An intelligent schedule management method, system, device and medium

Abstract

The application is suitable for the field of computer technology, and provides an intelligent schedule management method, which comprises the following steps: collecting multi-dimensional data related to schedule items of a user, adopting a federated learning technology to construct a dynamic user portrait model; identifying a multi-dimensional user context and inferring a potential schedule intention of the user based on the dynamic user portrait model and in combination with the multi-dimensional data; calculating multi-dimensional competition indexes of each schedule item in terms of time, attention, physical strength and external resources, and generating a plurality of candidate schedule schemes based on the multi-dimensional competition indexes and labeling a priority weight for each schedule item in each candidate schedule scheme; generating a preliminary schedule scheme by adopting a generative model based on the plurality of candidate schedule schemes; generating a visual interface for the preliminary schedule scheme for the user to adjust and confirm, and generating a target schedule scheme. The application greatly reduces manual input of the user, improves user experience, and retains and enhances the final control right and decision-making experience of the user.

Description

Technical Field

[0001] This application belongs to the field of computer technology, and in particular relates to an intelligent schedule management method, system, device and medium. Background Technology

[0002] In modern society, efficient management of personal and collaborative schedules has become increasingly important for improving work efficiency and quality of life. Traditional schedule management tools, such as electronic calendar applications, rely mainly on users manually entering schedule items and setting reminders. They are essentially passive recording and reminder tools with limited functionality and intelligence.

[0003] In existing technologies, some improvement solutions attempt to introduce simple rule reminders or pattern learning based on historical records, but certain shortcomings remain: First, at the intelligence level, existing tools lack deep learning of user behavior patterns and the ability to perceive real-time contexts, failing to proactively infer users' potential intentions and provide forward-looking scheduling suggestions, resulting in insufficient intelligence. Second, in terms of optimization, existing solutions primarily focus on the allocation of time resources, neglecting the fact that scheduling is essentially a collaborative optimization problem involving multiple dimensions of resources such as time, attention, physical strength, and external resources (e.g., meeting rooms, transportation), leading to a singular management dimension. Furthermore, regarding the flexibility of the scheduling structure, rigid scheduling based on fixed time points is exceptionally fragile in the face of sudden changes, lacking an inherent flexible adjustment mechanism.

[0004] In summary, existing technical solutions have significant room for improvement in terms of proactive intelligence, multi-dimensional resource collaborative optimization, and flexible adaptability, and there is an urgent need to propose a more advanced intelligent schedule management method. Summary of the Invention

[0005] This application provides an intelligent schedule management method, system, device, and medium that can solve one of the problems of the prior art mentioned above.

[0006] In a first aspect, embodiments of this application provide an intelligent schedule management method, including: Collect multi-dimensional data related to user schedules and use federated learning technology to build a dynamic user profile model; Based on the dynamic user profile model and combined with the multi-dimensional data, multi-dimensional user contexts are identified, and potential user schedule intentions are inferred. Calculate the multidimensional competition index of each schedule item in terms of time, attention, physical strength and external resources, and generate multiple candidate schedule schemes based on the multidimensional competition index, and assign priority weights to each schedule item in each candidate schedule scheme; Based on multiple candidate scheduling schemes, a generative model is used to generate a preliminary scheduling scheme; A visual interface is generated for the preliminary schedule plan, which is then adjusted and confirmed by the user to generate the target schedule plan.

[0007] Furthermore, the collection of multi-dimensional data related to user schedules and events, using federated learning technology, constructs a dynamic user profile model, including: The user terminal device performs anonymization and feature engineering on the multi-dimensional data to extract structured feature vectors that reflect user behavior patterns. Based on the structured feature vectors and user historical interaction feedback data, the local user model is trained so that the local user model can predict user schedule preferences and user behavior patterns. A global user profile model is generated by aggregating encrypted local user model parameter updates from multiple user terminal devices through a central server. The global user profile model is securely updated and distributed to each user terminal device, and then merged with the local user model of each user terminal device to form a dynamic user profile model. The dynamic user profile model includes a user time preference vector, a user behavior pattern prediction vector, and a resource consumption baseline. The multi-dimensional data includes schedule data, application log data, device context data, and biosensor data.

[0008] Furthermore, the identification of multi-dimensional user contexts based on the dynamic user profile model and the multi-dimensional data includes: Based on the device context data and combined with the dynamic user profile model, the physical context is identified, where the physical context is the objective physical environment in which the user is located. Based on the application log data and combined with the dynamic user profile model, digital contexts are identified, where the digital context refers to the user's engagement state in the application's digital space. Based on the biosensor data, combined with the dynamic user profile model, the physical context, and the digital context, the psychological context is inferred, which refers to the user's internal cognitive and emotional state.

[0009] Furthermore, the calculation of the multidimensional competition index of each schedule item in terms of time, attention, physical strength, and external resources includes: Calculate the time conflict index between schedule items based on the overlap of their time intervals. Based on the attention consumption value of the schedule items and the user's real-time cognitive state, calculate the attention competition index of the schedule items. Based on the physical exertion value of the scheduled tasks and the user's real-time energy status, calculate the physical exertion competition index of the scheduled tasks. Based on the dependence of schedule items on external resources and the real-time availability of resources, the external competition index of schedule items is calculated.

[0010] Furthermore, the step of generating multiple candidate scheduling schemes based on the multidimensional competition index, and assigning priority weights to each schedule item in each candidate scheduling scheme, includes: Use the planned execution time and priority weight of multiple schedule items as decision variables; Multiple optimization objectives are set, including minimizing resource conflicts, maximizing user preference satisfaction, and maximizing schedule stability. Using multi-objective evolutionary algorithms or Pareto optimization algorithms, a set of candidate scheduling schemes is generated under the constraints of time boundaries, resource availability, and user state security thresholds for each item. Calculate the dynamic priority weight for each schedule item in the multiple candidate schedule schemes. The priority weight is a function of the inherent attributes of the item, the real-time competitive situation, and its contribution to the global optimization objective, wherein the real-time competitive situation is the importance of the item under the multi-dimensional competition index.

[0011] Furthermore, the step of generating a preliminary schedule plan based on multiple candidate schedule plans using a generative model includes: Data transformation and knowledge encapsulation are performed on each candidate schedule plan to generate seed knowledge; Based on the seed knowledge, the generation process of the generative model is guided and constrained by at least one of the following methods: enhanced generation, conditional vector guidance, or model initialization. In the generative model, at least one of the direct optimization strategy or the integrated innovation strategy is used to generate a preliminary schedule.

[0012] Furthermore, the step of generating a visual interface for the preliminary schedule plan, allowing the user to adjust and confirm it, and generating the target schedule plan, includes: The suggested start and end times allocated to each schedule item in the preliminary schedule plan are defined as the core interval of that schedule item. Based on the priority weight, item type, and historical user behavior data of each scheduled item, elastic buffers located before and after the core interval are dynamically calculated and allocated to form fuzzy time boundaries for the scheduled items. On the visual timeline interface, each schedule item is rendered as a time period color block, wherein the time period color block corresponds to the core interval; When a user clicks on the time period color block of the corresponding core interval, elastic color blocks are added to both ends of the time period color block of each core interval to represent the elastic buffer zone. The visual timeline interface allows users to dynamically adjust the color blocks for the specified time periods and detects and reports time conflicts in real time during user operations.

[0013] Secondly, embodiments of this application provide an intelligent schedule management system, including: The first processing module is used to collect multi-dimensional data related to user schedules and uses federated learning technology to build a dynamic user profile model. The second processing module is used to identify multi-dimensional user contexts and infer potential user schedule intentions based on the dynamic user profile model and the multi-dimensional data. The third processing module is used to calculate the multi-dimensional competition index of each schedule item in terms of time, attention, physical strength and external resources, and based on the multi-dimensional competition index, generate multiple candidate schedule schemes, and assign priority weights to each schedule item in each candidate schedule scheme. The fourth processing module is used to generate a preliminary schedule plan based on multiple candidate schedule plans using a generative model. The fifth processing module is used to generate a visual interface for the preliminary schedule plan, allowing users to adjust and confirm it, and to generate the target schedule plan.

[0014] Thirdly, embodiments of this application provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described intelligent schedule management method.

[0015] Fourthly, embodiments of this application provide a computer-readable storage medium, including a computer program stored in the computer-readable storage medium, which, when executed by a processor, implements the intelligent schedule management method described above.

[0016] The beneficial effects of the embodiments in this application compared with the prior art are: This application discloses an intelligent schedule management method. Based on multi-dimensional data related to schedule items, and under the premise of strictly protecting user privacy, a dynamic user profile model is constructed. This dynamic user profile model can comprehensively, accurately, and dynamically reflect user habits, preferences, and status. Furthermore, based on the dynamic user profile model, combined with real-time acquired multi-dimensional data, instant interpretation and dynamic reasoning are performed to predict the user's current situation and schedule intentions, thereby inferring the user's potentially unexpressed schedule needs, greatly reducing manual input and improving user experience. In addition, multi-dimensional competitive evaluation of schedule items is conducted, and a multi-objective optimization algorithm is used to calculate dynamically changing priority weights for each schedule item, enabling the generation of schedule plans to be adjusted according to the items, thereby reducing fatigue and improving efficiency. The generated schedule plan is transformed into a visual timeline interface for user interaction and selection, and a flexible buffer is allocated to the start and end times of each schedule item, transforming rigid start and end times into flexible time ranges, preserving and enhancing the user's ultimate control and decision-making experience. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating an intelligent schedule management method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of an intelligent schedule management system provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0019] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0020] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0021] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0022] As used in this application specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrases "if determined" or "if detected [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once detected [the described condition or event]," or "in response to detection [the described condition or event]."

[0023] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0024] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0025] Please see Figure 1 As shown, the present invention is an intelligent schedule management method, comprising the following steps: S100: Collect multi-dimensional data related to user schedules and use federated learning technology to build a dynamic user profile model; In some embodiments, step S100 above includes: The user terminal device performs anonymization and feature engineering on the multi-dimensional data to extract structured feature vectors that reflect user behavior patterns. Based on the structured feature vectors and user historical interaction feedback data, the local user model is trained so that the local user model can predict user schedule preferences and user behavior patterns. A global user profile model is generated by aggregating encrypted local user model parameter updates from multiple user terminal devices through a central server. The global user profile model is securely updated and distributed to each user terminal device, and then merged with the local user model of each user terminal device to form a dynamic user profile model. The dynamic user profile model includes a user time preference vector, a user behavior pattern prediction vector, and a resource consumption baseline. The multi-dimensional data includes schedule data, application log data, device context data, and biosensor data.

[0026] In this embodiment, a dynamic user profile model is constructed based on multi-dimensional data related to schedule items, while strictly protecting user privacy. This dynamic user profile model can comprehensively, accurately, and dynamically reflect user habits, preferences, and status, and can be used for subsequent processes such as context awareness, intent inference, and personalized decision-making. The multi-dimensional data includes schedule item data, application log data, device context data, and biosensor data.

[0027] Specifically, the schedule data consists of tasks or events and their corresponding basic attributes. This is achieved by accessing user-authorized calendar applications and task management tools, such as the system calendar, Outlook, and Google Calendar, and Todoist and Microsoft To Do. This process collects calendar events and to-do lists that users actively enter or synchronize, such as meetings, appointments, birthdays, family dinners, etc. In addition, the collected schedule data needs to be parsed to obtain the corresponding basic attributes, such as title, start and end time, location, participants, estimated duration, user-preset priority, and related keyword descriptions.

[0028] Specifically, application log data refers to application usage data from user terminal devices. This data is collected through compliant interfaces that connect to user terminal devices, including application usage sequences, single usage durations, application categories, and communication interaction metadata. By collecting application usage sequences, single usage durations, and application categories from user mobile devices, the types and durations of applications used by the user at different times can be obtained, enabling analysis of user workflows and attention allocation. The communication interaction metadata specifically includes metadata from user emails and instant messaging tools, such as send and receive times, contacts, and subject keywords, allowing analysis of user communication frequency, key contacts, and their response habits. In some embodiments, with user authorization, browser bookmarks, frequently visited website categories, or appointment-related application operation records can also be analyzed as supplementary signals for uncovering potential user interests and future needs.

[0029] Specifically, device contextual data refers to data obtained through sensors built into or connected to the user terminal device. For example, the user's location movement trajectory points can be obtained through GPS, Wi-Fi, or base station positioning built into the user terminal device, such as "home," "office," "gym," and other frequently visited locations. Continuous trajectory points can be used to analyze the user's commuting or resting patterns. In addition, external sensors of the user terminal device can be collected as indirect evidence of contextual data. For example, external sensors such as in-vehicle Bluetooth devices and Bluetooth headset connections can be collected to obtain the user terminal device's screen opening / closing status and charging status, inferring whether the user is currently driving, in a meeting, or sleeping. Furthermore, motion sensor data such as accelerometer and gyroscope data are also included to obtain the user's motion status.

[0030] Specifically, biosensor data refers to usage data from wearable devices authorized by the user, such as smart bracelets or smartwatches, which can be used to obtain information such as heart rate, steps, and sleep stages, providing physiological basis for subsequent analysis of physical resources and inference of psychological situations.

[0031] In this embodiment, to ensure data availability and user privacy, the extracted data requires preprocessing. The core operations are performed locally on the user's terminal device, such as a mobile phone, to enhance data processing privacy and security. Specifically, once the raw data is collected, it is immediately anonymized on the user's terminal device. For example, sensitive Personal Identification Information (PII) in the address book, such as contact names, is hashed into irreversible anonymous IDs. Precise GPS coordinates (such as latitude and longitude) are converted in real-time into semantic location tags, such as "company" or "XX shopping mall," or blurred into a geographic grid code at the hundred-meter level, ensuring that it cannot be traced back to the individual.

[0032] Furthermore, the cleaned unstructured schedule data, application log data, device context data, and biosensor data undergo local feature engineering to be converted into structured feature vectors, which serve as input features for subsequent machine learning. For example, a day is divided into 96 15-minute segments, and features such as "percentage of office application usage," "number of screen lights," and "duration of sitting still" are calculated for each segment; a week's communication records are statistically analyzed to obtain social features such as "average daily call frequency with colleague A."

[0033] Furthermore, the anonymized, structured feature vectors after the above feature processing need to be encrypted and securely transmitted through a secure channel during subsequent federated learning. The original data collected is always stored in the user's terminal device to ensure the security of data transmission and data privacy.

[0034] In this embodiment, when constructing the dynamic user profile model, a neural network model is deployed as a local user model on each user terminal device. The local user profile model uses the structured feature vector obtained after anonymization and feature engineering as input, and uses user historical interaction feedback data as a supervision signal for training. Its training objective is to enable the model to accurately predict user schedule preferences and user behavior patterns based on the input features. The user historical feedback data specifically includes the adoption, modification, ignoring, and explicit rating of schedule suggestions. In a preferred embodiment, the local user model has three prediction heads: a time preference prediction head, a behavior pattern prediction head, and a resource consumption prediction head, which quantify the prediction of user schedule preferences and user behavior patterns.

[0035] Furthermore, the construction of the dynamic user profile model also adopts federated learning technology to build a general global user profile model to cover a wider range of public behavior patterns. On this basis, multiple local user models are fused together, and the final dynamic user profile model not only has general knowledge, but also deeply integrates the uniqueness of individual users. Moreover, as new data is continuously generated and the federated learning cycle continues, the dynamic user profile model can dynamically adapt to the evolution of user behavior habits.

[0036] Specifically, a trusted central server periodically initiates federated learning rounds, receiving parameter updates from encrypted local user models across multiple user terminal devices. Using a federated averaging algorithm, these updates are aggregated to generate a global user profile model. This global user profile model is then securely distributed to the original user terminal devices, where it is merged with their local user models to form a dynamic user profile model. During this fusion process, weighted averaging or knowledge distillation techniques are employed to combine the general knowledge of the global user profile model with the unique user habits observed in the local user models. Thus, the final dynamic user profile model receives a general knowledge injection through federated learning at regular intervals and is fine-tuned daily through local incremental learning.

[0037] More specifically, when the central server initiates a new round of federated learning, it broadcasts the current global model parameters to the participating user terminal devices. Each user terminal device, after training, calculates the difference between its local model parameters and the received global model parameters, using this as a parameter update. This local parameter update is encrypted using secure aggregation or homomorphic encryption techniques to ensure that the central server cannot decrypt the update from a single device before aggregation. The encrypted update is then uploaded to the central server. Once the central server receives a sufficient number of encrypted parameter updates from various devices, it executes the decryption step of the secure aggregation protocol to obtain the aggregated model update.

[0038] It's worth noting that the central server does not receive any raw data from users. It only periodically collects encrypted parameter updates uploaded by each user's terminal device. Furthermore, a secure aggregation protocol based on secret sharing is used. Each user terminal device splits its local parameter updates into multiple secret shares and uploads them to the server separately. After the server collects a sufficient number of shares, it can only calculate the sum of updates from all devices, but cannot know the update content of any individual device, further ensuring the security of data transmission and data privacy.

[0039] More specifically, after receiving the global user profile model, the user terminal device performs model fusion. Taking the weighted average fusion method as an example, a dynamic weight α is assigned to the parameters of the local user model and the global user profile model. Then, the parameters of the fused dynamic user profile model = α × local parameters + (1-α) × global parameters. The weight α is dynamically adjusted based on the performance of the local user model on the validation data in the user's historical interaction feedback data during the training process. Specifically, if the local user model is more accurate in predicting recent user behavior, it is given a higher weight. In some embodiments, a knowledge distillation method is also used for fusion, with the global user profile model as the "teacher model" and the local user model as the "student model". This allows the local user model to learn and imitate the output of the global user profile model on local data, while retaining its own characteristics, ultimately generating a dynamic user profile model that combines general knowledge and personal characteristics.

[0040] Furthermore, the completed dynamic user profile model can receive the input features mentioned above and then output a user time preference vector, a user behavior pattern prediction vector, and a resource consumption baseline. The user time preference vector represents the probability distribution of a user's preferences for various activities at different times of the day. Specifically, in the time preference prediction head, features are extracted from the structured feature vector to obtain a time preference embedding vector for training, generating a model like V_pref(“2025-10-27 19:00”) = {“Fitness”: 0.65, “Social”: 0.20, “Rest”:} The time preference probability distribution of 0.15}, where the time preference embedding vector includes strong periodic features, behavioral context features, schedule signals, and personal feature states. The strong periodic features are specifically obtained by absolute and cyclic encoding of the time period t to be predicted. Absolute encoding encodes time period t as a time of day, day of the week, and whether it is a weekday / holiday, while relative encoding encodes time period t as its position within the user's personal cycle, such as the day after payday or the day before a project deadline. The behavioral context features are the behavioral summary features of the M most recent time slices before time period t, including the activity categories... The embedding representations include cumulative duration, activity transition frequency, and main activity sequences; the schedule signal is the presence of strongly constrained events in the schedule data around time period t, such as meetings, which directly suppress the probability of other activities; the personal feature state is the user state related to time period t extracted from the structured feature vector, such as the user's location, etc. The time preference prediction head is usually a multilayer perceptron, which performs dimensionality reduction and nonlinear transformation on the above time preference embedding vectors to output the probability distribution of different activities. It also needs to be combined with aggregation in federated learning to improve the robustness of user time preference vector prediction.The user behavior prediction vector is used to predict the activity sequence and its probability of the user in the future time period. Specifically, taking time t as the end point, the historical activity sequence H = [h_{-L}, h_{-L+1},..., h_{-1}] of the previous L time slices is traced back. Here, h_j is a feature vector that integrates information such as activity type, location, and application usage intensity. The above historical activity sequence H is processed by an encoder network, such as a multi-layer GRU or Transformer Encoder. This encoder network learns the temporal dependence and transition patterns of historical activities, and outputs a behavior preference state vector and the encoding of each historical time slice. Then, based on the output of the encoder, a decoder gradually predicts the future sequence. For the first future time slice i = 1 to be generated, the decoder outputs a probability distribution P(a_i| C, a_{<i}) on the set of activity categories A and an estimated confidence scalar according to its current state, selects the activity type with the highest probability as a_i, and the corresponding probability value is used as a component of the confidence c_i after calibration. The predicted a_i and its related features, such as the predicted location at the next moment, are used as part of the input together with the decoder state to predict the next time slice i + 1. This process repeats until N predictions are generated. The parameters in this prediction process also participate in federated learning. By aggregating the transition patterns of a large number of users, the model can learn a general "activity script" and enhance the prediction ability for unknown users or rare scenarios. The resource consumption baseline is used to estimate the consumption levels of different activity types on the user's attention and physical strength. Specifically, by maintaining a standardized activity ontology library, the activities that may appear in the user's schedule are classified to generate multiple activity types, such as meetings, phone calls, aerobic fitness, social dinners, etc. Each activity type is encoded as a high-dimensional activity type embedding vector. This activity type embedding vector not only identifies the activity itself but also encodes the implicit attributes of the activity in the embedding space, such as "cognitive intensity", "social interactivity", "physical strength requirement level", etc. The above activity type embedding vector is trained and optimized together with other parameters of the model during the federated learning process. Then, a quantitative description of the specific execution conditions of the activity is obtained from the structured feature vector as the context feature vector, which specifically includes the activity duration, activity occurrence time, and activity social scenario. The activity social scenario specifically marks social data such as whether the activity is online or offline, the number of participants, and whether it involves cross-regions. Finally, the above activity type embedding vector and context feature vector are concatenated and trained in the local user model in combination with the global user portrait model, and finally the physical strength consumption value and attention consumption value of different activity types are output.

[0041] S200. Based on the dynamic user portrait model, combined with the multi-dimensional data, identify multi-dimensional user situations and infer the potential schedule intentions of the user; In some embodiments, step S200 above includes: Based on the device context data and combined with the dynamic user profile model, the physical context is identified, where the physical context is the objective physical environment in which the user is located. Based on the application log data and combined with the dynamic user profile model, digital contexts are identified, where the digital context refers to the user's engagement state in the application's digital space. Based on the biosensor data, combined with the dynamic user profile model, the physical context, and the digital context, the psychological context is inferred, which refers to the user's internal cognitive and emotional state.

[0042] In this embodiment, based on the dynamic user profile model constructed above, real-time multi-dimensional data is combined for instant interpretation and dynamic reasoning to predict the user's current situation and schedule intentions.

[0043] Specifically, in the physical context recognition module, based on device context data and combined with a dynamic user profile model, the objective physical environment in which the user is located is understood. Specifically, real-time data from devices such as GPS / Wi-Fi positioning, Bluetooth connection lists, accelerometers, gyroscopes, ambient light sensors, and microphones is acquired. Using user time preference vectors, user behavior pattern prediction vectors, and resource consumption baselines from the dynamic user profile model as prior knowledge, the corresponding physical context is inferred. GPS / Wi-Fi positioning provides the user's precise location, and real-time coordinates are converted into semantic tags such as "office workstation," "living room," "commuting," and "XX coffee shop." Furthermore, by fusing accelerometer and gyroscope data, a lightweight... The classification model can determine in real time whether a user is in a "stationary", "walking", "running", or "driving" state. By combining ambient light, noise levels, and Bluetooth connection status, it makes a preliminary inference about the environment, such as "bright and quiet", "dim and noisy", "in-vehicle environment", or "in a meeting". At the same time, it obtains the probability distribution of various activities of the user at the current moment through the user time preference vector, the activity sequence of the predicted future time period in the user behavior pattern prediction vector, and the attention consumption value and physical consumption value of each activity type in the resource consumption baseline, as auxiliary clues for location and environment judgment. Finally, it outputs a structured physical context vector, such as: [Location label: office; Activity state: stationary; Environmental context: bright and quiet; Device connection: charging]. If the user is stationary, the system accurately matches the known permanent location using real-time coordinates and directly assigns a high-confidence semantic label. If the user is in a moving state such as "walking," "running," or "driving," a probabilistic fusion method is used to combine the rough location inferred from sensor data with the high-probability activity type in the user's time preference vector to infer the user's location label and assign dynamic semantics to the location label. For example, if the user's sensor data shows that the user is in a large shopping mall and is moving, the system combines the high-probability "dining" activity in the user behavior prediction vector to generate the dynamic semantic "The user is in a commercial area and may be currently or about to engage in dining activities."

[0044] More specifically, the digital context recognition module analyzes the user's activity trajectory in the digital space to reveal their current task focus and cognitive input. Specifically, it acquires real-time usage data of the user's terminal device, i.e., application log data, and uses the user time preference vector, user behavior prediction vector, and resource consumption baseline from the dynamic user profile model as prior knowledge to verify and infer task focus and cognitive input. It matches application usage events in the application log data with the activity sequences predicted in the user behavior prediction vector. For example, if the user behavior prediction vector predicts that the user is about to start "project report writing," and at this moment the user opens a document on their terminal device and loads a specific project file, then the current task can be confirmed as "project report writing," and a task description can be output, specifically {task type, confidence level}. Furthermore, the system acquires the user's real-time interactive event stream for the task. Taking "project report writing" as an example, it listens for and aggregates data such as the frequency, rhythm, and intensity of keyboard taps, mouse movements / clicks, and touch gestures. It calculates the operational entropy and burst index within the time window as the cognitive engagement index. At the same time, it uses the attention consumption value of this task type output by the dynamic user profile model as the benchmark value. This benchmark value represents the cognitive engagement level that the user usually needs for this task. The system compares the real-time calculated cognitive engagement index with the attention consumption value and finally outputs a quantitative engagement index, which includes the engagement index and engagement deviation level. The final output is a structured digital context vector, such as [Current task: Project proposal writing (confidence: 0.95); engagement index: 78; engagement deviation level: 78 / 100].

[0045] In some embodiments, the method further includes processing communication interaction metadata to obtain the user's stress level on unprocessed digital information. Specifically, it obtains the unread counts of email clients and instant messaging tools, the sender of the most recent message, and the receiving timestamp from application log data. Simultaneously, it obtains the user time preference vector and user behavior prediction vector from a dynamic user profile model. Based on the user time preference vector, it assesses the probability of the user's preference for communication-related activities in the current time period. Combining this with the prediction probability of the activity sequence for communication-related activities in the user behavior prediction vector, it generates a weighted stress index using a weighted average algorithm. Specifically, W(t) = a × (1 - P_pr ef(t))+b×(1-P_pred(t)), where P_pref(t) represents the preference probability of communication activities under the user time preference vector, and P_pred(t) represents the prediction probability of the activity sequence for communication activities in the user behavior prediction vector. If no prediction result related to communication activities appears in the user time preference vector and the user behavior prediction vector, its probability value is assigned to 1, indicating that the user's pressure on unprocessed digital information is at its maximum in this case. Finally, a digital to-do pressure vector is output, which includes {total unread count, weighted pressure index, and unread count of high priority contacts}.

[0046] More specifically, in the psychological situation inference module, real-time biosensor data, as well as physical and digital situation vectors, are combined with a dynamic user profile model to estimate the user's internal cognitive and emotional state, and finally output a structured psychological situation vector, such as: [Cognitive load index: 65%; Emotion / stress level: mild stress; Energy reserve estimate: 55].

[0047] Specifically, the user time preference vector provides the probability of a user's tendency to engage in various activities at the current moment, reflecting their inherent expected rhythm; the user behavior prediction vector provides a prediction of activities in the next few minutes based on recent behavior, reflecting the user's expected trajectory; and the resource consumption baseline provides the attention and physical exertion values ​​required for different activities, used to calculate energy consumption and energy recovery.

[0048] More specifically, the cognitive load index is used to estimate the instantaneous occupancy rate of a user's current cognitive resources, such as working memory and executive control. Based on the current task in the digital context vector, the cognitive engagement index and its deviation level, and heart rate variability (HRV) from real-time biosensor data, a weighted calculation method is used to obtain the cognitive load index. The emotion / stress level is used to qualitatively assess the user's current emotional valence and stress arousal level. A classification model is used, with heart rate variability (HRV) and skin conductance (GSR) from real-time biosensor data as biometric sources, and the weighted stress index from the digital to-do stress vector as the contextual stress source as input features. The final output is the stress classification result, specifically including classification labels and their confidence levels. The classification labels of the model are such as {relaxed, calm, mild stress, high stress}. Energy reserve estimation is used to predict a user's current physiological and cognitive energy reserves available for subsequent activities. Specifically, it obtains circadian rhythms and recent sleep quality from historical data of wearable devices and calculates the daily baseline energy value Base(t) = B_peak × C(t) × S, where B_peak represents the personalized morning peak, which is the average resting heart rate variability of the user in a relaxed, non-task state for 2-3 hours after waking up each day; C(t) represents the circadian rhythm function, which simulates the natural fluctuation of energy levels throughout the day based on a personalized baseline curve fitted to the user's historical activity energy data, with a value between [0.7, 1.0], and lows in the afternoon and late at night; S is the sleep quality decay factor, usually S∈[0.85, 1.05], and S>1 when sleep quality is good. The system acquires the cumulative attention and physical exertion values ​​for each activity type up to the current day. Based on the duration of each user's activity, it applies a recovery rate model based on a resource consumption baseline to estimate the recovered energy value. This recovery rate model is a regression model, and its training data is obtained by analyzing historical data on the recovery trajectory of indicators such as HRV and subjective energy level after users complete high-exertion activities. Simultaneously, it adjusts the energy reserve estimate using the current mood / stress level, outputting the energy reserve estimate as: Daily Baseline Value - Cumulative Expenditure + Recovery Estimate ± Mood State Adjustment.

[0049] In this embodiment, the aforementioned multidimensional user context is combined with a dynamic user profile model to infer potential unexpressed scheduling needs of users. The inference process employs an encoder-decoder architecture and a neural network model incorporating an attention mechanism. The encoder receives and encodes a sequence of multidimensional context vectors from the past N time segments, the current multidimensional context vector, and the dynamic user profile model. Simultaneously, the attention mechanism is integrated into the decoder. When the decoder predicts the intent for each future time segment, it dynamically and selectively reviews and focuses on different parts of the historical context sequence output by the encoder. For example, when predicting "whether to exercise tonight," the attention mechanism focuses more on relevant historical segments such as "energy consumption this afternoon" and "number of times I've exercised this week," rather than "meeting content this morning." Finally, the decoder generates a sequence based on the encoder's comprehensive output and attention-weighted historical information. Its output is a probability distribution of potential scheduling intentions for the next T time segments based on the intent space.

[0050] Specifically, the intent space is a predefined, scalable collection containing various intent categories, such as {deep work, collaborative meetings, rest and relaxation, commuting, fitness, social entertainment, dining, learning and self-improvement, housework, and healthcare}.

[0051] S300. Calculate the multidimensional competition index of each schedule item in terms of time, attention, physical strength and external resources, and generate multiple candidate schedule schemes based on the multidimensional competition index, and assign priority weights to each schedule item in each candidate schedule scheme. In this embodiment, the schedule items are evaluated in a multi-dimensional competitive manner, and a multi-objective optimization algorithm is used to calculate the dynamically changing priority weight of each schedule item for the personalized generation of subsequent schedule plans.

[0052] In some embodiments, calculating the multidimensional competition index of each schedule item in terms of time, attention, physical strength, and external resources includes: Calculate the time conflict index between schedule items based on the overlap of their time intervals. Based on the attention consumption value of the schedule items and the user's real-time cognitive state, calculate the attention competition index of the schedule items. Based on the physical exertion value of the scheduled tasks and the user's real-time energy status, calculate the physical exertion competition index of the scheduled tasks. Based on the dependence of schedule items on external resources and the real-time availability of resources, the external competition index of schedule items is calculated.

[0053] In this embodiment, the schedule items correspond to all the items determined by the user in the schedule item data, as well as all the schedule items in the potential schedule intentions inferred in step S200 above. Each item has time attributes such as start time and end time, as well as the corresponding activity type, such as deep work, collaborative meeting, rest and relaxation, commuting, fitness exercise, etc.

[0054] More specifically, each schedule item is transformed into a time conflict graph G_t = (V, E, T), where V represents the vertex set, representing all schedule items; E represents the edge set, if there is an intersection between the time occupancy intervals of two schedule items i and j, then an edge e(i,j) is established between them; T represents the time conflict index between two schedule items, specifically, T(i,j) = Overlap(i,j) × (Importance_i(t) + Importance_j(t)) / 2, where Overlap(i,j) represents the overlap duration, and Importance(t) is a function used to measure the importance of a time point t located in the overlap interval within the original time occupancy interval. In one embodiment, Importance(t) = 1 - (t - d_c) / h_l, where d_c represents the center point of the time occupancy interval, and h_l is specifically half the length of the time occupancy interval.

[0055] More specifically, the activity type of the schedule item is obtained, and the attention consumption value is obtained from the dynamic user profile model based on the activity type. The attention consumption value is normalized into the attention consumption rate per unit time. In one embodiment, the attention consumption rate is calculated in combination with the historical execution time of the activity type. For example, if the attention consumption value of an activity type is 80 and the historical execution time is 60 minutes, then the attention consumption rate is 80 / 60≈1.33 units / minute.

[0056] Furthermore, since human cognitive resources are limited and relatively stable in a short period of time, at any given moment, their cognitive resources are either occupied or available, and the sum of the two should always equal 1. Therefore, by obtaining the user's cognitive load index from step S200 above and determining the user's cognitive resource occupancy, the user's resource availability can be determined accordingly. This resource availability is set as the user's attention budget, and combined with the user's personalized baseline curve, an attention budget curve is constructed. Specifically, based on the cognitive load index C, the user's real-time attention budget B_budget is obtained, where B_budget = 1 - C. Then, starting from the current time t_now, the relative shape of the attention budget curve is extended into the future. At the same time, weighted correction is performed through the personalized baseline curve to ensure that the attention budget curve can simultaneously reflect cognitive patterns and physiological energy reserves.

[0057] Furthermore, the consumption of attention budget curves by two scheduled items i and j under the same time arrangement is simulated to obtain the attention competition index. Specifically, the consumption of attention budget by scheduled items i and j is calculated as B_consume(t) = B_budget(t) - (b_i + b_j) × Δt at a simulation step size Δt, where B_budget(t) represents the attention budget at the start time t of the simulation, and b_i and b_j represent the attention consumption rates of scheduled items i and j, respectively. More specifically, the recovered attention budget is obtained by combining the budget recovery rate, and the attention competition index B is obtained from this. B = min(B_budget(t), B_consume(t) + ρ × (B(t) - B_consume(t)) × Δt), where ρ represents the budget recovery rate, which is positively correlated with (B_budget(t) - B_consume).

[0058] More specifically, using the energy reserve estimate obtained in step S200 above, the user's real-time energy state is determined. Using the same method as the attention competition index, the consumption of real-time energy by two scheduled events i and j under the same time arrangement is simulated to finally obtain the physical fitness competition index. Specifically, the consumption of real-time energy by scheduled events i and j under the simulation step size Δt is calculated as C_consume(t) = C_budget(t) - (c_i + c_j) × Δt, where C_budget(t) represents the energy reserve estimate at the start time t of the simulation, and c_i and c_j represent the energy consumption rates of scheduled events i and j, respectively. Combined with the energy recovery rate, the recovered energy state is obtained, and the physical fitness competition index C is obtained accordingly: C = min(C_budget(t), C_consume(t) + τ × (C_budget(t) - C_consume(t)) × Δt), where τ represents the energy recovery rate, which is positively correlated with (C_budget(t) - C_consume).

[0059] Furthermore, the external resources that the scheduled items depend on are obtained, such as "specific meeting room A", "company shuttle bus", "gym lane", "expert time", etc., as well as the real-time availability status of external resources. This can be predicted based on historical data to establish a mapping relationship between external resources and scheduled items, thereby constructing a bipartite graph G=(U,V,E), where U represents all scheduled items {i}, V represents all external resources {r}, and E represents the edge set. If scheduled item i depends on external resource r, an edge e(i,r) is established between i and r. This edge is attached with the time window of scheduled item i's demand for external resource r as an edge attribute. For resource r, the time windows of all connected items {i_k} {[t_start_k, For any two items i and j, if their demand time windows for the same resource r overlap, then i and j are determined to have a hard conflict on resource r. Resource r is then treated as an external shared resource for items i and j. Thus, for each resource r, a set of conflicting items C_r is generated, containing all pairs of items (i,j) that conflict on that resource. For a resource r that causes i and j to conflict, its conflict intensity s(r) = λ1 × Scarcity(r) + λ2 × (1 - Substitutability(r)) is calculated, where Scarcity(r) represents scarcity, which can be quantified as 1 / (the total number of available resource instances of resource r); Substitutability(r) represents substitutability, which is a value between [0,1]. The minimum similarity between resource r and other currently available similar resources in key attributes such as capacity, location, and device is calculated. A completely unique resource with no substitutes has a Substitutability of 0, while a resource with multiple identical backups has a Substitutability of 1. λ1 and λ2 are configurable weights. The external competition index of an event is measured by the maximum or weighted sum of the conflict intensity of conflicting event pairs in the conflicting event set C_r. For example, in one embodiment, E(i) = max_{r∈R_conflict(i)}s(r), where R_conflict represents the set of all event pairs related to event i in the conflicting event set C_r. This formula indicates that the value with the highest conflict intensity is selected from the R_conflict event pair set as the external competition index of event i.

[0060] In some embodiments, the step of generating multiple candidate scheduling schemes based on the multidimensional competition index and assigning priority weights to each schedule item in each candidate scheduling scheme includes: Use the planned execution time and priority weight of multiple schedule items as decision variables; Multiple optimization objectives are set, including minimizing resource conflicts, maximizing user preference satisfaction, and maximizing schedule stability. Using multi-objective evolutionary algorithms or Pareto optimization algorithms, a set of candidate scheduling schemes is generated under the constraints of time boundaries, resource availability, and user state security thresholds for each item. Calculate the dynamic priority weight for each schedule item in the candidate schedule scheme. The priority weight is a function of the inherent attributes of the item, the real-time competitive situation, and its contribution to the global optimization objective, wherein the real-time competitive situation is the importance of the item under the multi-dimensional competition index.

[0061] In this embodiment, the results of the four-dimensional competition analysis of time, attention, physical strength and external resources are combined to calculate a dynamic weight that changes over time for each item.

[0062] Specifically, a multi-objective optimization algorithm is used to dynamically plan each schedule item. The decision variables are the planned execution time s_i and priority weight w_i of each schedule item. Then, an optimization objective function is constructed to quantify multiple optimization objectives, including minimizing resource conflicts, maximizing user preferences, and maximizing schedule stability.

[0063] Specifically, minimizing resource conflict F1 means minimizing the weighted sum of conflicts of all items in the above four-dimensional competition: MinΣ(α×T+β×B+γ×C+δ×E), where α, β, γ, and δ are the configuration weights of each dimension, and α+β+γ+δ=1.

[0064] More specifically, maximizing user preference F2 means maximizing the matching degree between the schedule and the user time preference vector and the probability distribution of potential schedule intentions in the dynamic user profile model, while maximizing schedule stability F3 means minimizing the adjustment range of confirmed schedules.

[0065] In addition, constraints need to be set for solving the above multi-objective problem. These constraints include time boundary constraints, resource availability constraints, and user state safety threshold constraints. The time boundary constraint means that the time boundary of event i must be greater than or equal to the start time and end time of the event. These start and end times are calculated from historical data. The resource availability constraint means that for any resource r, at any time t, at most one item can occupy it. The user state safety threshold constraint means that the attention budget curve and energy reserve estimate cannot be continuously lower than the preset safety threshold.

[0066] Furthermore, a multi-objective evolutionary algorithm or constrained Pareto optimization method is used to solve the problem, generating a set of Pareto optimal solutions. Each solution represents a trade-off among multiple objectives, ultimately generating a set of candidate scheduling schemes, and calculating the corresponding dynamic priority weights for each schedule item in the candidate scheduling schemes.

[0067] More specifically, the dynamic priority weight of each schedule item in the candidate schedule scheme is a function of the item's inherent attributes, real-time competitive status, and contribution to the global optimization goal. The item's inherent attributes represent its relatively stable importance related to the individual user, including the user's preset basic priority and its matching degree with the user's time preference vector. Specifically, the tags assigned by the user when creating the item, such as "high," "medium," and "low," are obtained and mapped to a baseline value m_i, such as 0.9, 0.6, and 0.3. Simultaneously, the user's time preference vector during the scheduled execution period of the item is obtained from the dynamic user profile model, and the matching degree p_i between the item's activity type a_i and the corresponding component in the user's time preference vector is calculated. For example, if the 3:00 PM fitness activity is scheduled during a time period with a probability of 0.8 for "fitness" in the user's time preference vector, then p_i = 0.8. Therefore, the item's inherent attribute I_i = m × m_i + n × p_i, where m and n are weights that can be configured by the user or learned by the system. The real-time competitive situation is used to measure the importance of the competition index in each dimension. Specifically, C_i = r_t × T + r_b × B + r_c × C + r_e × E, where r_* is the weight of each dimension. The higher the value of C_i, the more likely that item i is a "hotspot" of resource competition in the current arrangement, and moving or adjusting it may have a greater effect on alleviating global conflicts. The contribution to the global optimization objective is used to measure the value of the corresponding item within the scheme. It is obtained by recording the impact of the arrangement of each item on the optimization objective space during the solution process. For example, for scheme P, the time of item i can be slightly disturbed, and the degree of degradation of scheme P on multiple optimization objectives can be observed. If the degradation is more severe, it means that i contributes more to maintaining the good performance of scheme P, and its contribution G_i(P) is higher. Therefore, based on the function w_i(P) = F(I_i, C_i(P), G_i(P)), the priority weight of each schedule item in the candidate schedule scheme is calculated. The function F usually adopts a weighted linear combination or a multiplicative combination.

[0068] S400. Based on multiple candidate scheduling schemes, a generative model is used to generate a preliminary scheduling scheme; In some embodiments, step S400 above includes: Data transformation and knowledge encapsulation are performed on each candidate schedule plan to generate seed knowledge; Based on the seed knowledge, the generation process of the generative model is guided and constrained by at least one of the following methods: enhanced generation, conditional vector guidance, or model initialization. In the generative model, at least one of the direct optimization strategy or the integrated innovation strategy is used to generate a preliminary schedule.

[0069] In this embodiment, the candidate scheduling schemes obtained through multi-objective optimization calculation in step S300 are used as seed schemes for the generative model, thereby enabling the generative model to perform local fine-tuning and optimization on the seed schemes, avoiding blind exploration of the generative model in the huge solution space and improving overall efficiency.

[0070] Specifically, for each candidate scheduling scheme S_k, k=1...K, where K is the number of candidate scheduling schemes obtained in step S300, it includes a time-series scheduling matrix T_k, a resource allocation vector R_k, a target score vector O_k, and a priority weight vector W_k. The time-series scheduling matrix T_k is an N×2 matrix, where N is the total number of items in the candidate scheduling scheme, and each row [start_i, end_i] represents the planned start and end times of item i. The resource allocation vector R_k is an N-dimensional vector recording the ID or selection index of the specific resource allocated to each item. The target score vector O_k is an M-dimensional vector, where M is the number of targets, corresponding to the scores on each optimization target in step S300. The priority weight vector W_k is an N-dimensional vector recording the dynamic priority weight w_i of each item.

[0071] More specifically, the day is discretized into a fixed number of slots. In one embodiment, this is discretized into 96 15-minute slots. For each candidate scheduling scheme S_k, a slot-item allocation sequence Seq_k is generated. Each slot contains a vector representing all items occupying that slot and their occupancy intensity, which is measured by the priority weight w_i of the items. Then, the statistical features of each candidate scheduling scheme's target score vector O_k and priority weight vector W_k, such as mean, variance, and quantiles, are input into a small encoder network to generate a scheme-level condition vector c_k, which characterizes the overall style of the scheme, such as "aggressive and efficient" or "conservative and comfortable." All candidate scheduling schemes are treated as nodes, and a scheme similarity graph is constructed. The node features are c_k, and the edge weights are calculated based on the difference in the temporal arrangement matrix T_k.

[0072] Furthermore, the processed slot-item allocation sequence Seq_k, scheme-level condition vector c_k, and scheme similarity graph are used as seed knowledge for knowledge injection into the generative model. The seed knowledge is integrated into the generation process of the generative model by at least one of the following methods: retrieval-enhanced generation, condition vector guidance, or model initialization.

[0073] Specifically, for the retrieval-enhanced generative injection method, at each step of the generative model decoding, the most relevant plan fragments are retrieved in real time from the candidate schedule plan set {S_k} based on the currently partially generated schedule context. The retrieval criteria are the semantic similarity between the scheduled items and the corresponding items in the candidate schedule plans; or the contextual similarity between the current time and the corresponding time slice in the candidate schedule plans. The retrieved relevant plan fragments are then converted into labeled sequences, which are used as context and fed into the decoder of the generative model along with the original conditional input, guiding it to continue the local pattern that conforms to the high-quality seed plan.

[0074] Specifically, for the conditional vector-guided injection method, the scheme-level conditional vectors {c_k} of all candidate scheduling schemes are pooled to form an aggregated guiding vector c_agg. c_agg is then concatenated with other conditional vectors to form the complete input to the generative model encoder. Furthermore, a learnable bias term is introduced into the decoder attention mechanism of the generative model. This bias term encourages the model to pay more attention to slots that are frequently occupied by candidate scheduling schemes when allocating event times.

[0075] Specifically, the model initialization injection method is an injection method adapted to the diffusion model architecture of generative models. It can take the noisy state of one or more candidate schedule schemes S_k as the starting point of the diffusion process, guide the generation process to denoise and refine near the high-quality schemes. The generative model thus generates a batch of diverse candidate schedule scheme drafts. Then, a scheme evaluator is used to quickly score each candidate schedule scheme draft. The scheme evaluator is independently trained with the O_k and W_k distributions of the candidate schedule schemes as reference benchmarks. It is used to evaluate the "Pareto approximation" of the candidate schedule scheme drafts and reorder the drafts, giving priority to the output of the drafts that are closest to the Pareto front.

[0076] For the generative model that incorporates seed knowledge, a direct optimization strategy or a fusion innovation strategy can be flexibly adopted to generate a preliminary schedule plan. Specifically, the direct optimization strategy involves selecting the candidate schedule plan with the highest optimization objective score from the multiple candidate schedule plans generated in step S300 as the baseline seed plan. The generative model then performs minimal editing on the baseline seed plan to fix any minor conflicts that may remain, such as correcting extremely minor resource conflicts or adapting to fine-tuning instructions proposed by the user in the subsequent interaction in step S500, such as "postpone the meeting by 10 minutes". In this strategy, the generative model uses the baseline seed plan as the initial state of the decoder, and the decoder is trained to maintain the arrangement of the baseline seed plan with a high probability, only regenerating at identified problem points or user-specified modification points. The fusion innovation strategy analyzes the target score vector O_k of different candidate scheduling schemes in step S300 above to obtain the corresponding advantage dimensions. For example, scheme A scores highly on "minimizing commuting time" and scheme B scores highly on "maximizing deep working time". Subsequently, through decoupling learning technology, the scheme vector c_k is decoupled into independent factors related to each target. Then, based on the user-defined preference combination or advanced strategy, such as "minimizing commuting time while retaining deep working time", the corresponding advantage factors are extracted from different schemes and recombined into a new fusion condition vector. The generative model uses this fusion condition vector as a guide to generate a scheduling scheme based on the user-specified preference target.

[0077] S500: Generate a visual interface for the preliminary schedule plan, allowing the user to adjust and confirm, and generate the target schedule plan.

[0078] In some embodiments, step S500 above includes: The suggested start and end times allocated to each schedule item in the preliminary schedule plan are defined as the core interval of that schedule item. Based on the priority weight, item type, and historical user behavior data of each scheduled item, elastic buffers located before and after the core interval are dynamically calculated and allocated to form fuzzy time boundaries for the scheduled items. On the visual timeline interface, each schedule item is rendered as a time period color block, wherein the time period color block corresponds to the core interval; When a user clicks on the time period color block of the corresponding core interval, elastic color blocks are added to both ends of the time period color block of each core interval to represent the elastic buffer zone. The visual timeline interface allows users to dynamically adjust the color blocks for the specified time periods and detects and reports time conflicts in real time during user operations.

[0079] In this embodiment, the initial schedule is transformed into a visual timeline interface for user interaction and selection. Flexible buffers are allocated to the start and end times of each schedule item, transforming the rigid start and end times into flexible time ranges, thus preserving and enhancing the user's ultimate control and decision-making experience.

[0080] Specifically, the core interval is the optimal time period for the items generated in the initial scheduling plan to be executed, usually corresponding to its suggested start and end times, representing the best execution window for the item. The elastic buffer is a scalable time range dynamically allocated before and after the core interval. The existence of the elastic buffer allows the scheduling plan to absorb small-scale unplanned disturbances.

[0081] Specifically, a hardness coefficient H_i∈[0,1] is added to the core interval to indicate the ease or difficulty of moving or compressing the interval. The activity type determines whether the event is a fixed-time event, such as a flight or an exam. Such events are usually immovable or uncompressible, so their hardness coefficient is set to 1, and their elastic buffer is set to 0. For other autonomous and flexible activity types, the hardness coefficient is determined according to their dynamic priority weight. In one embodiment, H_i is positively correlated with w_i, and w_i can also be set as the corresponding hardness coefficient.

[0082] More specifically, for schedule items with a hardness coefficient less than 1, flexible buffer zones are allocated based on priority weights, item types, and historical user behavior data. Specifically, a baseline buffer duration is set according to the activity type of each item, such as 15 minutes for meetings, 20% of the historical average commute time for commuting items, and 10 minutes for deep work items. The baseline buffer duration varies depending on the activity type. Then, a priority factor is calculated based on the dynamic priority weight w_i, indicating that higher priority items have a larger buffer zone: F_priority(w_i) = 1 + κ × w_i, where κ represents the gain coefficient. Furthermore, historical user behavior data is analyzed to obtain the user's habitual delay time and completion time deviation when performing similar items T_i. The flexible buffer zone is then extended. If a user has a habitual delay time of X minutes after the item's start time, X minutes are implicitly added to the flexible buffer zone before the core interval. If the actual time for a user to complete such items often exceeds the estimate, the flexible buffer zone after the core interval is expanded proportionally to the excess time, thus forming fuzzy time boundaries for each schedule item.

[0083] Furthermore, a visual interface is set up, in which a timeline is composed of time period color blocks for each schedule item. The elastic buffer corresponding to the core interval is set as an additional attribute on the timeline interface. When the user clicks on the time period color block of the corresponding core interval, the time period color block pops up and elastic color blocks are attached to both ends to represent the elastic buffer, so as to remind the user of the adjustable time space of the item.

[0084] In addition, users can directly drag the entire time period color block to move items, or stretch the two ends of the color block to manually adjust the size of the core area or buffer zone. When a hard collision occurs during dragging, the interface will provide an immediate prompt through vibration, red color, and other means. After the user completes the review and adjustment, the final target schedule plan will be generated through the confirmation button on the interface, and the user will be reminded to perform the corresponding schedule items with the target schedule plan.

[0085] Please see Figure 2 As shown, the present invention also provides an intelligent schedule management system, the system comprising: First processing module 201: Used to collect multi-dimensional data related to user schedules and use federated learning technology to build a dynamic user profile model; The second processing module 202 is used to identify multi-dimensional user contexts and infer potential user schedule intentions based on the dynamic user profile model and the multi-dimensional data. The third processing module 203 is used to calculate the multidimensional competition index of each schedule item in terms of time, attention, physical strength and external resources, and generate multiple candidate schedule schemes based on the multidimensional competition index, and assign priority weights to each schedule item in each candidate schedule scheme. Fourth processing module 204: used to generate a preliminary schedule plan based on multiple candidate schedule plans using a generative model; The fifth processing module 205 is used to generate a visual interface for the preliminary schedule plan, which can be adjusted and confirmed by the user to generate the target schedule plan.

[0086] It is understandable that, such as Figure 1 The content of the intelligent schedule management method embodiments shown is applicable to the embodiments of this intelligent schedule management system. The specific functions implemented by the embodiments of this intelligent schedule management system are the same as those shown below. Figure 1 The intelligent schedule management method shown in the embodiment is the same, and the beneficial effects achieved are the same as those described above. Figure 1 The beneficial effects achieved by the illustrated intelligent schedule management method embodiment are also the same.

[0087] It should be noted that the information interaction and execution process between the above systems are based on the same concept as the method embodiments of the present invention. For details on their specific functions and technical effects, please refer to the method embodiments section, which will not be repeated here.

[0088] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0089] Please see Figure 3 As shown, this embodiment of the invention also provides a computer device 3, including: a memory 302 and a processor 301, and a computer program 303 stored on the memory 302. When the computer program 303 is executed on the processor 301, it implements the intelligent schedule management method as described in any of the above methods.

[0090] The computer device 3 may be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer device 3 may include, but is not limited to, a processor 301 and a memory 302. Those skilled in the art will understand that... Figure 3 The computer device 3 is merely an example and does not constitute a limitation on the computer device 3. It may include more or fewer components than shown in the figure, or combine certain components, or different components, such as input / output devices, network access devices, etc.

[0091] The processor 301 may be a Central Processing Unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0092] In some embodiments, the memory 302 may be an internal storage unit of the computer device 3, such as a hard disk or memory of the computer device 3. In other embodiments, the memory 302 may be an external storage device of the computer device 3, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the computer device 3. Furthermore, the memory 302 may include both internal and external storage units of the computer device 3. The memory 302 is used to store the operating system, applications, bootloader, data, and other programs, such as the program code of the computer program. The memory 302 can also be used to temporarily store data that has been output or will be output.

[0093] This invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the intelligent schedule management method as described in any of the above methods.

[0094] In this embodiment, if the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include at least: any entity or device capable of carrying computer program code to a photographic device / computer device, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium. Examples include USB flash drives, portable hard drives, magnetic disks, or optical disks. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electrical carrier signals or telecommunication signals.

[0095] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such 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 application, and should all be included within the protection scope of this application.

Claims

1. An intelligent schedule management method, characterized in that, include: Collect multi-dimensional data related to user schedules and use federated learning technology to build a dynamic user profile model; Based on the dynamic user profile model and combined with the multi-dimensional data, multi-dimensional user contexts are identified, and potential user schedule intentions are inferred. Calculate the multidimensional competition index of each schedule item in terms of time, attention, physical strength and external resources, and generate multiple candidate schedule schemes based on the multidimensional competition index, and assign priority weights to each schedule item in each candidate schedule scheme; Based on multiple candidate scheduling schemes, a generative model is used to generate a preliminary scheduling scheme; A visual interface is generated for the preliminary schedule plan, which is then adjusted and confirmed by the user to generate the target schedule plan.

2. The method as described in claim 1, characterized in that, The process involves collecting multi-dimensional data related to user schedules and events, and using federated learning technology to construct a dynamic user profile model, including: The user terminal device performs anonymization and feature engineering on the multi-dimensional data to extract structured feature vectors that reflect user behavior patterns. Based on the structured feature vectors and user historical interaction feedback data, the local user model is trained so that the local user model can predict user schedule preferences and user behavior patterns. A global user profile model is generated by aggregating encrypted local user model parameter updates from multiple user terminal devices through a central server. The global user profile model is securely updated and distributed to each user terminal device, and then merged with the local user model of each user terminal device to form a dynamic user profile model. The dynamic user profile model includes a user time preference vector, a user behavior pattern prediction vector, and a resource consumption baseline. The multi-dimensional data includes schedule data, application log data, device context data, and biosensor data.

3. The method as described in claim 2, characterized in that, The process of identifying multi-dimensional user contexts based on the dynamic user profile model and the multi-dimensional data includes: Based on the device context data and combined with the dynamic user profile model, the physical context is identified, where the physical context is the objective physical environment in which the user is located. Based on the application log data and combined with the dynamic user profile model, digital contexts are identified, where the digital context refers to the user's engagement state in the application's digital space. Based on the biosensor data, combined with the dynamic user profile model, the physical context, and the digital context, the psychological context is inferred, which refers to the user's internal cognitive and emotional state.

4. The method as described in claim 1, characterized in that, The calculation of the multidimensional competition index of each schedule item in terms of time, attention, physical strength, and external resources includes: Calculate the time conflict index between schedule items based on the overlap of their time intervals. Based on the attention consumption value of the schedule items and the user's real-time cognitive state, calculate the attention competition index of the schedule items. Based on the physical exertion value of the scheduled tasks and the user's real-time energy status, calculate the physical exertion competition index of the scheduled tasks. Based on the dependence of schedule items on external resources and the real-time availability of resources, the external competition index of schedule items is calculated.

5. The method as described in claim 4, characterized in that, The process of generating multiple candidate scheduling schemes based on the multidimensional competition index, and assigning priority weights to each schedule item in each candidate scheduling scheme, includes: Use the planned execution time and priority weight of multiple schedule items as decision variables; Multiple optimization objectives are set, including minimizing resource conflicts, maximizing user preference satisfaction, and maximizing schedule stability. Using multi-objective evolutionary algorithms or Pareto optimization algorithms, a set of candidate scheduling schemes is generated under the constraints of time boundaries, resource availability, and user state security thresholds for each item. Calculate the dynamic priority weight for each schedule item in the multiple candidate schedule schemes. The priority weight is a function of the inherent attributes of the item, the real-time competitive situation, and its contribution to the global optimization objective, wherein the real-time competitive situation is the importance of the item under the multi-dimensional competition index.

6. The method as described in claim 1, characterized in that, The step of generating a preliminary schedule plan based on multiple candidate schedule plans using a generative model includes: Data transformation and knowledge encapsulation are performed on each candidate schedule plan to generate seed knowledge; Based on the seed knowledge, the generation process of the generative model is guided and constrained by at least one of the following methods: enhanced generation, conditional vector guidance, or model initialization. In the generative model, at least one of the direct optimization strategy or the integrated innovation strategy is used to generate a preliminary schedule.

7. The method as described in claim 1, characterized in that, The process of generating a visual interface for the preliminary schedule plan, allowing users to adjust and confirm it, and generating the target schedule plan includes: The suggested start and end times allocated to each schedule item in the preliminary schedule plan are defined as the core interval of that schedule item. Based on the priority weight, item type, and historical user behavior data of each scheduled item, elastic buffers located before and after the core interval are dynamically calculated and allocated to form fuzzy time boundaries for the scheduled items. On the visual timeline interface, each schedule item is rendered as a time period color block, wherein the time period color block corresponds to the core interval; When a user clicks on the time period color block of the corresponding core interval, elastic color blocks are added to both ends of the time period color block of each core interval to represent the elastic buffer zone. The visual timeline interface allows users to dynamically adjust the color blocks for the specified time periods and detects and reports time conflicts in real time during user operations.

8. An intelligent schedule management system, characterized in that, include: The first processing module is used to collect multi-dimensional data related to user schedules and uses federated learning technology to build a dynamic user profile model. The second processing module is used to identify multi-dimensional user contexts and infer potential user schedule intentions based on the dynamic user profile model and the multi-dimensional data. The third processing module is used to calculate the multidimensional competition index of each schedule item in terms of time, attention, physical strength and external resources, and based on the multidimensional competition index, generate multiple candidate schedule schemes, and assign priority weights to each schedule item in each candidate schedule scheme. The fourth processing module is used to generate a preliminary schedule plan based on multiple candidate schedule plans using a generative model. The fifth processing module is used to generate a visual interface for the preliminary schedule plan, allowing users to adjust and confirm it, and to generate the target schedule plan.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 7.

Images