Multi-user interaction calendar management method based on artificial intelligence

Abstract

The invention relates to the technical field of artificial intelligence calendar management, and discloses a multi-user interaction calendar management method based on artificial intelligence. The method comprises the following steps: acquiring a multi-user calendar event original data set containing a timestamp, an event type label and a participating member identifier; performing spatio-temporal feature extraction on the data to generate enhanced event data containing time distribution and spatial correlation features; and constructing an event relation graph based on an attention mechanism, and inputting enhanced data to complete multi-user event relevance modeling. And after a new calendar event request of a target user is received, searching an associated event set with potential conflicts from the atlas, detecting conflicts by adopting a dynamic conflict resolution strategy, generating a report containing conflict identifiers and levels, and calculating event priority weights according to the conflict levels. And automatically adjusting the time scheduling of the newly added event and the associated event based on the weight, updating a graph structure, finally outputting coordinated multi-user calendar arrangement, and generating an event change log containing all adjustment records.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence calendar management technology, specifically to a multi-user interactive calendar management method based on artificial intelligence. Background Technology

[0002] With the widespread adoption of digital office and collaborative living scenarios, calendar management tools have evolved from simple time recording for single users to multi-user collaborative scheduling. Whether it's scheduling project meetings for corporate teams, planning cross-departmental collaborative tasks, or coordinating travel plans and gatherings for family groups, calendar tools are essential for synchronizing time among multiple users and avoiding conflicts. However, current mainstream calendar management technologies still have many limitations and struggle to meet the efficient scheduling needs of complex multi-user interaction scenarios.

[0003] Existing calendar management tools are mostly designed with a single user at their core. Even those that support multi-user sharing often only offer static information synchronization. For example, some tools allow users to share their personal calendars with other members, but when multiple members add or modify events, the system can only issue a simple reminder when the event times completely overlap, failing to proactively identify potential conflicts between events. For instance, if a team member adds a "project review meeting" event, and this meeting depends on another member completing a previously arranged "data preparation" task, the existing system cannot identify the conflict that the incomplete "data preparation" task might cause the "project review meeting" to be unable to proceed normally through correlation analysis. It can only issue a reminder when the two times overlap, leading to low collaboration efficiency.

[0004] Current technologies lack the ability to deeply mine and model the multi-dimensional features of calendar events. Current calendar event data mostly includes basic information such as timestamps, event names, and participating members. The system does not extract and analyze the temporal distribution patterns of events (e.g., certain types of meetings often concentrate on Monday mornings) or spatial relationships (e.g., events involving the same team members often involve the same office location), making it difficult to construct deep relationship networks between multi-user events. This makes it difficult for the system to accurately determine the dependencies and scope of influence between events when faced with massive amounts of multi-user events, resulting in insufficient accuracy and comprehensiveness in conflict detection. For example, when a user adds a "customer visit" event, the system cannot combine the time and location of the customer's previous "requirements communication" events with other team members to determine whether the new event conflicts with the customer's other schedules. The user must manually check each event, increasing operational costs and the probability of errors.

[0005] Existing conflict resolution methods largely rely on manual adjustments by users, lacking dynamic and intelligent automated scheduling mechanisms. When the system detects event conflicts, it typically only lists the conflicting events, leaving users to negotiate and modify the times themselves. It doesn't consider factors such as the importance of the events, the number of participating members, or the event type to prioritize them, nor can it proactively recommend the optimal adjustment plan based on the available time of multiple users. For example, if a team has two conflicting events—a "departmental meeting" and an "urgent client meeting"—the system cannot automatically assign higher priority based on factors such as the "urgent client meeting" involving core business objectives and key clients. This requires repeated communication and confirmation from team members, prolonging the scheduling cycle and potentially causing business delays.

[0006] Current technologies lack a complete recording and traceability mechanism for event change processes. In multi-user collaborative scenarios, the addition, modification, and deletion of events are frequent, and each operation may have a cascading impact on other users' schedules. However, current systems mostly only record the final event scheduling result, without saving information such as the reasons for each adjustment, the status before and after the adjustment, and related events. When scheduling problems occur later, users cannot trace the source of the change, making it difficult to identify responsibility and optimize scheduling strategies. For example, if a team's project meeting time is adjusted multiple times, and some members are absent due to not receiving the adjustment notification in time, users cannot use the system to find out who initiated each adjustment and the basis for the adjustment, making problem troubleshooting difficult and further affecting the collaborative experience. Summary of the Invention

[0007] The purpose of this invention is to provide a multi-user interactive calendar management method based on artificial intelligence to solve the problems mentioned in the background art.

[0008] To achieve the above objectives, the present invention provides a multi-user interactive calendar management method based on artificial intelligence, the method comprising: Obtain the raw dataset of calendar events from multiple users, where each user's raw calendar event data includes timestamps, event type labels, and participant identifiers; Spatiotemporal features are extracted from the original dataset of calendar events to generate enhanced event data containing temporal distribution features and spatial correlation features; Construct an event relationship graph based on an attention mechanism, and input enhanced event data into the event relationship graph to model the multi-user event correlation. It receives new calendar event requests from target users in real time and retrieves a set of related events that may conflict with the new calendar event requests from the event relationship graph. A dynamic conflict resolution strategy is adopted to detect conflicts in the set of related events, generate an event conflict report containing conflict event identifiers and conflict levels, and calculate the priority weight of each event in the set of related events based on the conflict levels in the event conflict report. Based on the priority weight calculation results, the time scheduling of newly added calendar event requests and events in the associated event set is automatically adjusted, and the time scheduling adjustment results are fed back to the event relationship graph for graph structure update; Output the final coordinated multi-user calendar event schedule and generate an event change log containing all adjustment records.

[0009] Preferably, the spatiotemporal feature extraction of the original calendar event dataset includes: Identify periodic time patterns in the raw data of calendar events and extract time feature vectors including daily, weekly, and monthly cycles; Analyze the social relationship network among participating member identifiers and construct a member interaction frequency matrix; Tensor fusion of temporal feature vectors and interaction frequency matrix is ​​performed to form a spatiotemporal joint feature representation; Normalization is performed on the spatiotemporal joint feature representation to obtain standardized augmented event data.

[0010] Preferably, the construction of the event relationship graph based on the attention mechanism includes: The enhanced event data is mapped to graph node feature vectors, the attention coefficient between any two node feature vectors is calculated, and a directed weighted edge between nodes is established based on the attention coefficient. By iteratively updating the node feature vectors through a multi-layer graph convolutional network, an event relationship graph containing global event association information is finally output.

[0011] Preferably, the execution of the dynamic conflict resolution strategy includes: Extract event nodes from the event relationship graph that share participating members with the new calendar event request; Calculate the overlap index of each event node in the time dimension, correct the overlap index based on the historical response delay data of participating members, and convert the corrected overlap index into a conflict level quantification value. Combine the conflict level quantification values ​​of each event to generate an event conflict report.

[0012] Preferably, the priority weight calculation includes: Obtain the identity attribute data of all participating members involved in the conflict event, extract the decision preference characteristics of each member in historical events, calculate the initial priority score by combining the conflict level quantification value and decision preference characteristics, apply the member identity attribute data to the weighted adjustment of the initial priority score, and output the final priority weight with time sensitivity.

[0013] Preferably, the automatic time scheduling adjustment includes: An optimization function is established with the goal of minimizing time conflicts. The final priority weight is used as a constraint on the optimization function, and feasible scheduling schemes are searched within a preset time sliding window. Resource consumption is evaluated for multiple feasible scheduling schemes found, and the scheduling scheme with the best resource consumption evaluation is selected as the adjustment result.

[0014] Preferably, the map structure update includes: Detect all event nodes involved in the time scheduling adjustment; Recalculate the feature vector representation of the affected event nodes; Update the corresponding node and edge weights in the event relationship graph; Verify the global consistency of the map; Record map version change information to the event change log.

[0015] Preferably, the generation of the event change log includes: Capture the original event state before each time scheduling adjustment and record all intermediate decision-making steps during the adjustment process; Mark each participating member's confirmation status of the adjustment result, attach the system-generated explanation of the adjustment reason, and compress and store the complete change record in chronological order.

[0016] Preferably, the method further includes: The system monitors participants' modifications to calendar events in real time; when an unverified modification is detected, a conflict pre-detection process is triggered; the pre-detection results are compared with the prediction results of the event relationship graph; and the graph modeling parameters are automatically adjusted based on the degree of difference in the comparison.

[0017] Preferably, the conflict pre-detection process includes: Construct a temporary event node for the current modification operation and simulate the global impact of adding the temporary event node to the event relationship graph; Calculate the set of potential secondary conflict events, generate a pre-inspection report containing preventative adjustment recommendations, and store the key metrics in the pre-inspection report into the model training dataset.

[0018] Compared with the prior art, the beneficial effects of the present invention are: By acquiring raw datasets of calendar events from multiple users and extracting enhanced event data containing temporal distribution and spatial correlation features, this approach overcomes the limitations of existing technologies that rely solely on basic event information. By mining the temporal distribution patterns of events, the system can identify the time preferences of different event types, such as team meetings often concentrated at fixed times each week and project deadlines mostly distributed at the end of the month, providing richer analytical dimensions for subsequent event correlation modeling and conflict detection. Simultaneously, the extraction of spatial correlation features can link the spatial information of events involving members of the same team, family, or collaborative group, such as office locations and activity venues, helping the system more accurately determine the spatial dependencies between events and avoid implicit conflicts caused by spatial factors. For example, existing technologies struggle to identify events where the same member is scheduled to be in two different locations at similar times, while this method can promptly detect such conflicts through spatial correlation feature analysis.

[0019] The event relationship graph construction based on the attention mechanism enables accurate modeling of deep relationships between multi-user events. The attention mechanism allows the system to automatically focus on information more strongly related to the target event when processing massive amounts of event data. For example, for a user's newly added "product launch" event, the system can prioritize identifying potential conflicts between this event and related events such as "product development progress" and "marketing plan" by allocating attention weights in the graph, rather than indiscriminately traversing all events, significantly improving the efficiency and accuracy of conflict retrieval. This relationship modeling capability allows the system to go beyond explicit conflicts caused by overlapping times and also identify implicit conflicts due to event dependencies and resource consumption. For instance, if the "product launch" event needs to be completed after the "product sample preparation" event, the system can detect potential conflicts between the "product launch" and "product sample preparation" events caused by the delay in the graph, providing early warnings.

[0020] The combination of dynamic conflict resolution strategies and priority weight calculation enables intelligent and personalized conflict handling. Upon detecting a conflict, the system does not simply list the events but generates a report including the conflict level. It calculates priority weights based on factors such as event type, participating members, and the scope of related impact, making conflict handling more targeted. For example, in an enterprise scenario, when a conflict arises between a "core customer negotiation" event and an "internal training" event, the system can assign a higher priority to the "core customer negotiation" event based on the importance of the event type and the number of key participating members, providing a basis for subsequent automatic scheduling. In a family scenario, when a conflict arises between a "parent-teacher meeting" and a "friends' dinner," the system can combine the user's historical behavioral preferences to determine the user's priority tendency for family events and provide adjustment suggestions that better suit the user's needs, avoiding the inefficiency of existing technologies that rely on manual negotiation and reducing communication costs among multiple users.

[0021] The linkage between automatic time scheduling adjustments and real-time updates to the event relationship graph ensures dynamic collaboration and continuous optimization of multi-user calendar events. The system automatically adjusts the times of newly added events and related events based on priority weight calculations, eliminating the need for manual user intervention and significantly improving scheduling efficiency. The scheduling adjustments are fed back to the event relationship graph for updates, enabling the graph to reflect the latest association status of multi-user events in real time, providing accurate data for conflict detection of subsequent new events. For example, when the system automatically adjusts the "Project Review Meeting" time from Monday morning to Tuesday afternoon, the graph will simultaneously update the association between this event and the "Data Processing" task, as well as other events involving participating members. When other users subsequently add events related to this project, the system can accurately determine conflicts based on the updated graph, avoiding misjudgments caused by outdated graph information.

[0022] The generation of event change logs provides a complete traceability basis for multi-user calendar event management. The log records all adjustments, including the event status before and after the adjustment, the reason for the adjustment, and related events. When scheduling problems arise later, users can quickly trace the source of the change through the logs to investigate the cause of the problem. For example, if a member is absent because they were not informed of a meeting time adjustment, users can query the change log to find the initiator of the adjustment, the adjustment time, and the notification details, clarifying the responsibility. Simultaneously, the log data can also serve as a reference for subsequent optimization of scheduling strategies, such as analyzing high-frequency conflict types and common adjustment directions, providing practical scenario-based support for system iteration and further improving the intelligence level of multi-user collaborative management. Attached Figure Description

[0023] Figure 1 This is a schematic diagram illustrating the working principle of the multi-user interactive calendar management method based on artificial intelligence as described in this invention. Figure 2 A flowchart for constructing an event relationship graph; Figure 3 The flowchart for calculating priority weights. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] Please see Figure 1This invention provides an AI-based multi-user interactive calendar management method. The method includes: acquiring a raw dataset of calendar events from multiple users to begin processing. Each data unit in the raw dataset contains a timestamp, event type label, and participant identifier. The system preprocesses and enhances the raw dataset, analyzing temporal patterns and spatial relationships through a dedicated spatiotemporal feature extraction module. The spatiotemporal feature extraction module outputs enhanced event data, which is then input into an attention-based event relationship graph construction module. The event relationship graph construction module uses graph neural network technology to model the relationships between events, generating a graph structure containing nodes and edges. In the real-time processing stage, the system receives new calendar event requests from target users, performs rapid retrieval using the event relationship graph, and identifies a set of related events that potentially conflict with the new calendar event request. The conflict resolution module employs a dynamic conflict resolution strategy to perform in-depth analysis of the related event set, calculates the conflict level, and generates an event conflict report. Based on the event conflict report, the system performs priority weight calculation, assigning a weight value to each event. The automatic time scheduling adjustment module reschedules events within the time domain based on the priority weight calculation results, outputting an optimized calendar arrangement.

[0026] Example 1: The original calendar event dataset serves as the starting point for system processing. Each data unit precisely contains a timestamp, event type label, and participant identifier. The spatiotemporal feature extraction process begins with the identification and parsing of periodic time patterns in the original calendar event data. The system employs a time series decomposition algorithm to decompose continuous timestamp information into trend components, periodic components, and residual components. For the extraction of periodic components, the system is configured with multi-scale period detectors. The daily period detector analyzes the distribution patterns of events within a 24-hour day, for example, identifying 9:00 AM to 11:00 AM as a high-frequency occurrence period for meeting events. The weekly period detector scans the differences in event distribution from Monday to Sunday, while the monthly period detector focuses on event clustering within a natural month or financial month cycle. The extracted daily, weekly, and monthly period feature vectors are all quantized into fixed-dimensional numerical vectors. These vectors together constitute a multi-period time feature representation that comprehensively describes the time patterns of events. Simultaneously with time feature extraction, the system performs parallel analysis of spatial correlation features. The core of spatial correlation features lies in constructing a social relationship network among participant identifiers. The construction of the social relationship network is based on raw data of historical calendar events. An interaction frequency matrix is ​​initialized by statistically analyzing the frequency of co-occurrence of each pair of participating member identifiers in past events. This matrix is ​​a symmetric matrix where each row and column corresponds to a unique participating member identifier, and the values ​​of the matrix elements directly reflect the degree of collaboration between two members. To more precisely characterize relationship strength, the system not only considers co-occurrence frequency but also introduces event type weighting factors. For example, co-occurrence in strategic decision-making meetings has a higher weight than co-occurrence in ordinary information synchronization meetings, thus enabling the interaction frequency matrix to more accurately reflect the actual collaboration patterns within the organization.

[0027] After extracting the temporal feature vectors and constructing the inter-member interaction frequency matrix, the system enters the feature fusion stage, integrating information from the temporal and spatial dimensions. The fusion operation is achieved through tensor fusion technology, specifically by elevating the multi-period temporal feature representation and the inter-member interaction frequency matrix into higher-order tensors and calculating their tensor product to generate a spatiotemporal joint feature representation. This spatiotemporal joint feature representation is a high-dimensional data structure that can simultaneously capture the temporal distribution characteristics of events and their spatial location within the member relationship network. For example, a project review event occurring on a Friday afternoon involving multiple core members across departments will be encoded in the spatiotemporal joint feature representation with both the temporal characteristics of the period before the weekend and the spatial characteristics of strong cross-departmental connections. Due to differences in source dimensions and units, the numerical range of the generated spatiotemporal joint feature representation may fluctuate significantly, potentially leading to bias if used directly for subsequent model training; therefore, normalization is necessary. The normalization module employs a min-max scaling method to linearly transform each feature dimension in the spatiotemporal joint feature representation to the range of zero to one. For feature dimensions exhibiting long-tail characteristics, the system prioritizes Z-score normalization for preprocessing to ensure they conform to a standard normal distribution. The normalization process strictly adheres to preset scaling parameters, which are calculated and fixed from the training dataset and applied to all subsequent online data processing flows, guaranteeing consistency in feature processing. The output after normalization is called standardized augmented event data. Standardized augmented event data has a uniform numerical range, providing a stable and efficient input for constructing attention-based event relationship graphs.

[0028] The entire spatiotemporal feature extraction module is designed as a scalable distributed computing task, deployed on an Apache Spark-based computing framework to handle the high-concurrency processing needs of massive amounts of raw calendar event data from multiple users. Internally, the module employs a combination of micro-batch processing and a sliding window mechanism. The sliding window advances at fixed time intervals (e.g., hourly), processing newly added raw calendar event data within each window and incrementally updating the periodic time pattern and inter-member interaction frequency matrix to ensure the real-time performance and accuracy of the spatiotemporal features. The feature extraction pipeline includes multiple stages: data validation, format conversion, parallel computing, and result aggregation. Each stage has corresponding fault tolerance and retry mechanisms to ensure the robustness of the feature extraction process. The final standardized enhanced event data not only contains the original attributes of calendar events but also deeply encodes the inherent correlations and patterns of events in the spatiotemporal dimension.

[0029] Example 2: See Figure 2In this system, each independent event in the standardized augmented event data is mapped to a node in the event relationship graph. The node initialization process uses a fully connected neural network layer as the embedding layer to convert the temporal distribution features and spatial correlation features contained in the standardized augmented event data into a fixed-dimensional node feature vector. The node feature vector is set to 256 dimensions, which balances representational power and computational complexity. Each dimension of the node feature vector encodes a certain latent spatiotemporal attribute of the event. The weight parameters of the embedding layer are initialized during the pre-training phase using historical calendar event data through unsupervised learning, making semantically similar events closer in the vector space. After completing the initialization and mapping of the node feature vectors, the system begins to calculate the correlation strength between nodes, i.e., the attention coefficient. The attention coefficient is calculated using an additive attention mechanism. For any two nodes in the event relationship graph, their node feature vectors are input into a single-layer feedforward neural network. The output of the feedforward neural network undergoes a non-linear transformation through a LeakyReLU activation function, and finally, it is normalized on neighboring nodes through a softmax function to produce a scalar value as the attention coefficient. This additive attention process allows the model to capture complex nonlinear relationships between node pairs. The attention coefficient does not depend on the inner product of node feature vectors, but is determined by a learnable weight matrix, resulting in stronger expressive power. The calculated attention coefficient explicitly characterizes the importance of the source node to the target node; for example, a project manager's schedule approval event may generate a high attention coefficient for multiple execution events of its subordinates. Based on the calculated attention coefficient, the system establishes directed weighted edges between nodes in the event relationship graph. The direction of the directed weighted edges follows the temporal causality or the direction of influence transmission, typically pointing from an earlier event to a later event, or from an event involving a high-priority member to an event involving a low-priority member. The weight values ​​of the directed weighted edges are directly taken from the normalized attention coefficient, with weight values ​​ranging between zero and one. The closer the weight value is to one, the stronger the correlation between the two event nodes. The initial topology of the event relationship graph is thus formed; it is a directed graph with self-loops, where each node retains edges connecting to itself to maintain the importance of its own information in subsequent information transmission. The graph structure is stored in memory in the form of an adjacency list, which supports efficient traversal and dynamic updates of a node's neighbors.

[0030] The initial construction of the event relationship graph only reflects the direct local associations between events. To capture deeper global event association information, the system employs a multi-layer graph convolutional network to iteratively update the node feature vectors. Each layer of the graph convolutional network performs a message passing process, where each node aggregates the feature information of its direct neighbors (connected by directed weighted edges). The aggregation operation is a weighted summation, with the weights being the attention coefficients of the directed weighted edges. The aggregated neighbor features are concatenated or added to the node's own feature vector, then linearly transformed through a learnable weight matrix, and non-linearity is introduced by applying the GELU activation function. Multi-layer graph convolutional networks typically stack two to three layers; too many layers may lead to oversmoothing, where the features of all nodes tend to be consistent. After each iteration, the node feature vector incorporates information from a wider range of neighbors. Through multi-layer propagation, each node's features contain contextual information from multi-hop neighbors in the graph, achieving the fusion of local and global features. The iterative update process of the node feature vectors is accompanied by a gradient descent optimization algorithm, with the training objective being to minimize a loss function based on event co-occurrence relationships. The loss function encourages frequently occurring consecutive event nodes in the calendar to have similar representations in the vector space, while the representations of irrelevant event nodes are far apart. During training, the Adam optimizer is used to adjust the parameters of the weight matrix in the graph convolutional network and the parameters of the feedforward neural network in the attention mechanism. The training period is set to one hundred epochs, and an early stopping strategy is employed to prevent overfitting. The final event relationship graph output after training convergence contains global event association information. The node feature vectors in the event relationship graph are high-order representations refined by the graph convolutional network, capable of simultaneously encoding the event's own attributes, local neighborhood structure, and global position within the entire graph.

[0031] Once the event relationship graph is constructed, it needs to be persistently stored to support real-time queries and updates. The graph data is stored in a graph database system. The graph database uses a native graph storage model, using nodes, directed weighted edges, and their attributes as basic storage units, and provides an efficient graph traversal query language. For new calendar event requests, the system can insert the new event as a temporary node into the event relationship graph in real time. Through a simplified incremental calculation process, it quickly estimates the attention coefficient between the new event and existing nodes and updates the local directed weighted edges, thus avoiding the need to rebuild the entire event relationship graph each time. This meets the high real-time requirements of a multi-user interactive calendar management system.

[0032] Example 3: See Figure 3The execution of the dynamic conflict resolution strategy begins when the system receives a new calendar event request from a target user. The system immediately accesses the constructed event relationship graph, a graph structure built based on an attention mechanism that contains global event association information. The system extracts all event nodes from the event relationship graph that share participating members with the new calendar event request. This extraction process is achieved through index queries on the graph database, with query conditions based on exact matching of participating member identifiers. The existence of shared participating members implies that these event nodes have potential competition for resources (i.e., member time). The extracted set of event nodes constitutes the initial set of associated events, which is the focus of in-depth conflict analysis.

[0033] For each event node in the associated event set, the system needs to calculate its overlap metric with the new calendar event request in the time dimension. The overlap metric is calculated based on the event's time interval, defined by the start and end timestamps. A quantified score is used to characterize the degree of interference between two events on the timeline, and its calculation formula is as follows: in: and These represent the start and end timestamps of the new calendar event request, respectively. and These represent the start and end timestamps of an existing event in the associated event set, respectively. (Function) The function is used to retrieve the earlier of two end timestamps. Used to retrieve the later of two start timestamps. (Molecular part) This calculation measures the actual intersection length of the time intervals of two events. When there is no intersection, this value is less than zero; this is the overlap index. Set to zero. The denominator part. This calculates the total length of the outer envelope of the two event time intervals. (Overlap index) The range of its value is between zero and one. A value of zero indicates that the events have no temporal overlap. A value of one indicates that the time intervals of the two events completely overlap. This fractionalized indicator provides a more nuanced reflection of the severity of the conflict than a simple Boolean overlap assessment.

[0034] The overlap metric provides a preliminary conflict assessment based on an ideal time schedule. However, in real-world work scenarios, the historical behavioral patterns of participants significantly influence the perceived intensity of conflict. The system needs to access a historical response delay database to refine the simple time overlap metric. This database records each participant's actual behavioral data for past calendar events, including deviations from the planned event entry and exit times. For each shared participant, the system calculates their historical average response delay, which is an exponential moving average of multiple event delays, giving higher weight to recent behavior. The refinement process uses a correction function that modifies the original overlap metric. The algorithm takes the historical average response delay vector of all shared members involved as input and outputs a corrected overlap index. The correction function reduces the overlap index for events involving habitually late members, as historical data suggests these members have a greater flexibility in their actual schedules; conversely, for always punctual members, the corrected overlap index is maintained or even slightly amplified, as time conflicts have a more direct and severe impact on them. Corrected overlap index It needs to be converted into a discrete, more easily understood and processed quantification of conflict levels. The system defines a mapping function to convert continuous... The values ​​are mapped to a preset conflict level. The mapping function uses piecewise linear interpolation to... The value range from zero to one is divided into four consecutive intervals, corresponding to four conflict level quantization values: no conflict, low conflict, medium conflict, and high conflict. For example, Less than Mapping to be conflict-free Between and The mapping between them is low-collision. Between and The mapping between them is a conflict. Greater than Mapped to high collision. Collision level quantization values ​​are encoded using integers; no collision is... Low conflict The conflict between China and the United States is High conflict For each event in the set of related events, the system generates a tuple containing the event's unique identifier and its corresponding conflict severity quantification. All these tuples are combined to form a structured document, the event conflict report. The event conflict report not only lists events with potential conflicts but, more importantly, provides a quantified rating of the conflict severity.

[0035] After an event conflict report is generated, the priority weight calculation module is immediately activated. The purpose of priority weight calculation is to establish a comparable order of priority among conflicting events. The module first obtains the identity attribute data of all participating members involved in the conflicting event. This identity attribute data is stored in the organization's member information database, including the member's job level, department, and role weight factor in the specific project. The system also extracts each member's decision-making preference characteristics in historical events. These characteristics are derived by analyzing the member's past acceptance rate of calendar adjustment suggestions, the frequency of proactively initiating schedule changes, and the strength of their preference across different event types. Initial priority score. The calculation incorporates the conflict level quantification value. and decision preference feature vector The computational model is a linear combination model. ,in and These are weight coefficients obtained in advance through machine learning training, representing the quantified conflict level. It is the main influencing factor, the decision preference feature vector As an adjustment factor, the initial priority score. This reflects the basic prioritization of events based on conflict severity and members' historical behavior, but it does not fully consider the immediate weight of members' identities within the current organizational context. The initial priority scores are weighted and adjusted using member identity attribute data, and the identity attribute data is normalized into an identity weight factor. Identity weight factor It is the product of the member's job level weight, the project's urgency weight, and the department's importance weight. Final priority weight. From the initial priority score With identity weight factor Multiplying them together, we get This final priority weight It is a time-sensitive value, and this time sensitivity is reflected in the quantification of conflict level. It is calculated based on time overlap, and the identity weight factor... This may include the project's time urgency. Final priority weight. The output provides a quantitative basis for subsequent automatic time scheduling adjustments, enabling the system to seek an overall optimal calendar arrangement scheme under multiple constraints.

[0036] Example 4: The automatic time scheduling adjustment module establishes an optimization function aimed at minimizing time conflicts. The core of this function is to find the new start time for a set of events, minimizing the total weighted conflict value among all potentially conflicting event pairs. The total weighted conflict value is defined as the sum of the products of the final priority weight of each pair of conflicting events and its modified overlap index. Minimizing this sum means the system tends to prioritize resolving high-priority, high-overlapping, severe conflicts. The decision variables of the optimization function are the adjustable start timestamp of each event. Constraints include the fixed duration of the event, the hard constraint that each participant can only participate in one event at a time, the soft constraint of working hours on weekdays, and the order relationship implied by the final priority weights. Events with higher final priority weights enjoy greater stability in scheduling, and their original times are less likely to be changed. The optimization problem is solved within a preset time sliding window, typically spanning seven calendar days, with the window's starting point aligned with the proposal time of any new calendar event request. The system employs a heuristic search algorithm within a sliding time window to find feasible scheduling schemes. The search process begins by generating an initial solution, which sorts all conflicting events in descending order of their final priority weight. The event with the highest weight retains its original scheduled time, the event with the second highest weight is placed in the nearest available time slot, and so on. The search algorithm uses a variant of the simulated annealing algorithm, introducing random perturbations to escape local optima. These perturbation operations include swapping the time slots of two events and randomly moving an event to a new available time slot. The algorithm records every feasible scheduling scheme encountered during the search process. A feasible scheduling scheme must satisfy all hard constraints, i.e., no two events with shared participants overlap in time.

[0037] For each feasible scheduling scheme found, a resource utilization assessment is required. This assessment considers not only time conflicts but also physical resource conflicts and member satisfaction deviations. Physical resource conflict checks involve the availability of shared resources such as meeting rooms and presentation equipment; the system maintains a separate resource calendar database for cross-validation. Member satisfaction deviations are measured by calculating the sum of the absolute differences between the new and original planned times for each event; a smaller deviation indicates less disruption to members' original schedules. The resource utilization assessment ultimately produces a comprehensive score, which is a weighted linear combination of time conflict value, resource conflict level, and member satisfaction deviation. The weighting coefficients are pre-set according to the organizational strategy. Table 1 shows the resource utilization assessment results for a scheme with three feasible scheduling options, illustrating the relationship between the assessment dimensions and the score calculation.

[0038] Table 1: Resource Utilization Assessment Table for Feasible Scheduling Schemes Scheduling scheme number Total weight conflict value Resource Conflict Level Member satisfaction deviation (person-hours) Overall Score Option A 15.2 Low 3.5 88.5 Option B 12.8 middle 2.1 85.1 Option C 10.1 high 5.8 82.7 The automatic time scheduling adjustment module selects the scheduling scheme with the highest comprehensive score in resource utilization assessment as the final adjustment result. Scheme A in Table 1 is the optimal choice due to its lowest resource conflict level and acceptable time conflict value. The adjustment result is encapsulated into a structured scheduling instruction set, which details the event identifier, new start timestamp, new end timestamp, and adjustment reason code for each event that needs to be moved. After the time scheduling adjustment result is generated, the graph structure update module is activated. The primary task of the graph structure update module is to detect all event nodes involved in the time scheduling adjustment. The detection process is completed by parsing the scheduling instruction set. All event nodes corresponding to the event identifiers listed in the scheduling instruction set are marked as affected nodes. Affected nodes include not only event nodes that are directly moved, but also adjacent nodes whose context changes due to the movement of related events. The system creates a data copy for each affected node, saving its original state in the event relationship graph, including node feature vectors, inbound edge sets, and outbound edge sets.

[0039] The system recalculates the feature vector representations of the affected event nodes, focusing on updating time-related features. The time distribution feature dimension in the node feature vector needs to be reconstructed based on the new start and end timestamps, and periodic time pattern features need to be re-extracted. The spatial association feature dimension usually remains stable unless the list of event participants changes. The feature vector recalculation uses the same spatiotemporal feature extraction algorithm as in Example 1 to ensure consistency of feature representation. The updated node feature vectors will more accurately reflect the meaning of the event in the adjusted new time context. The corresponding nodes and edge weights in the event relationship graph need to be updated synchronously. The node update operation involves writing the recalculated node feature vectors back into the node attributes of the event relationship graph. The edge weight update is more complex, requiring recalculation of the attention coefficients between the affected node and other associated nodes based on the updated node feature vectors. The recalculation process focuses on the local subgraph centered on the affected node, calculating the attention coefficients between all node pairs within the subgraph, and updating the weight values ​​of the directed weighted edges accordingly. For edges whose attention coefficient becomes zero, the system deletes the edges; for node pairs that generate significant attention coefficients, the system adds new directed weighted edges. This series of update operations is completed within a single database transaction, ensuring the atomicity of the event graph update.

[0040] Global consistency verification of the event relationship graph is performed after the update transaction is committed. This verification checks for isolated nodes, edge weights within the valid range of zero to one, and any abnormal increase in connected components. The verification algorithm traverses the entire updated graph, calculates basic graph statistics such as average degree and clustering coefficient, and compares them with the baseline values ​​before the update. Deviations exceeding a preset threshold trigger alarms and rollback mechanisms. Verified event relationship graph versions are marked as stable versions. Graph version change information is recorded in the event change log, including the old version number, new version number, change timestamp, list of affected nodes, and edge change summary. The event change log is appended to a distributed log system, with each entry containing a complete operation sequence, supporting subsequent auditing and fault recovery. The execution of the graph structure update module enables the event relationship graph to evolve dynamically, maintaining a high degree of synchronization with the current calendar schedule, providing an accurate underlying data model for the system to handle subsequent interaction requests.

[0041] Example 5: The generation of the event change log begins with the system capturing the original event state before the time scheduling adjustment. This capture is achieved by creating a data snapshot. Before applying the scheduling instruction set output by the automatic time scheduling adjustment module, the system immediately queries the current attribute values ​​of all affected event nodes in the event relationship graph, including the event start timestamp, event end timestamp, list of participating members, event type label, and associated physical resource information. These attribute values ​​are serialized and stored in a temporary buffer, marked as "pre-adjustment state." The system records all intermediate decision-making steps during the adjustment process, covering the complete logical chain from event conflict report generation and priority weight calculation to feasible scheduling scheme evaluation. Detailed parameters and results for each weight calculation, constraint check, and scheme ranking are appended to an audit trail. The audit trail is stored in structured JSON format, ensuring machine readability and traceability for subsequent analysis. Marking the confirmation status of each participating member regarding the adjustment result is a crucial step in event change log generation. The system sends notifications to all participating members affected by the scheduling adjustment via a message push interface. The notifications include detailed information about the event change, the reason for the adjustment, and available confirmation options. Participating members can provide feedback on their confirmation status using the "Accept," "Reject," or "Suggest Modification" buttons on the user interface. The system records the member's identifier, operation timestamp, and specific selection. For members who do not respond within a preset time window (e.g., 24 hours), the system marks their confirmation status as "Pending" and may trigger a notification. An automatically generated explanation of the adjustment reason is appended to each change record. This explanation is not simple code but descriptive text synthesized by a natural language generation module. Based on the conflict level quantification value in the event conflict report, the member's identity attribute data, and resource conflict level, this module generates an easy-to-understand explanation, such as "Due to a time conflict with Project Manager Zhang San's high-priority project review meeting, and your low role weight in that meeting, your team's weekly meeting has been rescheduled to one hour later."

[0042] All captured raw states, recorded intermediate decision steps, marked confirmation states, and additional explanations of adjustment reasons are organized in strict chronological order, based on Coordinated Universal Time (UTC) timestamps. The system employs a highly efficient log compression storage algorithm to process complete change records. This algorithm identifies consecutive modifications to the same event attribute, retaining only the initial raw value and the latest final value, and recording the sequence number of all change operations. This differential encoding technique significantly reduces storage space usage while allowing the reconstruction of event states at any historical point in time by replaying the operation sequence. The compressed log blocks are written to a distributed log storage system that provides high availability and data persistence guarantees. The real-time monitoring module operates independently of the log generation module, continuously monitoring modification operations performed by participating members on calendar events. The monitoring module captures modification intentions by listening to user interface operation events and application programming interface (API) call requests. When an unverified modification is detected, such as user Li Si attempting to move his "quarterly planning meeting" from the scheduled next Tuesday at 2 PM to 10 AM on the same day, the monitoring module immediately suspends the execution of the modification request. The system triggers a conflict pre-check process. The first step in this process is to construct a temporary event node representing the current modification operation. The temporary event node contains the new timestamp after modification, the event type label, and the identifier of the participating member, but its status is marked as "pre-checking" and it has not yet been formally integrated into the event relationship graph.

[0043] The system simulates the global impact of adding temporary event nodes to an event relationship graph. The simulation is conducted in a copy of the event relationship graph. The copy graph is loaded with the latest stable version, and then the temporary event node is inserted. The system recalculates the attention coefficient between the temporary event node and existing nodes in the graph based on the temporary event node's attributes, establishing new directed weighted edges or updating the weights of existing edges. A graph traversal algorithm is initiated, starting from the temporary event node and probing neighboring nodes within a two- or three-hop range along directed weighted edges, assessing the direct conflicts and indirect ripple effects that the modification operation may trigger. Calculating the set of potential secondary conflict events requires analyzing the timing and resource dependencies of these neighboring nodes. For example, bringing forward the "quarterly planning meeting" might prevent participant Wang Wu from attending the previously scheduled "customer needs interview," while the delay of the "customer needs interview" might affect the subsequent "internal discussion meeting on solution design." Based on the simulation results, the system generates a pre-inspection report containing preventative adjustment suggestions. The pre-inspection report visualizes the conflict chain that the modification operation may cause and provides alternative solutions for each identified potential conflict point. For the above examples, preventative adjustment suggestions might include "suggesting that the quarterly planning meeting be moved to Tuesday morning at 11:00 AM to avoid conflict with participants in the customer needs interview" or "suggesting that the customer needs interview be rescheduled to Wednesday morning." The pre-emptive report is presented to the member initiating the modification, Li Si, through the user interface. Li Si can then decide whether to adopt the suggestion, cancel the modification, or force submission based on this information. Regardless of Li Si's choice, key metrics from the pre-emptive report, such as the predicted number of conflicts, the number of affected members, and the total potential time delay, are extracted and stored in a dedicated model training dataset. This dataset is used to subsequently optimize the modeling parameters of the event relationship graph and the judgment threshold of the conflict detection model. The system periodically compares the accumulated pre-emptive report data with the actual modification results to analyze the differences between predictions and reality. Through a machine learning training process, the weights of the attention coefficient calculation network in the attention-based event relationship graph are adjusted, making the graph more accurate in predicting the global impact of similar future modifications. The entire real-time monitoring and conflict pre-emptive process forms a closed loop of learning from practical feedback. The system continuously improves itself by observing real user behavior, enhancing the intelligence level of calendar coordination and user satisfaction.

[0044] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0045] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A multi-user interactive calendar management method based on artificial intelligence, characterized in that, Includes the following steps: Obtain the raw dataset of calendar events from multiple users, where each user's raw calendar event data includes timestamps, event type labels, and participant identifiers; Spatiotemporal features are extracted from the original dataset of calendar events to generate enhanced event data containing temporal distribution features and spatial correlation features; Construct an event relationship graph based on an attention mechanism, and input enhanced event data into the event relationship graph to model the multi-user event correlation. It receives new calendar event requests from target users in real time and retrieves a set of related events that may conflict with the new calendar event requests from the event relationship graph. A dynamic conflict resolution strategy is adopted to detect conflicts in the set of related events, generate an event conflict report containing conflict event identifiers and conflict levels, and calculate the priority weight of each event in the set of related events based on the conflict levels in the event conflict report. Based on the priority weight calculation results, the time scheduling of newly added calendar event requests and events in the associated event set is automatically adjusted, and the time scheduling adjustment results are fed back to the event relationship graph for graph structure update; Output the final coordinated multi-user calendar event schedule and generate an event change log containing all adjustment records.

2. The multi-user interactive calendar management method based on artificial intelligence according to claim 1, characterized in that, The spatiotemporal feature extraction of the original calendar event dataset includes: Identify periodic time patterns in the raw data of calendar events and extract time feature vectors including daily, weekly, and monthly cycles; Analyze the social relationship network among participating member identifiers and construct a member interaction frequency matrix; Tensor fusion of temporal feature vectors and interaction frequency matrix is ​​performed to form a spatiotemporal joint feature representation; Normalization is performed on the spatiotemporal joint feature representation to obtain standardized augmented event data.

3. The multi-user interactive calendar management method based on artificial intelligence according to claim 2, characterized in that, The construction of the event relationship graph based on the attention mechanism includes: The enhanced event data is mapped to graph node feature vectors, the attention coefficient between any two node feature vectors is calculated, and a directed weighted edge between nodes is established based on the attention coefficient. By iteratively updating the node feature vectors through a multi-layer graph convolutional network, an event relationship graph containing global event association information is finally output.

4. The multi-user interactive calendar management method based on artificial intelligence according to claim 3, characterized in that, The execution of the dynamic conflict resolution strategy includes: Extract event nodes from the event relationship graph that share participating members with the new calendar event request; Calculate the overlap index of each event node in the time dimension, correct the overlap index based on the historical response delay data of participating members, and convert the corrected overlap index into a conflict level quantification value. Combine the conflict level quantification values ​​of each event to generate an event conflict report.

5. The multi-user interactive calendar management method based on artificial intelligence according to claim 4, characterized in that, The priority weight calculation includes: Obtain the identity attribute data of all participating members involved in the conflict event, extract the decision preference characteristics of each member in historical events, calculate the initial priority score by combining the conflict level quantification value and decision preference characteristics, apply the member identity attribute data to the weighted adjustment of the initial priority score, and output the final priority weight with time sensitivity.

6. The multi-user interactive calendar management method based on artificial intelligence according to claim 5, characterized in that, The automatic time scheduling adjustment includes: An optimization function is established with the goal of minimizing time conflicts. The final priority weight is used as a constraint on the optimization function, and feasible scheduling schemes are searched within a preset time sliding window. Resource consumption is evaluated for multiple feasible scheduling schemes found, and the scheduling scheme with the best resource consumption evaluation is selected as the adjustment result.

7. The multi-user interactive calendar management method based on artificial intelligence according to claim 6, characterized in that, The map structure update includes: Detect all event nodes involved in the time scheduling adjustment; Recalculate the feature vector representation of the affected event nodes; Update the corresponding node and edge weights in the event relationship graph; Verify the global consistency of the map; Record map version change information to the event change log.

8. The multi-user interactive calendar management method based on artificial intelligence according to claim 7, characterized in that, The generation of the event change log includes: Capture the original event state before each time scheduling adjustment and record all intermediate decision-making steps during the adjustment process; Mark each participating member's confirmation status of the adjustment result, attach the system-generated explanation of the adjustment reason, and compress and store the complete change record in chronological order.

9. The multi-user interactive calendar management method based on artificial intelligence according to claim 1, characterized in that, Also includes: Real-time monitoring of participants' modifications to calendar events; When an unverified modification is detected, a conflict pre-check process is triggered. The pre-detection results are compared with the prediction results of the event relationship map; the map modeling parameters are automatically adjusted based on the degree of difference in the comparison.

10. The multi-user interactive calendar management method based on artificial intelligence according to claim 9, characterized in that, The conflict pre-detection process includes: Construct a temporary event node for the current modification operation and simulate the global impact of adding the temporary event node to the event relationship graph; Calculate the set of potential secondary conflict events, generate a pre-inspection report containing preventative adjustment recommendations, and store the key metrics in the pre-inspection report into the model training dataset.

Images