Home intelligent service mining method, related device and computer program product

Abstract

The invention provides a home intelligent service mining method, a related device and a computer program product. The home intelligent service mining method comprises the steps that a home equipment action list is acquired, and the home equipment action list comprises equipment actions of multiple pieces of equipment in a home and execution time corresponding to the equipment actions; the household equipment action list is input into an intelligent service mining model, an equipment control sequence group is obtained, and the equipment control sequence group comprises preorder actions with the first execution time and postorder actions with the later execution time; and based on the device control sequence group, generating an intelligent service, the trigger condition of the intelligent service being a preorder action, and the trigger result of the intelligent service being automatically executing a subsequent action. The home intelligent service mining method can construct an application scene in a self-adaptive manner and accurately match actual use requirements.

Description

Technical Field

[0001] This application relates to the field of smart home technology, and more specifically, to a method for mining smart home services, related devices, and computer program products. Background Technology

[0002] Currently, various proactive service and proactive management methods have emerged in the field of whole-house intelligence. Among them, the customized model method based on predefined scenarios has strong adaptability and can accurately match the business needs of known scenarios, directly meeting the service and management requirements under specific scenarios. However, existing methods of this kind require algorithm design, data mining and model training based on predefined scenarios, and lack the ability to adaptively build scenarios. In actual whole-house intelligence operation, there are a large number of small and micro scenarios that are not predefined, making it difficult to achieve adaptive scenario construction. Summary of the Invention

[0003] This application provides a method, related device, and computer program product for discovering smart home services, which are designed to adaptively construct application scenarios and accurately match actual usage needs.

[0004] In view of this, the first aspect of this application provides a method for mining home intelligent services, comprising:

[0005] Get the list of home device actions, which includes the device actions of multiple devices in the home and the corresponding execution times of the device actions;

[0006] Input the list of home device actions into the smart service mining model to obtain the device control sequence group. The device control sequence group includes the preceding action with the earlier execution time and the following action with the later execution time.

[0007] Based on the device control sequence group, intelligent services are generated. The trigger condition for an intelligent service is the preceding action, and the trigger result of an intelligent service is the automatic execution of the following action.

[0008] In the above technical solution, the home device action list is a collection of data recording the specific actions of multiple devices within a single home and their corresponding execution times. The intelligent service mining model is a model with the ability to identify temporal action associations, capable of mining related action combinations from the home device action list. The device control sequence group is an action combination containing a preceding action with a later execution time and a subsequent action with a later execution time. The intelligent service is an automated device control logic generated based on the device control sequence group, automatically executing the subsequent action when the preceding action is triggered.

[0009] In the above technical solution, a list of home device actions is obtained and summarized to form a data set containing the specific actions of multiple devices and their corresponding execution times. This collects discrete device runtime sequence data in a home setting, enabling basic data processing without relying on predefined scenarios. The list of home device actions is then input into an intelligent service mining model to obtain a device control sequence group containing preceding and subsequent actions. This automatically identifies the temporal correlation characteristics between device actions without requiring manual analysis of linkage logic based on predefined scenarios. Finally, intelligent services are generated based on the device control sequence group. Even without relying on predefined scenarios, this model can automatically mine the device action relationships in unknown small and micro-scenes within a home setting, generating intelligent services that accurately match actual home usage habits. This enables the adaptive construction of home intelligent services.

[0010] In some technical solutions of this application, optionally, the intelligent service mining model is pre-trained through the following process:

[0011] Obtain the sample home device action list, which includes the sample device actions of multiple sample devices in the sample home and the sample execution time of the sample device actions;

[0012] From the sample home device action list, positive samples and negative samples are extracted. Positive samples contain at least two first sample device actions, which are executed at adjacent times and are related in the sample home device action list. Negative samples contain at least two second sample device actions, which are executed at adjacent times and are not related in the sample home device action list.

[0013] The original model is trained based on positive and negative samples to obtain the intelligent service mining model.

[0014] In the above technical solution, the sample household device action list is data on device actions and their corresponding execution times from multiple sample households used for model training. Positive samples are training data containing at least two sample device actions with adjacent execution times and a correlation. Negative samples are training data containing at least two sample device actions with adjacent execution times but no correlation. The original model is an initial model that has not been trained with samples but possesses basic correlation recognition potential.

[0015] In the above technical solution, a training dataset is formed by summarizing the action lists of sample home devices to create a multi-home scenario training dataset. This collects discrete sample operation data, providing a comprehensive and scenario-free data foundation for model training. Positive and negative samples are automatically extracted from the action lists of sample home devices to accurately select core training data and help the model establish a universal standard for distinguishing between associations and non-associations. The original model is trained based on the positive and negative samples to obtain an intelligent service mining model. This enables the intelligent service mining model to accurately mine associations of home device actions, significantly improving model training efficiency and mining accuracy, and achieving generalized and efficient training of the intelligent service mining model.

[0016] In some technical solutions of this application, optionally, the association relationship refers to the logical causal relationship or cooperative relationship based on the family life scenario that exists between at least two first sample device actions with adjacent execution times in the sample family device action list;

[0017] The relationship includes at least one of the following:

[0018] The execution of a preceding action directly leads to the execution of a subsequent action;

[0019] The execution of a preceding action causes a change in the environmental state, while a subsequent action is a passive response to the change in the environmental state.

[0020] In the above technical solution, the specific connotation of the association relationship is defined, and the logical basis for screening positive and negative samples is clarified, which helps to improve the quality of sample data. At the same time, a clear definition of the association relationship can help the original model learn the interaction patterns between device actions more accurately, reduce the interference of invalid samples, and thus improve the training efficiency and recognition accuracy of the intelligent service mining model.

[0021] In some technical solutions of this application, optionally, positive and negative samples are extracted from the sample home device action list, including:

[0022] A sliding time window of predetermined duration is used to slide through the list of home device actions, resulting in multiple device action groups. The execution time of the device actions in each device action group falls within the sliding time window.

[0023] After adding prompt words to the device action group, input it into the large model to obtain the judgment correlation and confidence probability between each device action in the device action group;

[0024] Based on the determination of correlation and confidence probability, positive and negative samples are selected from multiple device action groups.

[0025] In the above technical solution, the sliding time window is a time interval of predetermined duration that can slide continuously along the time axis, used to segment continuous device action data. A device action group is a set of device actions captured by the sliding time window, where the execution time of all actions falls within the window. The prompt word is the instruction text used to guide the large model in determining the correlation. The large model is a high-parameter model with strong semantic understanding and correlation recognition capabilities. The correlation determination is the judgment result of the correlation attributes between device actions output by the large model. The confidence probability is a quantitative value (between 0 and 1) of the large model's confidence in its own judgment result.

[0026] In the above technical solution, multiple device action groups are obtained by sliding a time window through the sample data. The data is organized according to the time dimension to focus on the correlation of device actions in the same time period. This can automatically capture the short-sequence device action correlation features in unknown small and micro scenes, avoiding interference from unrelated actions across time periods. The device action groups are then input into a large model for mining to obtain the judgment correlation relationship and confidence probability between each device action, realizing the automated and accurate judgment of the correlation relationship and improving the judgment efficiency. Based on the judgment correlation relationship and confidence probability, positive and negative samples are screened, which can provide high-reliability core training data for model training, further improving the intelligent service mining model's ability to learn the correlation features of unknown small and micro scenes, and realizing the automated and efficient screening of high-quality training samples.

[0027] In some technical solutions of this application, optionally, positive and negative samples are selected from multiple device action groups based on the determination of correlation and confidence probability, including:

[0028] If it is determined that there is a correlation between the various device actions in the device action group and the confidence probability is greater than the first threshold, then the device action group is selected as a positive sample.

[0029] If it is determined that there is no correlation between the various device actions in the device action group and the confidence probability is greater than the first threshold, then the device action group is selected as a negative sample.

[0030] In the above technical solution, the first threshold is a preset confidence probability threshold used to screen high-confidence samples.

[0031] In the above technical solution, determining the correlation between equipment action groups and clarifying their correlation attributes provides a general basis for sample classification without needing to set classification rules based on predefined scenarios. Furthermore, comparing confidence probabilities with a first threshold to filter high-confidence samples effectively filters low-confidence correlation results and eliminates accidentally misjudged sample data without requiring subjective judgment based on predefined scenarios. By quantifying the screening criteria, the sample screening process remains repeatable and objective without relying on predefined scenarios. This ensures that both positive and negative samples entering the training process can support the intelligent service mining model in learning general correlation patterns, improving its scenario-adaptive mining capabilities and guaranteeing accurate correlation mining even in unknown small or micro-scenarios. This makes sample screening more objective and reliable.

[0032] Optionally, in some technical solutions of this application, after extracting positive and negative samples from the sample home device action list, the training process of the intelligent service mining model further includes:

[0033] If the number of positive or negative samples is less than a predetermined threshold, the extracted positive or negative samples are input into the sample augmentation model to augment the model so that the number of positive and negative samples is not less than the predetermined threshold.

[0034] In the above technical solution, the predetermined threshold is a pre-set critical value for the minimum number of samples required to ensure the model training effect. The sample augmentation model is a model with sample generation capabilities, capable of expanding from a small number of existing samples to generate similar high-quality samples.

[0035] In the above technical solution, the number of positive or negative samples is compared with a predetermined threshold to accurately identify sample-scarce scenarios. The extracted insufficient samples are then input into the sample expansion model to obtain a sufficient number of high-quality samples of the same type, quickly supplementing the samples to the predetermined threshold. This avoids the problems of poor generalization ability and low mining accuracy of the intelligent service mining model caused by the scarcity of samples in unknown small or micro scenarios. By expanding the samples, the cost of sample acquisition can be reduced, ensuring that the model training always has sufficient high-quality training data to support it. This enables the intelligent service mining model to learn general correlation rules covering unknown small and micro scenarios, further strengthening the scenario adaptive construction capability of the intelligent service mining model and achieving efficient supplementation of scarce samples and stable supply of model training data.

[0036] In some technical solutions of this application, optionally, an intelligent service mining model is obtained by training the original model based on positive and negative samples, including:

[0037] Extract memory summary information from the positive and negative samples in the first time period, which is the predetermined duration before the training time point of the original model.

[0038] Based on positive samples, negative samples, and memory summary information, the original model is trained to obtain the intelligent service mining model.

[0039] In the above technical solution, the first time period is a predetermined time interval prior to the time point of training the original model. The memory summary information is a condensed information extracted from the positive and negative samples of the first time period, reflecting the correlation patterns of historical device actions.

[0040] In the above technical solution, memory summary information is extracted from positive and negative samples within the first time period to condense the general association patterns in historical data and reduce the interference of redundant data on training. Then, the original model is trained based on positive samples, negative samples, and memory summary information, so that the original model inherits the general association patterns in history and avoids the problem of insufficient learning of association patterns in unknown small and micro scenarios caused by relying solely on current samples for training. Through this training method, the training efficiency of the model can be improved, the adaptability of the intelligent service mining model to the long-term usage habits of various family scenarios (including unknown small and micro scenarios) can be enhanced, the mining accuracy of the intelligent service mining model can be improved, and the mining capability of the intelligent service mining model can be made long-term and accurate.

[0041] Optionally, in some technical solutions of this application, intelligent services are generated based on device control sequence groups, including:

[0042] Multiple device control sequence groups obtained within a predetermined period are subjected to synonym identification, and device control sequence groups identified as having the same meaning are grouped into one category.

[0043] If the number of device control sequence groups in a class exceeds a predetermined number, intelligent services are generated based on the device control sequence groups in the class.

[0044] In the above technical solution, the predetermined period is a pre-set time interval used to statistically analyze the patterns of occurrence of device control sequence groups. Synonym recognition is the process of determining whether different device control sequence groups express the same device association. The predetermined number is a pre-set threshold value used to determine whether a device control sequence group represents a high-frequency user demand.

[0045] In the above technical solution, multiple device control sequence groups obtained within a predetermined period are subjected to synonym recognition. Device control sequence groups expressing the same device association are grouped into one category, thereby unifying the association of different expressed actions and improving the comprehensiveness of coverage of diverse action expressions in unknown small and micro scenarios. Then, the number of each type of device control sequence group is counted, and device control sequence groups corresponding to categories with more than a predetermined number are selected to generate intelligent services. Data quantification is used to distinguish between accidental associations and high-frequency needs, and single accidental device action associations are eliminated. Through this selection and generation method, it can be ensured that the generated intelligent services are the real high-frequency needs of users in various scenarios, reducing the possibility of false triggering in unknown scenarios, improving the practicality and user acceptance of intelligent services, and achieving accurate selection of high-frequency real needs and high-quality generation of intelligent services.

[0046] Optionally, in some technical solutions of this application, after inputting the list of home device actions into the smart service mining model to obtain the device control sequence group, the method further includes: if the user's future action is consistent with the preceding action in the device control sequence group, displaying a first query message, the first query message being used to ask the user whether to perform the subsequent action;

[0047] After generating intelligent services based on the device control sequence group, the method further includes: if the user's future action is consistent with the preceding action in the device control sequence group, automatically execute the subsequent action.

[0048] In the above technical solution, the user's future action refers to the actual device action performed by the user during subsequent use after the intelligent service is generated. The first inquiry message is a prompt message used to ask the user whether they agree to execute the subsequent action.

[0049] In the above technical solution, before the intelligent service is officially generated, if it is detected that the user's future action is consistent with the preceding action of the device control sequence group, the first inquiry message is displayed to obtain user confirmation. This respects the user's willingness to use the service in unknown small or micro-scenarios and reduces the user's resistance to automatically executing the service. After the intelligent service is generated, if it is detected that the user's future action is consistent with the preceding action of the device control sequence group, the subsequent action of the device control sequence group is automatically executed, improving ease of use. This better suits the usage habits of home users in various scenarios, avoids the risk of misoperation caused by direct automatic execution, improves the acceptance of intelligent services in unknown scenarios, ensures the smooth implementation of intelligent services built with scene adaptation, and realizes the humanized and progressive implementation of intelligent services.

[0050] Optionally, in some technical solutions of this application, the device control sequence group is written using a model language;

[0051] The first query message is displayed, including:

[0052] Convert the device control sequence group from model language into natural language;

[0053] Based on the device control sequence group converted into natural language, the first query message is displayed.

[0054] In the above technical solutions, model language is a machine-recognizable structured language used in the output results of the intelligent service mining model. Natural language is the everyday communication language that humans can understand.

[0055] In the above technical solution, the device control sequence group written in the model language is converted into natural language to solve the language barrier between the model output and the user's cognition, making the device action association information easier for the user to understand; then, the first query message is displayed based on the converted natural language to ensure that the user clearly knows the device action to be performed; thus, it can avoid misunderstandings of intelligent services by users due to language barriers, enhance user trust, improve interaction efficiency, optimize user experience, and achieve clear and efficient intelligent service interaction.

[0056] Optionally, in some technical solutions of this application, the list of home device actions may also include environmental data corresponding to the device actions;

[0057] The device control sequence group includes the preceding action, the environmental data corresponding to the preceding action, and the following action;

[0058] The triggering conditions for intelligent services are the preceding actions and the environmental data corresponding to the preceding actions.

[0059] In the above technical solution, environmental data refers to environmental parameter data (such as weather, temperature, humidity, and light) related to the equipment's action execution scenario.

[0060] In the above technical solution, environmental data corresponding to device actions is supplemented into the list of home device actions to enrich the basis for data mining and provide environmental dimension support for accurately identifying scene-related features, avoiding the problem that single action data cannot distinguish different scenes. Furthermore, the device control sequence group is expanded into a combination that includes preceding actions, corresponding environmental data, and subsequent actions. The intelligent service generated based on the device control sequence group upgrades the triggering condition to a combination of preceding actions and environmental data, effectively avoiding scene misjudgment caused by single action triggering. Through this expansion and optimization method, the triggering of intelligent services can be more in line with the dynamic changes of the home environment, enhancing adaptability and practicality, ensuring the accuracy of scene adaptive construction, and realizing the precision and scene-specific adaptation of intelligent service triggering.

[0061] A second aspect of this application discloses a home smart service discovery device, comprising:

[0062] The acquisition module is used to acquire a list of home device actions, which includes the device actions of multiple devices in the home and the corresponding execution times of the device actions;

[0063] The processing module is used to input the list of home device actions into the smart service mining model to obtain the device control sequence group, which includes the preceding action and the following action.

[0064] The generation module is used to generate intelligent services based on the device control sequence group. The trigger condition for the intelligent service is the preceding action, and the trigger result of the intelligent service is the automatic execution of the following action.

[0065] The third aspect of this application proposes a home intelligent service mining controller, comprising: a memory for storing programs or instructions; and a processor for executing the programs or instructions to implement the steps of any of the home intelligent service mining methods described above. Therefore, it can achieve all the same technical effects, and to avoid repetition, it will not be described further here.

[0066] The fourth aspect of this application provides a readable storage medium having a program or instructions stored thereon. When the program or instructions are executed by a processor, they implement the steps of the home intelligent service mining method provided in any of the above technical solutions, thus achieving all the same technical effects. To avoid repetition, further details are omitted here.

[0067] The fifth aspect of this application provides a computer program product, which is stored in a storage medium and executed by at least one processor to implement the steps of the home intelligent service mining method provided in any of the above technical solutions. Therefore, it can also achieve all the same technical effects, and will not be described again here to avoid repetition.

[0068] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0069] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which:

[0070] Figure 1 A flowchart illustrating a method for mining smart home services provided in this application embodiment;

[0071] Figure 2 A schematic diagram illustrating the pre-training process of the intelligent service mining model provided in this application embodiment;

[0072] Figure 3 for Figure 2A specific flowchart of the S40;

[0073] Figure 4 for Figure 2 Another specific flowchart of the S40;

[0074] Figure 5 for Figure 2 A specific flowchart of S60;

[0075] Figure 6 for Figure 1 A specific flowchart of S140 in China;

[0076] Figure 7 A flowchart illustrating another method for mining smart home services provided in this application embodiment;

[0077] Figure 8 for Figure 7 A specific flowchart of S130 in China;

[0078] Figure 9 A structural block diagram of a home smart service discovery device provided in this application embodiment;

[0079] Figure 10 This is a structural block diagram of a home smart service mining controller provided in an embodiment of this application. Detailed Implementation

[0080] To better understand the above-mentioned objectives, features, and advantages of this application, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0081] Many specific details are set forth in the following description in order to provide a full understanding of this application. However, this application may also be implemented in other ways different from those described herein. Therefore, the scope of protection of this application is not limited to the specific embodiments disclosed below.

[0082] The following reference Figures 1 to 10 This application describes a method, related apparatus, and computer program product for discovering home smart services, based on some embodiments.

[0083] like Figure 1 As shown, the first aspect of this application provides a method for mining home smart services, including:

[0084] S100: Obtain the list of home device actions, which includes the device actions of multiple devices in the home and the corresponding execution time of the device actions;

[0085] S120: Input the list of home device actions into the smart service mining model to obtain the device control sequence group. The device control sequence group includes the preceding action with the earlier execution time and the following action with the later execution time.

[0086] S140: Based on the device control sequence group, generate intelligent services. The trigger condition for intelligent services is the preceding action, and the trigger result of intelligent services is the automatic execution of the following action.

[0087] In the above embodiments, the home device action list is a collection of data recording the specific actions of multiple devices within a single home and their corresponding execution times. The intelligent service mining model is a model with the ability to identify temporal action associations, capable of mining related action combinations from the home device action list. The device control sequence group is an action combination containing a preceding action with a later execution time and a subsequent action with a later execution time. The intelligent service is an automated device control logic generated based on the device control sequence group, automatically executing the subsequent action when the preceding action is triggered.

[0088] In the above embodiments, a list of home device actions is obtained and summarized to form a data set containing the specific actions of multiple devices and their corresponding execution times. This collects discrete device runtime sequence data in a home setting, enabling basic data processing without relying on predefined scenarios. The list of home device actions is then input into an intelligent service mining model to obtain a device control sequence group containing preceding and subsequent actions. This automatically identifies the temporal correlation characteristics between device actions without requiring manual analysis of linkage logic based on predefined scenarios. Finally, intelligent services are generated based on the device control sequence group. Even without relying on predefined scenarios, this model can automatically mine the device action relationships in unknown small and micro-scenes within a home setting, generating intelligent services that accurately match actual home usage habits. This enables the adaptive construction of home intelligent services.

[0089] It is understandable that the preceding and following actions can be either actual actions of a specific device, such as "opening the curtains," or changes in the attributes of a specific device, such as changing from "standby mode" to "running mode," changing an air conditioner from "cooling mode" to "dehumidification mode," or changing the temperature from 26°C to 24°C.

[0090] It is understandable that different devices have different actions, therefore different preceding or following actions can be attributed to different devices. For example, a list of home device actions might include data such as "Bedroom window (Device A01) opens at 08:00:01" and "Bedroom air purifier (Device B01) closes at 08:00:03". That is, the device control sequence group could be "Device A01 opens (preceding action), Device B01 closes (following action)". The generated smart service could be "Bedroom window opened (trigger condition), automatically turn off bedroom air purifier (trigger result)".

[0091] like Figure 2 As shown, in some embodiments of this application, optionally, the intelligent service mining model is pre-trained through the following process:

[0092] S20: Obtain the sample home device action list, which includes the sample device actions of multiple sample devices in the sample home and the sample execution time corresponding to the sample device actions;

[0093] S40: Extract positive and negative samples from the sample home device action list. The positive sample contains at least two first sample device actions, which are executed at adjacent times and are related in the sample home device action list. The negative sample contains at least two second sample device actions, which are executed at adjacent times and are not related in the sample home device action list.

[0094] S60: Based on positive and negative samples, train the original model to obtain the intelligent service mining model.

[0095] In the above embodiments, the sample household device action list is data on device actions and their corresponding execution times from multiple sample households used for model training. Positive samples are training data containing at least two sample device actions with adjacent execution times and a correlation. Negative samples are training data containing at least two sample device actions with adjacent execution times but no correlation. The original model is an initial model that has not been trained with samples but possesses basic correlation recognition potential.

[0096] In the above embodiments, the sample household device action list is obtained and summarized to form a training data set for multiple household scenarios. Discrete sample operation data is collected to provide a comprehensive and scenario-free data foundation for model training. Positive and negative samples are automatically extracted from the sample household device action list to accurately screen core training data and help the model establish a universal standard for distinguishing between association and non-association. The original model is trained based on positive and negative samples to obtain an intelligent service mining model, which enables the intelligent service mining model to have the ability to accurately mine associations of household device actions, greatly improving the model training efficiency and mining accuracy, and realizing the generalized and efficient training of the intelligent service mining model.

[0097] For example, the original model can be a base model on the order of 7 billion (Billions). The sample list of household device actions can include data such as "Bedroom window (Device A01) opens at 08:00:01, Bedroom air purifier (Device B01) closes at 08:00:03". That is, a positive sample can be "Device A01 opens, Device B01 closes, there is a trigger relationship between the two", and a negative sample can be "Device A01 opens, Bedroom light (Device C01) turns on, there is no correlation between the two". After training the original model based on these positive and negative samples, an intelligent service mining model with the ability to mine household device action associations is obtained.

[0098] In some embodiments of this application, optionally, the association relationship refers to the logical causal relationship or cooperative relationship based on the family life scenario that exists between at least two first sample device actions that are executed in adjacent times in the sample family device action list;

[0099] The relationship includes at least one of the following:

[0100] The execution of a preceding action directly leads to the execution of a subsequent action;

[0101] The execution of a preceding action causes a change in the environmental state, while a subsequent action is a passive response to the change in the environmental state.

[0102] In the above embodiments, the specific connotation of the association relationship is defined, and the logical basis for screening positive and negative samples is clarified, which helps to improve the quality of sample data. At the same time, a clear definition of the association relationship can help the original model learn the interaction patterns between device actions more accurately, reduce the interference of invalid samples, and thus improve the training efficiency and recognition accuracy of the intelligent service mining model.

[0103] For example, there is a direct correlation between a preceding action and a subsequent action. In the positive sample, "Device A01 (bedroom window) opens, Device B01 (bedroom air purifier) ​​closes," indicating a triggering relationship; opening the window directly prompts the subsequent action of closing the air purifier. Another example is a correlation where a preceding action causes a change in environmental conditions, leading to a passive response from the subsequent action. In the positive sample, "Device D01 (living room air conditioner) turns on cooling mode, Device E01 (living room humidifier) ​​turns on," the air conditioner's cooling mode lowers the indoor humidity (a change in environmental conditions), triggering the humidifier's passive response to turn on; a response relationship exists between the two. Figure 3 As shown, in some embodiments of this application, optionally, positive and negative samples are extracted from the sample home device action list, including:

[0104] S42: A sliding time window of predetermined duration is slid through the list of home device actions to obtain multiple device action groups. The execution time of the device actions in each device action group falls into the sliding time window.

[0105] S44: Add the device action group to the prompt words and input it into the large model to obtain the judgment correlation and confidence probability between each device action in the device action group;

[0106] S46: Based on the determination of correlation and confidence probability, select positive and negative samples from multiple device action groups.

[0107] In the above embodiments, the sliding time window is a time interval of predetermined duration that can slide continuously along the time axis, used to segment continuous device action data. A device action group is a set of device actions captured by the sliding time window, where the execution time of all actions falls within the window. The prompt word is the instruction text used to guide the large model in determining the correlation. The large model is a high-parameter model with strong semantic understanding and correlation recognition capabilities. The correlation determination is the judgment result of the correlation attributes between device actions output by the large model. The confidence probability is a quantitative value (between 0 and 1) of the large model's confidence in its own judgment result.

[0108] In the above embodiments, multiple device action groups are obtained by sliding a time window through the sample data. The data is organized according to the time dimension to focus on the association of device actions in the same time period. This can automatically capture the short-sequence device action association features in unknown small and micro scenes, avoiding interference from unrelated actions across time periods. The device action groups are then input into a large model for mining to obtain the judgment association relationship and confidence probability between each device action, realizing the automated and accurate judgment of the association relationship and improving the judgment efficiency. Based on the judgment association relationship and confidence probability, positive and negative samples are screened, which can provide highly reliable core training data for model training, further improving the intelligent service mining model's ability to learn the association features of unknown small and micro scenes, and realizing the automated and efficient screening of high-quality training samples.

[0109] In some embodiments of this application, optionally, positive and negative samples are selected from multiple device action groups based on the determined correlation and confidence probability, including:

[0110] If it is determined that there is a correlation between the various device actions in the device action group and the confidence probability is greater than the first threshold, then the device action group is selected as a positive sample.

[0111] If it is determined that there is no correlation between the various device actions in the device action group and the confidence probability is greater than the first threshold, then the device action group is selected as a negative sample.

[0112] In the above embodiments, the first threshold is a preset confidence probability threshold used to screen high-confidence samples.

[0113] In the above embodiments, determining the association relationship of device action groups and clarifying the association attributes of device action groups can provide a general basis for sample classification without setting classification rules based on predefined scenarios. Then, comparing the confidence probability with the first threshold to screen high-confidence samples can effectively filter low-confidence association relationship results and eliminate sample data with accidental misjudgments without the need for subjective judgment based on predefined scenarios. By quantifying the screening criteria, the sample screening process can still be made repeatable and objective without the premise of predefined scenarios. This ensures that both positive and negative samples entering the training process can support the intelligent service mining model in learning general association rules, improve the scene adaptive mining capability of the intelligent service mining model, and ensure that the intelligent service mining model can still accurately mine association relationships in unknown small scenarios and micro scenarios, making the sample screening more objective and more credible.

[0114] For example, the first threshold is set to 0.8. The device action group determination results output by the large model may include "Bedroom window (device A01) is open, bedroom air purifier (device B01) is closed, a triggering relationship exists, confidence probability 0.92", etc. That is, if the confidence probability 0.92 > 0.8, the corresponding device action group is regarded as a positive sample; if the confidence probability of a certain determination relationship result is < 0.8, then it is not included in the positive and negative samples; the filtered samples can be used for training the intelligent service mining model.

[0115] like Figure 4 As shown, in some embodiments of this application, optionally, after extracting positive and negative samples from the sample home device action list, the training process of the intelligent service mining model further includes:

[0116] S48: If the number of positive or negative samples is less than a predetermined threshold, the extracted positive or negative samples are input into the sample expansion model to expand the model so that the number of positive and negative samples is not less than the predetermined threshold.

[0117] In the above embodiments, the predetermined threshold is a preset critical value for the minimum number of samples required to ensure the training effect of the model. The sample augmentation model is a model with sample generation capabilities, capable of expanding from a small number of existing samples to generate similar high-quality samples.

[0118] In the above embodiments, the number of positive or negative samples is compared with a predetermined threshold to accurately identify sample-scarce scenarios. The extracted insufficient samples are then input into the sample expansion model to obtain a sufficient number of high-quality samples of the same type, quickly supplementing the samples to the predetermined threshold. This avoids the problems of poor generalization ability and low mining accuracy of the intelligent service mining model caused by the scarcity of samples in unknown small or micro scenarios. By expanding the samples, the cost of sample acquisition can be reduced, ensuring that the model training always has sufficient high-quality training data to support it. This enables the intelligent service mining model to learn the general correlation rules covering unknown small and micro scenarios, further strengthening the scenario adaptive construction capability of the intelligent service mining model and achieving efficient supplementation of scarce samples and stable supply of model training data.

[0119] For example, the sample expansion model can be a large model with sample generation capabilities, with a predetermined threshold set at 1000 samples. The actual collected positive samples may include 300 data points such as "the bedroom window (device A01) is open, the bedroom air purifier (device B01) is closed, and a trigger relationship exists." These 300 positive samples are input into the sample expansion model to generate 700 similar positive samples, ultimately bringing the total number of positive samples to 1000; the same logic is used to expand negative samples when they are insufficient.

[0120] like Figure 5As shown, in some embodiments of this application, optionally, an intelligent service mining model is trained based on positive and negative samples to obtain the original model, including:

[0121] S62: Extract memory summary information from the positive and negative samples in the first time period, which is the predetermined duration before the training time point of the original model.

[0122] S64: Based on positive samples, negative samples, and memory summary information, train the original model to obtain the intelligent service mining model.

[0123] In the above embodiments, the first time period is a predetermined time interval prior to the time point when the original model was trained. The memory summary information is a condensed information extracted from the positive and negative samples of the first time period, reflecting the correlation patterns of historical device actions.

[0124] In the above embodiments, memory summary information is extracted from positive and negative samples within the first time period to condense the general association patterns in historical data and reduce the interference of redundant data on training. Then, the original model is trained based on the positive samples, negative samples, and memory summary information, so that the original model inherits the general association patterns of history and avoids the problem that the original model is not good at learning the association patterns of unknown small and micro scenarios due to training only on the current samples. Through the training method, the training efficiency of the model can be improved, the adaptability of the intelligent service mining model to the long-term usage habits of various family scenarios (including unknown small and micro scenarios) can be enhanced, the mining accuracy of the intelligent service mining model can be improved, and the mining capability of the intelligent service mining model can be made long-term and accurate.

[0125] For example, the original model can be a base model of 7B, with the first time period set to one week prior to the time point when the original model was trained. Positive and negative samples within this first time period can contain data such as "Bedroom window (device A01) opens at 08:00:01, bedroom air purifier (device B01) closes at 08:00:03, a trigger relationship exists." The memory summary information extracted from these positive and negative samples is "Daily 07:30-08:30, the average confidence of the association relationship of device B01 closing within 3 seconds after device A01 opens is 0.9." During training, this information is input into the original model along with the positive and negative samples to help the original model learn historical association patterns.

[0126] For example, the original model can also be on the order of 10B, 32B, etc.

[0127] like Figure 6 As shown, in some embodiments of this application, optionally, intelligent services are generated based on device control sequence groups, including:

[0128] S142: Perform synonym recognition on multiple groups of equipment control sequences obtained within a predetermined period, and classify the groups of equipment control sequences that are identified as having the same meaning into one category;

[0129] S144: If the number of device control sequence groups in the class exceeds a predetermined number, generate intelligent services based on the device control sequence groups in the class.

[0130] In the above embodiments, the predetermined period is a preset time interval used to statistically analyze the patterns of occurrence of device control sequence groups. Synonym identification is a process of determining whether different device control sequence groups express the same device association. The predetermined number is a preset threshold value used to determine whether a device control sequence group represents a high-frequency user demand.

[0131] In the above embodiments, multiple device control sequence groups obtained within a predetermined period are subjected to synonym recognition. Device control sequence groups expressing the same device association are grouped into one category, thereby unifying the association of different expressed actions and improving the comprehensiveness of coverage of diverse action expressions in unknown small and micro scenarios. Then, the number of each type of device control sequence group is counted, and device control sequence groups corresponding to categories with more than a predetermined number are selected to generate intelligent services. Data quantification is used to distinguish between accidental associations and high-frequency needs, and single accidental device action associations are eliminated. Through this selection and generation method, it can be ensured that the generated intelligent services are the real high-frequency needs of users in various scenarios, reducing the possibility of false triggering in unknown scenarios, improving the practicality and user acceptance of intelligent services, and achieving accurate selection of high-frequency real needs and high-quality generation of intelligent services.

[0132] For example, the predetermined period is set to 1 week, and the predetermined number is set to 5 times. Within 1 week, multiple device control sequence groups are mined, which may include "Bedroom window (device A01) open, bedroom air purifier (device B01) close", etc. Similar action groups are grouped into one category through synonym recognition. If the action group appears 6 times (more than 5 times) in the statistical category, the intelligent service "Detected that the bedroom window is open, automatically turn off the bedroom air purifier" is generated.

[0133] like Figure 7 As shown, in some embodiments of this application, optionally, after inputting the list of home device actions into the smart service mining model to obtain the device control sequence group, the method further includes:

[0134] S130: If the user's future action is consistent with the previous action in the device control sequence group, display the first query message. The first query message is used to ask the user whether to execute the subsequent action.

[0135] After generating intelligent services based on device control sequence groups, the method also includes:

[0136] S150: If the user's future action is consistent with the preceding action in the device control sequence group, the subsequent action will be executed automatically.

[0137] In the above embodiments, the user's future action refers to the actual device action performed by the user during subsequent use after the intelligent service is generated. The first inquiry message is a prompt message used to ask the user whether they agree to perform the subsequent action.

[0138] In the above embodiments, before the intelligent service is officially generated, if it is detected that the user's future action is consistent with the preceding action of the device control sequence group, a first inquiry message is displayed to obtain user confirmation. This respects the user's willingness to use the service in unknown small or micro-scenarios and reduces the user's resistance to automatically executing the service. After the intelligent service is generated, if it is detected that the user's future action is consistent with the preceding action of the device control sequence group, the subsequent action of the device control sequence group is automatically executed, improving ease of use. This better suits the usage habits of home users in various scenarios, avoids the risk of misoperation caused by direct automatic execution, improves the acceptance of intelligent services in unknown scenarios, ensures the smooth implementation of intelligent services built with scenario adaptation, and realizes the humanized and progressive implementation of intelligent services.

[0139] For example, before a formal smart service is generated, if it is detected that a user has opened a bedroom window (consistent with the preceding action of the device control sequence group), the first query message is displayed: "The bedroom window has been detected to be open. Should we turn off the air purifier in the same room for you?". After the action group has accumulated 5 occurrences and a formal smart service has been generated, if the user opens the bedroom window again, the air purifier will automatically turn off without requiring a query.

[0140] like Figure 8 As shown, in some embodiments of this application, the device control sequence group is optionally written in a model language;

[0141] The first query message is displayed, including:

[0142] S132: Convert the device control sequence group from model language into natural language;

[0143] S134: Based on the device control sequence group converted into natural language, display the first query message.

[0144] In the above embodiments, the model language is a machine-recognizable structured language used in the output of the intelligent service mining model. Natural language is the everyday language that humans can understand.

[0145] In the above embodiments, the device control sequence group written in the model language is converted into natural language to solve the language barrier between the model output and the user's cognition, making the device action association information easier for the user to understand; then, the first query message is displayed based on the converted natural language to ensure that the user clearly knows the device action to be performed; this can avoid misunderstandings of intelligent services by users due to language barriers, enhance user trust, improve interaction efficiency, optimize user experience, and achieve clear and efficient intelligent service interaction.

[0146] Optionally, in some embodiments of this application, the list of home device actions may also include environmental data corresponding to the device actions;

[0147] The device control sequence group includes the preceding action, the environmental data corresponding to the preceding action, and the following action;

[0148] The triggering conditions for intelligent services are the preceding actions and the environmental data corresponding to the preceding actions.

[0149] In the above embodiments, the environmental data is environmental parameter data (such as weather, temperature, humidity, light, etc.) related to the device's action execution scenario.

[0150] In the above embodiments, environmental data corresponding to device actions are supplemented into the list of home device actions to enrich the basis for data mining and provide environmental dimension support for accurately identifying scene-related features, avoiding the problem that single action data cannot distinguish different scenes. Furthermore, the device control sequence group is expanded into a combination that includes a preceding action, corresponding environmental data, and a subsequent action. The intelligent service generated based on the device control sequence group upgrades the triggering condition to a combination of the preceding action and environmental data, effectively avoiding scene misjudgment caused by a single action trigger. Through this expansion and optimization method, the triggering of intelligent services can be more in line with the dynamic changes of the home environment, enhancing adaptability and practicality, ensuring the accuracy of scene adaptive construction, and realizing the precision and scene-specific adaptation of intelligent service triggering.

[0151] For example, the list of home device actions includes "Device A01 (window) opened at 08:00:01, corresponding environmental data: rain outside". The action group obtained by mining is "Preceding action: window opened, corresponding environmental data: rain outside, subsequent action: close window". Based on this, a smart service is generated, the trigger condition of which is "the window is detected to be open and it is raining outside", and the trigger result is "automatically close the window"; if the environmental data is "sunny outside", it will not be triggered.

[0152] like Figure 9 As shown, a second aspect of this application discloses a home smart service discovery device 300, comprising:

[0153] The acquisition module 302 is used to acquire a list of home device actions, which includes the device actions of multiple devices in the home and the execution time of the device actions.

[0154] Processing module 304 is used to input the list of home device actions into the smart service mining model to obtain a device control sequence group, which includes the preceding action with the earlier execution time and the following action with the later execution time.

[0155] The generation module 306 is used to generate intelligent services based on the device control sequence group. The triggering condition of the intelligent service is the preceding action, and the triggering result of the intelligent service is to automatically execute the following action.

[0156] like Figure 10 As shown, the third aspect of this application provides a home smart service mining controller 400, including: a memory 402 for storing programs or instructions; and a processor 404 for executing programs or instructions to implement the steps of the home smart service mining method provided in any of the above embodiments, thus achieving all the same technical effects. To avoid repetition, it will not be described again here.

[0157] The fourth aspect of this application provides a readable storage medium having a program or instructions stored thereon. When the program or instructions are executed by a processor, they implement the steps of the home smart service mining method provided in any of the above embodiments, thus achieving all the same technical effects. To avoid repetition, they will not be described again here.

[0158] The fifth aspect of this application provides a computer program product stored in a storage medium. The computer program product is executed by at least one processor to implement the steps of the home intelligent service mining method provided in any of the above-described technical solutions, thus achieving all the same technical effects. To avoid repetition, further details are omitted here. The method can be implemented in various different ways depending on specific features and / or example applications. For example, these methods can be implemented by a combination of hardware, firmware, and / or software. For example, in a hardware implementation, the processor can be implemented in one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, electronic devices, other device units for performing the above functions, and / or combinations thereof.

[0159] A computer-readable storage medium can be a tangible device that holds and stores instructions for use by an instruction execution device. A computer-readable storage medium can be an electronic storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing, but is not limited thereto. A non-exhaustive list of more specific examples of computer-readable storage media includes: portable computer floppy disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), or flash memory, static random-access memory (SRAM), portable optical disc read-only memory (CD-ROM), digital video disc (DVD), memory cards, floppy disks, encoding mechanical devices (e.g., punched cards or grooves with raised structures for recording instructions), and any suitable combination of the foregoing. The computer-readable storage medium used herein should not be construed as the transmission signal itself, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media, or electrical signals transmitted through wires.

[0160] In the claims, description, and accompanying drawings of this application, the term "plural" refers to two or more objects. Unless otherwise explicitly defined, the terms "upper," "lower," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used solely for the convenience of describing this application and simplifying the descriptive process, and are not intended to indicate or imply that the device or element referred to must have the described specific orientation, or be constructed and operated in a specific orientation. Therefore, these descriptions should not be construed as limitations on this application. The terms "connection," "installation," "fixing," etc., should be interpreted broadly. For example, "connection" can be a fixed connection between multiple objects, a detachable connection between multiple objects, or an integral connection; it can be a direct connection between multiple objects or an indirect connection between multiple objects through an intermediate medium. For those skilled in the art, the specific meaning of the above terms in this application can be understood based on the specific circumstances described above.

[0161] In the claims, description, and accompanying drawings of this application, the terms "one embodiment," "some embodiments," "specific embodiment," etc., refer to a specific feature, structure, material, or characteristic described in connection with an embodiment or example, which is included in at least one embodiment or example of this application. In the claims, description, and accompanying drawings of this application, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0162] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for mining smart home services, characterized in that, include: Obtain a list of home device actions, which includes the device actions of multiple devices in the home and the execution time of the corresponding device actions; The list of home device actions is input into the smart service mining model to obtain a device control sequence group, which includes the preceding action with the earlier execution time and the following action with the later execution time. Based on the device control sequence group, an intelligent service is generated. The triggering condition of the intelligent service is the preceding action, and the triggering result of the intelligent service is to automatically execute the following action.

2. The method for mining home intelligent services according to claim 1, characterized in that, The intelligent service mining model is pre-trained through the following process: Obtain a sample home device action list, which includes sample device actions of multiple sample devices in the sample home and the sample execution time corresponding to the sample device actions; From the sample home device action list, positive samples and negative samples are extracted. The positive samples contain at least two first sample device actions, which are executed at adjacent times and are related in the sample home device action list. The negative samples contain at least two second sample device actions, which are executed at adjacent times and are not related in the sample home device action list. Based on the positive and negative samples, the original model is trained to obtain the intelligent service mining model.

3. The method for mining home intelligent services according to claim 2, characterized in that, The association refers to the logical causal relationship or cooperative relationship based on the family life scenario that exists between at least two first sample device actions with adjacent execution times in the sample family device action list; The association includes at least one of the following: The execution of the preceding action directly leads to the execution of the following action; The execution of the preceding action causes a change in the environmental state, and the subsequent action is a passive response to the change in the environmental state.

4. The method for mining home intelligent services according to claim 2, characterized in that, Extracting positive and negative samples from the sampled home device action list includes: A sliding time window of predetermined duration is slid through the list of home device actions to obtain multiple device action groups. The execution time of the device action in each device action group falls within the sliding time window. After adding prompt words to the device action group, input it into the large model to obtain the judgment correlation and confidence probability between each device action in the device action group; Based on the determined correlation and the confidence probability, the positive sample and the negative sample are selected from the plurality of device action groups.

5. The method for mining home intelligent services according to claim 4, characterized in that, The step of selecting the positive and negative samples from the plurality of device action groups based on the determined correlation and the confidence probability includes: If it is determined that there is a correlation between the various device actions in the device action group and the confidence probability is greater than the first threshold, then the device action group is selected as the positive sample. If it is determined that there is no correlation between the various device actions in the device action group and the confidence probability is greater than the first threshold, then the device action group is selected as the negative sample.

6. The method for mining home intelligent services according to claim 2, characterized in that, After extracting positive and negative samples from the sampled home device action list, the training process of the intelligent service mining model further includes: If the number of positive samples or negative samples is less than a predetermined threshold, the extracted positive samples or negative samples are input into the sample expansion model for expansion, so that the number of positive samples and negative samples is not less than the predetermined threshold.

7. The method for mining home intelligent services according to claim 2, characterized in that, The process of training the original model based on the positive and negative samples to obtain the intelligent service mining model includes: Memory summary information is extracted from the positive and negative samples within the first time period, where the first time period is a predetermined duration prior to the training time point of the original model. Based on the positive samples, the negative samples, and the memory summary information, the original model is trained to obtain the intelligent service mining model.

8. The method for mining home smart services according to any one of claims 1 to 7, characterized in that, The generation of intelligent services based on the device control sequence group includes: The multiple device control sequence groups obtained within a predetermined period are subjected to synonym identification, and the device control sequence groups identified as having the same meaning are grouped into one category; If the number of device control sequence groups in the class exceeds a predetermined number, the intelligent service is generated based on the device control sequence groups in the class.

9. The method for mining home intelligent services according to any one of claims 1 to 7, characterized in that, After inputting the list of home device actions into the smart service mining model to obtain the device control sequence group, the method further includes: if the user's future action is consistent with the preceding action in the device control sequence group, displaying a first query message, the first query message being used to ask the user whether to execute the subsequent action; After generating intelligent services based on the device control sequence group, the method further includes: automatically executing the subsequent action if the user's future action is consistent with the preceding action in the device control sequence group.

10. The method for mining home intelligent services according to claim 9, characterized in that, The device control sequence group is written in a model language; The display of the first query message includes: The device control sequence group is converted from the model language into natural language. The first query message is displayed based on the device control sequence group converted into the natural language.

11. The method for mining home intelligent services according to any one of claims 1 to 7, characterized in that, The list of home device actions also includes environmental data corresponding to the device actions. The device control sequence group includes the preceding action with the earlier execution time, the environmental data corresponding to the preceding action, and the following action with the later execution time. The triggering condition for the intelligent service is the preceding action and the environmental data corresponding to the preceding action.

12. A home intelligent service discovery device, characterized in that, include: The acquisition module is used to acquire a list of home device actions, which includes the device actions of multiple devices in the home and the execution time of the device actions. The processing module is used to input the list of home device actions into the intelligent service mining model to obtain a device control sequence group, wherein the device control sequence group includes the preceding action with the earlier execution time and the following action with the later execution time. The generation module is used to generate intelligent services based on the device control sequence group. The triggering condition of the intelligent service is the preceding action, and the triggering result of the intelligent service is to automatically execute the following action.

13. A home intelligent service mining controller, characterized in that, include: Memory, used to store programs or instructions; A processor, configured to implement the steps of the home smart service mining method as described in any one of claims 1 to 11 when executing the program or instructions.

14. A readable storage medium having a program or instructions stored thereon, characterized in that, When the program or instructions are executed by the processor, they implement the steps of the home smart service mining method as described in any one of claims 1 to 11.

15. A computer program product, characterized in that, The computer program product is stored in a storage medium, and the computer program product is executed by at least one processor to implement the steps of the home smart service mining method as described in any one of claims 1 to 11.

Images