Intelligent reminding method, device and equipment based on large model

Abstract

The invention discloses an intelligent reminding method, device and equipment based on a large model, and relates to the field of natural language process.The method comprises the steps that first dialogue content input by a user is recognized through an exclusive model agent, and a first condition execution rule corresponding to a reminding task is obtained, the exclusive model agent is obtained by performing parameter fine tuning on a pre-trained large model; converting a first condition execution rule corresponding to the reminding task into a corresponding target CRON expression; establishing connection channels with a plurality of user terminals of the user; and under the condition that the triggering opportunity of the target CRON expression is met, based on a pre-configured execution mode of a reminding task and the target CRON expression, actively triggering the reminding task, and sequentially triggering the reminding task according to a multi-stage reminding matrix, so as to actively send reminding information to a user through a connection channel of a user terminal corresponding to each stage of reminding. According to the method, the reminding matrix is constructed, so that the cross-device cooperation capability is improved.

Description

Technical Field

[0001] This application relates to the field of natural language processing technology, and in particular, to an intelligent reminder method, device, and equipment based on a large model. Background Art

[0002] In the related intelligent systems, there are problems such as shallow semantic understanding, mechanized task management, and insufficient multi-device collaboration. Summary of the Invention

[0003] The purpose of this application is to provide an intelligent reminder method, device, equipment, medium, and product based on a large model.

[0004] To achieve the above purpose, this application provides the following solutions: In the first aspect, this application provides an intelligent reminder method based on a large model, including: Identifying the first conditional execution rule corresponding to the reminder task through a dedicated model agent, where the dedicated model agent is obtained by fine-tuning the parameters of a pre-trained large model; Converting the first conditional execution rule corresponding to the reminder task into a corresponding target CRON expression; Establishing a connection channel with multiple user terminals of the user; When the triggering opportunity of the target CRON expression is met, based on the pre-configured execution mode of the reminder task and the target CRON expression, actively triggering the reminder task, and the reminder task is triggered sequentially according to a multi-level reminder matrix, so as to actively send reminder information to the user through the connection channels of the user terminals corresponding to each level of reminder.

[0005] In the second aspect, this application provides an intelligent reminder device based on a large model, including: An identification module, configured to identify the first conditional execution rule corresponding to the reminder task through a dedicated model agent, where the dedicated model agent is obtained by fine-tuning the parameters of a pre-trained large model; A conversion module, configured to convert the first conditional execution rule corresponding to the reminder task into a corresponding target CRON expression; An establishment module, configured to establish a connection channel with multiple user terminals of the user; An execution module, configured to, when the triggering opportunity of the target CRON expression is met, actively trigger the reminder task based on the pre-configured execution mode of the reminder task and the target CRON expression, and the reminder task is triggered sequentially according to a multi-level reminder matrix, so as to actively send reminder information to the user through the connection channels of the user terminals corresponding to each level of reminder.

[0006] In a third aspect, the present application provides a computer device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor executes the computer program to implement the steps of the large model-based intelligent reminder method described in any one of the above.

[0007] In a fourth aspect, the present application provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the large model-based intelligent reminder method described in any one of the above.

[0008] In a fifth aspect, the present application provides a computer program product, including a computer program, and when the computer program is executed by a processor, it implements the steps of the large model-based intelligent reminder method described in any one of the above.

[0009] According to the specific embodiments provided by the present application, the following technical effects are disclosed in the present application: The present application provides a large model-based intelligent reminder method, device, equipment, medium and product. By identifying the first conversation content of the user, the first conditional execution rule corresponding to the reminder task is obtained, and the first conditional execution rule is converted into a target CRON expression, so that the user intention can be identified more accurately; and when the triggering opportunity of the target CRON expression is met, the reminder task is actively triggered to actively send a reminder message to the user, realizing the automation and initiative of reminder task management, without manual configuration; by constructing a cross-device reminder matrix, cross-device real-time synchronous reminder can be realized, improving the cross-device collaboration ability and the accuracy, efficiency, real-time performance and reliability of the reminder. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required to be used in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application, and those of ordinary skill in the art can obtain other drawings based on these drawings without creative efforts.

[0011] Figure 1 It is a schematic flow chart of a large model-based intelligent reminder method provided by an embodiment of the present application; Figure 2 It is a schematic flow chart of a large model-based intelligent reminder method provided by another embodiment of the present application; Figure 3 It is a schematic flow chart of a large model-based intelligent reminder method provided by another embodiment of the present application; Figure 4A flowchart of an intelligent reminder method based on a large model provided by another embodiment of the present application; Figure 5 A flowchart of an intelligent reminder method based on a large model provided by another embodiment of the present application; Figure 6 A flowchart of an intelligent reminder method based on a large model provided by another embodiment of the present application; Figure 7 A flowchart of an intelligent reminder method based on a large model provided by another embodiment of the present application; Figure 8 A flowchart of an intelligent reminder method based on a large model provided by another embodiment of the present application; Figure 9 A schematic diagram of functional modules of an intelligent reminder device based on a large model provided by an embodiment of the present application; Figure 10 A schematic diagram of the structure of a computer device provided by an embodiment of the present application. Detailed implementation manners

[0012] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0013] To make the above objects, features, and advantages of the present application more obvious and understandable, the present application will be further described in detail below in conjunction with the accompanying drawings and specific implementation manners.

[0014] In an exemplary embodiment, as Figure 1 shown, an intelligent reminder method based on a large model is provided. This method is executed by a computer device, and specifically, it can be executed independently by a computer device such as a terminal or a server, or jointly executed by a terminal and a server. In the embodiments of the present application, the following steps 102 to 108 are included. Among them: Step 102: Identify the first conversation content input by the user through the exclusive model agent to obtain the first conditional execution rule corresponding to the reminder task. The exclusive model agent is obtained by fine-tuning the parameters of the pre-trained large model; Among them, the Q&A pair data can be first trained into the base of the large model in a fine-tuning manner to obtain a trained exclusive model, and then an exclusive model agent is generated based on the trained exclusive model. The large model refers to a deep learning model with a huge number of parameters, a large scale of training data, and high consumption of computing resources. It is usually based on architectures such as Transformer and can handle complex natural language understanding (NLU), natural language generation (NLG), multi-modal tasks (such as image-text generation), etc. The exclusive model agent can be an agent that can perceive the environment and take actions to achieve specific goals generated based on the trained exclusive model. It has autonomy, adaptability, and interaction capabilities. The agent perceives changes in the environment (such as through sensors or data input), makes judgments and decisions based on the knowledge and algorithms it has learned, and then executes actions to affect the environment or achieve a predetermined goal. The exclusive model agent can dynamically monitor the environment, call multi-modal tools and trigger tasks, adjust strategies according to real-time feedback. For example, when the user modifies the reminder time or the user's location changes, the conditional execution rules generated by the exclusive model agent will also change accordingly.

[0015] In the user's real-time conversation, the exclusive model agent real-time identifies the text content input by the user and extracts the reminder intention and conditional execution rules. The reminder intention includes TRUE (true) and FALSE (false). When the reminder intention is TRUE, it means that the user needs to be reminded. When the reminder intention is FALSE, it means that the user does not need to be reminded.

[0016] The first conversation content can be the original intention expressed by the user in real time. Exemplarily, the first conversation content can be "Remind me to drink water in three quarters of an hour.", "If it rains when school is over, tell mom to pick me up.", "Alarm when the temperature exceeds 50 degrees." The above-mentioned conditional execution rule is also known as the IF-THEN rule or the trigger-action rule. It is an automated execution mechanism based on logical judgment. Its core structure is: IF (condition is met) → THEN (execute action), that is, when a specific condition is detected to be true, a preset action or process is automatically triggered. Each of the conditional execution rules can correspond to a reminder task. Exemplarily, when the first conversation content input by the user is "If it rains when school is over, tell mom to pick me up.", the corresponding conditional execution rule output by the dedicated model agent can be "IF school_time == true AND weather == "rainy" THEN send_messag(mom, "Please pick up the child")". When the first conversation content input by the user is "Alarm when the temperature exceeds 50 degrees.", the corresponding conditional execution rule output by the dedicated model agent can be "IF sensor_temp>50 THEN trigger_alarm("High temperature warning").

[0017] Step 104: Convert the first conditional execution rule corresponding to the reminder task into a corresponding target CRON expression; Among them, the CRON expression is a string format used to define the trigger time of periodic tasks and is applied to system timing tasks (such as backups, reminders, data synchronization). It is like a "time password" that describes complex time rules with short codes. A standard CRON expression contains 5 to 7 fields (different systems may have extensions). Each field represents a time unit and is separated by a space. The following takes a 5-field CRON expression as an example for illustration: 1 2 3 4 5 │││││ * * * * * Among them, the first field represents minutes (0-59), the second field represents hours (0-23), the third field represents the date (1-31), the fourth field represents the month (1-12 or JAN-DEC), and the fifth field represents the week (0-7, both 0 and 7 represent Sunday, or SUN-SAT). The CRON expression includes basic time parameters, date modifiers, and special markers. The basic time parameters are numerical or alphabetical parameters for each field. For example, 12 in the second field represents 12 o'clock, and JAN in the fourth field represents January; the date modifiers can include asterisk (*), question mark (?), minus sign (-), comma (,), slash ( / ), L, W, #, LW combination, C. The symbols have different meanings when they appear in different fields. For example, the # character usually only appears in the week field and represents a certain working day of the month. 5#3 represents the third Friday of the month; the special markers include holiday markers, make-up holiday markers, and season markers, etc.

[0018] Step 106: Establish a connection channel between the server and multiple user terminals of the user; The connection channel refers to a two-way communication link established between a server or other terminal device and multiple terminal devices of the same user. Each user terminal has an independent connection channel, and the user terminal can be devices at different levels.

[0019] Step 108: When the triggering opportunity that meets the target CRON expression occurs, based on the execution mode of the reminder task and the target CRON expression configured in advance, actively trigger the reminder task. The reminder task is triggered sequentially according to a multi-level reminder matrix, so as to actively send reminder information to the user through the connection channels of the user terminals corresponding to each level of reminder.

[0020] Among them, the triggering opportunity may include triggering time, triggering location, device status, etc. The triggering time may be a preset time, the triggering location may include home, company address, etc., and the device status may be device battery level, device online or offline, etc.; the confirmation of the triggering opportunity can specifically be based on dynamic adjustment of tasks such as geographical location and device status (mobile phone battery level) (such as "trigger a reminder when arriving at a specified location"). Exemplarily, if the current time is the preset time, the user location is "company", and the mobile phone battery level is greater than 20%, then trigger the reminder.

[0021] Flexible task scheduling and lifecycle management can be achieved by configuring the execution mode of the reminder task; the reminder task can be dynamically orchestrated. By configuring the set execution mode, the execution mode of the reminder task can include immediate execution and automatic destruction (the corresponding first conversation content can be "Turn off the lights in the living room immediately now"), maintaining the task until the exit condition is met (the corresponding first conversation content can be "Check the temperature every 10 minutes until it is lower than 25°C"), cumulative triggering, triggering after reaching the threshold number of times (the corresponding first conversation content can be "Alarm after detecting 3 abnormal vibrations or remind me until I have exercised three times").

[0022] The multi-level reminder matrix realizes different degrees of reminder effects through the coordinated response of different device levels. The differences between different levels of reminders can be reflected in four dimensions: reach intensity, sensory intrusion, coverage reliability, and response urgency.

[0023] Implementing the above steps 102 to 108 to identify the user's first conversation content, obtaining the first conditional execution rule corresponding to the reminder task, and converting the first conditional execution rule into a target CRON expression, so that the user's intention can be identified more accurately; and when the triggering time of the target CRON expression is met, actively trigger the reminder task to actively send a reminder message to the user, realizing the automation and initiative of reminder task management, without manual configuration; by constructing a cross-device reminder matrix, cross-device real-time synchronous reminder can be realized, improving the cross-device collaboration ability, and improving the accuracy, efficiency, real-time performance and reliability of the reminder.

[0024] In an exemplary embodiment of the present application, step 106 "establishing a connection channel between the user's multiple user terminals" includes: establishing a low-latency link for WebSocket duplex communication between the user's multiple user terminals, and the WebSocket duplex communication includes: Step S1: Establish a connection. Through an HTTP Upgrade handshake (including the Upgrade: websocket header), the HTTP protocol is switched to the WebSocket protocol, and binary frame format (including mask processing / fragment control) is used for data transmission. The frame header is only 2-14 bytes, significantly reducing the transmission overhead. The TCP long connection is kept alive to avoid repeated handshakes.

[0025] Step S2: Establish a latency optimization strategy. Use CompositeByteBuf to merge multiple buffers, set the number of WorkerGroup threads = the number of CPU (Central Processing Unit) cores × 2 to avoid context switching, enable the permessage-deflate extension for protocol compression (reducing bandwidth occupancy), and aggregate small data packets in an accumulation window (50ms / 32KB) for batch merging (effectively reducing the number of small packets in the network and reducing the transmission overhead).

[0026] Among them, 50ms / 32KB represents the time window and the space window. Small data packets within 50ms are sent with a delay, waiting for merging, and are immediately sent when reaching 32KB.

[0027] In the embodiments of the present application, by using CompositeByteBuf to merge multiple buffers, the number of memory copies and system call overhead can be reduced. By setting the number of WorkerGroup threads = the number of CPU cores × 2, the performance of multi-core CPUs can be fully utilized, and the context switching loss caused by too many threads can be avoided. By enabling the permessage-deflate extension for protocol compression, the bandwidth occupancy can be reduced. By aggregating small data packets through the cumulative window for batch merging, the number of small packets in the network can be effectively reduced, and the transmission overhead can be reduced.

[0028] The embodiments of the present application can achieve compressing the cross-device synchronization delay to 300 ms, and the arrival rate of critical scenario reminders is 99.99%.

[0029] In another exemplary embodiment of the present application, as Figure 2 shown, step 104 can be replaced by the following steps 1041 to 1043: Step 1041: Use a parameter parser to parse the structure in the first condition execution rule corresponding to the reminder task to obtain the constituent structure; Among them, using a parameter parser to parse a structure such as "every working day [time] [exclude holidays]", the structure can include regular conditions, actions, exception conditions, etc. The regular conditions can include time conditions (minutes, hours, days, months, weeks, etc.), location conditions, device status conditions, weather conditions, etc. The actions can include closing, opening, checking, alarming, etc. The exception conditions can include time exceptions (exclude specific dates or time periods), status exceptions (skip when the device is already on, or do not execute when the user is at home), or environmental exceptions (cancel watering if it is raining), etc.

[0030] Step 1042: Based on a pre-established rule mapping table, assemble the constituent structure into a first CRON expression through a hierarchical replacement strategy, where the rule mapping table is used to associate the first condition execution rule and the CRON expression; Among them, the hierarchical replacement strategy can be a regularized method for processing different levels of parameters in stages and according to priorities. The hierarchical replacement strategy can include basic time parameter filling, date modifier processing, and special marker injection. The following table 1 is the rule mapping table. Based on the rule mapping table, the constituent structure in the first condition rule can be assembled into a first CRON expression through the hierarchical replacement strategy.

[0031] Table 1 Rule mapping table

[0032] Step 1043: Standardize the heterogeneous event source data in the first CRON expression to obtain the target CRON expression.

[0033] Among them, heterogeneous event source data can be a collection of event data with different protocols, formats, structures or semantics collected from different devices. These data are naturally heterogeneous due to differences in sources and need to be standardized before they can be analyzed and responded to by a unified system. Standardization processing can be to convert heterogeneous event source data into a unified format; for example, the collected MQTT (Message Queuing Telemetry Transport) long-connected IoT (Internet of Things) devices, HTTP Websocket mobile APPs, and third-party systems’ heterogeneous event source data can be converted into a unified format.

[0034] In the embodiments of the present application, accurate conversion from business rules to executable tasks is achieved through rule parsing, hierarchical replacement and standardized processing.

[0035] In another exemplary embodiment of the present application, the user terminal includes at least one intelligent terminal layer device, a near field device layer device and a remote service layer device, such as Figure 3 As shown, in step 108, "when the triggering timing of the target CRON expression is met, based on the pre-configured execution mode of the reminder task and the target CRON expression, executing the reminder task" can be replaced by the following steps 1081 to 1083: Based on the pre-configured execution mode of the reminder task, the following three phases are performed in sequence: Step 1081: In the first stage, at least one of a pop-up window reminder, a vibration reminder, and a light-emitting diode LED reminder is executed by at least one of the smart terminal layer devices; Among them, the smart terminal layer device can be a user's personal device such as a mobile phone and / or a watch; the first stage can be within 300 milliseconds (unit: ms), and the first level reminder can be triggered within 300ms, and the mobile phone and / or watch will pop up a reminder (for example, it can be "It's time to drink water"), a vibration reminder, and an LED reminder.

[0036] Step 1082: In the second stage, at least one of a voice broadcast reminder and a screen flashing reminder is performed by at least one of the near field device layer devices; Among them, the near-field device layer device can be a user's peripheral device such as a smart speaker and a car system; the second stage can be within 5 seconds (unit: s), and the second-level reminder can be triggered within 5 seconds, the smart speaker will give a voice broadcast reminder, and the screen of the car system will flash a screen reminder.

[0037] Step 1083: In the third stage, at least one of automatic dialing reminder and email follow-up reminder is performed through at least one of the remote service layer devices.

[0038] Among them, the remote service layer device can be an outbound telephone call system or an email system; the third stage can be 1 minute, and the third level reminder can be triggered within 1 minute. The outbound telephone call system performs automatic dialing reminder, and the email system performs email follow-up reminder.

[0039] In the embodiments of the present application, efficient implementation of reminder tasks can be achieved through phased, graded, and multi-modal reminder methods, and reliable execution of reminder tasks can be achieved through multi-device collaboration.

[0040] In another exemplary embodiment of the present application, Figure 4 As shown, the intelligent reminder method further includes steps 1011 to 1016: Step 1011: Select the second conversation content in the domain data conversation record, time series event table, and log; Among them, the domain data conversation record can be a children's watch question and answer record (for example, it can be "wake me up at 3:20 pm"); the time series event table can be an event log that records timestamps (such as historical reminder records); the log can be a system operation log, which may be used to verify conditional execution rules; the second conversation content can be the user's historical question and answer record.

[0041] Step 1012: generating a second conditional execution rule based on the spoken speech time in the second dialogue content; The spoken time may refer to a non-standard time description expressed by a user through natural language, such as "quarters of an hour later" or "tomorrow morning".

[0042] Step 1013: Execute a rule based on the second conversation content and the second condition to generate question-answer pair data; The question-answer pair data may be a data pair consisting of a question and an answer generated by binding the user's second conversation content (i.e., question) with the generated second conditional execution rule (i.e., answer); illustratively, the question-answer pair data may include: Question: "Remind me to drink water after 35 minutes" Answer: IF time==15:30 THEN alert("drink water") It should be noted that the intelligent reminder method of the embodiment of the present application supports cross-domain adaptation, pairs the time expression (question) in the user's original conversation with the automatically generated IF-THEN rule (answer) to generate question-answer pair data, which can, for example, be migrated from the field of children's watches to industrial operation and maintenance, smart homes, medical care and other fields.

[0043] Step 1014: Train the Q&A pair data into the large model base in a fine-tuning manner to obtain an exclusive model; Among them, the large model base can be fine-tuned with the Q&A pair data to obtain an exclusive model; Step 1015: Set up an agent in the form of a Prompt prompt based on the exclusive model to obtain the exclusive model agent; Among them, Prompt (prompt) is the input instruction or question provided by the user to the model, used to guide the model to generate a specific type of response. Its core role is to delimit the task scope for the model, clarify the output format or inject domain knowledge.

[0044] Step 1016: Associate the exclusive model agent with multi-modal perception tools.

[0045] Among them, multi-modal data includes data such as voice, text, images, and videos. Multi-modal perception tools can be tools that can process one or more modalities of data. The multi-modal perception tools include speech recognition tools for processing voice data, vision APIs (Application Programming Interfaces) that can process image data, pop-up notification tools for processing text data and image data, and weather APIs that can process voice data, text data, and image data.

[0046] The association of multi-modal tools needs to integrate speech recognition and speech synthesis, and can standardize the interaction through REST APIs. After the data returned by the multi-modal tools is secondarily processed by the exclusive model agent, multi-modal responses can be generated according to a preset template to ensure end-to-end controllability and user experience consistency.

[0047] In this application, through domain data fusion, rule-guided data generation, model fine-tuning, agent construction, and multi-modal expansion, the development of highly adaptable and implementable domain agents is realized.

[0048] The embodiments of this application solve the limitations of related systems in complex intent recognition, task management, and cross-device collaboration through deep semantic parsing, dynamic task engines, and multi-modal collaboration technologies. The core of the embodiments of this application lies in improving the semantic understanding ability. The parsing accuracy of complex expressions (such as "two hours after taking medicine") is significantly improved, dynamic logic is quickly generated, conditional statements are converted with one key, and the setting time can be reduced by 80%. It realizes the dynamic generation and adjustment of tasks, and constructs a low-latency cross-device reminder matrix. Efficient collaboration between multiple devices, with a synchronous response at the 300ms level, significantly improves the accuracy, efficiency, and real-time performance of the reminder system, and is applicable to various scenarios such as medical monitoring and business management.

[0049] In another exemplary embodiment of the present application, as Figure 5 shown, step 1012 can be replaced by the following steps 10121 to 10127: Step 10121: Convert the colloquial time in the second conversation content into absolute time; Among them, absolute time is usually the time expressed in specific numbers or standard time units. Exemplarily, if the colloquial time is "after three quarters", and the current time is 14:45, then the absolute time corresponding to "after three quarters" is 15:30.

[0050] Step 10122: Identify the conditional keywords in the second conversation content; Among them, the "conditional keywords" include "if", "when...", etc.; the conditional keywords in the input text can be parsed using a natural language processing model (Natural Language Processing, NLP).

[0051] Step 10123: Based on the absolute time and the conditional keywords, use a bidirectional attention mechanism to identify the entity associations and temporal logics in the second conversation content to obtain the semantic roles in the second conversation content; Among them, the bidirectional attention mechanism (Bidirectional Attention Mechanism) is a key attention mechanism design in natural language processing. By simultaneously considering the forward (from left to right) and backward (from right to left) context information, it can more comprehensively capture the dependencies between elements in the sequence. Identifying the entity associations in the second conversation content means identifying the hidden relationships between different entities (people, objects, events) in the text and establishing logical connections; identifying the temporal logic means analyzing the time order or dependencies between events to ensure that actions are triggered at the correct time points; the bidirectional attention mechanism can be used to analyze the absolute time and the conditional keywords to identify the entity associations and temporal logics in the second conversation content; semantic roles include conditions and actions. Exemplarily, if the second conversation content is "turn off all lights after 10 pm", the identified condition in the second conversation content is "22 o'clock", and the action is "turn off all lights".

[0052] Step 10124: Based on the semantic roles, construct a structured logical framework for conditions and actions; Among them, subsequently, based on the semantic role annotation, a logical framework for conditions - actions is constructed (condition "22 o'clock", action "turn off all lights"), and fuzzy expressions (such as "the value is too high") can also be mapped to precise threshold comparison expressions (for example, it can be "IF temperature > 30°C THEN...").

[0053] Step 10125: Generate the third conditional execution rule based on the structured logic framework and the rule template engine; Among them, the structured logic can be converted into executable code (i.e., the third conditional execution rule) through the rule template engine.

[0054] Step 10126: Optimize the rule weights and conflict resolution strategies in the third conditional execution rule by using a genetic algorithm to obtain the fourth conditional execution rule; Exemplarily, the third conditional execution rule may include Rule A and Rule B. Rule A: "Turn on the lights when arriving home" (weight 0.6), Rule B: "Turn off all lights after 10 pm" (weight 0.8). The conflict is that if you arrive home after 10 pm, should you turn on the lights or turn them off? The genetic algorithm can use historical data to test the effects of each set of weights, retain the combinations with good effects, eliminate the poor ones, generate a new generation of weights, and finally obtain the optimal weights. For example, the weight of Rule B can be increased to 0.9 and given priority for execution.

[0055] Step 10127: Use test cases to perform boundary verification on the fourth conditional execution rule. After the verification passes, determine the fourth conditional execution rule as the second conditional execution rule.

[0056] Among them, extreme and abnormal examples can be used to test the fourth conditional execution rule, so that the generated fourth conditional execution rule can meet the system execution requirements while retaining semantic accuracy. The test cases may include time boundary test cases, numerical boundary test cases, and logical boundary test cases. The time boundary test case may be to set "reminder at 23:59 every day" and verify whether there will be an error when crossing midnight; the numerical boundary test case may be "alarm when temperature > 30°C" and verify what will happen if the temperature is exactly 30.0°C; the logical boundary test case may be that the user says "if I am not at home and it is raining...", but what should be done when both "I am not at home" and "it is raining" do not hold. In the case of failed verification, the fourth conditional execution rule can be corrected according to the test results or exception handling can be added. The method of the embodiment of the present application can handle compound condition nesting (such as "A and (B or C)").

[0057] In the embodiment of the present application, by performing temporal normalization, semantic parsing, rule generation, and optimization verification on the second conversation content, the second conditional execution rule is obtained, so that the constructed second conditional execution rule can be more accurate.

[0058] In another exemplary embodiment of the present application, as Figure 6 shown, the Q&A pair data includes a question text set and an answer text set, and step 1014 can be replaced by the following steps 10141 to 10148: Step 10141: Structurally process the Q&A pair data to obtain a first data set containing an input sequence and an output sequence. The input sequence is obtained by embedding the question text set into a preset input template, and the output sequence is obtained by embedding the answer text set into a preset output template; Among them, the Q&A pair data can be structurally processed. In the format of "Question: {text}" as the input template and "Answer: {text}" as the output template according to the type of generative model. Combine the Q&A in the context with the relevant text as the input, and the output is the answer text or the position index of the answer; for multi-turn dialogue data, the historical dialogue can be concatenated as the input. The input sequence can be "Question: {Remind me in three quarters}", and the output sequence can be "Answer: {IF current_time== 15:30 THEN alert('Drink water')}".

[0059] Step 10142: Perform data preprocessing on the first data set, including normalization processing, truncation processing, and alignment word segmentation processing, to obtain a second data set; Among them, the normalization processing includes unifying the case, etc. If the length is greater than 512 Tokens, truncation processing can be performed.

[0060] Step 10143: Perform data augmentation processing on the second data set, including reverse question processing and synonym replacement processing, to obtain a third data set; Exemplarily, from the answer "IF time==15:30 THEN remind to drink water", the reverse questions "What time should you remind me to drink water?", "What should you do at 15:30?", and "Help me set a reminder to drink water" can be derived. The question "Remind me to take medicine every 4 hours" can be synonymously replaced with "Notify me to take medicine every 4 hours". The second data set after data augmentation processing can also be converted into a third data set file in JSON format.

[0061] Step 10144: Use the LlamaFactory framework as the training tool, with the Llama3.2-8B model as the large model base; Among them, the Llama3.2-8B model is an open-source large model with 8 billion parameters, which can balance performance and computing power requirements. The LlamaFactory framework is an efficient training tool optimized for the Llama series, which can simplify the training process.

[0062] Step 10145: Based on the Low-Rank Adaptation technique LoRA, update the Attention layer parameters of the large model base through low-rank matrices, and insert lightweight modules into the adapter to freeze the original parameters of the Llama3.2-8B model; Among them, according to the size of the dataset, parameter-efficient techniques (LoRA, Low-Rank Adaptation) can be adopted to update the parameters of the Attention layer through low-rank matrix, and adapters are inserted into lightweight modules to freeze the original model.

[0063] Step 10146: Adopt prompt fine-tuning technology to guide the semantic alignment between the output dataset of the large model base and the third dataset with learnable vectors; Step 10147: Combine the AdamW optimizer, cosine decay learning rate, and gradient clipping constraint to perform mixed-precision training on the large model base to obtain a trained model; Among them, the norm of the gradient can be ≤ 1.0.

[0064] Step 10148: Verify the trained model, and if the verification passes, determine the trained model as the exclusive model.

[0065] Among them, after the weight deployment of the trained model, verification can be performed.

[0066] In the embodiments of the present application, through data preprocessing, data quality can be improved and computing efficiency can be increased; through data augmentation processing, the generalization ability of the model can be improved, edge scenarios can be covered, and data diversity can be increased; through LoRA and adapters, the number of training parameters can be reduced; through prompt fine-tuning technology, parameter adaptation can be performed more efficiently and semantic alignment can be achieved more precisely; through the AdamW optimizer, overfitting can be prevented; through learning rate adjustment and gradient clipping constraint, training stability can be improved and computing resources can be saved.

[0067] In another exemplary embodiment of the present application, as Figure 7 shown, "verifying the trained model" in step 10148 includes the following steps 201 to step 205: Step 201: Based on BLEU and / or ROUGE, verify the trained model for generation tasks; Among them, BLEU can measure the lexical overlap between the generated answers output by the trained model and the reference answers, and ROUGE can measure the longest common subsequence between the generated answers output by the trained model and the reference answers.

[0068] Step 202: Based on the exact match value and F1 (balancing accuracy and recall) value, verify the trained model for extraction tasks; Among them, for the Exact Match (EM), it is required that the generated answer is exactly the same as the reference answer to return the result. If the match is successful, it returns 1; if it fails, it returns 0. The F1 value can measure the classification performance of the trained model. It is the harmonic mean of precision and recall, which can balance the contradiction between the two and give a more comprehensive evaluation result.

[0069] Step 203: Based on the classification accuracy rate and / or the ROC-AUC value, verify the classification task of the trained model. Among them, the classification accuracy rate is used to measure the proportion of the number of samples correctly predicted by the trained model in the total number of samples; the ROC-AUC value is used to measure the ability of the trained model to distinguish positive and negative samples.

[0070] Step 204: Based on at least one of the generation task verification, the extraction task verification, and the classification task verification, and the verification results of the manual verification, obtain the comprehensive verification result. Among them, manual sampling evaluation can be performed, randomly extracting the generation results, and manually judging the logical consistency (such as whether the answer is off-topic).

[0071] It should be noted that staged verification and adversarial verification can also be carried out. The staged verification can be evaluated every 100 - 500 steps to monitor the loss and index fluctuations. The adversarial verification can check the distribution differences between the verification set and the training set to avoid data leakage.

[0072] Step 205: When the comprehensive verification result meets the preset conditions, determine that the verification is passed.

[0073] Exemplarily, the comprehensive verification result can be obtained based on the verification results of the generation task verification and the manual verification. When BLEU > 25, ROUGE > 40, the passing rate of the manual evaluation ≥ 80%, and the answer is logically correct and has no factual errors, the test is passed, and the training of the exclusive model ends.

[0074] In the embodiments of the present application, multi-task joint verification of the generation task, the extraction task, and the classification task can be carried out, and combined with the verification results of the manual verification to obtain the comprehensive verification result. Thus, the model can be verified from multiple perspectives. After the verification is passed, the trained model is determined as the exclusive model, thereby improving the accuracy and reliability of the answer output by the model and improving the generalization performance of the model.

[0075] In one embodiment, the Prompt includes the task type, the called tool, the step structured instruction template, the example pair, and the output format constraint. As Figure 8 shown, step 1015 includes steps 10151 to 10155: Step 10151: Identify the task type of the reminder task through structured tags in the Prompt prompt; Among them, the Prompt can be a pre-designed structured instruction template used to guide the exclusive model to understand the task, call tools, and generate a standardized output; its core function is similar to a "detailed operation manual for the model", ensuring that the agent accurately executes tasks in a multi-modal environment. Exemplarily, in the case where the second conversation content is "If the temperature is lower than 10 degrees tomorrow morning, remind me to wear a down jacket with voice and pop-up windows", the Prompt can clearly inform the model that the current task type is a "temperature condition reminder task".

[0076] Step 10152: Embed a tool registry in the Prompt, where the name, input format, and permissions of the called tool are defined in the tool registry; Among them, the Prompt can call the weather API to check the temperature tomorrow morning and prepare a voice synthesis tool and a pop-up notification tool for use in subsequent reminders with voice reminders and pop-up reminders.

[0077] Step 10153: Guide the exclusive model to operate step by step through the procedural instructions of the step structured instruction template in the Prompt; Step 10154: Insert example pairs in the Prompt to show the mapping relationship between input and output; Step 10155: Declare output format constraints in the Prompt.

[0078] Among them, the format constraint can be a JSON format constraint, forcing the model to generate results in a specified format (such as JSON format) for subsequent parsing.

[0079] When building an agent based on an exclusive model, the Prompt is designed as follows for the reminder agent to analyze the task type, call tools, integrate the step structured instruction template of the results, add example pairs, and perform JSON format constraints. The agent architecture adopts hierarchical decision-making: after the model parses the user's intention, it calls the engine to match tools through the tool call mechanism, triggers a retry or downgraded response when the execution fails, and records error logs to optimize the process. The multi-modal perception tool association needs to integrate speech recognition and speech synthesis (reply with voice), and standardize the interaction through the REST API. After the data returned by the multi-modal perception tool is processed by the model for the second time, a multi-modal response is generated according to a preset template to ensure end-to-end controllability and user experience consistency.

[0080] In the embodiments of the present application, by adding Prompts on the basis of the exclusive model, task orientation can be more precisely performed according to the task type identified; by integrating external multi-modal perception tools, the model capabilities can be extended; through step-structured instruction templates, complex tasks can be disassembled to achieve multi-step reasoning automation; through example pairs, the model can be analogously prompted to activate the model's pattern recognition ability, enabling the model to accelerate learning, improve the answer accuracy, and quickly adapt to new tasks; by imposing output format constraints, the output quality can be controlled.

[0081] Based on the same inventive concept, the embodiments of the present application further provide a large model-based intelligent reminder device for implementing the above-mentioned large model-based intelligent reminder method. The implementation solutions provided by this device for solving problems are similar to the implementation solutions described in the above method. Therefore, the specific limitations in one or more of the following embodiments of the large model-based intelligent reminder device can refer to the limitations on the large model-based intelligent reminder method in the above text, and will not be repeated here.

[0082] In an exemplary embodiment, as Figure 9 shown, a large model-based intelligent reminder device 300 is provided, including: An identification module 301, configured to identify the first conversation content input by the user through an exclusive model agent to obtain a first conditional execution rule corresponding to the reminder task, where the exclusive model agent is obtained by fine-tuning the parameters of a pre-trained large model; A conversion module 302, configured to convert the first conditional execution rule corresponding to the reminder task into a corresponding target CRON expression; An establishment module 303, configured to establish connection channels with multiple user terminals of the user; An execution module 304, configured to, when the triggering timing of the target CRON expression is met, actively trigger the reminder task based on the pre-configured execution mode of the reminder task and the target CRON expression, and the reminder task is triggered sequentially according to a multi-level reminder matrix, so as to actively send reminder information to the user through the connection channels of the user terminals corresponding to each level of reminder.

[0083] As an alternative implementation, the conversion module 302 includes: a parsing sub-module for parsing the structure in the first conditional execution rule corresponding to the reminder task by using a parameter parser to obtain a constituent structure; an assembly sub-module for assembling the constituent structure into a first CRON expression based on a pre-established rule mapping table through a hierarchical replacement strategy, where the rule mapping table is used to associate the first conditional execution rule with the CRON expression; and a first processing sub-module for standardizing the heterogeneous event source data in the first CRON expression to obtain the target CRON expression.

[0084] As an alternative implementation, the user terminal includes at least one device in the intelligent terminal layer, near-field device layer, and remote service layer. The execution module 304 includes: a first reminder sub-module for sequentially performing the steps in the following three stages based on the pre-configured execution mode of the reminder task: in the first stage, performing at least one of a pop-up reminder, vibration reminder, and light-emitting diode (LED) reminder through at least one of the devices in the intelligent terminal layer; a second reminder sub-module for performing at least one of a voice broadcast reminder and a screen flash reminder through at least one of the devices in the near-field device layer in the second stage; and a third reminder sub-module for performing at least one of an automatic dialing reminder and an email follow-up reminder through at least one of the devices in the remote service layer in the third stage.

[0085] As an alternative implementation, the device further includes: a selection module for selecting second conversation content from domain data conversation records, a time series event table, and logs; a first generation module for generating a second conditional execution rule based on the spoken time in the second conversation content; a second generation module for generating question-and-answer pair data based on the second conversation content and the second conditional execution rule; a training module for training the question-and-answer pair data in a fine-tuning manner into a large model base to obtain a dedicated model; a setting module for setting an agent in a Prompt prompting manner based on the dedicated model to obtain the dedicated model agent; and an association module for associating the dedicated model agent with a multi-modal perception tool.

[0086] As an alternative implementation, the first generation module includes: a conversion sub-module for converting the colloquial time in the second conversation content into absolute time; a first recognition sub-module for recognizing conditional keywords in the second conversation content; a second recognition sub-module for using a bidirectional attention mechanism to recognize entity associations and temporal logic in the second conversation content based on the absolute time and the conditional keywords to obtain semantic roles in the second conversation content; a construction sub-module for constructing a structured logic framework of conditions and actions based on the semantic roles; a first generation sub-module for generating a third conditional execution rule based on the structured logic framework and a rule template engine; a second generation sub-module for optimizing the rule weights and conflict resolution strategies in the third conditional execution rule using a genetic algorithm to obtain a fourth conditional execution rule; a first verification sub-module for performing boundary verification on the fourth conditional execution rule using test cases, and after the verification passes, determining the fourth conditional execution rule as the second conditional execution rule.

[0087] As an alternative implementation, the Q&A pair data includes a question text set and an answer text set. The training module includes: a second processing sub-module for performing structured processing on the Q&A pair data to obtain a first data set including an input sequence and an output sequence, where the input sequence is obtained by embedding the question text set into a preset input template, and the output sequence is obtained by embedding the answer text set into a preset output template; a third processing sub-module for performing data preprocessing on the first data set including normalization processing, truncation processing, and aligned word segmentation processing to obtain a second data set; a fourth processing sub-module for performing data augmentation processing on the second data set including reverse question processing and synonym replacement processing to obtain a third data set; an update sub-module for using the LlamaFactory framework as a training tool, with the Llama3.2-8B model as the large model base; based on the Low-Rank Adaptation technique LoRA, updating the parameters of the Attention layer of the large model base through low-rank matrices and inserting lightweight modules into the adapter to freeze the original parameters of the Llama3.2-8B model; an alignment sub-module for using the prompt fine-tuning technique to guide the output data set of the completed large model base to be semantically aligned with the third data set using learnable vectors; a training sub-module for performing mixed-precision training on the large model base in combination with the AdamW optimizer, cosine decay learning rate, and gradient clipping constraints to obtain a trained model; a second verification sub-module for verifying the trained model, and in the case where the verification passes, determining the trained model as the exclusive model.

[0088] As an alternative implementation, the second verification sub-module includes: a first verification unit for verifying the trained model for a generation task based on BLEU and / or ROUGE; a second verification unit for verifying the trained model for an extraction task based on an exact match value and an F1 value; a third verification unit for verifying the trained model for a classification task based on a classification accuracy rate and / or an ROC-AUC value; a fourth verification unit for obtaining a comprehensive verification result based on at least one of the generation task verification, the extraction task verification, and the classification task verification, and the verification result of manual verification; and a fifth verification unit for determining that the verification is passed when the comprehensive verification result meets a preset condition.

[0089] As an alternative implementation, the Prompt prompt includes a task type, a calling tool, a step-structured instruction template, an example pair, and an output format constraint. The setting module includes: a third recognition sub-module for recognizing the task type of the reminder task through a structured tag in the Prompt prompt; an embedding sub-module for embedding a tool registry in the Prompt, where the tool registry defines the name, input format, and permissions of the calling tool; a guiding sub-module for guiding the exclusive model to operate step by step through the process instructions of the step-structured instruction template in the Prompt; an inserting sub-module for inserting an example pair in the Prompt to show the mapping relationship between the input and the output; and a declaration sub-module for declaring the output format constraint in the Prompt.

[0090] In an exemplary embodiment, a computer device is provided. The computer device can be a server or a terminal, and its internal structure diagram can be as Figure 10 shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with an external terminal through a network connection. The computer program, when executed by the processor, implements an intelligent reminder method based on a large model.

[0091] Those skilled in the art can understand,Figure 10 The structure shown is only a block diagram of some structures related to the solution of this application, and does not constitute a limitation on the computer device to which the solution of this application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0092] In an exemplary embodiment, a computer device is further provided, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, the steps in the above method embodiments are implemented.

[0093] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0094] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0095] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data need to comply with relevant regulations.

[0096] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memories. Non-volatile memories can include read-only memory (ROM), magnetic tapes, floppy disks, flash memories, optical memories, high-density embedded non-volatile memories, resistive random access memories (ReRAM), magnetoresistive random access memories (MRAM), ferroelectric random access memories (FRAM), phase change memories (PCM), graphene memories, etc. Volatile memories can include random access memory (RAM) or external cache memories, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0097] The databases involved in the embodiments provided in the present application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided in the present application can be general-purpose processors, central processors, graphics processors, digital signal processors, programmable logics, data processing logics based on quantum computing, etc., without limitation.

[0098] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0099] Specific examples are used in this article to elaborate on the principles and implementation manners of the present application. The descriptions of the above embodiments are only used to help understand the method and its core idea of the present application; at the same time, for those of ordinary skill in the art, according to the idea of the present application, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to the present application.

Claims

1. An intelligent reminder method based on a large model, characterized in that, The intelligent reminder method based on the large model includes: Identifying the first conversation content input by the user through the exclusive model agent to obtain the first conditional execution rule corresponding to the reminder task, where the exclusive model agent is obtained by fine-tuning the parameters of the pre-trained large model; Converting the first conditional execution rule corresponding to the reminder task into a corresponding target CRON expression; Establishing a connection channel with multiple user terminals of the user; When the triggering opportunity of the target CRON expression is met, actively triggering the reminder task based on the pre-configured execution mode of the reminder task and the target CRON expression. The reminder task is triggered sequentially according to the multi-level reminder matrix, so as to actively send reminder information to the user through the connection channels of the user terminals corresponding to each level of reminder.

2. The intelligent reminder method based on a large model according to claim 1, wherein The converting the first conditional execution rule corresponding to the reminder task into a corresponding target CRON expression includes: Using a parameter parser to parse the structure in the first conditional execution rule corresponding to the reminder task to obtain the composition structure; Based on the pre-established rule mapping table, assembling the composition structure into a first CRON expression through a hierarchical replacement strategy, where the rule mapping table is used to associate the first conditional execution rule and the CRON expression; Performing standardization processing on the heterogeneous event source data in the first CRON expression to obtain the target CRON expression.

3. The intelligent reminder method based on a large model according to claim 1, wherein, The user terminal includes at least one type of intelligent terminal layer device, near-field device layer device, and remote service layer device. When the triggering opportunity of the target CRON expression is met, actively triggering the reminder task based on the pre-configured execution mode of the reminder task and the target CRON expression includes: Sequentially executing the steps of the following three stages based on the pre-configured execution mode of the reminder task: In the first stage, executing at least one of pop-up reminder, vibration reminder, and light-emitting diode (LED) reminder through at least one of the intelligent terminal layer devices; In the second stage, executing at least one of voice broadcast reminder and screen flashing reminder through at least one of the near-field device layer devices; In the third stage, performing at least one of automatic dialing reminder and email follow-up reminder through at least one of the remote service layer devices.

4. The intelligent reminder method based on a large model according to claim 1, wherein, The intelligent reminder method based on the large model further includes: Selecting the second conversation content in the domain data conversation record, time series event table, and log; Generating a second conditional execution rule based on the spoken time in the second conversation content; Generating question-and-answer pair data based on the second conversation content and the second conditional execution rule; Training the question-and-answer pair data into the large model base in a fine-tuning manner to obtain an exclusive model; Setting an agent in the form of a Prompt prompt based on the exclusive model to obtain the exclusive model agent; Associating the exclusive model agent with a multi-modal perception tool.

5. The intelligent reminder method based on a large model according to claim 4, wherein The generating a second conditional execution rule based on the spoken time in the second conversation content includes: Converting the spoken time in the second conversation content into an absolute time; Identify the conditional keywords in the second conversation content; Based on the absolute time and the conditional keywords, use the bidirectional attention mechanism to identify the entity associations and temporal logics in the second conversation content to obtain the semantic roles in the second conversation content; Based on the semantic roles, construct a structured logic framework of conditions and actions; Based on the structured logic framework and the rule template engine, generate the third conditional execution rule; Use a genetic algorithm to optimize the rule weights and conflict resolution strategies in the third conditional execution rule to obtain the fourth conditional execution rule; Use test cases to perform boundary verification on the fourth conditional execution rule. After the verification passes, determine the fourth conditional execution rule as the second conditional execution rule.

6. The intelligent reminder method based on a large model according to claim 4, wherein The Q&A pair data includes a question text set and an answer text set. Training the Q&A pair data into the large model base in a fine-tuning manner to obtain a dedicated model includes: Perform structured processing on the Q&A pair data to obtain a first data set including an input sequence and an output sequence. The input sequence is obtained by embedding the question text set into a preset input template, and the output sequence is obtained by embedding the answer text set into a preset output template; Perform data preprocessing including normalization processing, truncation processing, and aligned tokenization processing on the first data set to obtain a second data set; Perform data augmentation processing including reverse question processing and synonym replacement processing on the second data set to obtain a third data set; Use the LlamaFactory framework as the training tool and the Llama3.2-8B model as the large model base; Based on the low-rank adaptation technology LoRA, update the Attention layer parameters of the large model base through low-rank matrices and insert lightweight modules in the adapter to freeze the original parameters of the Llama3.2-8B model; Use the prompt fine-tuning technology to guide the semantic alignment of the output data set of the large model base and the third data set with learnable vectors; Perform mixed-precision training on the large model base by combining the AdamW optimizer, cosine decay learning rate, and gradient clipping constraints to obtain a trained model; Verify the trained model. If the verification passes, determine the trained model as the dedicated model.

7. The intelligent reminder method based on a large model according to claim 6, wherein The verification of the trained model includes: Based on BLEU and / or ROUGE, perform generation task verification on the trained model; Based on the exact match value and F1 value, perform extraction task verification on the trained model; Based on the classification accuracy and / or ROC-AUC value, perform classification task verification on the trained model; Based on at least one of the generation task verification, the extraction task verification, the classification task verification, and the verification results of manual verification, obtain a comprehensive verification result; If the comprehensive verification result meets the preset conditions, determine that the verification passes.

8. The intelligent reminder method based on a large model according to claim 4, wherein The Prompt includes a task type, a calling tool, a step-structured instruction template, example pairs, and output format constraints. Setting up an agent in the form of a Prompt based on the exclusive model to obtain the exclusive model agent includes: Identifying the task type of the reminder task through structured tags in the Prompt; Embedding a tool registry in the Prompt, where the name, input format, and permissions of the calling tool are defined in the tool registry; Guiding the exclusive model to operate step by step through the process instructions of the step-structured instruction template in the Prompt; Inserting example pairs in the Prompt to show the mapping relationship between input and output; Declaring output format constraints in the Prompt.

9. An intelligent reminder device based on a large model, characterized in that, The intelligent reminder device based on the large model includes: An identification module for identifying the first conversation content input by the user through the exclusive model agent to obtain the first conditional execution rule corresponding to the reminder task, where the exclusive model agent is obtained by fine-tuning the parameters of the pre-trained large model; A conversion module for converting the first conditional execution rule corresponding to the reminder task into a corresponding target CRON expression; A connection module for establishing a connection channel with multiple user terminals of the user; An execution module for actively triggering a reminder task based on the execution mode of the reminder task pre-configured and the target CRON expression when the triggering opportunity of the target CRON expression is met, and the reminder task is triggered sequentially according to a multi-level reminder matrix to actively send a reminder message to the user through the connection channel of the user terminal corresponding to each level of reminder.

10. A computer device, comprising: A memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the intelligent reminder method based on the large model according to any one of claims 1-8.

Images