Digital calendar event prediction reminding method and system based on artificial intelligence

Abstract

The invention discloses a digital calendar event prediction reminding method and system based on artificial intelligence, and relates to the technical field of artificial intelligence, and the method comprises the steps: collecting calendar event record data, and executing semantic coding through a BERT model to generate a feature vector; constructing an event pattern recognition model based on the long and short term memory neural network to perform time sequence analysis; generating an event prediction probability distribution diagram by using a Transform model; screening high-probability candidate events in combination with the current calendar data; evaluating an event dependency relationship through a graph neural network to generate a reminding priority sequence; and pushing a prediction prompt according to the predicted occurrence time point. By fusing BERT semantic coding, LSTM time sequence analysis and the graph neural network, intelligent prediction and graded reminding of calendar events are realized, the prediction accuracy and reminding timeliness are improved, the false alarm rate is reduced, schedule conflicts are effectively avoided, and the working efficiency of users is improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method and system for predicting and reminding digital calendar events based on artificial intelligence. Background Art

[0002] Current digital calendar applications can usually only provide basic event recording and fixed-time reminder functions, and lack the ability to deeply mine and intelligently analyze user historical event data. Existing systems often use simple rule matching or statistical methods for event reminders, which cannot effectively identify the complex association patterns and temporal dependencies between events, resulting in low levels of intelligence and personalization of reminders.

[0003] In addition, traditional calendar reminder methods usually handle each event in isolation, without considering the semantic association and contextual information between events, and lack an adaptive evaluation mechanism for the importance of events. As a result, users often receive a large number of disordered reminder messages of varying importance, reducing the practicality of reminders and user experience. Summary of the invention

[0004] In view of the problems existing in the existing digital calendar system in terms of intelligent prediction, event association analysis and reminder priority management, the present invention is proposed.

[0005] Therefore, the problem to be solved by the present invention is how to realize intelligent semantic understanding, pattern mining and predictive analysis of calendar events based on deep learning and knowledge graph technology, and adaptively generate personalized reminder strategies according to the dependencies between events.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: In the first aspect, an embodiment of the present invention provides an artificial intelligence-based digital calendar event prediction and reminder method, which includes collecting calendar event record data, performing semantic encoding on the calendar event record data through a pre-trained BERT model, and generating an event semantic feature vector; constructing an event pattern recognition model based on a long short-term memory neural network, performing time series analysis on the event semantic feature vector, and extracting the event pattern feature vector; inputting the event pattern feature vector into a Transformer model, calculating the feature weight relationship, and generating an event prediction probability distribution map; based on the event prediction probability distribution map, combined with the current calendar data, screening out candidate events whose prediction probability exceeds the event importance threshold, and marking the expected occurrence time point; performing correlation analysis on the candidate events, evaluating the dependency relationship between events through a graph neural network, and generating an event reminder priority sequence; generating reminder information according to the event reminder priority sequence according to the expected occurrence time point, and pushing the event prediction reminder through the calendar application interface.

[0007] As a preferred solution of the artificial intelligence-based digital calendar event prediction and reminder method described in the present invention, wherein: performing semantic encoding on calendar event record data through a pre-trained BERT model includes the following steps: parsing the calendar event record data into structured data to generate an event field sequence; performing natural language processing on the event field sequence to identify time entities, location entities and personnel entities to generate an event segmentation sequence; adding field separators to the event segmentation sequence, aligning the length of the field sequence, and generating a normalized event sequence; inputting the normalized event sequence into the pre-trained BERT model to extract a semantic encoding vector; constructing a calendar event knowledge graph; combining the relationship information of the calendar event knowledge graph, using a contrastive learning method to enhance the features of the semantic encoding vector to obtain an event semantic feature vector.

[0008] As a preferred solution of the artificial intelligence-based digital calendar event prediction and reminder method described in the present invention, the event pattern recognition model is constructed based on the long short-term memory neural network, including the following steps: sorting the event semantic feature vectors by timestamp to construct a time series feature matrix; inputting the time series feature matrix into the forgetting gate of the long short-term memory neural network to output the feature forgetting weight; inputting the time series feature matrix and the feature forgetting weight into the memory gate to output the memory unit state vector; inputting the memory unit state vector into the Fourier transformer and the correlation analyzer to extract the event period feature vector and the event association strength vector respectively; inputting the memory unit state vector into the spatiotemporal attention layer to output the event spatiotemporal distribution vector, inputting the event spatiotemporal distribution vector into the multi-scale sliding window decomposer and the feature aggregation network to generate a fused feature vector; combining the event period feature vector, the event association strength vector and the fused feature vector to generate an event pattern feature vector.

[0009] As a preferred solution of the artificial intelligence-based digital calendar event prediction and reminder method described in the present invention, generating an event prediction probability distribution map includes the following steps: dividing the event pattern feature vector into a query vector, a key vector and a value vector, and constructing a multi-head attention layer input matrix; performing a scaled dot product operation on the multi-head attention layer input matrix to generate an attention weight matrix; multiplying the value vector by the attention weight matrix to generate a weighted feature vector; inputting the weighted feature vector into a feedforward neural network, and generating a temporal context feature vector through residual connection and layer normalization; constructing a bidirectional prediction head, the bidirectional prediction head includes a forward prediction layer and a backward prediction layer, and performing forward feature mapping and backward feature mapping on the temporal context feature vector respectively; performing probability fusion on the output features of the forward prediction layer and the backward prediction layer to generate an event prediction probability distribution map.

[0010] As a preferred solution of the artificial intelligence-based digital calendar event prediction and reminder method described in the present invention, generating an event reminder priority sequence includes the following steps: mapping events in the candidate event set into graph network nodes, calculating the similarity between nodes based on the event semantic feature vector, and constructing an event relationship graph; using a graph convolutional network to perform spatial domain aggregation on the node features in the event relationship graph to generate a node embedding vector; constructing a temporal attention layer, sorting the node embedding vectors according to the expected occurrence time points, and calculating the temporal dependency strength between event nodes; based on the temporal dependency strength, constructing directed edges for event nodes to generate an event dependency directed graph; performing topological sorting on the event dependency directed graph, and using the sorting result as the event reminder priority sequence.

[0011] As a preferred solution of the artificial intelligence-based digital calendar event prediction and reminder method described in the present invention, the feature enhancement of the semantic coding vector using a contrastive learning method includes the following steps: feature mapping the relationship information between the semantic coding vector and the calendar event knowledge graph to generate a relationship enhancement matrix; performing subgraph sampling on the relationship enhancement matrix to construct sample pairs based on the node relationships in the graph; constructing a feature projection space based on a projection head network, mapping the sample pairs to the feature projection space to obtain projection feature pairs; calculating the contrast loss of the projection feature pairs in the feature projection space, and optimizing the parameters of the projection head network using the gradient descent method; applying the optimized projection head network to the semantic coding vector to generate initial enhanced features; inputting the initial enhanced features into a multi-head self-attention network to capture the long-range dependencies between events and outputting attention features; performing a residual connection between the attention features and the semantic coding vector, and obtaining an event semantic feature vector after processing through a normalization layer.

[0012] As a preferred solution of the artificial intelligence-based digital calendar event prediction and reminder method described in the present invention, the calendar event knowledge graph includes time relationships, spatial relationships, interpersonal relationships and event type relationships.

[0013] The beneficial effects of the present invention are as follows: the present invention extracts and enhances semantic features through the pre-trained BERT model combined with the calendar event knowledge graph, and innovatively adopts the contrastive learning method to construct positive and negative sample pairs, so that the feature representations of related events are more similar; in terms of event pattern recognition, a three-way parallel feature extraction strategy based on LSTM is designed to comprehensively capture event features from three dimensions: periodicity, correlation, and spatiotemporal distribution; in the prediction link, a bidirectional prediction head is innovatively constructed to enhance the model's prediction capability, and the event importance threshold is dynamically adjusted through an adaptive threshold generator; finally, a graph neural network is used to analyze the dependencies between candidate events, and intelligent reminder priority sorting based on topological sorting is realized. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0015] Figure 1 The present invention is an overall flow chart of the digital calendar event prediction and reminder method based on artificial intelligence.

[0016] Figure 2 The data processing flow chart of the digital calendar event prediction and reminder method based on artificial intelligence.

[0017] Figure 3 Event pattern recognition flow chart for the artificial intelligence-based digital calendar event prediction and reminder method. DETAILED DESCRIPTION

[0018] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0019] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0020] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0021] Example 1, reference Figure 1~Figure 3 , which is the first embodiment of the present invention, and provides a digital calendar event prediction and reminder method based on artificial intelligence. The overall flow chart is as follows Figure 1 As shown, including, S1: Collect calendar event record data, perform semantic encoding on the calendar event record data through the pre-trained BERT model, and generate event semantic feature vectors; Specifically, the data processing flow chart is as follows: Figure 2 As shown, the following steps are included: S1.1: Parse calendar event record data into structured data and generate event field sequences.

[0022] Specifically, field information such as event title, event description, start time, end time, location, participants, etc. is extracted from the calendar event record data, and a unified data structure template is used to store the extracted field information to generate a standardized event field sequence.

[0023] S1.2: Perform natural language processing on the event field sequence, identify time entities, location entities, and person entities, and generate event segmentation sequences.

[0024] Specifically, a named entity recognition model is used to process the text content in the event field sequence to identify entities such as time expressions, place names, organization names, and personal names; the identified entities are standardized to generate an event word segmentation sequence containing standardized entities.

[0025] S1.3: Add field separators to the event word sequence, align the field sequence lengths, and generate a normalized event sequence.

[0026] S1.4: Input the normalized event sequence into the pre-trained BERT model and extract the semantic encoding vector.

[0027] S1.5: Build a knowledge graph of calendar events.

[0028] Among them, the calendar event knowledge graph includes time relations, spatial relations, interpersonal relations and event type relations.

[0029] Specifically, an event semantic network is constructed using a triple structure, with event identifiers as nodes and associations between events as edges; multi-granularity classification is performed on the nodes of the event semantic network, and the temporal link relationship between nodes is calculated based on temporal rules and semantic similarity to construct a temporal relationship subgraph; geographic location attributes are extracted from the nodes of the event semantic network, and the spatial dependency relationship between nodes is calculated based on geographic information coding and location distance to construct a spatial relationship subgraph; person entities are identified for the nodes of the event semantic network, and the social collaboration relationship between nodes is calculated based on the social network analysis method to construct an interpersonal relationship subgraph; semantic clustering is performed on the nodes of the event semantic network according to the event description information, and the category inheritance relationship between nodes is calculated based on the hierarchical clustering tree to construct a type relationship subgraph; the temporal relationship subgraph, spatial relationship subgraph, interpersonal relationship subgraph and type relationship subgraph are fused, and a unified event knowledge representation is generated based on a heterogeneous graph neural network to construct a calendar event knowledge graph.

[0030] S1.6: Combined with the relational information of the calendar event knowledge graph, the contrastive learning method is used to enhance the features of the semantic encoding vector to obtain the event semantic feature vector.

[0031] Further, the following steps are included: S1.6.1: Perform feature mapping between the semantic encoding vector and the relationship information of the calendar event knowledge graph to generate a relationship enhancement matrix.

[0032] S1.6.2: Perform subgraph sampling on the relationship enhancement matrix and construct sample pairs based on the node relationships in the graph.

[0033] S1.6.3: Construct a feature projection space based on the projection head network, map the sample pairs to the feature projection space, and obtain projected feature pairs.

[0034] Specifically, a multilayer perceptron MLP1 is initialized as a projection head network, wherein MLP1 includes two fully connected layers FC1 and FC2, and the input dimension of FC1 is the same as the dimension of the sample pair; the sample pair is input into FC1 for feature dimensionality reduction to obtain an intermediate feature vector, and the dimension of the intermediate feature vector is smaller than the dimension of the sample pair; batch normalization is performed on the intermediate feature vector to generate a normalized feature vector, and a nonlinear transformation is introduced through a ReLU activation function; the normalized feature vector is input into FC2 for feature transformation, and a projection feature is output, and the projection feature dimension is the same as the preset feature projection space dimension; L2 norm normalization is performed on the projection feature to generate a unitized projection feature; the unitized projection features of the positive sample pair and the negative sample pair are respectively input into the feature projection space, the cosine similarity between the sample pairs is calculated, and a similarity matrix is ​​generated; based on the similarity matrix, the projection features of the positive sample pair and the projection features of the negative sample pair are combined to form a projection feature pair.

[0035] S1.6.4: Calculate the contrast loss of the projected feature pairs in the feature projection space and use gradient descent to optimize the parameters of the projection head network.

[0036] It should be noted that in contrastive learning, the purpose of contrastive loss is to make the distance between similar (positive sample) feature pairs closer and the distance between dissimilar (negative sample) feature pairs farther.

[0037] Furthermore, a positive sample pair set P1 is defined, wherein the positive sample pairs include event pairs with a time interval less than a preset threshold T1, event pairs with the same event type, event pairs with the same participants, and event pairs with similar geographical locations; a negative sample pair set N1 is defined, wherein the negative sample pairs include event pairs with a time interval greater than a preset threshold T1, event pairs with different event types, event pairs with no intersection of participants, and event pairs with unrelated geographical locations; the cosine similarity S1 between the projection features in the positive sample pair P1 and the cosine similarity S2 between the projection features in the negative sample pair N1 are calculated; the temperature parameter T2 is introduced to scale the cosine similarity, and the cosine similarities S1 and S2 are divided by the temperature parameter T2 respectively; the contrast loss value L1 is calculated based on the InfoNCE loss function, wherein the numerator is the exponential form of the scaled cosine similarity of the positive sample pair, and the denominator is the sum of the scaled cosine similarity exponents of the positive sample pair and all negative sample pairs; the stochastic gradient descent method is used to minimize the contrast loss value L1, and the parameters of the projection head network are updated to obtain the optimized projection head network.

[0038] S1.6.5: Apply the optimized projection head network to the semantic encoding vector to generate initial enhanced features.

[0039] Specifically, the semantic coding vector is input into the first fully connected layer FC1 of the optimized projection head network, and a linear transformation is performed to obtain an intermediate representation vector; the intermediate representation vector is batch normalized to eliminate the feature distribution offset and obtain a standardized feature vector; the standardized feature vector is passed through the ReLU activation function, and a nonlinear transformation is introduced to obtain an activated feature vector; the activated feature vector is input into the second fully connected layer FC2 to generate a transformed feature vector; the transformed feature vector is normalized by L2 norm to obtain a unit length feature vector; the unit length feature vector is feature fused with the original semantic coding vector, and the initial enhanced feature is generated by weighted summation.

[0040] S1.6.6: Input the initial enhanced features into the multi-head self-attention network to capture the long-range dependencies between events and output the attention features.

[0041] S1.6.7: Perform a residual connection between the attention feature and the semantic encoding vector, and obtain the event semantic feature vector after processing through the normalization layer.

[0042] The sample pairs include positive sample pairs and negative sample pairs, wherein the positive sample pairs are event pairs with a direct relationship, and the negative sample pairs are event pairs without a direct relationship.

[0043] Preferably, the present invention enhances the semantic features of events by combining the relational information of the knowledge graph and the contrastive learning method. The design point is: adopting a multi-stage feature enhancement strategy, first constructing positive and negative sample pairs based on the node relationship in the graph structure for contrastive learning, so that the feature representations of related events are more similar; then using a multi-head self-attention mechanism to capture the long-range dependencies between events, and retaining the original semantic information through residual connections. This feature enhancement method can simultaneously utilize the structured relational information of the knowledge graph and the semantic associations between events, improving the distinguishability and generalization of event feature representations.

[0044] S2: Build an event pattern recognition model based on the long short-term memory neural network, perform temporal analysis on the event semantic feature vector, and extract the event pattern feature vector.

[0045] Specifically, the event pattern recognition flow chart is as follows: Figure 3 As shown, the following steps are included: S2.1: Sort the event semantic feature vectors by timestamp and construct a temporal feature matrix.

[0046] S2.2: Input the time series feature matrix into the forget gate of the long short-term memory neural network and output the feature forgetting weight.

[0047] S2.3: Input the time series feature matrix and feature forgetting weight into the memory gate and output the memory unit state vector.

[0048] S2.4: Input the memory unit state vector into the Fourier transformer and the correlation analyzer to extract the event period feature vector and the event association strength vector respectively.

[0049] Specifically, the memory unit state vector is analyzed in the frequency domain through the Fourier transformer, and the main frequency components are extracted as the periodic characteristics of the event. The timing correlation coefficient in the event sequence is calculated through the correlation analyzer, and the event correlation matrix is ​​constructed based on the sliding time window to extract the correlation strength characteristics.

[0050] It should be noted that in order to avoid the influence of data noise on the extraction of periodic features, the memory unit state vector is firstly smoothed before Fourier transform.

[0051] S2.5: Input the memory unit state vector into the spatiotemporal attention layer, output the event spatiotemporal distribution vector, input the event spatiotemporal distribution vector into the multi-scale sliding window decomposer and feature aggregation network to generate a fused feature vector.

[0052] Specifically, the memory unit state vector is input into the temporal attention sublayer and the spatial attention sublayer respectively to obtain the temporal attention weight and the spatial attention weight; the memory unit state vector is weighted based on the temporal attention weight and the spatial attention weight to generate the spatiotemporal fusion feature; the spatiotemporal fusion feature is processed by the fully connected layer and the activation function to obtain the event spatiotemporal distribution vector; sliding windows of different scales are constructed to perform multi-scale scanning on the event spatiotemporal distribution vector to extract the local feature sequence; maximum pooling and average pooling operations are performed on the local feature sequence to obtain a multi-scale feature matrix; a feature aggregation network is used to perform channel attention calculation on the multi-scale feature matrix to obtain the feature importance weight; based on the feature importance weight, the multi-scale feature matrix is ​​weighted and summed to generate a fused feature vector.

[0053] S2.6: Combine the event period feature vector, the event association strength vector and the fusion feature vector to generate an event pattern feature vector.

[0054] Preferably, the present invention screens and memorizes the time series features through the gating mechanism of the LSTM network, and adopts a three-way parallel feature extraction strategy: first, the Fourier transform is used to analyze the periodic pattern, second, the correlation strength between events is obtained through the correlation analyzer, and third, the spatiotemporal attention mechanism and the multi-scale sliding window are combined to realize the adaptive extraction of the spatiotemporal distribution features of events. This multi-dimensional feature extraction method can fully capture the periodicity, correlation and spatiotemporal distribution features in the event sequence, thereby effectively identifying complex event patterns.

[0055] S3: Input the event pattern feature vector into the Transformer model, calculate the feature weight relationship, and generate an event prediction probability distribution map.

[0056] Specifically, the following steps are included: S3.1: Split the event pattern feature vector into query vector, key vector and value vector, and construct the multi-head attention layer input matrix.

[0057] S3.2: Perform a scaled dot product operation on the multi-head attention layer input matrix to generate the attention weight matrix.

[0058] Among them, the attention weight matrix represents the association strength between event features.

[0059] S3.3: Multiply the value vector by the attention weight matrix to generate a weighted feature vector.

[0060] Among them, the weighted feature vector integrates the temporal dependencies between events.

[0061] S3.4: Input the weighted feature vector into the feedforward neural network, generate the temporal context feature vector through residual connection and layer normalization.

[0062] S3.5: Construct a bidirectional prediction head, which includes a forward prediction layer and a backward prediction layer, and performs forward feature mapping and backward feature mapping on the temporal context feature vector respectively.

[0063] Specifically, the forward prediction layer predicts future events based on historical time series features, and the backward prediction layer traces back historical events based on future time series features, and enhances the prediction ability of the model through bidirectional feature mapping. Both the forward prediction layer and the backward prediction layer adopt a multi-layer perceptron structure and output a prediction probability vector.

[0064] S3.6: Probabilistically fuse the output features of the forward prediction layer and the backward prediction layer to generate an event prediction probability distribution map.

[0065] Among them, the event prediction probability distribution diagram represents the probability of occurrence of various events in the future time window.

[0066] Specifically, the output probabilities of the forward prediction layer and the backward prediction layer are weighted averaged to generate a fused probability distribution; according to the preset time window size, the fused probability distribution is mapped to the time axis to construct an event prediction probability distribution graph.

[0067] S4: Based on the event prediction probability distribution diagram and combined with the current calendar data, select candidate events whose prediction probability exceeds the event importance threshold and mark the expected time of occurrence.

[0068] Specifically, the event prediction probability distribution map is divided into multiple time windows, the event prediction probability in each time window is normalized to generate a standardized probability vector; the confirmed event information in the current calendar data is collected, the time occupancy is extracted, and the time resource constraint matrix is ​​constructed; based on the historical event completion and user feedback data, an adaptive threshold generator is trained, and the adaptive threshold generator outputs the importance threshold corresponding to the event type; the event prediction probability in the standardized probability vector is compared with the importance threshold of the corresponding event type to screen out high-probability events; the high-probability events are matched with the time resource constraint matrix, the events with time conflicts are eliminated, and a set of candidate events is generated; according to the probability peak position in the event prediction probability distribution map, the expected occurrence time point is marked for each event in the candidate event set.

[0069] S5: Perform correlation analysis on candidate events, evaluate the dependencies between events through graph neural networks, and generate event reminder priority sequences.

[0070] Specifically, the following steps are included: S5.1: Map the events in the candidate event set into graph network nodes, calculate the similarity between nodes based on the event semantic feature vectors, and construct an event relationship graph.

[0071] S5.2: Use graph convolutional network to aggregate node features in the event relationship graph in the spatial domain and generate node embedding vectors.

[0072] S5.3: Construct a temporal attention layer, sort the node embedding vectors according to the expected occurrence time points, and calculate the temporal dependency strength between event nodes.

[0073] S5.4: Based on the temporal dependency strength, construct directed edges for event nodes and generate an event dependency directed graph.

[0074] The weight of the directed edge represents the degree of dependency between events.

[0075] S5.5: Perform topological sorting on the event dependency directed graph and use the sorting result as the event reminder priority sequence.

[0076] S6: Generate reminder information according to the event reminder priority sequence according to the expected occurrence time point, and push the event prediction reminder through the calendar application interface.

[0077] Accordingly, the event reminder priority sequence is divided into time windows according to the expected occurrence time points, and events with similar times are organized into event clusters; based on the priority distribution of the event clusters, the reminder time interval is calculated for the events in each event cluster to generate a reminder time series; according to the number of events and priority distribution in the event cluster, a reminder display template is assigned to each event to generate reminder content; a reminder buffer pool is constructed, and the reminder time series and reminder content are organized into reminder information packages and stored in the reminder buffer pool; the trigger time of the reminder information package in the reminder buffer pool is monitored, and the reminder information package is sent to the push queue when the trigger time is reached; the calendar application program interface is called to read the reminder information package in the push queue, and the reminder content is pushed to the user through the system notification bar.

[0078] In summary, the present invention extracts and enhances semantic features through the pre-trained BERT model combined with the calendar event knowledge graph, and innovatively adopts the contrastive learning method to construct positive and negative sample pairs, so that the feature representations of related events are more similar; in terms of event pattern recognition, a three-way parallel feature extraction strategy based on LSTM is designed to comprehensively capture event features from three dimensions: periodicity, correlation, and spatiotemporal distribution; in the prediction link, a bidirectional prediction head is innovatively constructed to enhance the model's prediction ability, and the event importance threshold is dynamically adjusted through an adaptive threshold generator; finally, a graph neural network is used to analyze the dependencies between candidate events, and intelligent reminder priority sorting based on topological sorting is realized.

[0079] Example 2, reference Figure 1~Figure 3 , which is the second embodiment of the present invention, provides a digital calendar event prediction and reminder method based on artificial intelligence. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0080] To verify the effectiveness of the AI-based digital calendar event prediction and reminder method, the research team selected calendar event data generated by 300 employees of an Internet company from January 2023 to December 2023 for experimental analysis. The experimental data set contains about 75,000 calendar event records, covering various types such as meetings, tasks, appointments, reminders, etc. The experimental environment uses an Intel Xeon Gold 6248R processor, 384GB memory, and NVIDIA A100 GPU.

[0081] In the data preprocessing stage, the research team used the pre-trained Chinese BERT-base model (12 layers, 768-dimensional hidden layer, 12 attention heads) to semantically encode calendar events. The knowledge graph was constructed using Neo4j graph database storage, which contains a total of 92,463 entity nodes and 347,852 relationship edges. The contrastive learning in the feature enhancement stage uses a 128-dimensional projection space, the temperature coefficient is set to 0.07, and the batch size is 256. The multi-head self-attention network is configured with 8 attention heads and a dropout rate of 0.1.

[0082] The event pattern recognition model uses a two-layer LSTM structure with a hidden layer dimension of 512 and a time window size of 28 days. The Fourier transform uses a 512-point FFT to extract the first 10 main frequency components as periodic features. The spatiotemporal attention layer is configured with 16 attention heads, and the sliding window scale includes four granularities of 1 day, 3 days, 7 days, and 14 days.

[0083] The Transformer prediction model uses a 6-layer encoder structure, each layer contains 8 attention heads, and the feedforward network dimension is 2048. The forward and backward prediction layers of the bidirectional prediction head both use a three-layer MLP structure, and the hidden layer dimensions are {512, 256, 128} respectively. The adaptive generator of event importance thresholds is implemented based on the XGBoost framework and trained using historical 6-month event completion data.

[0084] After a 12-month system performance evaluation, the method has achieved significant results in all key indicators. As shown in Table 1, the method of the present invention is compared and analyzed with two existing mainstream calendar intelligent reminder methods: Table 1 Comparison of indicators between the method of the present invention and the prior art It can be seen from the experimental data that the method of the present invention has significantly improved the core performance indicators compared with the existing technology, especially in the four aspects of event prediction accuracy, reminder timeliness, false alarm rate, and average prediction advance. Although the system resource usage is relatively high, considering the significant improvement in prediction performance, this resource overhead is acceptable.

[0085] In actual applications, the system's prediction effects on different types of events also vary. For regular meetings and fixed tasks with obvious periodicity, the prediction accuracy rate reaches 96.7%; for temporary meetings and sudden affairs, the prediction accuracy rate is 88.5%; for comprehensive events that rely on multi-party coordination, the prediction accuracy rate is 90.1%. The event dependency analysis constructed by the graph neural network shows that on average each event has a direct or indirect dependency relationship with 3.7 other events. This related information plays an important role in improving the accuracy of predictions.

[0086] User feedback data shows that the system's intelligent reminder function has significantly improved work efficiency. Statistics show that after using the system, users have reduced schedule conflicts by an average of 27.3%, increased advance preparation time by 65.4%, and reduced the omission rate of important matters by 81.2%. Especially for managers and project coordinators, the system helps them better balance and coordinate various affairs, and the satisfaction rate of reminder response has reached 93.5%.

[0087] After one year of continuous monitoring and optimization, the system has demonstrated good stability and scalability in large-scale practical application environments.

[0088] Embodiment 3 is the third embodiment of the present invention, which provides a digital calendar event prediction reminder system based on artificial intelligence, including a data collection module, an event pattern recognition module, a candidate event screening module, and a push module; The data collection module is used to collect calendar event record data and obtain historical event data in the user's calendar; The event pattern recognition module performs semantic encoding on the collected calendar event record data to generate an event semantic feature vector; The candidate event screening module selects candidate events whose predicted probabilities exceed the event importance threshold based on the event prediction probability distribution diagram and combines the current calendar data, and marks the expected occurrence time point; The push module generates reminder information according to the event reminder priority sequence and the expected occurrence time point, and pushes the event prediction reminder through the calendar application interface.

[0089] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A digital calendar event prediction and reminder method based on artificial intelligence, characterized in that: include, Collect calendar event record data, perform semantic encoding on the calendar event record data through a pre-trained BERT model, and generate an event semantic feature vector; Building an event pattern recognition model based on a long short-term memory neural network, performing time series analysis on the event semantic feature vector, and extracting the event pattern feature vector; Input the event pattern feature vector into the Transformer model, calculate the feature weight relationship, and generate an event prediction probability distribution graph; According to the event prediction probability distribution map, combined with the current calendar data, candidate events whose prediction probabilities exceed the event importance threshold are screened out, and the expected occurrence time points are marked; Performing correlation analysis on the candidate events, evaluating dependencies between events through a graph neural network, and generating an event reminder priority sequence; The event reminder priority sequence is used to generate reminder information according to the expected occurrence time point, and the event prediction reminder is pushed through the calendar application interface.

2. The method for predicting and reminding digital calendar events based on artificial intelligence as claimed in claim 1, characterized in that: The performing semantic encoding on the calendar event record data by the pre-trained BERT model comprises the following steps: Parse calendar event record data into structured data and generate event field sequences; Performing natural language processing on the event field sequence, identifying time entities, location entities and person entities, and generating an event segmentation sequence; Adding field separators to the event word segmentation sequence, aligning the length of the field sequence, and generating a normalized event sequence; Input the normalized event sequence into a pre-trained BERT model to extract a semantic encoding vector; Build a knowledge graph of calendar events; Combined with the relationship information of the calendar event knowledge graph, a contrastive learning method is used to enhance the features of the semantic encoding vector to obtain an event semantic feature vector.

3. The method for predicting and reminding digital calendar events based on artificial intelligence as claimed in claim 1, characterized in that: The event pattern recognition model based on the long short-term memory neural network comprises the following steps: Sort the event semantic feature vectors by timestamp and construct a time series feature matrix; Input the time series feature matrix into the forget gate of the long short-term memory neural network, and output the feature forgetting weight; Input the time series feature matrix and the feature forgetting weight into a memory gate, and output a memory unit state vector; Inputting the memory unit state vector into a Fourier transformer and a correlation analyzer to extract an event period feature vector and an event association strength vector respectively; Input the memory unit state vector into the spatiotemporal attention layer, output the event spatiotemporal distribution vector, input the event spatiotemporal distribution vector into the multi-scale sliding window decomposer and feature aggregation network to generate a fused feature vector; The event period feature vector, the event association strength vector and the fusion feature vector are combined to generate an event pattern feature vector.

4. The method for predicting and reminding digital calendar events based on artificial intelligence as claimed in claim 1, characterized in that: Generating an event prediction probability distribution graph comprises the following steps: Split the event pattern feature vector into query vector, key vector and value vector, and construct the input matrix of the multi-head attention layer; Performing a scaled dot product operation on the multi-head attention layer input matrix to generate an attention weight matrix; Multiplying the value vector by the attention weight matrix to generate a weighted feature vector; Inputting the weighted feature vector into a feedforward neural network, generating a temporal context feature vector through residual connection and layer normalization; Constructing a bidirectional prediction head, the bidirectional prediction head comprising a forward prediction layer and a backward prediction layer, performing forward feature mapping and backward feature mapping on the temporal context feature vector respectively; The output features of the forward prediction layer and the backward prediction layer are probabilistically fused to generate an event prediction probability distribution map.

5. The method for predicting and reminding digital calendar events based on artificial intelligence as claimed in claim 1, characterized in that: Generating an event reminder priority sequence comprises the following steps: Map the events in the candidate event set into graph network nodes, calculate the similarity between nodes based on the event semantic feature vectors, and construct an event relationship graph; A graph convolutional network is used to perform spatial domain aggregation on node features in the event relationship graph to generate a node embedding vector; Construct a temporal attention layer, sort the node embedding vectors according to the expected occurrence time points, and calculate the temporal dependency strength between event nodes; Based on the temporal dependency strength, construct directed edges for event nodes to generate an event dependency directed graph; A topological sort is performed on the event dependency directed graph, and the sorting result is used as an event reminder priority sequence.

6. The method for predicting and reminding digital calendar events based on artificial intelligence as claimed in claim 2, characterized in that: The feature enhancement of the semantic coding vector by using the contrastive learning method comprises the following steps: Perform feature mapping between the semantic encoding vector and the relationship information of the calendar event knowledge graph to generate a relationship enhancement matrix; Performing subgraph sampling on the relationship enhancement matrix, and constructing sample pairs based on node relationships in the graph; Constructing a feature projection space based on a projection head network, mapping the sample pairs to the feature projection space, and obtaining projected feature pairs; Calculating the contrast loss of the projection feature pair in the feature projection space, and optimizing the parameters of the projection head network using a gradient descent method; Applying the optimized projection head network to the semantic encoding vector to generate initial enhanced features; Inputting the initial enhanced features into a multi-head self-attention network to capture long-range dependencies between events and output attention features; The attention feature is residually connected with the semantic encoding vector, and the event semantic feature vector is obtained after processing through a normalization layer.

7. The method for predicting and reminding digital calendar events based on artificial intelligence as claimed in claim 2, characterized in that: The calendar event knowledge graph includes time relations, spatial relations, interpersonal relations and event type relations.

8. A digital calendar event prediction and reminder system based on artificial intelligence, using the digital calendar event prediction and reminder method based on artificial intelligence as claimed in any one of claims 1 to 7, characterized in that: It includes data collection module, event pattern recognition module, candidate event screening module, and push module; The data collection module is used to collect calendar event record data and obtain historical event data in the user's calendar; The event pattern recognition module performs semantic encoding on the collected calendar event record data to generate an event semantic feature vector; The candidate event screening module screens out candidate events whose predicted probabilities exceed the event importance threshold according to the event prediction probability distribution diagram and combines the current calendar data, and marks the expected occurrence time point; The push module generates reminder information according to the event reminder priority sequence and the expected occurrence time point, and pushes the event prediction reminder through the calendar application interface.

Images