obtain K training sets, where K is an integer greater than 5. Generative artificial intelligence models are trained using the K training sets until the models converge to obtain K probability prediction branches for item pickup. These branches are then combined to obtain the item pickup probability predictor.

[0083] Specifically, a "generative AI model" is a machine learning model capable of generating or predicting output results based on input features, preferably a deep neural network, a generative adversarial network (GAN), or other generative models suitable for behavioral probability modeling. A "pickup probability predictor" is a specialized generative AI model trained for the pickup probability prediction task, used to infer from the input user profile and pickup-related factors to obtain the user's pickup probability in a specific scenario.

[0084] Specifically, firstly, the obtained sample dataset is divided into K equal parts (K is an integer greater than 5, preferably 10). Using sampling with replacement, several samples from the K parts are randomly selected each time to construct the first training set. This sampling process is repeated K times to obtain K overlapping training sets. Then, a generative artificial intelligence model is trained based on each training set. For example, during training, appropriate loss functions such as cross-entropy loss and mean squared error are used until the model converges, resulting in K independent probability prediction branches for item pickup. Next, these K probability prediction branches are integrated, for example, using a voting method, averaging method, or weighted ensemble method, to construct the final probability predictor for item pickup. This predictor can output the corresponding probability prediction value for any input user profile and item pickup association factors.

[0085] Through the above process, by utilizing sampling with replacement and multi-model ensemble, the generalization ability and robustness of the pickup probability predictor are effectively improved, while reducing the risk of overfitting that may occur with a single model. The resulting pickup probability predictor can accurately predict the pickup probability under different user profiles and scenario-related factors, providing highly reliable decision support for applications such as intelligent reminders and operational optimization.

[0086] S400: Using the pickup probability predictor, optimize the SMS sending time within the SMS sending time zone to determine the optimal time node, and send a reminder SMS to the user at the optimal time node, wherein the optimal time node is the time node corresponding to the maximum predicted pickup probability.

[0087] Specifically, the SMS sending time zone refers to a pre-defined range or set of time periods within which SMS messages can be sent. This range can be determined based on a combination of factors, including user behavior habits and site operating hours, and is used to limit the possible time range for SMS sending. Within the SMS sending time zone, a pickup probability predictor can be used to find the optimal SMS sending time, maximizing the likelihood that the sent SMS messages will prompt users to pick up their packages.

[0088] In some embodiments, the pickup probability predictor is used to optimize the SMS sending time within the SMS sending time zone to determine the optimal time node, including:

[0089] The SMS sending time zone is divided according to a preset time window to determine multiple candidate sending times; based on a preset pickup association factor, the association information of the multiple candidate sending times within a future preset time window is collected to obtain multiple real-time association factors; using the pickup probability predictor, the pickup probability is predicted for the multiple real-time association factors, multiple predicted pickup probabilities are output, and the candidate sending time corresponding to the maximum predicted pickup probability is set as the optimal time node.

[0090] Specifically, the preset time window refers to a pre-defined interval of SMS sending time granularity (such as every 10 minutes or every 30 minutes) based on business needs or user behavior characteristics, used to divide the SMS sending time zone into multiple candidate time periods. Pre-defined pickup correlation factors refer to relevant variables affecting user pickup behavior, including but not limited to user's historical pickup time, geographical location, weather, holidays, courier type, user preferences, etc., used to help predict the probability of a user picking up their package within a certain time period. Real-time correlation factors are multi-dimensional dynamic data related to user pickup behavior collected in real-time within the future preset time window for each candidate sending time node.

[0091] Specifically, firstly, the SMS sending time zone (e.g., 8:00-20:00) is divided into several candidate sending time nodes (e.g., 8:00, 8:15, 8:30, ..., 20:00) according to a preset time window (e.g., every 15 minutes), with each node representing a candidate SMS sending time. Then, for each candidate sending time node, real-time data for that time node within a future preset time window is collected based on preset pickup correlation factors. For example, data such as the user's historical pickup probability during that time period, weather conditions, remaining capacity of the parcel locker, and holiday information are collected to obtain multiple real-time correlation factors.

[0092] Furthermore, using a pickup probability predictor, the pickup probability is predicted for each of these multiple real-time correlation factors, outputting multiple predicted pickup probabilities, and setting the candidate sending time corresponding to the highest predicted pickup probability as the optimal time node.

[0093] For example, if the pickup probability predictor predicts that the pickup probability is 0.6 when the SMS is sent at 8:00, 0.7 when it is sent at 8:30, and 0.65 when it is sent at 9:00, then 8:30 can be set as the optimal time node, and a reminder SMS should be sent to the user at this time.

[0094] Through the above process, the optimal time to send reminder SMS messages can be accurately determined within the SMS sending time zone. By fully considering various factors influencing user package pickup behavior, such as weather, parcel volume at the station, and their actual conditions at different times, and combining this with a pickup probability predictor, the effectiveness of SMS reminders can be maximized. This ensures users receive reminders at the most likely pickup time, thereby improving pickup efficiency. Simultaneously, this has significant effects on optimizing parcel processing workflows, increasing parcel processing speed, and improving user experience. Users receive reminders at appropriate times, reducing parcel dwell time and enhancing service satisfaction.

[0095] In some implementations, the pickup probability predictor is used to predict the pickup probability of the plurality of real-time correlation factors, including:

[0096] The number of selection branches for the pickup probability prediction is P, configured according to the urgency of pickup. A first real-time correlation factor is randomly selected from the multiple real-time correlation factors. P pickup probability prediction branches are randomly selected from the K pickup probability prediction branches of the pickup probability predictor. Pickup probability prediction is performed on the first real-time correlation factor respectively to obtain P predicted pickup probabilities. The average value is used to calculate the first predicted pickup probability. Multiple predicted pickup probabilities are then analyzed sequentially.

[0097] Specifically, based on the preset mapping relationship or corresponding rules, the number of probability prediction branches P (P≤K) to be selected for this prediction is automatically configured according to the urgency of the current user's pickup. That is, P takes a larger value when the urgency is high and P takes a smaller value when the urgency is low.

[0098] Furthermore, among the multiple real-time correlation factors to be analyzed, one is randomly selected as the first real-time correlation factor. Then, among the K prediction branches of the pickup probability predictor, P branches are randomly selected, and the first real-time correlation factor is input into each of these P branches to obtain P pickup probability prediction results. Finally, the average of these P prediction results is taken as the final predicted pickup probability for the first real-time correlation factor.

[0099] Following the steps described above, the process of branch selection, prediction, and mean calculation can be repeated for the remaining real-time related factors to ultimately obtain multiple predicted pick-up probabilities.

[0100] The above-described method significantly improves the robustness and generalization ability of pickup probability prediction by using multi-branch ensemble prediction and mean fusion, while reducing errors caused by fluctuations in single-branch prediction. Dynamically adjusting the number of branches based on pickup urgency helps balance prediction efficiency and accuracy, enabling intelligent and flexible optimization of customer service.

[0101] In some implementations, the number P of selection branches for the pickup probability prediction is configured according to the urgency of pickup, including:

[0102] The ratio of the urgency of item retrieval to the preset urgency of item retrieval is set as the branch adjustment coefficient; the product of the branch adjustment coefficient and the initial number of branches is rounded down to obtain the selection number P, wherein the initial number of branches is 2.

[0103] Specifically, the preset pickup urgency level refers to a pre-defined standard urgency threshold, serving as a reference for urgency grading and branch adjustment. The branch adjustment coefficient is the ratio of the current pickup urgency level to the preset pickup urgency level, used to dynamically adjust the number of predicted branches. The initial number of branches is the default minimum pickup probability predicted branch number, typically 2.

[0104] Specifically, first, the ratio of the current urgency level to the preset urgency level is calculated to obtain the branch adjustment coefficient. For example, if the current urgency level is 4 and the preset urgency level is 2, then the branch adjustment coefficient is 4 / 2 = 2. Then, the branch adjustment coefficient is multiplied by the initial number of branches (e.g., 2), and the result is rounded to obtain the number of branches P to be selected in this prediction. For example, if the branch adjustment coefficient is 2 and the initial number of branches is 2, then P = 2 × 2 = 4.

[0105] Optionally, an upper limit can be set for P based on the actual application environment to prevent excessive number of branches from wasting computing resources.

[0106] The above process dynamically adjusts the number of branches by the ratio of the urgency of the item to the preset urgency, thus achieving an adaptive balance between prediction accuracy and efficiency under different urgency scenarios.

[0107] In summary, the customer service optimization method based on generative artificial intelligence provided by this invention has the following technical effects:

[0108] By acquiring the attribute information and dwell time of pending packages, and combining this with the current site's operational load and the user's historical average pickup time, the frequency of sending reminder SMS messages is dynamically calculated, and the time period for SMS sending is defined accordingly. Based on user profile data, information is retrieved from the package operation database according to the set pickup-related factors to extract sample datasets for model training. The generative artificial intelligence model is trained using the sample datasets to construct a pickup probability prediction model for specific users. Using this prediction model, probability optimization analysis is performed within the preset SMS sending time period to select the time point with the highest pickup probability as the optimal time node for SMS sending. Reminder SMS messages are sent to target users at the optimal time node to improve the actual pickup response rate after SMS delivery, thereby achieving the technical effects of accurate reminders, improved processing efficiency, and improved user experience.

[0109] Example 2, as Figure 2 This is a schematic diagram of the customer service optimization system based on generative artificial intelligence according to the present invention. For example, Figure 1 The flowchart of the customer service optimization method based on generative artificial intelligence of this invention can be illustrated as follows: Figure 2 The structure shown is implemented.

[0110] Based on the same concept as the customer service optimization method based on generative artificial intelligence in the above embodiments, the present invention also provides a customer service optimization system based on generative artificial intelligence, which includes:

[0111] The SMS configuration module 11 is used to read the package attributes and dwell time of the delayed package, and set the real-time sending frequency of reminder SMS messages in combination with the site load and the user's historical average pickup time, and configure the SMS sending time zone.

[0112] The information retrieval module 12 is used to retrieve information from the parcel operation database based on user profiles and preset pickup association factors to obtain a sample dataset.

[0113] The generation training module 13 is used to train a generative artificial intelligence model using the sample dataset as training data and to build a pickup probability predictor.

[0114] The prediction and sending module 14 is used to use the pickup probability predictor to optimize the SMS sending time within the SMS sending time zone to determine the optimal time node, and send a reminder SMS to the user at the optimal time node, wherein the optimal time node is the time node corresponding to the maximum predicted pickup probability.

[0115] In some embodiments, the SMS configuration module 11 includes:

[0116] The site load calculation unit is used to read the current number of packages at the site and set the ratio of the current number of packages at the site to a preset quantity threshold as the site load.

[0117] The urgency assessment unit is used to assess the urgency of package retrieval based on the package volume, item value, storage environment constraints, dwell time, site load, and the user's historical average retrieval time, and to determine the urgency level of the retrieval.

[0118] The SMS sending frequency configuration unit is used to compensate and round the preset SMS sending frequency according to the urgency of the package pickup to obtain the real-time sending frequency, and configure the SMS sending time zone based on the real-time sending frequency.

[0119] In some implementations, the execution steps of the urgency assessment unit in the SMS configuration module 11 include:

[0120] Using the coefficient of variation method and historical parcel operation data, weights are assigned to the parcel volume, item value, storage environment constraints, dwell time, site load, and user's historical average parcel pickup time. An urgency assessment function is constructed based on the weight assignment results, wherein the urgency of pickup is positively correlated with parcel volume, item value, storage environment constraints, dwell time, site load, and user's historical average parcel pickup time.

[0121] After dimensionless processing of the package volume, item value, storage environment constraints, dwell time, site load, and user's historical average pickup time, the urgency of pickup is evaluated and determined using the urgency assessment function.

[0122] In some embodiments, the information retrieval module 12 includes:

[0123] The user profile acquisition unit is used to acquire user profiles, wherein the user profiles include basic attribute features and mail receiving and sending behavior features.

[0124] A preset pickup association factor configuration unit is used to configure preset pickup association factors, wherein the preset pickup association factors include at least time nodes, weather conditions, and the number of packages to be picked up at the station.

[0125] The sample association factor retrieval and pickup probability statistics unit is used to retrieve information in the parcel operation database according to the preset pickup association factors, with the user profile as the comparison constraint, to obtain the sample association factor set, and to count the pickup ratio of different sample association factors within a preset time window, which is set as the sample pickup probability, to obtain the sample pickup probability set.

[0126] The sample dataset generation unit is used to map and combine the sample association factor set and the sample selection probability set to obtain the sample dataset.

[0127] In some embodiments, the training generation module 13 includes:

[0128] The sample dataset equal division and training set construction unit is used to divide the sample dataset into K equal parts, select K times with replacement to construct the first training set, and iteratively select K times to obtain K training sets, where K is an integer greater than 5.

[0129] The generative artificial intelligence model training and predictor building unit is used to train the generative artificial intelligence model using the K training sets until the model converges, obtain K probability prediction branches for item pickup, and combine them to obtain the probability predictor for item pickup.

[0130] In some embodiments, the prediction and transmission module 14 includes:

[0131] The candidate sending time determination unit is used to divide the SMS sending time zone according to a preset time window and determine multiple candidate sending times.

[0132] The real-time correlation factor acquisition unit is used to collect correlation information of the multiple candidate sending times within a future preset time window based on preset pickup correlation factors, and obtain multiple real-time correlation factors.

[0133] The optimal time node determination unit is used to use the pickup probability predictor to predict the pickup probability of the multiple real-time correlation factors, output multiple predicted pickup probabilities, and set the candidate sending time corresponding to the maximum predicted pickup probability as the optimal time node.

[0134] In some implementations, the execution steps of the optimal time node determination unit in the prediction and transmission module 14 include:

[0135] The number of branches selected, P, for predicting the probability of pickup are configured according to the urgency of pickup.

[0136] The first real-time correlation factor is randomly selected from the plurality of real-time correlation factors.

[0137] Within the K probability prediction branches of the pickup probability predictor, P pickup probability prediction branches are randomly selected, and the pickup probability is predicted for the first real-time correlation factor respectively to obtain P predicted pickup probabilities. The average value is used to calculate the first predicted pickup probability, and multiple predicted pickup probabilities are obtained by sequential analysis.

[0138] In some implementations, the execution steps of the optimal time node determination unit in the prediction and transmission module 14 further include:

[0139] The ratio of the urgency level of the pickup to the preset urgency level is set as the branch adjustment coefficient.

[0140] The product of the branch adjustment coefficient and the initial number of branches is rounded down to obtain the selection number P, where the initial number of branches is 2.

[0141] It should be understood that the focus of the embodiments mentioned in this specification is their difference from other embodiments. The specific embodiments in the aforementioned Embodiment 1 are also applicable to the customer service optimization system based on generative artificial intelligence described in Embodiment 2. For the sake of brevity, they will not be elaborated further here.

[0142] It should be understood that the embodiments disclosed in this invention and the above description enable those skilled in the art to implement this invention. However, this invention is not limited to the embodiments mentioned above. It should be understood that those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this invention, and should all be included within the protection scope of this invention.

Claims

1. A customer service optimization method based on generative artificial intelligence, characterized in that, include: Read the package attributes and dwell time of the delayed package, and set the real-time sending frequency of reminder SMS messages based on the site load and the user's historical average pickup time, and configure the SMS sending time zone; Using user profiles as comparison constraints, information is retrieved from the parcel operation database according to preset pickup association factors to obtain a sample dataset. The user profiles include basic attribute features and sending and receiving behavior features, and the preset pickup association factors include at least time nodes, weather conditions, and the number of parcels waiting to be picked up at the site. Using the aforementioned sample dataset as training data, a generative artificial intelligence model is trained to construct a pickup probability predictor, including: The sample dataset is divided into K equal parts, and selected K times with replacement to construct the first training set. The selection is iteratively repeated K times to obtain K training sets, where K is an integer greater than 5. Using the K training sets, generative artificial intelligence models are trained separately until the models converge, resulting in K probability prediction branches for item pickup, which are then combined to obtain the item pickup probability predictor. Using the pickup probability predictor, the optimal time node is determined by optimizing the SMS sending time within the SMS sending time zone, and a reminder SMS is sent to the user at the optimal time node, wherein the optimal time node is the time node corresponding to the maximum predicted pickup probability. The step of using the pickup probability predictor to optimize the SMS sending time within the SMS sending time zone to determine the optimal time node includes: The SMS sending time zone is divided according to a preset time window to determine multiple candidate sending times; Based on preset pickup association factors, the association information of the multiple candidate sending times within a future preset time window is collected to obtain multiple real-time association factors. Using the pickup probability predictor, the pickup probability is predicted for each of the multiple real-time correlation factors, and multiple predicted pickup probabilities are output. The candidate sending time corresponding to the highest predicted pickup probability is set as the optimal time node.

2. The customer service optimization method based on generative artificial intelligence according to claim 1, characterized in that, The package attributes include package volume, item value, and storage environment constraints.

3. The customer service optimization method based on generative artificial intelligence according to claim 2, characterized in that, The setting of the real-time sending frequency of reminder SMS messages and the configuration of the SMS sending time zone include: Read the current number of packages at the site, and set the ratio of the current number of packages at the site to a preset quantity threshold as the site load; The urgency of package retrieval is assessed based on factors such as package volume, item value, storage environment constraints, dwell time, site load, and the user's historical average retrieval time to determine the urgency level. The preset SMS sending frequency is compensated and rounded according to the urgency of the pickup to obtain the real-time sending frequency, and the SMS sending time zone is configured based on the real-time sending frequency.

4. The customer service optimization method based on generative artificial intelligence according to claim 3, characterized in that, The urgency of package pickup is assessed based on factors such as package volume, item value, storage environment constraints, dwell time, site load, and the user's historical average pickup time, including: Using the coefficient of variation method and combined with historical parcel operation data, weights are configured for parcel volume, item value, storage environment constraints, dwell time, site load, and user's historical average parcel pickup time. An urgency assessment function is constructed based on the weight configuration results, wherein the urgency of pickup is positively correlated with parcel volume, item value, storage environment constraints, dwell time, site load, and user's historical average parcel pickup time. After dimensionless processing of the package volume, item value, storage environment constraints, dwell time, site load, and user's historical average pickup time, the urgency of pickup is evaluated and determined using the urgency assessment function.

5. The customer service optimization method based on generative artificial intelligence according to claim 1, characterized in that, Using user profiles as comparison constraints, information is retrieved from the package operation database according to preset pickup association factors to obtain a sample dataset, including: Obtain user profiles; Configure preset pickup association factors; Using the user profile as a comparison constraint, information is retrieved in the parcel operation database according to the preset pickup association factors to obtain a sample association factor set, and the pickup ratio of different sample association factors within a preset time window is statistically analyzed and set as the sample pickup probability to obtain a sample pickup probability set. The sample dataset is obtained by mapping and combining the sample association factor set and the sample selection probability set.

6. The customer service optimization method based on generative artificial intelligence according to claim 1, characterized in that, Using the aforementioned pickup probability predictor, the pickup probability is predicted for each of the multiple real-time correlation factors, including: Configure the number P of the probability prediction branches to be selected according to the urgency of the pickup. Randomly select a first real-time correlation factor from the plurality of real-time correlation factors; Within the K probability prediction branches of the pickup probability predictor, P pickup probability prediction branches are randomly selected, and the pickup probability is predicted for the first real-time correlation factor respectively to obtain P predicted pickup probabilities. The average value is used to calculate the first predicted pickup probability, and multiple predicted pickup probabilities are obtained by sequential analysis.

7. The customer service optimization method based on generative artificial intelligence according to claim 6, characterized in that, The number P of selection branches for the pickup probability prediction is configured according to the urgency of pickup, including: The ratio of the urgency level of the pickup to the preset urgency level of the pickup is set as the branch adjustment coefficient; The product of the branch adjustment coefficient and the initial number of branches is rounded down to obtain the selection number P, where the initial number of branches is 2.

8. A customer service optimization system based on generative artificial intelligence, characterized in that, The customer service optimization method based on generative artificial intelligence as described in any one of claims 1 to 7 includes: The SMS configuration module is used to read the package attributes and dwell time of the delayed package, and set the real-time sending frequency of reminder SMS messages based on the site load and the user's historical average pickup time, and configure the SMS sending time zone; The information retrieval module is used to retrieve information from the parcel operation database based on user profiles and preset pickup association factors to obtain sample datasets. A training module is used to train a generative artificial intelligence model using the sample dataset as training data, and to build a probability predictor for item pickup. The prediction and sending module is used to use the pickup probability predictor to optimize the SMS sending time within the SMS sending time zone to determine the optimal time node, and send a reminder SMS to the user at the optimal time node, wherein the optimal time node is the time node corresponding to the maximum predicted pickup probability. The information retrieval module includes: User profile acquisition unit, wherein the user profile includes basic attribute features and mail receiving and sending behavior features; A preset pickup association factor configuration unit, wherein the preset pickup association factors include at least time nodes, weather conditions, and the number of packages to be picked up at the site; The training generation module includes: The sample dataset equal division and training set construction unit is used to divide the sample dataset into K equal parts, select them K times with replacement to construct the first training set, and iteratively select them K times to obtain K training sets, where K is an integer greater than 5; The generative artificial intelligence model training and predictor building unit is used to train the generative artificial intelligence model using the K training sets respectively until the model converges, and obtain K probability prediction branches for item pickup, which are then combined to obtain the probability predictor for item pickup. The prediction and transmission module includes: The candidate sending time determination unit is used to divide the SMS sending time zone according to a preset time window and determine multiple candidate sending times; The real-time correlation factor acquisition unit is used to collect the correlation information of the multiple candidate sending times within a future preset time window based on preset item picking correlation factors, and obtain multiple real-time correlation factors. The optimal time node determination unit is used to use the pickup probability predictor to predict the pickup probability of the multiple real-time correlation factors, output multiple predicted pickup probabilities, and set the candidate sending time corresponding to the maximum predicted pickup probability as the optimal time node.

Images