Equipment control method and device, medium and electronic equipment

Abstract

The invention provides an equipment control method and device, a medium and electronic equipment, and relates to the field of equipment control, and the equipment control method comprises the steps: obtaining natural language data of a target user; the natural language data are input into a first semantic recognition model and a second semantic recognition model respectively, a first intention list output by the first semantic recognition model and a second intention list output by the second semantic recognition model are obtained, and the first semantic recognition model is different from the second semantic recognition model; according to the first intention list and the second intention list, a target intention is determined, and the target intention comprises to-be-controlled target equipment and a target control intention for the target equipment; and controlling the target equipment according to the target control intention. According to the method and the device, the natural language data of the user can be processed in parallel based on the two different semantic recognition models, and the obtained target intention better conforms to the actual intention of the user for equipment control, so that the equipment control is more accurate.

Description

Technical Field

[0001] This disclosure relates to the field of equipment control technology, and in particular to an equipment control method, apparatus, medium and electronic equipment. Background Technology

[0002] With the development of Internet of Things (IoT) technology, more and more devices are being connected to the internet, forming a vast IoT ecosystem. Among related technologies, there are those that allow control of devices within the IoT based on user voice or text input at the control terminal. Summary of the Invention

[0003] This disclosure provides a device control method, apparatus, medium, and electronic device to offer users a more precise device control solution that better matches their actual control intentions.

[0004] According to a first aspect of the present disclosure, a device control method is provided, comprising: Acquire natural language data from the target user; The natural language data is input into a first semantic recognition model and a second semantic recognition model respectively to obtain a first intent list output by the first semantic recognition model and a second intent list output by the second semantic recognition model. The first semantic recognition model and the second semantic recognition model are different. Based on the first intent list and the second intent list, a target intent is determined, wherein the target intent includes the target device to be controlled and the target control intent for the target device; The target device is controlled according to the stated target control intent.

[0005] Optionally, determining the target intent based on the first intent list and the second intent list includes: Each intent in the first intent list and the second intent list is scored to obtain the confidence level of each intent; The target intent is determined based on the confidence level of each intent.

[0006] Optionally, each intent in the first intent list and the second intent list is scored to obtain the confidence level of each intent, including: For each intent, determine the slot information for that intent; Based on the slot information of the intent, determine the target correction value corresponding to the intent; Based on the target correction value, the preset confidence level is corrected to obtain the confidence level of the intent.

[0007] Optionally, determining the target correction value corresponding to the intent based on the slot information of the intent includes: Based on the slot information of the intent, determine at least one of the following: device information, control intent information, and number of slots corresponding to the intent; The target correction value is determined based on at least one of the device information corresponding to the intent, the control intent information, and the number of slots.

[0008] Optionally, determining the target correction value based on the device information corresponding to the intent includes: Determine the matching result between the device corresponding to the device information and the device associated with the target user, and / or obtain the first historical usage frequency of the device corresponding to the device information; The target correction value is determined based on the matching results and / or the first historical usage frequency.

[0009] Optionally, based on the number of slots corresponding to the intent, a target correction value for correcting the preset confidence level is determined, including: Determine the proportion of the number of slots corresponding to the stated intent to the preset number of slots; The target correction value is determined based on the stated proportion.

[0010] Optionally, determining the target correction value based on the control intent information corresponding to the intent includes: Determine whether the control intent corresponding to the control intent information is a pre-set easily confused intent, and / or obtain the second historical usage frequency of the control intent corresponding to the control intent information; The target correction value is determined based on the confusing intent and / or the second historical usage frequency.

[0011] Optionally, determining the target intent based on the confidence level of each intent includes: Based on the confidence level of each intent, the intent with the highest confidence level in the first intent list and the second intent list is determined as the target intent.

[0012] Optionally, determining the target intent based on the confidence level of each intent includes: Based on the confidence level of each intent, determine whether there is an intent of a target type, wherein the confidence level of the intent of the target type is greater than or equal to a confidence level threshold; When it is determined that an intent of the target type exists, the intent with the highest confidence among the intents of the target type is determined as the target intent.

[0013] Optionally, the method further includes: When it is determined that the intent of the target type does not exist, a prompt message is output to indicate that semantic recognition has failed.

[0014] Optionally, the first semantic recognition model is used to obtain the first intent list in the following manner: Semantic recognition is performed on the natural language data to obtain key information units; According to the preset mapping table, the key information units are matched to obtain the standard intent information corresponding to each key information unit. The standard intent information is information in the standard intent used for device control. The first intent list is obtained based on the standard intent information corresponding to each key information unit.

[0015] Optionally, natural language data of the target user can be obtained, including: Acquire the voice data of the target user; The speech data is preprocessed to obtain the natural language data of the target user. The preprocessing operation includes at least a format conversion operation to convert the speech data into text data.

[0016] Optionally, after controlling the target device according to the target control intent, the method further includes: The first semantic recognition model and the second semantic recognition model are trained based on the feedback data of the target user to the target intent and the execution results of controlling the target device.

[0017] According to a second aspect of the present disclosure, a device control apparatus is provided, comprising: The acquisition module is configured to acquire the natural language data of the target user. The acquisition module is configured to input the natural language data into a first semantic recognition model and a second semantic recognition model respectively, and obtain a first intent list output by the first semantic recognition model and a second intent list output by the second semantic recognition model, wherein the first semantic recognition model and the second semantic recognition model are different; The determination module is configured to determine a target intent based on the first intent list and the second intent list, wherein the target intent includes a target device to be controlled and a target control intent for the target device; The control module is configured to control the target device according to the target control intent.

[0018] According to a third aspect of the present disclosure, a computer-readable storage medium is provided, having stored thereon computer program instructions that, when executed by a processor, implement the device control method provided in the first aspect of the present disclosure.

[0019] According to a fourth aspect of the present disclosure, an electronic device is provided, comprising: Storage device for storing computer programs; An execution device is used to execute the computer program to implement the device control method provided in the first aspect of the embodiments of this disclosure.

[0020] The technical solutions provided by the embodiments of this disclosure may include the following beneficial effects: This disclosure obtains a first intent list and a second intent list by inputting the natural language data of the target user into a first semantic recognition model and a second semantic recognition model, respectively. Based on the first intent list and the second intent list, the target intent is determined, and then the target device to be controlled is controlled according to the control intent in the target intent. In this way, the user's natural language data can be processed in parallel based on two different semantic recognition models, resulting in a target intent that more closely matches the user's actual intention to control the device, thus making device control more precise.

[0021] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0022] The accompanying drawings, which are incorporated in and form a part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure.

[0023] Figure 1 This is a flowchart illustrating a device control method according to an exemplary embodiment.

[0024] Figure 2 This is a block diagram illustrating a device control system according to an exemplary embodiment.

[0025] Figure 3 This is a flowchart illustrating another device control method according to an exemplary embodiment.

[0026] Figure 4 This is a flowchart illustrating another device control method according to an exemplary embodiment.

[0027] Figure 5 This is a block diagram illustrating a device control apparatus according to an exemplary embodiment.

[0028] Figure 6 This is a block diagram illustrating an electronic device according to an exemplary embodiment.

[0029] Figure 7 This is a block diagram illustrating an apparatus for device control according to an exemplary embodiment. Detailed Implementation

[0030] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0031] With the development of Internet of Things (IoT) technology, more and more devices are being connected to the Internet, forming a vast IoT ecosystem. Users can control the devices they wish to control via voice or manually entered text.

[0032] For example, a user can input voice or text into an IoT device control system. The system uses Natural Language Processing (NLP) to recognize the user's semantic information and sends the processed semantic information to the IoT platform. Based on the information about the IoT devices that the user has bound and their capabilities, the IoT platform uses preset rules to determine the device to be controlled and its control information from a fixed instruction set. Then, the IoT platform sends the control information to the corresponding device to be controlled, completing the entire control chain.

[0033] However, existing IoT device control systems typically rely on fixed instruction sets or limited semantic understanding capabilities to identify control commands, which restricts support for long-tail IoT devices. Furthermore, older IoT device control systems may suffer from inaccurate control of emerging IoT devices due to outdated updates.

[0034] Reference Figure 1 , Figure 1 This is a flowchart illustrating a device control method according to an exemplary embodiment, such as... Figure 1 As shown, the equipment control method includes the following steps.

[0035] In step S101, the natural language data of the target user is obtained.

[0036] In step S102, natural language data is input into the first semantic recognition model and the second semantic recognition model respectively to obtain the first intention list output by the first semantic recognition model and the second intention list output by the second semantic recognition model. The first semantic recognition model and the second semantic recognition model are different.

[0037] In step S103, the target intent is determined according to the first intent list and the second intent list. The target intent includes the target device to be controlled and the target control intent for the target device.

[0038] In step S104, the target device is controlled according to the target control intention.

[0039] For example, the target user is a user who outputs data such as voice or text, intending to control an IoT device associated with the target user. Natural language data may include voice data or text data, or it may be preprocessed voice or text data.

[0040] For example, the first semantic recognition model and the second semantic recognition model are different models. These two models may be based on different algorithms or training data to improve the overall recognition accuracy and robustness. For instance, the first semantic recognition model may be a natural language processing model such as a Recurrent Neural Network (RNN), a Long Short-Term Memory (LSTM) network, or a Transformer, while the second semantic recognition model may be a Large Language Model (LLM) such as ChatGPT or BERT.

[0041] For example, one of the first semantic recognition model and the second semantic recognition model can be the original natural language processing model of the IoT device control system, while the other adopts a different natural language processing model. Alternatively, both the first semantic recognition model and the second semantic recognition model can adopt a different natural language processing model than the original natural language processing model of the IoT device control system; this is not limited here.

[0042] For example, the first semantic recognition model can obtain a first list of intents based on natural language data, and the second semantic recognition model can obtain a second list of intents based on natural language data. The semantics may be relatively clear or unclear. When the semantics are unclear, the recognition results of the semantic recognition model may be multiple; that is, the first list of intents or the second list of intents may contain only one intent or multiple intents.

[0043] For example, an intent can be an action performed by a user with a specific purpose or need conveyed through natural language, such as "turn on the light" or "set an alarm clock." A target intent can be the intent determined from a first intent list and a second intent list that most likely matches the target user's actual semantic intent. Based on each intent in the first and second intent lists, an evaluation can be conducted to determine the target intent that most closely matches the target user's actual intent. Then, based on the target intent, which includes the target device to be controlled and the target control intent for the target device, the target device can be controlled.

[0044] For example, each intent includes the device the user intends to control and the control intent for that device. The target intent includes the target device to be controlled and the target control intent for that target device. The control intent can include the control action to be performed on the device, the timing of the control action, the execution duration, etc. For example, if the user inputs the instruction "Turn on the air conditioner at 6 PM, set the temperature to 26 degrees Celsius, minimum fan speed, and turn it off after 30 minutes," then "6 PM" is the control timing, "air conditioner" is the device to be controlled, "turn on the air conditioner," "set the temperature to 26 degrees Celsius, minimum fan speed" are the control actions to be performed on the device, and "turn it off after 30 minutes" is the execution duration.

[0045] It is understood that the examples here are only used to illustrate the information contained in control intents, and do not mean that control intents include all the information contained therein, nor do they mean that they do not include other information besides the examples above. There are no restrictions on control intents here.

[0046] This disclosure obtains a first intent list and a second intent list by inputting the natural language data of the target user into a first semantic recognition model and a second semantic recognition model, respectively. Based on the first intent list and the second intent list, the target intent is determined, and then the target device to be controlled is controlled according to the control intent in the target intent. In this way, the user's natural language data can be processed in parallel based on two different semantic recognition models, resulting in a target intent that more closely matches the user's actual intention to control the device, thus making device control more precise.

[0047] As an optional embodiment, determining the target intent based on a first intent list and a second intent list includes: Each intent in the first intent list and the second intent list is scored to obtain the confidence level of each intent; Determine the target intent based on the confidence level of each intent.

[0048] For example, confidence level can be used to characterize how credible it is that the intent is the user's actual intention. The higher the confidence level of an intent, the greater the likelihood that the intent is the user's true intention. Confidence level scores are typically a value between 0 and 1, with values ​​closer to 1 indicating higher confidence.

[0049] For example, to score each intent in the first intent list and the second intent list, another model with a scoring mechanism can be used to receive the first and second intent lists output by the first and second semantic recognition models, and score the intents in the first and second intent lists respectively based on preset scoring rules. Alternatively, the first semantic recognition model can score each intent in the first intent list based on preset scoring rules, and the second semantic recognition model can score each intent in the second intent list based on preset scoring rules, thus obtaining the confidence level of each intent.

[0050] For example, the target intent can be determined based on the intents in two lists and their confidence scores. Comparison, merging, or voting mechanisms can be used to select the most likely intent. For instance, if the highest-confidence intents in the intent lists output by the two models differ, the intent with the highest confidence in both lists might be chosen as the target intent. Similarly, if both models identify the same intent with high confidence, that intent can be used as the target intent. Furthermore, if all intents in the two lists are dissimilar, the target intent can be determined through voting by multiple models with scoring mechanisms.

[0051] Here, by using two different semantic recognition models, user instructions can be understood from multiple perspectives, improving the accuracy of intent recognition. Furthermore, if one semantic recognition model misidentifies the intent, while the other model can still provide the correct intent, the robustness of the system is enhanced. This disclosure can also be easily extended to new devices and actions; the semantic recognition models can be appropriately trained, offering flexibility and scalability.

[0052] As an optional embodiment, each intent in the first intent list and the second intent list is scored to obtain the confidence level of each intent, including: For each intent, determine the slot information for that intent; Based on the slot information of the intent, determine the target correction value corresponding to the intent; Based on the target correction value, the preset confidence level is adjusted to obtain the confidence level of the intent.

[0053] For example, in natural language processing, a slot refers to a key information field that constitutes an intent, such as time, location, or device name. These fields are typically used to further refine and execute the intent. For instance, in the intent "Turn on the air conditioner at 6 PM and turn it off half an hour later," "turn on" and "turn off" might be two slots.

[0054] For example, the target correction value is used to adjust the preset confidence level. Based on the slot information of the intent, the target correction value corresponding to the intent is determined. Then, the preset confidence level is adjusted according to the target correction value to obtain the confidence level of the intent. The preset confidence level can be a basic confidence value, which can be any number from 1 to 0, such as 0.5. Alternatively, the preset confidence level can be set according to the actual situation. For example, if the target correction value only contains deductions, the confidence level can be 1; if the target correction value only contains bonuses, the confidence level can be 0; if the target correction value contains both bonuses and deductions, the confidence level can be 0.5, 0.6, 0.8, etc.

[0055] For example, for each intent, the slot information of the intent can be determined based on each slot hit in the intent. Then, based on the slot information of the intent and some preset rules, a target correction value corresponding to the intent can be determined to adjust the preset confidence level to more accurately reflect the user's intent. By correcting the preset confidence level according to the target correction value, the confidence level of the intent can be obtained.

[0056] This disclosure determines the target correction value corresponding to the intent by using the slot information of the intent, and corrects the preset reliability. Even when there is ambiguity or vagueness in the natural language data, it can more accurately identify the user's intent and improve the accuracy of user intent recognition.

[0057] As an optional embodiment, determining the target correction value corresponding to the intent based on the slot information of the intent includes: Based on the slot information of the intent, determine at least one of the following: the device information corresponding to the intent, the control intent information, and the number of slots; The target correction value is determined based on at least one of the following: the device information corresponding to the intent, the control intent information, and the number of slots.

[0058] For example, device information is data related to a device, such as device type, device name, device parameters, and device functions. In an IoT environment, device information may include identifiers such as the device ID or device name, device location, current status, or device capabilities, etc. A slot refers to a key information field that constitutes an intent. Correspondingly, the number of slots represents the number of key information fields that constitute an intent. For example, "turn on the bedroom light" includes three slots: "turn on," "bedroom," and "light."

[0059] For example, the control intent information corresponding to the intent is the control intent for the device to be controlled in the intent, including the control action for the device, the control time for the device, the setting target for the device's functional parameters, etc. For example, in "Turn on the air conditioner and set the temperature to 27 degrees Celsius in 10 minutes", "10 minutes later" is the control time for the device, "turn on" is the control action, "temperature" is the functional parameter, and "27 degrees Celsius" is the setting target for the device's functional parameters.

[0060] For example, by determining the device information, the number of slots, and the control intent information based on the slot information in the intent, it is possible to identify the device, control intent, and number of slots associated with the intent. Based on at least one of the device information, the number of slots, and the control intent information, a target correction value can be calculated to adjust the preset confidence level to obtain a more accurate intent confidence level.

[0061] This disclosure uses the equipment information, number of slots, and control intent information in the slot information to determine the target correction value for adjusting the preset confidence level, which can make the obtained confidence level more accurate, so as to more accurately determine whether the intent is the user's actual intent.

[0062] As an optional embodiment, determining the target correction value based on the device information corresponding to the intent includes: Determine the matching result between the device corresponding to the device information and the device associated with the target user, and / or obtain the first historical usage frequency of the device corresponding to the device information; The target correction value is determined based on the matching results and / or the first historical usage frequency.

[0063] For example, the matching result is whether the device corresponding to the device information is associated with the target user. If the matching result is yes, it can be assumed that the device to be controlled in the intent is a device associated with the target user, and the target user can control the device by outputting control commands through voice or text. If the matching result is no, it can be assumed that the device to be controlled in the intent is not a device associated with the target user, the target user may not have associated the device, or the device that the target user actually wants to control is not the device in the intent.

[0064] For example, the initial historical usage frequency of a device can be obtained based on the target user's historical cache data in the storage module. Each time the device is controlled, the usage count for that device can be increased by 1. For different devices, the frequency with which users control the device via voice output commands varies. The higher the control frequency of a device, the more likely the device the user needs to control this time is that device with a high control frequency. For example, the usage frequency of devices like lights and air conditioners may be higher than that of devices like rice cookers.

[0065] For example, the target correction value can be determined based on the correction value corresponding to the matching result or the correction value corresponding to the first historical usage frequency. Additionally, correction values ​​corresponding to other correction items are also included to determine the final target correction value.

[0066] For example, taking a preset confidence level of 1 as an example, for the matching result, if the device corresponding to the device information represented by the matching result is associated with the target user, no deduction operation can be performed. If the device corresponding to the device information represented by the matching result is not associated with the target user, a deduction operation can be performed, and the first correction value is deducted.

[0067] For example, for a first historical usage frequency, a second correction value can be obtained based on the historical usage frequency of each device. For instance, if the first historical usage frequency exceeds a preset frequency threshold, the second correction value is 0; if the first historical usage frequency does not exceed the preset frequency threshold, the second correction value is 0.1. Another example is to sort multiple devices by their historical usage frequencies from highest to lowest, and determine the second correction value by multiplying the weight corresponding to the rank by a preset deduction value. For instance, if the historical usage frequency of an air conditioner corresponds to rank 2, then based on the weight of 0.1 for the 2nd rank and the preset deduction value of 0.1, the second correction value is determined to be 0.01; if the rice cooker corresponds to rank 6, then based on the weight of 0.8 for the 6th rank, the second correction value is determined to be 0.08.

[0068] As an optional embodiment, a target correction value for correcting the preset confidence level is determined based on the number of slots corresponding to the intent, including: Determine the proportion of the number of slots corresponding to the intent to the preset number of slots; The target adjustment value is determined based on the proportion.

[0069] For example, the preset number of slots for device control can be set according to actual conditions. Taking the user's voice data as "Turn on the master bedroom air conditioner at 6 PM, set the temperature to 26 degrees Celsius, and turn it off after running for 30 minutes" as an example, the number of slots corresponding to this intent is 5, namely the device "air conditioner", the device location "master bedroom", the control actions "on" and "off", the control time "6 PM", and the control parameter "set the temperature to 26 degrees Celsius". If the preset number of slots is set to 5, the proportion corresponding to the above voice data is 1. If the user's voice data is adjusted to "turn on the air conditioner", the corresponding slot information includes 2 slots, namely the device "air conditioner" and the control action "turn on".

[0070] As shown in the example above, it is clear that the more slots matched in the user's natural language data, the clearer the user's intent, and the easier it is to obtain a control strategy that better aligns with the user's final intention based on the user's commands. Whether the target correction value is determined by deduction or addition, the higher the proportion of slots corresponding to the intent in the preset number of slots, the higher the confidence level of that intent.

[0071] For example, a third correction value corresponding to the number of slots can be determined based on the weight of the proportion of the number of slots corresponding to the intent in the preset number of slots, and a preset deduction or bonus value. Taking the deduction value as an example, if the proportion is 0.5, then based on the weight value of 0.5 for the proportion of 0.5 and the preset deduction value of 0.3, the correction value corresponding to the intent is determined to be 0.15; if the proportion is 0.8, then based on the weight value of 0.1 for the proportion of 0.8 and the preset deduction value of 0.3, the correction value corresponding to the intent is determined to be 0.03.

[0072] As an optional embodiment, determining the target correction value based on the control intent information corresponding to the intent includes: Determine whether the control intent corresponding to the control intent information is a pre-set easily confused intent, and / or obtain the second historical usage frequency of the control intent corresponding to the control intent information; The target correction value is determined based on the confusing intent and / or the frequency of second-historical usage.

[0073] For example, semantic recognition models have difficulty understanding easily confused intents when it comes to control intents, meaning the reliability of the recognition results is low. When the control intent is a pre-defined easily confused intent, the confidence score can be reduced.

[0074] For example, the second historical usage frequency of a control intent can be obtained based on the target user's historical cache data in the storage module. Each time the device is controlled, the usage count of the corresponding control method for that device can be increased by 1. Different control intents have different frequencies; the higher the control intent frequency, the more likely the user's current control intent for the device is that high-frequency control intent. For example, for an air conditioner, the frequency of adjusting temperature and fan speed parameters is generally greater than the frequency of adjusting airflow direction parameters.

[0075] For example, a target correction value can be determined based on whether the control intent corresponds to a pre-defined correction value for easily confused intents, and / or a correction value corresponding to the second historical usage frequency of the control intent. Additionally, correction values ​​corresponding to other correction items are also included to determine the final target correction value.

[0076] For example, the language corresponding to easily confused intents is "how to save electricity for the refrigerator" and the corresponding "refrigerator energy-saving mode". Deduction or addition values ​​can be set for each of the pre-defined easily confused intents. For instance, easily confused intents include "is the refrigerator in energy-saving mode?", "how to save electricity for the refrigerator", and "turn on refrigerator energy-saving mode". Since "turn on refrigerator energy-saving mode" is clearly a control command and relatively easy to understand, deduction values ​​of 0.3, 0.3, and 0.1 can be set respectively based on the model's comprehensibility. Then, a fourth correction value corresponding to the easily confused intent can be obtained. Alternatively, when the intent is determined to be easily confused, the preset deduction value is used as the fourth correction value.

[0077] For example, for the second historical usage frequency, a fifth correction value corresponding to the second historical usage frequency can be obtained based on the historical usage frequency corresponding to each control intention. For instance, when the fifth historical usage frequency exceeds a preset frequency threshold, the fifth correction value is 0; when the fifth historical usage frequency does not exceed the preset frequency threshold, the fifth correction value is 0.1. As another example, based on sorting the historical usage frequencies of multiple devices from high to low, the fifth correction value is determined by multiplying the weight corresponding to the position by a preset deduction value. For example, if the historical usage frequency of the air conditioner corresponds to position 2, then based on the weight of the 2nd position (0.1) and the preset deduction value (0.1), the fifth correction value is determined to be 0.01; if the rice cooker's position is 6, then based on the weight of the 6th position (0.8), the fifth correction value is determined to be 0.08.

[0078] For example, the target correction value can be determined by comprehensively considering the first correction value corresponding to the matching result, the second correction value corresponding to the first historical usage frequency, the third correction value corresponding to the number of slots, the fourth correction value corresponding to the easily confused intent, and the fifth correction value corresponding to the second historical usage frequency. In this way, more accurate, faster, and more personalized device control services can be provided, thereby improving the user experience.

[0079] As an optional embodiment, determining the target intent based on the confidence level of each intent includes: Based on the confidence level of each intent, the intent with the highest confidence level in the first intent list and the second intent list is determined as the target intent.

[0080] For example, in statistics, confidence level is used to represent the degree of credibility of an estimate of a parameter. Here, confidence level is used to characterize the credibility of an intent being the user's actual intent. The higher the confidence level of an intent, the higher the credibility of that intent being the user's actual intent. Therefore, the intent with the highest confidence level in the first intent list and the second intent list can be identified as the target intent.

[0081] Here, the target intent is determined by combining the first intent list and the second intent list. If the intent identified by one of the semantic recognition models has low accuracy, an intent that is more likely to match the user's actual intent can be provided based on the intent output by the other model with higher accuracy, thereby enhancing the robustness of the system and the accuracy of device control.

[0082] As an optional embodiment, determining the target intent based on the confidence level of each intent includes: Based on the confidence level of each intent, determine whether there is an intent of the target type, where the confidence level of the intent of the target type is greater than or equal to the confidence level threshold; When an intent of a target type is identified, the intent with the highest confidence among the target type intents is identified as the target intent.

[0083] For example, the confidence threshold can be set according to actual needs. There may also be cases where the confidence levels of intents in both intent lists are low. In this case, we can first determine the intents with confidence levels greater than or equal to the confidence threshold. When there is an intent of the target type with a confidence level greater than or equal to the confidence threshold, the intent with the highest confidence level in both the first and second intent lists can be identified as the target intent.

[0084] As an optional embodiment, the method further includes: when it is determined that there is no intent of the target type, outputting a prompt message to indicate that semantic recognition has failed.

[0085] For example, when there is no intent of the target type with a confidence level greater than or equal to the confidence threshold, the confidence levels of the intents in both intent lists are low. In this case, instead of controlling the device based on the intent with the highest confidence level, a prompt message can be output to inform the user that semantic recognition has failed. The user can also be further prompted to input more semantically clear device control commands.

[0086] As an optional embodiment, the first semantic recognition model is used to obtain a first list of intents in the following manner: Semantic recognition is performed on natural language data to obtain key information units; According to the preset mapping table, the key information units are matched to obtain the standard intent information corresponding to each key information unit. The standard intent information is the information in the standard intent used for device control. Based on the standard intent information corresponding to each key information unit, a first intent list is obtained.

[0087] For example, semantic recognition is a process in natural language processing used to understand the meaning of user-input text and identify key information within it. Key information units can represent text fragments extracted during semantic recognition, which contain the essential information needed to perform intent recognition.

[0088] For example, a preset mapping table is a predefined database or mapping system used to map key information units to standard intent information. The key information units cannot be directly used by the IoT platform to control devices. The standard intent information in the preset mapping table corresponds to standard intents within the device control dimension; that is, the standard intent information can be used by the IoT platform for identification, enabling the IoT platform to send corresponding control commands to the device based on the standard intent information. The standard intent information can be in the form of instructions; that is, the preset mapping table stores instruction sets, including standard instructions, which are a set of standard control function instructions. Different slots can correspond to different instruction sets. Based on standard instructions, control commands can be received and executed in a unified manner, thereby achieving device operation and monitoring.

[0089] For example, by filling the corresponding standard intent information for each identified key information unit into the corresponding slot in the preset slots, a complete intent can be obtained. The key information unit can be the semantics of the user command determined by the semantic recognition model. However, semantic recognition may contain errors. For instance, if the user inputs "turn on the bedroom light," the semantic recognition might result in incorrect key information units such as "increase brightness by 10%" or "decrease brightness by 10%." Therefore, many different intents can be generated, forming a first intent list.

[0090] For example, the second semantic recognition model can perform knowledge retrieval based on its powerful semantic recognition capabilities, and then directly output multiple possible user intents to generate a second intent list.

[0091] As an optional embodiment, acquiring the natural language data of the target user includes: Acquire the target user's voice data; Preprocessing operations are performed on the voice data to obtain the natural language data of the target user. The preprocessing operations include at least a format conversion operation to convert the voice data into text data.

[0092] For example, to facilitate operation by the target user, the target user's voice data can be acquired based on the sound acquisition device, and the voice data can be preprocessed to obtain the target user's natural language data.

[0093] For example, preprocessing operations may include format conversion operations that convert speech data into text data, segmentation operations that perform semantic pre-segmentation on text data, data cleaning operations that remove characters that are irrelevant to device control, vectorization operations that convert text data into numerical form, etc., without limitation.

[0094] As an optional embodiment, after controlling the target device according to the target control intent, the method further includes: The first semantic recognition model and the second semantic recognition model are trained based on the feedback data of the target user on the target intent and the execution results of controlling the target device.

[0095] For example, feedback data may include data on the target user's satisfaction, accuracy, or other relevant reactions to the control actions performed by the system. This can be achieved by displaying a user satisfaction survey interface on the control device to obtain feedback data on the target user's intent. Execution results may include the actual effect of controlling the target device, including success, failure, or other status information. The user feedback data and the results of controlling the target device are used as training datasets to adjust and optimize the first semantic recognition model and the second recognition model.

[0096] The feedback-based continuous training mechanism ensures continuous improvement over time, enabling more accurate understanding and response to user intent. Furthermore, by analyzing execution results, it can identify and correct potential problems that lead to errors or failures, thereby reducing future mistakes.

[0097] As a concrete example, see Figure 2 This disclosure also provides an Internet of Things (IoT) device control system, including a semantic processing module, an evaluation module, and an execution module. The semantic processing module includes a first semantic processing model and a second semantic processing model. The first semantic processing model can be used to output a first intent list based on the natural language data of the target user. The second semantic processing model can be used to output a second intent list based on the natural language data of the target user. The evaluation module can be used to determine a target intent based on the first and second intent lists. The target intent includes the target device to be controlled and the target control intent for the target device. The execution module can be used to control the target device according to the target control intent.

[0098] See Figure 3 The IoT device control system may also include a preprocessing module and a data feedback module. The preprocessing module can be used to preprocess the voice or text commands of the target user. The data feedback module can train the first semantic processing model and the second semantic processing model based on the user's feedback and the execution results of the device controlled by the execution module.

[0099] See Figure 4 The IoT device control system proposed in this invention combines the semantic recognition model in traditional control systems with a semantic understanding system based on a large model. By introducing a large model, such as the GPT series model, the system can understand more complex and diverse user requests and map these requests to specific IoT device operations. The two semantic understanding systems work in parallel, processing and scoring user requests independently, and finally selecting the most appropriate response for execution through comprehensive evaluation.

[0100] The first semantic recognition model performs traditional NLP processing on the natural language data output from the preprocessing module to obtain key information units containing device information, control intent information, etc. Then, based on a preset mapping table, it converts each key information unit into an instruction that the IoT platform can understand. During the conversion, one key information unit may correspond to multiple corresponding instructions. Therefore, a first intent list can be generated based on the possible instructions and the instructions corresponding to different slots. The second semantic recognition model, based on knowledge retrieval, identifies possible intent information in the natural language data output from the preprocessing module and generates a second intent list.

[0101] The evaluation module evaluates the first and second intent lists based on preset evaluation rules, identifies the target intent that is most likely to match the user's actual intent, and then sends the target intent to the execution module. The execution module controls the target device according to the target control intent in the target intent.

[0102] Traditional control systems and a model-based semantic recognition system process user requests in parallel. The traditional control system relies on a pre-set mapping table for processing, while the model-based semantic recognition system relies on the large model for semantic understanding. By comparing the processing results and confidence levels of the two systems, the optimal solution is selected for execution after comprehensive evaluation.

[0103] The dual-system collaborative mechanism involves simultaneously transmitting user control requests to both a traditional control system and a large-scale model-based semantic recognition system. The traditional control system processes the request using pre-defined rules, while the large-scale model semantic recognition system leverages its powerful semantic recognition capabilities. Both systems can then independently score the recognized intent results, considering factors such as parsing accuracy and operational feasibility. Scoring can be performed using the semantic recognition model itself or based on other neural network models suitable for scoring. A comprehensive evaluation module, based on the scores from both systems and a pre-defined algorithm, determines which system's result to adopt and optimizes the processing strategy to adapt to different request types. After selecting the optimal result, the corresponding control operation is executed, including the device to be controlled and the control intent. The execution results are then used for feedback learning between the semantic recognition models of both systems, continuously optimizing system performance.

[0104] This disclosure utilizes large-scale deep learning technology to improve the system's ability to understand user natural language requests, especially long-tail requests that are difficult to match with specific device operations. This significantly enhances the flexibility and versatility of IoT device control systems, particularly for long-tail IoT devices that traditional control systems struggle to support, providing more precise and personalized control solutions. Furthermore, the understanding capabilities and decision-making logic of the large-scale model can be continuously optimized through user feedback and execution results, improving the support and accuracy for controlling long-tail IoT devices.

[0105] Figure 5 This is a block diagram illustrating a device control apparatus according to an exemplary embodiment. (Refer to...) Figure 5 The device includes an acquisition module 501, a obtaining module 502, a determining module 503, and a control module 504.

[0106] The acquisition module 501 is configured to acquire the natural language data of the target user. The module 502 is configured to input natural language data into the first semantic recognition model and the second semantic recognition model respectively, and obtain a first intention list output by the first semantic recognition model and a second intention list output by the second semantic recognition model. The first semantic recognition model and the second semantic recognition model are different. The determination module 503 is configured to determine the target intent based on the first intent list and the second intent list, wherein the target intent includes the target device to be controlled and the target control intent for the target device; Control module 504 is configured to control the target device according to the target control intent.

[0107] As an optional embodiment, the determining module 503 is also configured to: Each intent in the first intent list and the second intent list is scored to obtain the confidence level of each intent; Determine the target intent based on the confidence level of each intent.

[0108] As an optional embodiment, the determining module 503 is also configured to: For each intent, determine the slot information for that intent; Based on the slot information of the intent, determine the target correction value corresponding to the intent; Based on the target correction value, the preset confidence level is adjusted to obtain the confidence level of the intent.

[0109] As an optional embodiment, the determining module 503 is also configured to: Based on the slot information of the intent, determine at least one of the following: the device information corresponding to the intent, the control intent information, and the number of slots; The target correction value is determined based on at least one of the following: the device information corresponding to the intent, the control intent information, and the number of slots.

[0110] As an optional embodiment, the determining module 503 is also configured to: Determine the matching result between the device corresponding to the device information and the device associated with the target user, and / or obtain the first historical usage frequency of the device corresponding to the device information; The target correction value is determined based on the matching results and / or the first historical usage frequency.

[0111] As an optional embodiment, the determining module 503 is also configured to: Determine the proportion of the number of slots corresponding to the intent to the preset number of slots; The target adjustment value is determined based on the proportion.

[0112] As an optional embodiment, the determining module 503 is also configured to: Determine whether the control intent corresponding to the control intent information is a pre-set easily confused intent, and / or obtain the second historical usage frequency of the control intent corresponding to the control intent information; The target correction value is determined based on the confusing intent and / or the frequency of second-historical usage.

[0113] As an optional embodiment, the determining module 503 is also configured to: Based on the confidence level of each intent, the intent with the highest confidence level in the first intent list and the second intent list is determined as the target intent.

[0114] As an optional embodiment, the determining module 503 is also configured to: Based on the confidence level of each intent, determine whether there is an intent of the target type, where the confidence level of the intent of the target type is greater than or equal to the confidence level threshold; When an intent of a target type is identified, the intent with the highest confidence among the target type intents is identified as the target intent.

[0115] As an optional embodiment, the device control unit is also configured to: When it is determined that there is no intent of the target type, a prompt message is output to indicate that semantic recognition has failed.

[0116] As an optional embodiment, the first semantic recognition model is used to obtain a first list of intents in the following manner: Semantic recognition is performed on natural language data to obtain key information units; According to the preset mapping table, the key information units are matched to obtain the standard intent information corresponding to each key information unit. The standard intent information is the information in the standard intent used for device control. Based on the standard intent information corresponding to each key information unit, a first intent list is obtained.

[0117] As an optional embodiment, the acquisition module 501 is also configured to: Acquire the target user's voice data; Preprocessing operations are performed on the voice data to obtain the natural language data of the target user. The preprocessing operations include at least a format conversion operation to convert the voice data into text data.

[0118] As an optional embodiment, the device control unit is also configured to: The first semantic recognition model and the second semantic recognition model are trained based on the feedback data of the target user on the target intent and the execution results of controlling the target device.

[0119] Regarding the device control apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the device control method, and will not be elaborated upon here.

[0120] Based on the same inventive concept, this disclosure also provides a computer-readable storage medium having computer program instructions stored thereon, which, when executed by a processor, implement the device control method of this disclosure.

[0121] Based on the same inventive concept, this disclosure also provides an electronic device, comprising: Storage device for storing computer programs; An execution device is used to execute the computer program to implement the device control method of this disclosure.

[0122] Figure 6 This is a block diagram illustrating an electronic device 600 according to an exemplary embodiment. For example, the electronic device 600 may be a mobile phone, computer, digital broadcasting terminal, messaging device, game console, tablet device, medical device, fitness equipment, personal digital assistant, etc.

[0123] Reference Figure 6 The electronic device 600 may include one or more of the following components: processing component 602, memory 604, power supply component 606, multimedia component 608, audio component 610, input / output interface 612, sensor component 614, and communication component 616.

[0124] Processing component 602 typically controls the overall operation of electronic device 600, such as operations associated with display, telephone calls, data communication, camera operation, and recording operations. Processing component 602 may include one or more processors 620 to execute instructions to perform all or part of the steps of the methods described above. Furthermore, processing component 602 may include one or more modules to facilitate interaction between processing component 602 and other components. For example, processing component 602 may include a multimedia module to facilitate interaction between multimedia component 608 and processing component 602.

[0125] Memory 604 is configured to store various types of data to support the operation of electronic device 600. Examples of such data include instructions for any application or method operating on electronic device 600, contact data, phonebook data, messages, pictures, videos, etc. Memory 604 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0126] Power supply component 606 provides power to various components of electronic device 600. Power supply component 606 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to electronic device 600.

[0127] Multimedia component 608 includes a screen that provides an output interface between the electronic device 600 and the user. In some embodiments, the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touchscreen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touches, swipes, and gestures on the touch panel. The touch sensors may sense not only the boundaries of the touch or swipe action but also the duration and pressure associated with the touch or swipe operation. In some embodiments, multimedia component 608 includes a front-facing camera and / or a rear-facing camera. When the electronic device 600 is in an operating mode, such as a shooting mode or a video mode, the front-facing camera and / or the rear-facing camera may receive external multimedia data. Each front-facing camera and rear-facing camera may be a fixed optical lens system or have focal length and optical zoom capabilities.

[0128] Audio component 610 is configured to output and / or input audio signals. For example, audio component 610 includes a microphone (MIC) configured to receive external audio signals when electronic device 600 is in an operating mode, such as call mode, recording mode, and voice recognition mode. The received audio signals may be further stored in memory 604 or transmitted via communication component 616. In some embodiments, audio component 610 also includes a speaker for outputting audio signals.

[0129] Input / output interface 612 provides an interface between processing component 602 and peripheral interface modules, such as keyboards, click wheels, buttons, etc. These buttons may include, but are not limited to, home buttons, volume buttons, start buttons, and lock buttons.

[0130] Sensor assembly 614 includes one or more sensors for providing state assessments of various aspects of electronic device 600. For example, sensor assembly 614 can detect the on / off state of electronic device 600, the relative positioning of components such as the display and keypad of electronic device 600, changes in position of electronic device 600 or a component of electronic device 600, the presence or absence of user contact with electronic device 600, orientation or acceleration / deceleration of electronic device 600, and temperature changes of electronic device 600. Sensor assembly 614 may include a proximity sensor configured to detect the presence of nearby objects without any physical contact. Sensor assembly 614 may also include a light sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, sensor assembly 614 may also include an accelerometer, gyroscope, magnetometer, pressure sensor, or temperature sensor.

[0131] Communication component 616 is configured to facilitate wired or wireless communication between electronic device 600 and other devices. Electronic device 600 can access wireless networks based on communication standards, such as WiFi, 2G, or 3G, or combinations thereof. In one exemplary embodiment, communication component 616 receives broadcast signals or broadcast-related information from an external broadcast management system via a broadcast channel. In one exemplary embodiment, communication component 616 also includes a near-field communication (NFC) module to facilitate short-range communication. For example, the NFC module may be implemented based on radio frequency identification (RFID) technology, Infrared Data Association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology, and other technologies.

[0132] In an exemplary embodiment, the electronic device 600 may be implemented by 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, or other electronic components to perform the methods described above.

[0133] In an exemplary embodiment, a non-transitory computer-readable storage medium including instructions is also provided, such as a memory 604 including instructions, which can be executed by a processor 620 of an electronic device 600 to perform the above-described method. For example, the non-transitory computer-readable storage medium may be a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage device, etc.

[0134] In another exemplary embodiment, a computer program product is also provided, the computer program product comprising a computer program executable by a programmable device, the computer program having a code portion for performing the device control method described above when executed by the programmable device.

[0135] Figure 7 This is a block diagram illustrating an apparatus 700 for device control according to an exemplary embodiment. For example, apparatus 700 may be provided as a server. (Refer to...) Figure 7 The device 700 includes a processing component 722, which further includes one or more processors, and memory resources represented by memory 732 for storing instructions, such as application programs, that can be executed by the processing component 722. The application programs stored in memory 732 may include one or more modules, each corresponding to a set of instructions. Furthermore, the processing component 722 is configured to execute instructions to perform the aforementioned device control method.

[0136] The device 700 may also include a power supply component 727 configured to perform power management of the device 700, a wired or wireless network interface 750 configured to connect the device 700 to a network, and an input / output interface 758. The device 700 can operate on an operating system, such as Windows Server, stored in memory 732. TM Mac OS X TM Unix TM Linux TM FreeBSD TM Or similar.

[0137] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of this disclosure. This disclosure is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the following claims.

[0138] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A device control method, characterized in that, include: Acquire natural language data from the target user; The natural language data is input into a first semantic recognition model and a second semantic recognition model respectively to obtain a first intent list output by the first semantic recognition model and a second intent list output by the second semantic recognition model. The first semantic recognition model and the second semantic recognition model are different. Based on the first intent list and the second intent list, a target intent is determined, wherein the target intent includes the target device to be controlled and the target control intent for the target device; The target device is controlled according to the stated target control intent.

2. The method according to claim 1, characterized in that, Based on the first intent list and the second intent list, the target intent is determined, including: Each intent in the first intent list and the second intent list is scored to obtain the confidence level of each intent; The target intent is determined based on the confidence level of each intent.

3. The method according to claim 2, characterized in that, Each intent in the first intent list and the second intent list is scored to obtain the confidence level of each intent, including: For each intent, determine the slot information for that intent; Based on the slot information of the intent, determine the target correction value corresponding to the intent; Based on the target correction value, the preset confidence level is corrected to obtain the confidence level of the intent.

4. The method according to claim 3, characterized in that, Based on the slot information of the intent, determine the target correction value corresponding to the intent, including: Based on the slot information of the intent, determine at least one of the following: device information, control intent information, and number of slots corresponding to the intent; The target correction value is determined based on at least one of the device information corresponding to the intent, the control intent information, and the number of slots.

5. The method according to claim 4, characterized in that, Determining the target correction value based on the device information corresponding to the intent includes: Determine the matching result between the device corresponding to the device information and the device associated with the target user, and / or obtain the first historical usage frequency of the device corresponding to the device information; The target correction value is determined based on the matching results and / or the first historical usage frequency.

6. The method according to claim 4, characterized in that, Based on the number of slots corresponding to the intent, a target correction value is determined for correcting the preset confidence level, including: Determine the proportion of the number of slots corresponding to the stated intent to the preset number of slots; The target correction value is determined based on the stated proportion.

7. The method according to claim 4, characterized in that, Based on the control intent information corresponding to the intent, the target correction value is determined, including: Determine whether the control intent corresponding to the control intent information is a pre-set easily confused intent, and / or obtain the second historical usage frequency of the control intent corresponding to the control intent information; The target correction value is determined based on the confusing intent and / or the second historical usage frequency.

8. The method according to any one of claims 2-7, characterized in that, Determining the target intent based on the confidence level of each intent includes: Based on the confidence level of each intent, the intent with the highest confidence level in the first intent list and the second intent list is determined as the target intent.

9. The method according to any one of claims 2-7, characterized in that, Determining the target intent based on the confidence level of each intent includes: Based on the confidence level of each intent, determine whether there is an intent of a target type, wherein the confidence level of the intent of the target type is greater than or equal to a confidence level threshold; When it is determined that an intent of the target type exists, the intent with the highest confidence among the intents of the target type is determined as the target intent.

10. The method according to claim 9, characterized in that, The method further includes: When it is determined that the intent of the target type does not exist, a prompt message is output to indicate that semantic recognition has failed.

11. The method according to any one of claims 1-7, characterized in that, The first semantic recognition model is used to obtain the first intent list in the following manner: Semantic recognition is performed on the natural language data to obtain key information units; According to the preset mapping table, the key information units are matched to obtain the standard intent information corresponding to each key information unit. The standard intent information is information in the standard intent used for device control. The first intent list is obtained based on the standard intent information corresponding to each key information unit.

12. The method according to any one of claims 1-7, characterized in that, Acquire the natural language data of the target user, including: Acquire the voice data of the target user; The speech data is preprocessed to obtain the natural language data of the target user. The preprocessing operation includes at least a format conversion operation to convert the speech data into text data.

13. The method according to any one of claims 1-7, characterized in that, After controlling the target device according to the target control intent, the method further includes: The first semantic recognition model and the second semantic recognition model are trained based on the feedback data of the target user to the target intent and the execution results of controlling the target device.

14. A device control apparatus, characterized in that, include: The acquisition module is configured to acquire the natural language data of the target user. The acquisition module is configured to input the natural language data into a first semantic recognition model and a second semantic recognition model respectively, and obtain a first intent list output by the first semantic recognition model and a second intent list output by the second semantic recognition model, wherein the first semantic recognition model and the second semantic recognition model are different; The determination module is configured to determine a target intent based on the first intent list and the second intent list, wherein the target intent includes a target device to be controlled and a target control intent for the target device; The control module is configured to control the target device according to the target control intent.

15. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the program instructions are executed by the processor, they implement the device control method according to any one of claims 1-13.

16. An electronic device, characterized in that, include: Storage device for storing computer programs; An execution device is used to execute the computer program to implement the device control method according to any one of claims 1-13.

Images