Application program calling method and device, computer device and storage medium

Abstract

The application relates to the technical field of artificial intelligence, and discloses an application calling method and device, a computer device and a storage medium, which comprise the following steps: collecting historical behavior data of a user for an application and historical calling data of the application, fusing the historical behavior data and the historical calling data, and generating comprehensive training data; performing transfer learning and fine-tuning training on a pre-trained large language model based on the comprehensive training data, and generating an application calling prediction model; obtaining current behavior data of a target user for a target application, inputting the current behavior data into the application calling prediction model, and generating a calling prediction result of the target application; and selecting a target calling scheme that is adapted to the target application from a preset application calling scheme library according to the calling prediction result, and executing the target calling scheme. The application can be applied to the application calling scene of medical health and financial technology, and can realize flexible and accurate calling of an application.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and more specifically to an application invocation method, apparatus, computer device, and computer-readable storage medium. Background Technology

[0002] Currently, with the rapid development of mobile internet technology and the widespread adoption of smart terminal devices, applications have become the core carriers for users to obtain information, handle affairs, and engage in social interactions. At present, mainstream mobile operating systems typically employ a static application launch mechanism, relying on users actively clicking application icons or system push notifications to trigger application loading and execution. However, this passive, responsive invocation model is ill-suited to complex and ever-changing user scenarios, leading to increasingly prominent problems such as application launch delays, redundant resource loading, and fragmented user experience.

[0003] In existing technologies, some solutions attempt to predict user application usage intentions through simple rule matching or statistical prediction models. For example, applications are preloaded based on time patterns (such as consistently opening transportation apps on weekday mornings) or geographical location information (such as recommending shopping apps after entering a shopping mall). However, these methods have significant limitations: First, rule engines struggle to capture fine-grained contextual features of user behavior and cannot handle complex behavioral patterns across scenarios and time periods; second, traditional statistical models lack the ability to model long-sequence dependencies and cannot effectively integrate the deep correlation between users' historical operation sequences and the underlying system call logs; third, existing solutions often employ independent prediction and execution architectures, lacking a dynamic adaptation mechanism between prediction results and resource scheduling strategies, resulting in the improvement in prediction accuracy failing to effectively translate into an improvement in actual user experience.

[0004] In the healthcare field, timely application activation is crucial for scenarios such as chronic disease management and emergency medical interventions. Current technologies for health monitoring applications typically rely on timed reminders or threshold alarm mechanisms for activation, failing to incorporate predictive service activation based on real-time physiological trends, medication adherence history, and external environmental factors (such as weather and air quality). For example, blood glucose management applications for diabetic patients only passively push suggestions after detecting abnormal values, rather than proactively predicting high blood glucose risk and pre-loading dietary intervention plans. This delayed response model not only reduces the timeliness of medical interventions but also increases the cognitive burden and operational costs for patients.

[0005] In the fintech field, scenarios such as high-frequency trading, real-time risk control, and personalized investment advisory place extremely high demands on the responsiveness and accuracy of application calls. Currently, mobile banking and securities trading applications mostly rely on user-initiated login or market fluctuation push notifications, lacking proactive service preparation based on user asset trends, market sentiment evolution, and macroeconomic events. For example, when a user's holdings face the risk of sharp exchange rate fluctuations, traditional solutions require the user to manually open the application before hedging tools can be loaded, missing the optimal decision-making window. Furthermore, financial applications involve sensitive permissions and multiple security verifications; frequent cold starts and redundant identity verification processes further exacerbate call delays, impacting user experience and business conversion rates.

[0006] Therefore, how to provide an application invocation method, apparatus, computer device, and computer-readable storage medium that can enable flexible and accurate invocation of applications is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0007] In view of the shortcomings of the prior art, the purpose of the present invention is to provide an application invocation method, apparatus, computer device and computer-readable storage medium, which aims to solve the problem of how to achieve flexible and accurate invocation of applications.

[0008] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides an application invocation method, comprising: Collect historical user behavior data and historical call data of the application, and fuse the historical behavior data and historical call data to generate comprehensive training data; Based on the comprehensive training data, the pre-trained large language model is subjected to transfer learning and fine-tuning training to generate an application call prediction model. Obtain the current behavior data of the target user for the target application, input the current behavior data into the application call prediction model, and generate the call prediction result of the target application; Based on the invocation prediction results, a target invocation scheme that is compatible with the target application is selected from a preset application invocation scheme library, and the target invocation scheme is executed to realize the invocation of the target application.

[0009] Secondly, the present invention provides an application invocation device, comprising: The collection module is used to collect historical user behavior data and historical call data of the application, and to perform fusion processing on the historical behavior data and the historical call data to generate comprehensive training data. The model generation module is used to perform transfer learning and fine-tuning training on the pre-trained large language model based on the comprehensive training data, and generate an application call prediction model. The result generation module is used to obtain the current behavior data of the target user for the target application, input the current behavior data into the application call prediction model, and generate the call prediction result of the target application. The filtering module is used to filter out a target calling scheme that is compatible with the target application from a preset application calling scheme library based on the calling prediction result, and execute the target calling scheme to realize the calling of the target application.

[0010] Thirdly, the present invention provides a computer device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the application invocation method described above.

[0011] Fourthly, the present invention provides a computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the application program invocation method described above.

[0012] Compared to existing technologies, this invention provides an application invocation method, apparatus, computer device, and computer-readable storage medium, comprising: collecting historical user behavior data and historical invocation data of the application; fusing the historical behavior data and the historical invocation data to generate comprehensive training data; performing transfer learning and fine-tuning training on a pre-trained large language model based on the comprehensive training data to generate an application invocation prediction model; acquiring current user behavior data of the target application; inputting the current behavior data into the application invocation prediction model to generate an invocation prediction result for the target application; and, based on the invocation prediction result, selecting a target invocation scheme adapted to the target application from a preset application invocation scheme library and executing the target invocation scheme to realize the invocation of the target application. This invention can adaptively select the optimal invocation timing and execution strategy according to the user's personalized characteristics, thereby reducing startup latency and resource waste while ensuring the accuracy of invocation decisions and the robustness of the execution process. In other words, this invention can achieve flexible and accurate application invocation. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 This is a schematic diagram of the application environment for an application invocation method provided in an embodiment of the present invention.

[0015] Figure 2 This is a flowchart illustrating an application invocation method according to an embodiment of the present invention.

[0016] Figure 3 This is a schematic diagram of a program module of an application calling device provided in an embodiment of the present invention.

[0017] Figure 4 This is a schematic diagram of the structure of a computer device provided in an embodiment of the present invention.

[0018] Figure 5 This is another structural schematic diagram of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0019] 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, not all, of the embodiments of the present invention. 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.

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

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

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

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

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

[0025] It should be understood that the sequence number of each step in the following embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0026] To illustrate the technical solution of the present invention, specific embodiments are described below.

[0027] An embodiment of the present invention provides an application invocation method that can be applied to, for example... Figure 1In the application environment shown, the client and server communicate via a network. The client includes, but is not limited to, handheld computers, desktop computers, laptops, ultra-mobile personal computers (UMPCs), netbooks, cloud computing devices, and personal digital assistants (PDAs). The server can be a standalone server or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0028] Please see Figure 2 An embodiment of the present invention provides an application invocation method, wherein the method includes the following steps: S100: Collect historical user behavior data and historical call data of the application, and perform fusion processing on the historical behavior data and the historical call data to generate comprehensive training data; S200. Based on the comprehensive training data, perform transfer learning and fine-tuning training on the pre-trained large language model to generate an application call prediction model. S300: Obtain the current behavior data of the target user for the target application, input the current behavior data into the application call prediction model, and generate the call prediction result of the target application; S400. Based on the call prediction result, a target call scheme that is compatible with the target application is selected from the preset application call scheme library, and the target call scheme is executed to realize the call to the target application.

[0029] In practical implementation, the method in this embodiment achieves a paradigm shift in applications from passive response to proactive prediction by constructing a four-stage technical closed loop of "data fusion - model building - intelligent prediction - dynamic execution," thereby achieving flexible and accurate invocation. At the data level, by fusing user historical behavior data with system-level call data, the limitation of information silos from a single data source is broken, enabling the model to simultaneously learn user intent semantics and resource constraints, establishing cross-modal relational representations. At the model level, a pre-trained large language model combined with domain-fine-tuned transfer learning strategy is adopted, which not only retains general semantic understanding capabilities but also injects terminal scenario knowledge into the comprehensive training data, significantly improving the accuracy of long-sequence behavior modeling and complex contextual reasoning. At the prediction level, call prediction results are generated based on real-time behavior data, realizing pre-emptive perception of user intent and advancing the application launch time from "after user triggering" to "the nascent stage of demand." At the execution level, through the dynamic filtering mechanism of the call solution library, abstract prediction results are transformed into implementable resource scheduling strategies, directly translating prediction accuracy into user experience gains. The four-stage collaborative operation enables the system to adaptively select the optimal invocation timing and execution strategy based on the user's personalized characteristics, thereby reducing startup latency and resource waste while ensuring the accuracy of invocation decisions and the robustness of the execution process.

[0030] Understandably, the application invocation method provided in this embodiment of the invention can be applied to application invocation scenarios related to the medical and health field. The following is a specific example: In the daily blood glucose management scenarios of diabetic patients, the system continuously collects behavioral data such as the patient's historical blood glucose monitoring records, insulin injection time, dietary intake logs, exercise trajectory, and sleep quality. Simultaneously, it collects system call data such as blood glucose meter Bluetooth connection logs, health data synchronization frequency, abnormal alarm trigger records, and medical device driver call sequence. After fusing and processing to generate comprehensive training data, the pre-trained large language model is fine-tuned to build a personalized application call prediction model for the patient. When the target patient exhibits continuous screen-swiping behavior data browsing health-related content around 10:15 AM on a weekday (historical data shows this is their peak post-meal blood glucose monitoring period), the model predicts that they will soon perform a blood glucose test and view trend analysis charts, generating a high-confidence call prediction result. The system then selects an appropriate solution from the call solution library: if the blood glucose meter is detected to be within Bluetooth range and has sufficient power, the main interface of the blood glucose management application is preloaded and the data from the last three days is synchronized; if the device is not connected, it switches to manual input mode and pre-fills the estimated timestamp. The final execution of this target invocation scheme allows patients to immediately enter the core function interface after clicking the application icon, without waiting for data loading, significantly shortening the time delay from the generation of intent to obtaining health feedback.

[0031] Understandably, the application invocation method provided in this embodiment of the invention can be applied to application invocation scenarios related to the fintech field. The following is a specific example: In intelligent wealth management scenarios, the system collects behavioral data such as users' historical asset query periods, risk assessment adjustment records, wealth management product browsing history, fixed investment deduction dates, and large fund arrival reminders. It also collects system call data such as account balance interface call frequency, product valuation calculation delays, transaction confirmation chain time, and compliance suitability matching rule verification logs. After generating comprehensive training data through fusion training, the pre-trained large language model is adapted to the financial domain to construct an asset allocation demand prediction model. When a target user, on the first weekend after receiving their year-end bonus via SMS at the end of the quarter (historical data shows this user typically rebalances assets during this period), continuously browses three fixed-income products with different maturities and their current risk assessment level is conservative, the model predicts they will soon perform a wealth management allocation operation, generating a call prediction result that includes product suitability scores and liquidity suggestions. Based on this result, the system selects the target call plan from the call plan library: if it detects that the user's current account balance is sufficient and the product is within the sales period, it preloads the wealth management purchase interface of the bank's official application, displays wealth management products that meet the suitability requirements in order of historical preferences, pre-fills the investment amount (based on the average of historical single investments), and highlights the risk level matching indicator; if it detects that the user's risk assessment is about to expire, it prioritizes launching the assessment update process and locks the transaction entry until the reassessment is completed. After executing this plan, users can immediately enter the compliant transaction path after generating configuration intentions. All product recommendations are verified in real time through the bank's internal suitability management system, strictly adhering to the principle of matching investors' risk tolerance with product risk levels, ensuring sales compliance while shortening decision-making delays.

[0032] Furthermore, in one embodiment, the application invocation method, wherein collecting historical user behavior data and historical application invocation data, and fusing the historical behavior data and the historical invocation data to generate comprehensive training data, specifically includes: Collect historical user behavior data and historical call data of the application from multiple heterogeneous data sources; A dual-channel data encoder is constructed to perform user intent semantic encoding on the historical behavior data and system state semantic encoding on the historical call data, respectively. The temporal correlation between the encoded historical behavior data and the historical call data is aligned through a cross-modal attention mechanism to generate a fused semantic vector sequence. Based on the fused semantic vector sequence, comprehensive training data is constructed.

[0033] In practice, the above steps in this embodiment achieve efficient alignment and deep representation of user behavior intent and system underlying state by constructing a multi-source heterogeneous data acquisition and dual-channel semantic coding fusion mechanism, thereby achieving the technical effect of improving training data quality and model prediction accuracy.

[0034] The specific implementation process of this embodiment is roughly as follows: 1. Multi-source heterogeneous data acquisition and cleaning steps: Historical user behavior data and application call data are collected from multiple heterogeneous data sources, including terminal device sensors, operating system logs, application tracking systems, and cloud service interfaces. The historical behavior data includes touch screen operation coordinate sequences, application dwell time, page jump paths, and device posture change events. The historical call data includes process creation logs, memory allocation records, network request timing sequences, and permission request trajectories. The collected raw data is processed by imputing missing values, removing outliers, and aligning timestamps to generate a standardized time-series data stream. 2. Dual-channel semantic encoding steps: A dual-channel data encoder is constructed, comprising a user behavior encoding channel and a system call encoding channel. In the user behavior encoding channel, a temporal convolutional network is used to extract the local operation patterns of the historical behavior data, and a bidirectional long short-term memory network is used to capture the contextual evolution of user intent, outputting the semantic encoding of user intent. In the system call encoding channel, a graph neural network is used to model the process dependencies of the historical call data, and a temporal Transformer is used to encode the dynamic changes of resource states, outputting the semantic encoding of system state. 3. Cross-modal timing alignment steps: The user intent semantic encoding and the system state semantic encoding are input into a cross-modal attention mechanism to calculate the attention weight matrix at the time of user behavior and the time of system call, and to identify the temporal correlation pattern between user operation intent and underlying resource response; through multi-head attention fusion and position encoding calibration, the temporal granularity and semantic space of the two types of semantic encoding are aligned to generate a fused semantic vector sequence. 4. Dynamic sample augmentation steps: An initial sample set is constructed based on the fused semantic vector sequence. A temporal mask prediction task is designed to perform self-supervised enhancement on the samples, randomly masking the semantic features of some time steps and training the model to reconstruct the missing information. A contrastive learning strategy is introduced to construct positive sample pairs (semantic sequences of adjacent time periods of the same user) and negative sample pairs (semantic sequences of different users or different applications) to optimize the discriminative representation of the fused semantic vector. 5. Steps for generating comprehensive training data: The enhanced fused semantic vector sequence is associated with user identifiers, application identifiers, and call result labels to construct comprehensive training data containing input feature sequences, predicted target labels, and sample weight coefficients.

[0035] Furthermore, in one embodiment, the application invocation method, wherein the step of performing transfer learning and fine-tuning training on the pre-trained large language model based on the comprehensive training data to generate an application invocation prediction model specifically includes: According to the preset training data partitioning strategy, the comprehensive training data is divided into transfer dataset, model training set, model validation set and model test set; The pre-trained large language model is transferred to the transfer dataset, and the transferred language model is fine-tuned using the model training set. The fine-tuned large language model is evaluated using the model validation set, and the large language model is optimized based on the evaluation results. The optimized large language model is tested on the model test set. If the test results meet the application call prediction requirements, the optimized large language model is determined as the application call prediction model.

[0036] In practice, the above steps in this embodiment achieve deep adaptation and performance optimization of the general-purpose large language model to the application calling prediction domain by constructing a phased and multi-level model training and evaluation system, thereby achieving the technical effect of improving the model's domain adaptability and prediction reliability.

[0037] The specific implementation process of this embodiment is roughly as follows: 1. Domain adaptation pre-training steps: According to the preset time series partitioning strategy, the comprehensive training data is stratified and sampled according to user identifiers, and divided into transfer dataset, model training set, model validation set and model test set to ensure the consistency of user behavior pattern distribution in each data subset; a pre-trained large language model for a general domain is obtained, and the parameters of its underlying Transformer block are frozen to retain general semantic understanding capabilities. Domain adaptive pre-training is performed on the transfer dataset, and the application calling domain-specific terms and behavior patterns are learned through the masked language modeling task to update the top-level parameters and generate a domain-adaptive basic model. 2. Steps for efficient parameter fine-tuning: A low-rank adapter module is injected into the domain adaptation base model to construct a parameter-efficient fine-tuning architecture. Using the model training set, the low-rank adapter module is trained with a layer-by-layer learning rate decay strategy. The bottom adapter is configured with a lower learning rate to stabilize general features, and the top adapter is configured with a higher learning rate to quickly adapt to the prediction task. A dynamic batch construction mechanism is introduced to adaptively adjust the batch size according to the sequence length, balancing training efficiency and memory usage. 3. Multi-task joint optimization steps: Design a multi-task learning framework to simultaneously optimize the application call intent classification task, call timing regression task, and call scheme generation task; construct a gradient coordination mechanism between tasks to monitor the convergence rate of the loss function of each task and automatically adjust the task weight coefficients to alleviate the negative transfer phenomenon; periodically evaluate the performance of each task through the model validation set, and trigger an early stopping mechanism when the validation set loss does not decrease for several consecutive epochs to save the current optimal checkpoints. 4. Steps to enhance robustness against adversity: Adversarial training is introduced during the fine-tuning process. Adversarial perturbation samples with fused semantic vector sequences are generated based on gradient information, and the model is trained to learn perturbation invariance features. A virtual adversarial loss function is constructed to constrain the model's prediction consistency on the original samples and adversarial samples, thereby improving the model's robustness to input noise and distribution shift. 5. Model Evaluation and Deployment Steps: The optimized large language model is tested on the model test set in multiple dimensions. The evaluation indicators include the accuracy of call intent prediction, the error of call timing prediction, the BLEU score of scheme generation, and the inference latency. When all indicators meet the preset application call prediction requirements, the optimized large language model is determined as the application call prediction model and exported as a lightweight inference format to adapt to the terminal deployment environment.

[0038] Further, in one embodiment, the application invocation method, wherein obtaining the current behavior data of the target user for the target application, inputting the current behavior data into the application invocation prediction model, and generating the invocation prediction result of the target application specifically includes: Obtain the call request for the target application input by the target user through the client, and perform format validation and integrity checks on the call request; When both format validation and integrity checks pass, the content of the call request is parsed, the current behavior data in the call request is obtained based on the parsing result, and the current behavior data is input into the application call prediction model to generate the call prediction result of the target application. If the format validation and / or integrity check fails, a prompt message is returned to the client.

[0039] In practice, the steps described above in this embodiment achieve high availability, data security, and response accuracy of the prediction service by constructing a collaborative processing mechanism for request verification, security filtering, and model inference, thereby achieving the technical effect of improving system robustness and user interaction reliability.

[0040] The specific implementation process of this embodiment is roughly as follows: 1. Real-time behavior stream capture and preprocessing steps: A lightweight behavior acquisition probe deployed on the client side captures the target user's current behavior data for the target application in real time, including touch screen event streams, sensor data changes, application lifecycle state transitions, and interface element interaction records. The current behavior data is cleaned and features extracted in real time, noise events are removed, timestamps are normalized, and redundant sequences are compressed to generate a standardized current behavior feature vector. 2. Request encapsulation and security verification steps: The standardized current behavior feature vector is encapsulated into a structured call request, with the user identity token, device fingerprint, and session identifier attached. The call request is then format-validated to verify whether the field types, lengths, and enumeration values ​​conform to the protocol specifications, and an integrity check is performed. Hash verification and digital signature verification are used to ensure that the data has not been tampered with. If both the format validation and integrity check pass, the subsequent parsing process is executed. If any check fails, a prompt message containing a specific error code is returned to the client and the processing is terminated. 3. Context-aware reasoning steps: The validated call request is parsed, and the current behavior feature vector and associated context information are extracted, including the current resource status of the device, network environment characteristics, and historical session cache. The current behavior feature vector and the context information are concatenated into a model input tensor, which is then input into the application call prediction model. The application call prediction model extracts behavior pattern features through an encoding layer, fuses context information through an attention mechanism, and outputs multi-dimensional raw prediction logits. 4. Prediction result decoding and confidence calibration steps: The original predicted logits are subjected to Softmax normalization and temperature scaling calibration to generate the probability distribution of each candidate calling scheme; the prediction confidence score and uncertainty estimate are calculated. When the confidence score exceeds the high confidence threshold, the calling prediction result is directly determined; when the confidence score is between the high and low thresholds, a multi-round inference fusion mechanism is triggered to aggregate the results of multiple forward propagation to reduce variance; when the confidence score is lower than the low confidence threshold, it is marked as a low confidence prediction and a manual review mark is attached. 5. Steps for structured output of results: The determined call prediction results are parsed into structured output, including the target application identifier of the predicted call, the suggested call time window, the confidence level, and the recommended call scheme code; the structured output is returned to drive the subsequent scheme selection and execution process.

[0041] Further, in one embodiment, the application invocation method, wherein the step of selecting a target invocation scheme adapted to the target application from a preset application invocation scheme library based on the invocation prediction result, and executing the target invocation scheme to realize the invocation of the target application, specifically includes: Based on the call prediction results, a multi-dimensional decision vector is constructed by extracting prediction confidence, resource sensitivity, and user preference entropy values. Based on the multi-dimensional decision vector, a target invocation scheme that is compatible with the target application is selected from a preset application invocation scheme library; The target invocation scheme is executed to invoke the target application.

[0042] Furthermore, the application invocation method, wherein executing the target invocation scheme to invoke the target application specifically includes: Analyze the target invocation scheme to obtain its resource requirement vector; Determine whether the resource requirement vector matches the client's resource configuration information; If the resource requirement vector matches the resource configuration information of the client, the target invocation scheme is executed in the client to invoke the target application.

[0043] Furthermore, in the application invocation method, the step of determining whether the resource requirement vector matches the client's resource configuration information further includes: If the resource requirement vector does not match the client's resource configuration information, then based on the multi-dimensional decision vector and the client's resource configuration information, a new target invocation scheme adapted to the target application is selected from the application invocation scheme library. The target invocation scheme obtained through re-filtering is executed in the client to invoke the target application.

[0044] In practical implementation, the above steps in this embodiment achieve efficient transformation of prediction results into executable strategies and flexible scheduling under system resource constraints by constructing a multi-dimensional decision-driven scheme selection and dynamic resource adaptation execution mechanism, thereby achieving the technical effect of improving the accuracy of invocation decisions and the robustness of the execution process.

[0045] The specific implementation process of this embodiment is roughly as follows: 1. Steps for constructing multi-dimensional decision vectors: Based on the call prediction results, a prediction confidence score, resource sensitivity index, and user preference entropy value are extracted to construct a multi-dimensional decision vector. The prediction confidence score reflects the model's certainty about the call intent, the resource sensitivity index characterizes the target application's resource dependence on CPU, memory, and network bandwidth, and the user preference entropy value quantifies the stability and predictability of the user's historical call behavior. The multi-dimensional decision vector is then normalized and dynamically weighted to generate standardized decision features. 2. Intelligent solution matching and filtering steps: Based on the standardized decision-making characteristics, a hierarchical filtering process is performed in the preset application invocation scheme library: first, the candidate scheme set is filtered according to the target application identifier; second, low-fit schemes are eliminated according to the prediction confidence threshold; third, the schemes are sorted according to the matching degree between resource sensitivity and current system load; and finally, the scheme with the best personalization is selected as the target invocation scheme based on the user preference entropy value. If the filtering result is empty, the scheme library expansion mechanism is triggered to generate an emergency invocation scheme based on the rule engine. 3. Steps for predicting resource availability: The target invocation scheme is parsed to obtain its resource demand vector, which includes a baseline resource threshold, an elastic resource limit, and a resource contention sensitivity weight. Real-time resource snapshots of the client are collected, including available memory capacity, CPU load rate, battery level, and network stability indicators, to construct resource configuration information. The probability that the resource demand vector and the resource configuration information can be satisfied is calculated to determine whether they meet the preset admission conditions. 4. Tiered Implementation and Dynamic Degradation Steps: If the probability of fulfillment exceeds the high confidence threshold, the resource requirement vector is determined to match the client's resource configuration information. The target invocation scheme is then executed in full on the client, the required resources are pre-allocated, and the application invocation process is initiated. If the probability of fulfillment is between the high and low thresholds, an elastic degradation mode is triggered. Based on the multi-dimensional decision vector, the resource compression ratio is recalculated, the preloaded data volume is adjusted, the animation frame rate is reduced, or the initialization of non-critical services is delayed. A degradation invocation scheme is generated and executed. If the probability of fulfillment is below the low confidence threshold, it is determined to not match the resource configuration information. Based on the preference entropy value dimension of the multi-dimensional decision vector, a historically successfully executed lightweight alternative scheme is retrieved from the application invocation scheme library, or the invocation parameters are dynamically reconstructed based on the upper limit constraint of the resource configuration information. A new, suitable target invocation scheme is obtained and executed. 5. Perform monitoring and anomaly rollback steps: Throughout the entire execution cycle of the call scheme, a resource consumption probe continuously monitors the deviation between actual resource consumption and prediction. When abnormal events such as memory leaks, thread deadlocks, or permission request failures are detected, a tiered rollback mechanism is triggered: for minor anomalies, execution parameters are dynamically adjusted; for moderate anomalies, the process switches to a backup execution branch; and for severe anomalies, the current state is saved and the call process is gracefully terminated, providing feedback to the user on the specific reasons for the blockage and suggested actions. The execution trajectory data and abnormal event records are fed back to the application call prediction model, supporting subsequent online fine-tuning of the model and optimization and updates to the scheme library.

[0046] As can be seen from the above method embodiments, the application invocation method provided by the present invention includes: collecting historical behavior data of users towards the application and historical invocation data of the application; fusing the historical behavior data and the historical invocation data to generate comprehensive training data; performing transfer learning and fine-tuning training on a pre-trained large language model based on the comprehensive training data to generate an application invocation prediction model; obtaining current behavior data of the target user towards the target application; inputting the current behavior data into the application invocation prediction model to generate an invocation prediction result for the target application; and, based on the invocation prediction result, selecting a target invocation scheme adapted to the target application from a preset application invocation scheme library and executing the target invocation scheme to realize the invocation of the target application. The present invention can adaptively select the optimal invocation timing and execution strategy according to the user's personalized characteristics, thereby reducing startup latency and resource waste while ensuring the accuracy of invocation decisions and the robustness of the execution process. In other words, the present invention can achieve flexible and accurate application invocation.

[0047] It should be understood that although this application provides the method operation steps as described in the embodiments or flowcharts, conventional or non-inventive labor may include more or fewer operation steps, and these operation steps are not necessarily executed sequentially according to the order of the embodiments or flowcharts. The order of steps listed in the embodiments or flowcharts is merely one way of executing many steps and does not represent the only execution order. It should be noted that there is no necessary sequential order between the above steps. Those skilled in the art can understand from the description of the embodiments of the present invention that the above steps may have different execution orders in different embodiments, that is, they may be executed in parallel or in exchange, etc. Moreover, at least some steps in the embodiments or flowcharts may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but may be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but may be executed in turn, alternately, or synchronously with other steps or at least a part of the sub-steps or stages of other steps.

[0048] Based on the above method embodiments, please refer to Figure 3 Another embodiment of the present invention also provides an application invocation device, wherein the device includes: Collection module 11 is used to collect historical behavior data of users for the application and historical call data of the application, and to perform fusion processing on the historical behavior data and the historical call data to generate comprehensive training data; Model generation module 12 is used to perform transfer learning and fine-tuning training on the pre-trained large language model based on the comprehensive training data, and generate an application call prediction model. The result generation module 13 is used to obtain the current behavior data of the target user for the target application, input the current behavior data into the application call prediction model, and generate the call prediction result of the target application. The filtering module 14 is used to filter out a target calling scheme that is compatible with the target application from a preset application calling scheme library based on the calling prediction result, and execute the target calling scheme to realize the calling of the target application.

[0049] Furthermore, in one embodiment, the application invocation device, wherein collecting historical user behavior data and historical application invocation data, and fusing the historical behavior data and the historical invocation data to generate comprehensive training data, specifically includes: Collect historical user behavior data and historical call data of the application from multiple heterogeneous data sources; A dual-channel data encoder is constructed to perform user intent semantic encoding on the historical behavior data and system state semantic encoding on the historical call data, respectively. The temporal correlation between the encoded historical behavior data and the historical call data is aligned through a cross-modal attention mechanism to generate a fused semantic vector sequence. Based on the fused semantic vector sequence, comprehensive training data is constructed.

[0050] Furthermore, in one embodiment, the application invocation device, wherein the step of performing transfer learning and fine-tuning training on the pre-trained large language model based on the comprehensive training data to generate an application invocation prediction model specifically includes: According to the preset training data partitioning strategy, the comprehensive training data is divided into transfer dataset, model training set, model validation set and model test set; The pre-trained large language model is transferred to the transfer dataset, and the transferred language model is fine-tuned using the model training set. The fine-tuned large language model is evaluated using the model validation set, and the large language model is optimized based on the evaluation results. The optimized large language model is tested on the model test set. If the test results meet the application call prediction requirements, the optimized large language model is determined as the application call prediction model.

[0051] Further, in one embodiment, the application invocation apparatus, wherein acquiring the current behavior data of the target user towards the target application, inputting the current behavior data into the application invocation prediction model, and generating the invocation prediction result of the target application specifically includes: Obtain the call request for the target application input by the target user through the client, and perform format validation and integrity checks on the call request; When both format validation and integrity checks pass, the content of the call request is parsed, the current behavior data in the call request is obtained based on the parsing result, and the current behavior data is input into the application call prediction model to generate the call prediction result of the target application. If the format validation and / or integrity check fails, a prompt message is returned to the client.

[0052] Further, in one embodiment, the application invocation device, wherein the step of selecting a target invocation scheme adapted to the target application from a preset application invocation scheme library based on the invocation prediction result, and executing the target invocation scheme to realize the invocation of the target application, specifically includes: Based on the call prediction results, a multi-dimensional decision vector is constructed by extracting prediction confidence, resource sensitivity, and user preference entropy values. Based on the multi-dimensional decision vector, a target invocation scheme that is compatible with the target application is selected from a preset application invocation scheme library; The target invocation scheme is executed to invoke the target application.

[0053] Furthermore, in the application invocation device, the step of executing the target invocation scheme to invoke the target application specifically includes: Analyze the target invocation scheme to obtain its resource requirement vector; Determine whether the resource requirement vector matches the client's resource configuration information; If the resource requirement vector matches the resource configuration information of the client, the target invocation scheme is executed in the client to invoke the target application.

[0054] Furthermore, in the application invocation device, the step of determining whether the resource demand vector matches the client's resource configuration information further includes: If the resource requirement vector does not match the client's resource configuration information, then based on the multi-dimensional decision vector and the client's resource configuration information, a new target invocation scheme adapted to the target application is selected from the application invocation scheme library. The target invocation scheme obtained through re-filtering is executed in the client to invoke the target application.

[0055] It should be noted that, in the device embodiments of the present invention, the information interaction and execution process between the above modules are based on the same concept as in the method embodiments of the present invention. For details on their specific functions and the resulting technical effects, please refer to the aforementioned method embodiments section, which will not be repeated here.

[0056] Based on the above method embodiments, another embodiment of the present invention also provides a computer device, which can be a server, and its internal structure diagram can be as follows. Figure 4 As shown. The computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements the functions or steps of the application call method server side as described in any of the above method embodiments.

[0057] Based on the above method embodiments, another embodiment of the present invention also provides a computer device, which can be a client, and its internal structure diagram can be as follows. Figure 5 As shown, the computer device includes a processor, memory, network interface, display screen, and input device connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface is used to communicate with external terminals via a network connection. When executed by the processor, the computer program implements the functions or steps of the application call method client side as described in any of the above method embodiments.

[0058] Those skilled in the art will understand that Figure 4 and Figure 5 The structural schematic diagram shown is only a schematic diagram of a part of the structure related to the present invention and does not constitute a limitation on the computer device on which the present invention is applied. The specific computer device may include more components than shown in the figure, or combine certain components, or have different component arrangements.

[0059] The processor referred to herein can be a CPU, but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

[0060] The memory includes readable storage media, internal memory, etc., where internal memory can be the RAM of a computer device. Internal memory provides an environment for the operation of the operating system and computer-readable instructions stored in the readable storage media. The readable storage media can be the hard drive of the computer device, or in other embodiments, it can be an external storage device of the computer device, such as a plug-in hard drive, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card. Furthermore, the memory can include both internal storage units and external storage devices of the computer device. The memory is used to store the operating system, applications, bootloader, data, and other programs, such as program code for computer programs. The memory can also be used to temporarily store data that has been output or will be output.

[0061] Based on the above method embodiments, another embodiment of the present invention provides a computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the application program invocation method as described in any of the above method embodiments. The computer-readable storage medium may be non-volatile or volatile.

[0062] It should be noted that the functions or steps that can be achieved by the computer-readable storage medium or computer device, and the technical effects brought about by the functions / steps, can be referred to the relevant descriptions in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0063] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc. The disclosed memory components or memories of the operating environment described herein are intended to include one or more of these and / or any other suitable types of memory.

[0064] Those skilled in the art will understand that, for the sake of convenience and brevity, the embodiments of the device of the present invention are only illustrated by the division of the above-mentioned functional units and modules. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of the present invention. The specific working process of the units and modules in the above device can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here. If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium.

[0065] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0066] In the embodiments provided by this invention, it should be understood that the disclosed apparatus / computer devices and methods can be implemented in other ways. For example, the apparatus / computer device embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0067] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0068] It should be noted that any software tools or components not belonging to this company appearing in the embodiments of this application are merely illustrative examples and do not represent actual use. All user personal information involved in the embodiments of this application has been authorized (with knowledge and consent) by the relevant parties or has been fully authorized by all parties, and the executing entity may obtain it through various legal and compliant means. The collection, storage, use, processing, transmission, provision, and disclosure of the information, data, and signals involved all comply with relevant laws and regulations and do not violate public order and good morals.

[0069] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. An application invocation method, characterized in that, include: Collect historical user behavior data and historical call data of the application, and fuse the historical behavior data and historical call data to generate comprehensive training data; Based on the comprehensive training data, the pre-trained large language model is subjected to transfer learning and fine-tuning training to generate an application call prediction model. Obtain the current behavior data of the target user for the target application, input the current behavior data into the application call prediction model, and generate the call prediction result of the target application; Based on the invocation prediction results, a target invocation scheme that is compatible with the target application is selected from a preset application invocation scheme library, and the target invocation scheme is executed to realize the invocation of the target application.

2. The application invocation method according to claim 1, characterized in that, The process of collecting historical user behavior data and historical application call data, fusing the historical behavior data and historical call data to generate comprehensive training data includes: Collect historical user behavior data and historical call data of the application from multiple heterogeneous data sources; A dual-channel data encoder is constructed to perform user intent semantic encoding on the historical behavior data and system state semantic encoding on the historical call data, respectively. The temporal correlation between the encoded historical behavior data and the historical call data is aligned through a cross-modal attention mechanism to generate a fused semantic vector sequence. Based on the fused semantic vector sequence, comprehensive training data is constructed.

3. The application invocation method according to claim 1, characterized in that, The step of performing transfer learning and fine-tuning training on the pre-trained large language model based on the comprehensive training data to generate an application invocation prediction model includes: According to the preset training data partitioning strategy, the comprehensive training data is divided into transfer dataset, model training set, model validation set and model test set; The pre-trained large language model is transferred to the transfer dataset, and the transferred language model is fine-tuned using the model training set. The fine-tuned large language model is evaluated using the model validation set, and the large language model is optimized based on the evaluation results. The optimized large language model is tested on the model test set. If the test results meet the application call prediction requirements, the optimized large language model is determined as the application call prediction model.

4. The application invocation method according to claim 1, characterized in that, The step of obtaining the target user's current behavior data for the target application, inputting the current behavior data into the application call prediction model, and generating the call prediction result for the target application includes: Obtain the call request for the target application input by the target user through the client, and perform format validation and integrity checks on the call request; When both format validation and integrity checks pass, the content of the call request is parsed, the current behavior data in the call request is obtained based on the parsing result, and the current behavior data is input into the application call prediction model to generate the call prediction result of the target application. If the format validation and / or integrity check fails, a prompt message is returned to the client.

5. The application invocation method according to claim 1, characterized in that, The step of selecting a target invocation scheme adapted to the target application from a preset application invocation scheme library based on the invocation prediction result, and executing the target invocation scheme to invoke the target application, includes: Based on the call prediction results, a multi-dimensional decision vector is constructed by extracting prediction confidence, resource sensitivity, and user preference entropy values. Based on the multi-dimensional decision vector, a target invocation scheme that is compatible with the target application is selected from a preset application invocation scheme library; The target invocation scheme is executed to invoke the target application.

6. The application invocation method according to claim 5, characterized in that, The execution of the target invocation scheme to invoke the target application includes: Analyze the target invocation scheme to obtain its resource requirement vector; Determine whether the resource requirement vector matches the client's resource configuration information; If the resource requirement vector matches the resource configuration information of the client, the target invocation scheme is executed in the client to invoke the target application.

7. The application invocation method according to claim 6, characterized in that, The step of determining whether the resource demand vector matches the client's resource configuration information further includes: If the resource requirement vector does not match the client's resource configuration information, then based on the multi-dimensional decision vector and the client's resource configuration information, a new target invocation scheme adapted to the target application is selected from the application invocation scheme library. The target invocation scheme obtained through re-filtering is executed in the client to invoke the target application.

8. An application invocation device, characterized in that, include: The collection module is used to collect historical user behavior data and historical call data of the application, and to perform fusion processing on the historical behavior data and the historical call data to generate comprehensive training data. The model generation module is used to perform transfer learning and fine-tuning training on the pre-trained large language model based on the comprehensive training data, and generate an application call prediction model. The result generation module is used to obtain the current behavior data of the target user for the target application, input the current behavior data into the application call prediction model, and generate the call prediction result of the target application. The filtering module is used to filter out a target calling scheme that is compatible with the target application from a preset application calling scheme library based on the calling prediction result, and execute the target calling scheme to realize the calling of the target application.

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

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

Images