Task processing method and device, electronic equipment and storage medium

Abstract

The application discloses a task processing method and device, electronic equipment and storage medium, and relates to the field of artificial intelligence. The specific technical scheme is: receiving a first input of a user on first information; in response to the first input, displaying a task processing result corresponding to a first task intent, the first task intent being a task intent output by a task intent recognition model based on the first information; the task intent recognition model being trained based on a task intent label and corresponding task intent description information.

Description

Technical Field

[0001] This application belongs to the field of artificial intelligence, and specifically relates to a task processing method, apparatus, electronic device, and storage medium. Background Technology

[0002] Currently, as the functions of electronic devices continue to improve, users' demands for the services they provide are also increasing. For example, users can use electronic devices to handle tasks such as filling out office forms, cross-language communication, and scheduling. However, currently, users need to give specific instructions to trigger the electronic device to execute the corresponding task. Sometimes, if the user's instructions are not clear enough, the electronic device cannot be triggered to execute the corresponding task, resulting in low task processing efficiency of the electronic device. Summary of the Invention

[0003] The purpose of this application is to provide a task processing method, apparatus, electronic device, and storage medium that can improve the task processing efficiency of electronic devices.

[0004] In a first aspect, embodiments of this application provide a task processing method, the method comprising:

[0005] Receive the user's first input on the first information;

[0006] In response to the first input, the task processing result corresponding to the first task intent is displayed. The first task intent is the task intent output by the task intent recognition model based on the first information. The task intent recognition model is trained based on the task intent label and the corresponding task intent description information.

[0007] Secondly, embodiments of this application provide a task processing apparatus, the apparatus comprising:

[0008] The receiving module is used to receive the user's first input on the first information;

[0009] The display module is used to respond to the first input and display the task processing result corresponding to the first task intent. The first task intent is the task intent output by the task intent recognition model based on the first information. The task intent recognition model is trained based on the task intent label and the corresponding task intent description information.

[0010] Thirdly, embodiments of this application provide an electronic device including a processor and a memory, wherein the memory stores programs or instructions executable on the processor, and the programs or instructions, when executed by the processor, implement the steps of the method described in the first aspect.

[0011] Fourthly, embodiments of this application provide a readable storage medium on which a program or instructions are stored, which, when executed by a processor, implement the steps of the method described in the first aspect.

[0012] Fifthly, embodiments of this application provide a chip, the chip including a processor and a communication interface, the communication interface being coupled to the processor, the processor being used to run programs or instructions to implement the method as described in the first aspect.

[0013] In a sixth aspect, embodiments of this application provide a computer program product stored in a storage medium, which is executed by at least one processor to implement the method described in the first aspect.

[0014] In this embodiment, a first input from a user regarding first information is received; in response to the first input, a task processing result corresponding to a first task intent is displayed. The first task intent is a task intent output by a task intent recognition model based on the first information; the task intent recognition model is trained based on task intent labels and corresponding task intent description information. Thus, even when the first information is not sufficiently clear, the user's task intent can be quickly and accurately identified through the task intent recognition model and the first information, thereby displaying the task processing result corresponding to the user's task intent, and improving the task processing efficiency of the electronic device. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating a task processing method provided in some embodiments of this application;

[0016] Figure 2A This is a schematic diagram of a task processing interface provided in some embodiments of this application;

[0017] Figure 2B This is a schematic diagram of a task processing interface provided in some embodiments of this application;

[0018] Figure 2C This is a schematic diagram of a task processing interface provided in some embodiments of this application;

[0019] Figure 3 This is a flowchart illustrating a task processing method provided in some embodiments of this application;

[0020] Figure 4 This is a flowchart illustrating a task processing method provided in some embodiments of this application;

[0021] Figure 5 This is a flowchart illustrating a task processing method provided in some embodiments of this application;

[0022] Figure 6AThis is a flowchart illustrating the task data construction process provided in some embodiments of this application;

[0023] Figure 6B This is a schematic diagram of the structure of a task intent recognition model provided in some embodiments of this application;

[0024] Figure 6C This is a schematic diagram illustrating the training process of a task intent recognition model provided in some embodiments of this application;

[0025] Figure 6D These are schematic diagrams illustrating user profile update strategies provided in some embodiments of this application;

[0026] Figure 6E These are schematic diagrams illustrating the task intent determination process provided in some embodiments of this application;

[0027] Figure 6F This is a schematic diagram illustrating the task intent determination process provided in some embodiments of this application;

[0028] Figure 7 These are schematic diagrams of the structure of a task processing device provided in some embodiments of this application;

[0029] Figure 8 These are schematic diagrams of the structure of a task processing device provided in some embodiments of this application;

[0030] Figure 9 These are schematic diagrams of the hardware structure of electronic devices provided in some embodiments of this application;

[0031] Figure 10 These are schematic diagrams of the structure of electronic devices provided in some embodiments of this application. Detailed Implementation

[0032] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0033] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0034] The terms "at least one," "at least one of," etc., used in the specification and claims of this application refer to any one, any two, or a combination of two or more of the included items. For example, at least one of a, b, and c can mean: "a," "b," "c," "a and b," "a and c," "b and c," and "a, b, and c," where a, b, and c can be single or multiple. Similarly, "at least two" refers to two or more items, and its meaning is similar to that of "at least one."

[0035] The following will explain the terminology used in the embodiments of this application.

[0036] Application (APP): An application is a computer program designed to perform one or more specific tasks. It runs in user mode, can interact with the user, and has a visual user interface.

[0037] In this embodiment, the electronic device may have multiple applications (APPs) installed, and icons of these applications may be displayed on the home screen of the electronic device. The applications in this embodiment may be embedded applications (i.e., system applications of the electronic device) or downloadable applications. Embedded applications are applications provided as part of the implementation of the electronic device. Downloadable applications are applications that can provide their own Internet Protocol Multimedia Subsystem (IMS) connectivity. These downloadable applications may be applications pre-installed on the electronic device or third-party applications downloaded and installed by the user. As an example, applications may include: social networking applications, image management applications (e.g., photo albums), map applications (e.g., maps), browser applications, music applications, etc.

[0038] The main screen of an electronic device can also be called the desktop or main interface. This main screen may include one or more sub-screens, which can be used to display control elements. Alternatively, the desktop of the electronic device may include one or more pages. Users can switch between different sub-screens by swiping left or right. A control is a graphical user interface (GUI) element; it is a software component contained within an application that controls all the data processed by the application and the interactive operations related to that data. Users can interact with controls through direct manipulation to read or edit information related to the application. Generally, controls can include icons (such as application icons or folder icons on an electronic device), buttons, menus, tabs, text boxes, dialog boxes, status bars, navigation bars, widgets, and other visual interface elements.

[0039] Interface: Refers to the medium through which humans interact with electronic devices. The interface allows users to send commands to the system via input devices and receive feedback information via output devices. Input devices can be keyboards, mice, touchscreens, etc.; monitors, speakers, etc.

[0040] User Interface (UI): The user interface (UI) is the bridge for interaction and communication between the system and the user. User interface design refers to the style setting of the graphical user interface, including color scheme, content arrangement, and layout. It includes not only the graphical interfaces of commonly seen desktop and mobile systems, but also new interaction methods such as touchscreens, motion sensing, and voice control. The three main principles of user interface design are: consistency, flexibility, and simplicity.

[0041] Controls: A control (also called a part, component, widget, or control) is a graphical user interface element. It is a basic building block of the user interface, such as a window or text box, and is displayed in the program interface of any application. Controls can be buttons, text boxes, labels, etc., and are used to control all the data processed by each application and the interactive operations on that data.

[0042] A control is an element in a graphical user interface that receives user input to perform corresponding processing or display relevant data. Controls can include, but are not limited to, virtual buttons, sliders, progress bars, and checkboxes.

[0043] Model: A model is a simulation or abstraction of certain characteristics and inherent relationships of objective reality. A model is a role. The concept of a model can be defined as follows: a thing is called a "model" because of its role or purpose in a specific situation—in that situation, it directly or indirectly carries certain attributes of another thing, and based on these attributes, it acts as a substitute or representation of that thing; thus, by using the attributes obtained from the model, the correlation between operations and the corresponding attributes of that thing can be achieved.

[0044] Model training: Model training refers to adjusting model parameters by learning from a large amount of data, enabling the model to accurately predict unknown data. Model training is a continuous process of adjusting model parameters, which requires making full use of the dataset to evaluate model performance in order to obtain a model with good performance and strong generalization ability.

[0045] The task processing method provided in this application will be described in detail below with reference to the accompanying drawings, through specific embodiments and application scenarios.

[0046] The task processing method provided in this application embodiment can identify the user's intent and execute the task corresponding to the user's intent during the interaction between the user and the electronic device.

[0047] In related technologies, during user interaction with electronic devices, if the electronic device needs to perform a task, the user must first issue a corresponding task command to the electronic device. The electronic device can only execute the task after recognizing the user's task intent from the task command. Therefore, the tasks executed by the electronic device are usually those indicated by the user's task intent identified from the task command. For example, in a conversational scenario, a user directly issues a task command to the phone's AI assistant, "AI Butler, open navigation for me," requesting the AI ​​assistant to open the navigation application interface. The phone can then determine from the user's task command that the user intends for the AI ​​assistant to open the navigation application interface, and thus execute the task of opening the navigation application interface through the AI ​​assistant.

[0048] However, in real life, during interactions between users and electronic devices, even without explicit task instructions, there may still be a need for the electronic device to perform a task. For example, if a user sends a message to a friend via a social media application saying "I have an interview at Building A tomorrow," the user might need the phone to create a reminder for the interview, or they might need navigation or a taxi to Building A. Alternatively, the user might not have any specific task need and is simply chatting with a friend. Therefore, how electronic devices can quickly and accurately identify the user's task intent and efficiently process the task is a problem that urgently needs to be solved.

[0049] Therefore, this application provides a task processing method that receives a user's first input of first information; in response to the first input, displays the task processing result corresponding to a first task intent, wherein the first task intent is a task intent output by a task intent recognition model based on the first information; the task intent recognition model is trained based on task intent labels and corresponding task intent description information. Thus, even when the first information is not sufficiently clear, the user's task intent can be quickly and accurately identified through the task intent recognition model and the first information, thereby displaying the task processing result corresponding to the user's task intent, and improving the task processing efficiency of the electronic device.

[0050] The task processing method provided in this application can be executed by an electronic device, or at least one of the functional modules and physical modules within the electronic device capable of implementing the task processing method. The specific execution subject can be determined according to actual usage requirements, and this application does not impose any limitations. The following explanation uses an electronic device task processing method as an example to illustrate the task processing method provided in this application.

[0051] Figure 1 This is a flowchart illustrating the task processing method provided in the embodiments of this application, as shown below. Figure 1 As shown, the data processing method provided in this application embodiment may include the following steps 101 and 102:

[0052] Step 101: The electronic device receives the user's first input on the first information.

[0053] In some embodiments of this application, the aforementioned first information may be information on a first interface of an electronic device, and the first interface may be the interface of a second application on the electronic device. In other words, the aforementioned first information may be information from the aforementioned second application.

[0054] In some embodiments of this application, the aforementioned second application may include, but is not limited to, any of the following: social networking applications, payment applications, navigation applications, shopping applications, video applications, photo album applications, music applications, etc. The specific application can be determined according to actual usage needs, and this application embodiment does not impose any limitations.

[0055] In some embodiments of this application, the first input is used to instruct the electronic device to identify the user's first task intent regarding the first information and to display the task processing result corresponding to the first task intent.

[0056] In some embodiments of this application, the first input mentioned above can be a user's click input, swipe input, press input, voice input, gesture input, or other feasible inputs, and this application does not limit this.

[0057] In some embodiments of this application, the above-mentioned gesture input may include, but is not limited to, at least one of the following: click gesture, swipe gesture, drag gesture, pressure recognition gesture, long press gesture, area change gesture, double press gesture, double tap gesture, specific gesture input or other possible gesture inputs. The specific gesture input form can be determined according to actual needs, and is not limited in some embodiments.

[0058] In some embodiments of this application, the above-mentioned click input can be single-click input, double-click input, or any number of clicks, or it can be long-press input or short-press input. In some embodiments, this is not limited.

[0059] In some embodiments of this application, the above-mentioned sliding input can be a sliding input in any direction, such as sliding up, sliding down, sliding left, or sliding right, etc., and in some embodiments, this is not limited.

[0060] Step 102: The electronic device responds to the first input and displays the task processing result corresponding to the first task intent.

[0061] In some embodiments of this application, the first task intent mentioned above is a task intent output by a task intent recognition model based on the first information mentioned above; the task intent recognition model can be trained based on task intent labels and corresponding task intent description information.

[0062] In some embodiments of this application, the task intent description information may include at least one of the following: first task intent description information and second task intent description information; the first task intent description information may be explicit intent description information, and the second task intent description information may be implicit intent description information.

[0063] The aforementioned first information may include at least one of the following: second information and third information; the aforementioned second information may be explicit intent description information, and the aforementioned third information may be implicit intent description information.

[0064] In some embodiments of this application, the first task intent description information may include explicit intent description information corresponding to multiple task intents, and the first task intent description information may include implicit intent description information corresponding to multiple task intents.

[0065] In some embodiments of this application, the first task intent description information may include explicit intent description information corresponding to multiple task intents. This can be understood as follows: the first task intent description information may include intent description information used to display and describe each of the multiple task intents.

[0066] In some embodiments of this application, the aforementioned second task intent description information may include implicit intent description information corresponding to multiple task intents. This can be understood as follows: the aforementioned second task intent description information may include intent description information used to implicitly describe each of the multiple task intents.

[0067] In some embodiments of this application, the above-mentioned task intent tags may include task intent tags for the above-mentioned multiple task intents.

[0068] In some embodiments of this application, the second information is used to explicitly describe the first task intent, and the third information is used to implicitly describe the first task intent.

[0069] In some embodiments of this application, the aforementioned multiple task intentions may include the aforementioned first task intention, the aforementioned first task intention description information may include the aforementioned second information, and the aforementioned second task intention description information may include the aforementioned third information.

[0070] Thus, since the task intent recognition model is trained based on task intent labels and corresponding first task intent description information used to explicitly describe the intent and second task intent description information used to implicitly describe the intent, electronic devices can quickly and accurately identify the user's task intent based on the first information, which includes at least one of the explicit and implicit intent description information, and then display the task processing result corresponding to the user's task intent, thereby improving the task processing efficiency of electronic devices.

[0071] In some embodiments of this application, the "displaying the task processing result corresponding to the first task intent" in step 102 above can be achieved through the following step 102a:

[0072] Step 102a: The electronic device displays the application interface of the first application.

[0073] In some embodiments of this application, the application interface may include the task processing result corresponding to the first task intent, the first application is the application that processes the task corresponding to the first task intent, the first information is the information in the second application, and the second application may be the same as or different from the first application.

[0074] For example, such as Figure 2AAs shown, the left screen 20a of the foldable phone displays a social networking application interface 21, and the right screen 20b displays a contacts application interface 22. The first message 211 can be a message sent to user A by a friend on the social networking application interface 21: "I am Xiao Wang, the housekeeper of Ziroom, and my phone number is 13430000111." User A can drag and drop the first message 211 onto the contacts application interface 22 on the right screen 20b. Upon receiving this drag input, the phone responds by identifying the user's primary task intent for the first message 211 as creating a new contact using an intent recognition model, and then combines this information with... Figure 2A ,like Figure 2B As shown, the task processing result 221 of the "Create Contact" task intent is displayed on the interface 22 of the Contacts application on the right screen 20b.

[0075] In this way, the electronic device can improve the flexibility of task processing by using the application interface of the first application, which includes the task processing results corresponding to the first task intent.

[0076] In the task processing method provided in this application embodiment, a first input from a user regarding first information is received; in response to the first input, a task processing result corresponding to a first task intent is displayed. The first task intent is a task intent output by a task intent recognition model based on the first information; the task intent recognition model is trained based on task intent labels and corresponding task intent description information. Thus, even when the first information is not sufficiently clear, the user's task intent can be quickly and accurately identified through the task intent recognition model and the first information, thereby displaying the task processing result corresponding to the user's task intent, and improving the task processing efficiency of the electronic device.

[0077] In some embodiments of this application, combined with Figure 1 ,like Figure 3 As shown, prior to step 102 above, the task processing method provided in this application embodiment may further include steps 103 and 104, and step 102 above can be implemented through step 102b below:

[0078] Step 103: The electronic device displays at least one task control corresponding to the first task intent.

[0079] In some embodiments of this application, each of the at least one task control described above indicates a task.

[0080] In some embodiments of this application, the at least one task indicated by the at least one task control can be the task corresponding to the first task intent.

[0081] In some embodiments of this application, when the number of tasks indicated by the at least one task control is one, the first task intent can be a single task intent. When the number of tasks indicated by the at least one task control is multiple, the first task intent can be multiple task intents.

[0082] In some embodiments of this application, the electronic device may display at least one task control on the interface of the first application described above.

[0083] In some embodiments of this application, where the first application is capable of handling each task indicated by the at least one task control, the electronic device may also display the at least one task control on the interface of the second application.

[0084] In some embodiments of this application, the shape of the task control can be any possible shape such as a circle, rectangle, triangle, rhombus, ring, or polygon, which can be determined according to actual usage requirements. This embodiment of the invention does not limit the shape.

[0085] For example, such as Figure 2A As shown, after the phone receives the drag input, it can respond to the drag input, combined with... Figure 2A ,like Figure 2C As shown, on the contact application interface 22 of the right screen 20b of the mobile phone, there are three task controls corresponding to the user's first task intention for the first information 211, namely, the new contact control 222, the update contact control 223, and the new contact control and dial a phone control 224.

[0086] Step 104: The electronic device receives a second input from the user to a first task control in at least one task control.

[0087] In some embodiments of this application, the first task control described above can be any one of the at least one task control described above.

[0088] In some embodiments of this application, the second input is used to instruct the processing of the task indicated by the first task control and to display the task processing result of the task indicated by the first task control.

[0089] In some embodiments of this application, the second input can be a user's click input, swipe input, press input, voice input, gesture input, or other feasible inputs, and this application does not limit this.

[0090] Step 102b: The electronic device responds to the second input and displays the task processing result corresponding to the first task intent.

[0091] In some embodiments of this application, the electronic device may respond to the second input described above, process the task indicated by the first task control, and display the task processing result of the task indicated by the first task control as the task processing result corresponding to the first task intent described above.

[0092] For example, such as Figure 2C As shown, when the mobile phone receives a press input from the user on the new contact control 222 on the interface 22 of the contacts application on the right screen 20b, it can respond to the press input, in conjunction with, as shown in 2C, such as Figure 2B As shown, in the interface 22 of the Contacts application, the task processing result 221 is the task intent.

[0093] Thus, by displaying at least one task control corresponding to the first task intent, receiving a second input from the user to the first task control among the at least one task controls, and responding to the second input, displaying the task processing result corresponding to the first task intent, the electronic device is able to display the processing result of the task indicated by the task control selected by the user, thereby improving the flexibility of the electronic device in handling tasks.

[0094] In some embodiments of this application, the task intent recognition method of the above-described task intent recognition model may include at least one of the following:

[0095] Methods for pre-setting task intent recall rules, methods for task intent recall in small neural network models, and methods for task intent recall in large neural network models;

[0096] The task intent recognition method of the above task intent recognition model is determined based on at least one of the first confidence level, the second confidence level, and the third confidence level;

[0097] Wherein, the first confidence level is the confidence level of the task intent that can be recalled using the aforementioned preset intent recall rule, the second confidence level is the confidence level of the task intent recalled using the aforementioned small neural network model intent recall method, and the third confidence level is the confidence level of the task intent recalled using the aforementioned large neural network model intent recall method.

[0098] In some embodiments of this application, when the task intent recognition method of the above-mentioned task intent recognition model includes the above-mentioned preset task intent recall rule, the above-mentioned task intent recognition model may include the preset task intent recall rule.

[0099] In some embodiments of this application, where the task intent recognition method of the above-mentioned task intent recognition model includes the task intent recall method of the above-mentioned small neural network model, the above-mentioned task intent recognition model may include a model with a small neural network model structure.

[0100] In some embodiments of this application, the aforementioned small neural network model can be a small model such as BERT.

[0101] In some embodiments of this application, where the task intent recognition method of the above-mentioned task intent recognition model includes the task intent recall method of the above-mentioned large neural network model, the above-mentioned task intent recognition model may include a model with a large neural network model structure.

[0102] In some embodiments of this application, the above-mentioned small neural network model can be a large model with more than 1B of model parameters.

[0103] In some embodiments of this application, since the method of recalling task intents using preset task intent recall rules is the fastest in terms of computation speed, followed by the method of recalling task intents using small neural network models, which is the slowest, the electronic device can use the aforementioned method of recalling task intents using preset task intent recall rules to recall task intents, and then determine whether the first confidence level of the task intents recalled using the aforementioned method of recalling task intents using preset task intent recall rules is within a preset confidence interval. If yes, it indicates that the confidence level of the task intents recalled using the aforementioned method of recalling task intents using preset task intent recall rules is reliable, and the task intents recalled using the aforementioned method of recalling task intents using preset task intent recall rules can be used as the aforementioned first task intent. If no, it indicates that the confidence level of the task intents recalled using the aforementioned method of recalling task intents using preset task intent recall rules is insufficient, and the electronic device can use the aforementioned method of recalling task intents using small neural network models to recall task intents, and then determine whether the second confidence level of the task intents recalled using the aforementioned method of recalling task intents using small neural network models is within a preset confidence interval. If yes, it indicates that the confidence level of the task intents recalled using the aforementioned method of recalling task intents using small neural network models is reliable, and the task intents recalled using the aforementioned method of recalling task intents using small neural network models can be used as the aforementioned first task intent. If not, the electronic device can use the task intent recall method of the above-mentioned large neural network model to recall the task intent, and then determine whether the third confidence level of the task intent recall method of the above-mentioned large neural network model is within the preset confidence interval.

[0104] In this way, the electronic device determines the task intent recognition method of the task intent recognition model by using the first confidence level of the task intent recalled by using the preset intent recall rule, the second confidence level of the task intent recalled by using the small neural network model intent recall method, and the third confidence level of the task intent recalled by using the large neural network model intent recall method, so that the task intent recognition model can accurately and quickly identify the task intent.

[0105] In some embodiments of this application, the above-described task intent recognition model may include a semantic feature extraction module, an intent feature extraction module, and a task intent recognition module connected in sequence; combined with Figure 1 ,like Figure 4 As shown, prior to step 102 above, the task processing method provided in this application embodiment may further include the following steps 105 to 108:

[0106] Step 105: The electronic device inputs the first information into the task intent recognition model, extracts semantic features from the first information through the semantic extraction module, and outputs the first semantic feature information of the first information.

[0107] In some embodiments of this application, the semantic extraction module can be the base of the task intent recognition model. The structure of the semantic extraction module can be the structure of the open-source BERT model, or the structure of other large open-source models. The specific structure can be determined according to actual needs, and no specific limitation is made here.

[0108] In some embodiments of this application, the first semantic feature information is used to characterize the semantics of the first information.

[0109] Step 106: The electronic device extracts intent features from the first semantic feature information through the intent feature extraction module to obtain the intent feature information corresponding to the first information.

[0110] In some embodiments of this application, the aforementioned intent feature information is used to characterize the user's task intent regarding the aforementioned first information.

[0111] Step 107: The electronic device uses the task intent recognition module to process the intent feature information to obtain M task intents and the corresponding score for each task intent.

[0112] Where M is a positive integer.

[0113] In some embodiments of this application, the M task intentions mentioned above can be the task intentions indicated by the aforementioned intention feature information.

[0114] In some embodiments of this application, the above score is used to characterize the confidence level of the corresponding task intent. The higher the score, the higher the confidence level of the corresponding task intent. The lower the score, the lower the confidence level of the corresponding task intent.

[0115] Step 108: The electronic device determines the task intent with the highest score among the M task intents as the first task intent.

[0116] In some embodiments of this application, the electronic device can sort the aforementioned M task intentions in descending order of score, and then determine one or more task intentions with the highest score as the aforementioned first task intention. In other words, the electronic device can select one or more task intentions with the highest confidence level from the aforementioned M task intentions as the aforementioned first task intention.

[0117] In this way, the electronic device inputs the first information into the task intent recognition model, extracts semantic features from the first information through the semantic extraction module, and outputs the first semantic feature information of the first information; extracts intent features from the first semantic feature information through the intent feature extraction module to obtain the intent feature information corresponding to the first information; performs task intent recognition processing on the intent feature information through the task intent recognition module to obtain M task intents and the score corresponding to each task intent; and determines the task intent with the highest score among the M task intents as the first task intent, thus accurately and quickly determining the user's task intent for the first information.

[0118] In some embodiments of this application, prior to step 105 above, the task processing method provided in this application may further include the following steps 109 to 111:

[0119] Step 109: The electronic device acquires at least one task intent tag and at least one task intent description information corresponding to each task intent tag.

[0120] In some embodiments of this application, the above-mentioned at least one task intent label can be a task intent label determined based on the intent vertical requirements of different tasks.

[0121] In some embodiments of this application, the aforementioned at least one task intent description information can be constructed through purely manual methods (such as brainstorming). The aforementioned at least one task intent description information can be explicit and / or implicit task intent expressions that humans can imagine in real life for a specific category. For example, the task intent description information corresponding to adjusting volume may include the explicit expression "turn the ringtone up 30%", or it may include implicit expressions such as "It's so noisy" and / or "I can't hear you."

[0122] Step 110: The electronic device expands each task intent description information to obtain at least one expanded task intent description information.

[0123] In some embodiments of this application, the electronic device may expand the task intent description information in at least one of the following ways:

[0124] Permutation and combination methods and model expansion methods.

[0125] In some embodiments of this application, the above-mentioned permutation and combination method refers to the electronic device segmenting the task intent description information to obtain at least one task intent description word, then obtaining at least one synonym of each task intent description word, and then permuting and combining all the synonyms of the task intent description word to obtain at least one expanded task intent description information.

[0126] In some embodiments of this application, the model expansion method refers to the electronic device expanding the task intent description information in a divergent manner through the task intent description information expansion model. The task intent description information expansion model can be an open-source large language model.

[0127] Step 111: The electronic device trains a task recognition model using at least one task intent label and at least one expanded task intent description.

[0128] In some embodiments of this application, the task intent recognition module includes M task intent recognition sub-modules, each task intent recognition sub-module being used to recognize a task intent; step 111 can be implemented through the following steps 111a and 111b:

[0129] Step 111a: The electronic device uses at least one task intent label and at least one expanded task intent description information to adjust the weight parameters of the intent feature extraction module and the weight parameters of the task intent recognition module.

[0130] In some embodiments of this application, the electronic device can input the at least one expanded task intent description information into the task intent recognition model, output at least one task intent corresponding to the at least one task intent description information through the task intent recognition model, and then calculate the similarity between the at least one task intent and the at least one task intent tag. If the similarity is less than or equal to a first similarity threshold, the weight parameters of the intent feature extraction module and the weight parameters of the task intent recognition module are adjusted while the weight parameters of the semantic extraction are maintained, until the similarity between the at least one task intent and the at least one task intent tag is greater than the first similarity threshold.

[0131] Step 111b: The electronic device uses each task intent label and the corresponding expanded task intent description information to adjust the weight parameters of the corresponding task intent recognition submodule.

[0132] In some embodiments of this application, after adjusting the weight parameters of the intent feature extraction module and the task intent recognition module, the electronic device inputs the expanded task intent description information corresponding to each task intent tag into the task intent recognition model. The task intent recognition model outputs the task intent corresponding to the task intent description information, and then calculates the similarity between the task intent corresponding to the task intent description information and the task intent tag. If the similarity is less than or equal to a second similarity threshold, the weight parameters of the corresponding task intent recognition submodule are adjusted until the similarity between the task intent corresponding to the task intent description information and the task intent tag is greater than the second similarity threshold.

[0133] Thus, the electronic device adjusts the weight parameters of the intent feature extraction module and the task intent recognition module by using at least one task intent label and at least one expanded task intent description information; and adjusts the weight parameters of the corresponding task intent recognition sub-module by using each task intent label and the corresponding expanded task intent description information, so that the task recognition model can accurately and quickly output the first task intent.

[0134] Thus, the electronic device acquires at least one task intent label and at least one task intent description information corresponding to each task intent label; expands each task intent description information to obtain at least one expanded task intent description information; and trains a task recognition model using at least one task intent label and at least one expanded task intent description information, so that the task recognition model can accurately and quickly output the first task intent.

[0135] In some embodiments of this application, prior to step 102a above, the task processing method provided in this application may further include the following steps 112 to 115:

[0136] Step 112: Electronic devices acquire user profile data.

[0137] In some embodiments of this application, the aforementioned user profile data can be used to characterize a user's usage of the application.

[0138] In some embodiments of this application, step 112 above can be implemented by the following steps 112a and 112b:

[0139] Step 112a: The electronic device collects gyroscope data at preset intervals.

[0140] In some embodiments of this application, the electronic device can acquire gyroscope data through a gyroscope sensor.

[0141] In some embodiments of this application, the gyroscope data may include the angular velocity of the electronic device on the horizontal axis, the angular velocity on the vertical axis, and the angular velocity on the longitudinal axis.

[0142] Step 112b: The electronic device obtains user profile data based on the gyroscope data of the electronic device.

[0143] In some embodiments of this application, step 112b above can be implemented by the following step 112b1:

[0144] Step 112b1: If the gyroscope data collected in two consecutive steps does not meet the first condition, the electronic device acquires user profile data.

[0145] In some embodiments of this application, the first condition described above is used to indicate that the gyroscope of the electronic device is stationary.

[0146] In some embodiments of this application, the first condition mentioned above includes:

[0147] The electronic device's angular velocities on the horizontal axis, vertical axis, and longitudinal axis are all less than or equal to the first angular velocity threshold.

[0148] In some embodiments of this application, if the gyroscope data collected in two consecutive periods does not meet the first condition described above, it indicates that the gyroscope of the electronic device rotated during the time interval between the two consecutive gyroscope data collections, meaning that the electronic device did not remain stationary during the time interval between the two consecutive gyroscope data collections. In other words, the user used the electronic device during that time interval, therefore, the user profile data will be updated, and the electronic device can acquire the user profile data.

[0149] In this way, the electronic device can acquire user profile data even when the gyroscope data collected in two consecutive transactions does not meet the first condition, thus enabling the electronic device to update the user profile data in a timely manner as the user uses the electronic device.

[0150] In this way, the electronic device collects gyroscope data at preset intervals; based on the gyroscope data, user profile data is obtained and can be updated regularly.

[0151] In some embodiments of this application, step 112b above can be implemented by the following steps 112b2 and 112b3:

[0152] Step 112b2: If the gyroscope data collected in two consecutive steps meet the first condition, the electronic device will increment the count value of the first counter by 1.

[0153] In some embodiments of this application, the first condition described above can be used to indicate that the gyroscope of an electronic device is stationary.

[0154] In some embodiments of this application, the first counter can be started before collecting gyroscope data from the electronic device, and the first counter can be used to count the number of consecutive stationary periods of the gyroscope of the electronic device.

[0155] In some embodiments of this application, if the gyroscope data collected in two consecutive intervals meets the first condition mentioned above, it indicates that the gyroscope of the electronic device did not rotate during the time interval between the two consecutive gyroscope data collections, that is, the electronic device remained stationary during the time interval between the two consecutive gyroscope data collections. Therefore, the electronic device can increment the count value of the first counter by 1 to update the number of consecutive stationary cycles of the gyroscope of the electronic device.

[0156] Step 112b3: The electronic device obtains user profile data based on the count value of the first counter and the time period in which the system time is located.

[0157] In some embodiments of this application, step 112b3 described above can be implemented by the following steps A1 or A2:

[0158] Step A1: If the count value of the first counter reaches the first value and the system time is in the first time period, the electronic device acquires the user profile data and clears the first counter to zero.

[0159] In some embodiments of this application, the first time period mentioned above may be a non-deep sleep period.

[0160] In some embodiments of this application, when the count value of the first counter reaches the first value and the system time is within the first time period, it indicates that the electronic device has been stationary for a first duration within the first time period, and the electronic device has not acquired user profile data during the first duration. Therefore, to avoid not updating user profile data for a long time during the first time period, the electronic device can acquire user profile data when the count value of the first counter reaches the first value and the system time is within the first time period.

[0161] In some embodiments of this application, after acquiring user profile data, the electronic device can reset the first counter to zero, so that the first counter can recount the number of consecutive stationary times of the gyroscope of the electronic device.

[0162] In this way, when the count value of the first counter reaches the first value and the system time is in the first time period, the electronic device can obtain user profile data, thus avoiding the failure to update user profile data for a long time during the first time period, and enabling the electronic device to update user profile data regularly during the first time period.

[0163] Step A2: If the count value of the first counter reaches the second value and the system time is in the second time period, the electronic device acquires the user profile data and clears the first counter.

[0164] In some embodiments of this application, the second value may be greater than the first value, and the second time period may be later than the first time period.

[0165] In some embodiments of this application, when the first time period is a non-deep sleep period, the second time period can be a deep sleep period that is later than the non-deep sleep period.

[0166] In some embodiments of this application, when the count value of the second counter reaches the second value and the system time is in the second time period, it indicates that the electronic device has been stationary for more than the second duration of the first time period, and the electronic device has not acquired user profile data during the second time period. Therefore, to avoid not updating user profile data for a long time during the second time period, the electronic device can acquire user profile data when the count value of the first counter reaches the second value and the system time is in the second time period. In this way, the electronic device can update the user profile less frequently during the second time period than during the first time period, thereby reducing the power consumption of the electronic device while still being able to update the user profile.

[0167] Thus, when the electronic device reaches the second value through the count value of the first counter and the system time is in the second time period, the electronic device acquires user profile data, enabling the electronic device to update the user profile less frequently in the second time period than in the first time period, thereby reducing the power consumption of the electronic device while updating the user profile.

[0168] In some embodiments of this application, after acquiring user profile data, the electronic device can reset the first counter to zero, so that the counter can recount the number of consecutive stationary times of the gyroscope of the electronic device.

[0169] Step 113: The electronic device determines at least one application preferred by the user and the user's preference value for each application in the at least one application based on user profile data.

[0170] In some embodiments of this application, the aforementioned user profile data includes historical behavioral data of a user on at least one application.

[0171] In some embodiments of this application, an electronic device can determine the duration of a user's use of at least one application based on historical behavioral data of at least one application, then identify at least one application whose usage duration is greater than or equal to a first usage duration as at least one application preferred by the user, and determine a preference value for each application among the at least one application preferred by the user based on the usage duration. The longer the user's usage duration of an application, the greater the user's preference value for that application.

[0172] Step 114: The electronic device determines from at least one application at least one third application that matches the task type of the task corresponding to the first task intent.

[0173] In some embodiments of this application, the above-described at least one application can handle different task types.

[0174] In some embodiments of this application, at least one third application matching the task type of the task corresponding to the first task intent can be an application capable of handling tasks of the aforementioned task type. In other words, the aforementioned at least one third application is capable of handling the task corresponding to the aforementioned first task intent.

[0175] Step 115: The electronic device identifies the application with the highest preference value among at least one third application as the first application.

[0176] In some embodiments of this application, the electronic device is able to determine the application with the largest preference value among the at least one third application as the first application for processing the task corresponding to the first task intent.

[0177] In this way, the electronic device determines at least one application preferred by the user and the user's preference value for each application in the at least one application based on user profile data; determines at least one third application from the at least one application that matches the task type of the task corresponding to the first task intent; and determines the application with the largest preference value among the at least one third application as the first application for processing the task corresponding to the first task intent, so that the electronic device can process the task corresponding to the first task intent through the application most preferred by the user, thereby improving the flexibility of processing the task required by the user.

[0178] In some embodiments of this application, prior to step 102 above, the task processing method provided in this application may further include the following steps 116 and 117:

[0179] Step 116: The electronic device inputs the first information and the task type corresponding to the first task intent into the task parameter extraction model, and extracts the task parameters that match the task type from the first information.

[0180] In some embodiments of this application, the above-mentioned task parameter extraction model can be an open-source multimodal large model.

[0181] In some embodiments of this application, the task parameter extraction model described above can be trained using at least one task type label and at least one task parameter sample corresponding to the at least one task type label.

[0182] In some embodiments of this application, the electronic device can be a terminal, the task intent recognition model described above can be deployed on the terminal, and the task parameter extraction model described above can be deployed on a server that is communicatively connected to the terminal.

[0183] In this way, by deploying the task intent recognition model on the terminal and the task parameter extraction model on the server that communicates with the terminal, the terminal can process the user's input privacy data, and the server can process high-computing data, thus improving computing efficiency while ensuring user privacy.

[0184] Step 117: The electronic device generates the task processing result based on the task parameters.

[0185] In some embodiments of this application, step 117 above can be implemented by step 117a as follows:

[0186] Step 117a: The electronic device calls the task parameters through the first application, processes the task corresponding to the first task intent, and obtains the task processing result.

[0187] In some embodiments of this application, the first application described above may be an application that processes the task corresponding to the first task intent described above.

[0188] In some embodiments of this application, the electronic device can use a first application to call the above-mentioned task parameters through a calling interface to process the task corresponding to the first task intent and obtain the above-mentioned task processing result.

[0189] In this way, the electronic device can accurately and quickly obtain the task processing result by calling the task parameters through the application that processes the task corresponding to the first task intent, thereby accurately and quickly processing the user's required tasks.

[0190] In this way, by inputting the first information and the task type corresponding to the first task intention into the task parameter extraction model, the electronic device can extract the task parameters that match the task type from the first information, accurately extract the task parameters that match the task type, and then quickly generate task processing results, thereby quickly processing the tasks required by the user.

[0191] The task processing method provided in this application embodiment will be further described in detail below with reference to specific implementation methods and taking a mobile phone as an example.

[0192] The task processing method provided in this embodiment can perform intent recognition for cross-app operation behaviors. Through a comprehensive intelligent system that organically combines key modules such as intent retrieval, intent ranking, and parameter extraction, it covers both explicit and implicit intents, helping users quickly and conveniently access the app services they need. At the same time, the edge-cloud collaborative link design ensures the privacy and security of user input content.

[0193] The improvements of the task processing method provided in this embodiment are as follows:

[0194] 1. Cross-app interaction for quick access to services: Users can quickly access the services they intend to access with simple operations such as dragging and dropping.

[0195] 2. Multiple data construction schemes: Combining multiple and various data construction methods to ensure the richness of task data.

[0196] 3. Multi-path intent recall: Multiple intent recall algorithms of various types are used in parallel to reduce the possibility of intent omission.

[0197] 4. Weighted Intent Scoring: Multi-dimensional weighted aggregated intent scores, taking into account both the accuracy and personalization of intent.

[0198] 5. Large-scale parameter extraction: Build a customized large-scale parameter extraction model specific to the task to ensure the correctness of the parameter extraction results.

[0199] 6. End-to-Cloud Collaboration System: Flexible switching between end-to-cloud links through top-level central control logic to ensure user privacy and security.

[0200] The following is combined Figure 5 The task processing method provided in this embodiment will be described. For example... Figure 5 As shown, the task processing method provided in this embodiment may include the following steps 501 to 510:

[0201] Step 501: The mobile phone constructs task data in advance through multiple channels.

[0202] For example, the task data is the task description information mentioned above.

[0203] Based on the task intent category requirements, data is constructed for each category, which will then be used for training the task intent recognition model and task parameter extraction model in step 502.

[0204] This invention provides a method for constructing task data through multiple approaches, mainly consisting of five steps: constructing a seed query statement, query expansion, query cleaning, label annotation, and label check. (See figure below.) Figure 6A As shown.

[0205] First, seed queries are constructed using purely manual methods (such as brainstorming). These queries are the explicit or implicit expressions of intent that humans can imagine in real life for a particular vertical category. For example, the intent to adjust volume might be expressed explicitly as "turn the ringer up 30%" or implicitly as "It's too noisy." Because it is a purely manual method, the number of seed queries is usually around 50 per intent.

[0206] Then, the seed query is expanded using multiple tools, mainly through four methods: manual expansion, brute-force permutation and combination, task log mining, and large model generation.

[0207] [Method 1] Manual expansion (taking navigation intent as an example):

[0208] Based on the key slots specified by the intent service, all parameter combinations are enumerated, such as: destination = value / no value, mode of transportation = driving / walking / no value. Therefore, there are a total of 6 parameter combinations, as shown in Table 1:

[0209]

[0210] Table 1

[0211] For each combination of parameters, try to diversify the query paradigms as much as possible, as shown in Table 2:

[0212]

[0213] Table 2

[0214] In the above process, it is usually necessary to combine task data and technical intuition to estimate the proportion of each combination or paradigm in advance, so as to make a decision on the final size of the data.

[0215]

Approach 2

[0216] This approach is based on the already completed step of manually expanding the "paradigm enumeration," and involves a process of "segmentation-lexicon enumeration-concatenation-sampling" of the paradigm statements. In the context of image style transfer, a typical phrase is "Please change the style of this image; I want a colorful Japanese anime style." The results of brute-force permutations and combinations are shown in Table 3.

[0217]

[0218] Table 3

[0219] In the above process, the "parameter data annotation" is automatically completed, because the "parameter label" is known in advance. It is especially suitable for skills with a large number of parameters that require fine-grained parameter annotation.

[0220] [Method 3] Task Log Mining:

[0221] If similar or identical tasks already exist online, data mining can be performed on the task logs, provided that national security and compliance laws are adhered to. Commonly used data mining techniques in the industry are not the main focus of this invention and will not be elaborated upon further.

[0222] [Approach 4] Large Model Generation (Taking Battery Management Intent as an Example):

[0223] This approach is based on a completed query seed and involves "divergent expansion" using a large model. By specifying the use case, explicit or implicit style, and seed query to the large model via a prompt, query expansion can be performed efficiently and quickly. A typical prompt expanded using a large model is as follows:

[0224] You are an excellent data generation assistant, particularly skilled in three areas, especially in smartphone scenarios:

[0225] 1. Given a smartphone skill <api>You can fully understand this by reading the Description, Parameters, and Scope. <api>How it operates.

[0226] 2. Given a batch of request data from smartphone users. <querys>Can you understand these? <querys>It is hoped that the above can be called. <api>.

[0227] 3. Reference <querys>,according to <api>Give full play to your imagination and automatically generate more <querys>Request data.

[0228] <Please Note That>

[0229] 1. <querys>The style of expression can be both "explicit" and "vague".

[0230] 2. "Explicit" means that the user is not implying this in a subtle or implicit way. <api>Instead, it explicitly requires smartphones to perform this. <api>.

[0231] 3. "Vague" means that the user did not explicitly request the smartphone to perform this action. <api>Instead, it subtly and implicitly suggests this. <api>.

[0232] 4. Before generating data, make sure you fully understand what "explicit" and "vague" mean.

[0233] 5. When generating data, please make full use of... <api>The parameter information in the database generates a rich variety of data through the permutation and combination of different parameter values ​​(such as enumerated values, Boolean values, and numerical values).

[0234] <api>

[0235] - device.optimize

[0236] - Description: Enhances phone performance.

[0237] - Parameter:

[0238] - "type": Optimization options. Options: 'power_consumption'.

[0239] - Scope: Use when phone meets 'power_consumption' and requiresoptimization.

[0240] <querys>

[0241] The phone consumes a lot of power.

[0242] The phone's power consumption is too high.

[0243] My phone battery drains too quickly.

[0244] <Task Start>

[0245] Let's solve this problem step by step to ensure we get the right answer.

[0246] refer to <querys>,according to <api>The rich and diverse data you generate is:

[0247] My phone battery drains so fast!

[0248] Is the phone's power consumption too high?

[0249] My phone's battery is draining too fast!

[0250] Please help me limit my phone's battery consumption.

[0251] Is it possible to save battery power on the phone by reducing power consumption?

[0252] The phone battery drains too quickly; it needs some optimization.

[0253] Please help me reduce my phone's battery consumption.

[0254] Optimize your phone's battery consumption.

[0255] How can I optimize my battery consumption?

[0256] This power consumption is a bit high, should we optimize it?

[0257] My phone's battery drains very quickly when I'm playing games. Can this be optimized?

[0258] My phone's battery is draining faster and faster, please fix it.

[0259] Implement energy-saving optimization strategies.

[0260] I hope that mobile phone battery usage can be further reduced.

[0261] At this point, the construction of the recall data has been completed. Next, query cleaning, label annotation, and label checking remain, thus completing the construction of the parameter extraction data. Since these are all standard, conventional processes and not the main improvements of this invention, they will not be elaborated upon further.

[0262] In summary, through step 501, the construction of recall data and parameter extraction data was completed using a multi-path approach. In the next step 502, these data can be used to train the task intent recognition model and the task parameter extraction model.

[0263] Step 502: The mobile phone uses the task data in advance to train the task intent recognition model and the task parameter extraction model.

[0264] In step 505, a task intent recognition model will be used, and in step 509, a task parameter extraction model will be used. Both of these need to be trained in advance using the task data from the previous step 501 before the task goes online.

[0265] Regarding the task parameter extraction model, since the commonly used model training techniques in the industry are all quite mature, whether for small models like BERT or large models with more than 1B of parameters, they can be directly selected and used as needed. This is not the main improvement point of this invention, so it will not be discussed further.

[0266] However, for task intent recognition models, a structural modification scheme is needed to enable them to handle classification problems involving multiple intent categories. This invention provides a task intent recognition model structure with independent decoupling between different categories, and an improved training scheme adapted to it.

[0267] [Task Intent Recognition Model Structure] A hierarchical structure is used to ensure that results are independent across different categories:

[0268] like Figure 6B As shown, it is mainly divided into three layers: model base, task information sharing layer, and vertical linear classification layer.

[0269] The first layer of the task intent recognition model is the model base, which can be a small model like BERT or a large model of 1B or more. Generally, authoritative open-source models from the industry can be used directly, and this is not the main improvement point of this invention, so it will not be discussed further. It should be noted that the model base is the semantic extraction module mentioned above.

[0270] The second layer of the task intent recognition model is the task information sharing layer. Without this layer, the results for each vertical category would largely depend on the performance of the model base, with little consideration for the semantic connections between vertical categories, whether during model training or inference. For example, the intents "navigation" and "ride-hailing" have very similar query paradigms in real life, as implied by users, such as "I'm going to the Times Building for an interview tomorrow." It's impossible to distinguish between them based solely on their literal meaning, so a correct recall would require hitting both intents. Without the task information sharing layer, the navigation intent would infer a score through its own independent model weights, and the ride-hailing intent would do the same. Regardless of whether the scores for both intents would be similarly high as expected, the training workload would be inherently doubled, offering no benefit. With the task information sharing layer, due to the large number of shared parameters among multiple intents, model training efficiency would significantly improve. Furthermore, vertical categories that are different but semantically similar would be forced to share key weights, thus improving model performance. It should be noted that the task information sharing layer is the intent extraction module mentioned above.

[0271] The third layer of the task intent recognition model is a vertical linear classification layer. As mentioned above, since the substructure of each vertical category is independent, the intent score, i.e., the intent recognition result, of each vertical category is also independent during the model inference stage. This successfully achieves "result independence," which is also a prerequisite for achieving "training independence" below. It should be noted that the vertical linear classification layer is the task intent recognition module mentioned above.

[0272] [Training Scheme for Task Intent Recognition Model] Segmented training ensures independent training across different categories:

[0273] like Figure 6C As shown, it is mainly divided into two stages: the unified training stage for all vertical categories and the independent fine-tuning stage for each vertical category.

[0274] The first stage of training the task intent recognition model is a unified training stage across all vertical categories. Using the recall data from step 501, the object being trained is the entire model (excluding the model base). As mentioned above, the purpose of this stage is to improve the overall performance of the task intent recognition model in a specific task by introducing a task information sharing layer, so that the key weights of the model are shared among the intents of each vertical category.

[0275] The second stage of training the task intent recognition model is the independent fine-tuning stage for each vertical category. At this stage, the task information sharing layer is frozen and not included in the training. This is because this layer has already learned the shared information of each vertical category and only needs to be fixed here as the "shared task knowledge" of each vertical category at the top level. At the same time, since the linear classification layers of each vertical category at the top level are completely independent and decoupled, each vertical category can be trained with its own and arbitrary datasets. Moreover, the training can be completely independent and unrelated. It is even possible to leave a vertical category completely untouched and not participate in the training. Only the vertical branch that needs to be optimized in a certain version iteration needs to be trained.

[0276] In summary, through step 502, a task parameter extraction model was trained using industry-standard methods. Furthermore, a task intent recognition model was trained using a task intent recognition model structure that decouples each vertical category independently, along with its specially improved training scheme. This latter model simultaneously ensures both task effectiveness and the ease of independent iterative optimization. At this point, the task intent recognition model and task parameter extraction model, which will be used in steps 505 and 509 respectively, have completed their preliminary preparations and are ready for subsequent formal use.

[0277] Step 503: Collect historical behavior data on the mobile device in advance to build an offline interest profile.

[0278] In step 506, the user's offline interest profile will be used, and the profile needs to be calculated using the user's historical behavior collected in advance on the device.

[0279] Historical user behavior can be easily collected using the built-in event tracking function of the mobile phone system. These events record a user's usage behavior across various apps on their phone, in the form of:

[0280]

[0281] Table 4. Example of historical behavior data table on the user side

[0282] By collecting the aforementioned user data and using industry-standard offline interest profiling algorithms, corresponding profiles can be easily calculated. Since these technologies are relatively mature and can be directly selected and used as needed, they are not the main improvement point of this invention and will not be elaborated upon further. The following section primarily describes three key issues: the form and physical meaning of the profile, the key characteristics of the profile, and how the profile influences the intent ranking in step 106.

[0283] [Question 1] The form and physical significance of the portrait

[0284] User offline interest profiles are essentially a dictionary data structure. Taking the "app usage preference" interest profile as an example, the dictionary keys are the app names, and the values ​​are the app preference scores (calculated by the various profiling statistical algorithms mentioned above, which inherently take into account positive information such as usage frequency and negative information such as exit behavior), in the form of:

[0285]

[0286] Table 5. A typical example of an interest profile regarding "app usage preferences"

[0287] As can be seen, the process of profile calculation involves specifying a time window, collecting behavioral data within that time window, and calculating the corresponding profile. The physical meaning of a profile is to score all key elements in a dictionary. Essentially, the input is a time window and the behavioral data within it, and the output is the profile result, with the goal of calculating preference scores.

[0288] [Question 2] Key characteristics of a portrait

[0289] The first key characteristic of a profile is that it "requires a specified time window." As mentioned above, given a profile result, it must refer to the profile result "within a specified time window." Changing the time window will result in a different result after recalculation. Therefore, profiles usually need to be updated every certain period of time.

[0290] The second key feature of the profile is "support for offline processing on the device". As mentioned above, a profile can be calculated as long as there is historical behavior data. Since historical behavior is recorded on the user's mobile device, both data collection and profile calculation can be directly supported offline on the device, which inherently protects user privacy and security.

[0291] [Question 3] How does the profile affect the intent ranking in step 506?

[0292] As mentioned above, although the statistical algorithms for user profiles are relatively mature and will not be elaborated upon in this invention, specifically, based on the first key characteristic of user profiles, "the need to specify a certain time window," a user profile update strategy needs to be designed so that each user's on-device interest profile can be automatically updated in a timely manner. If the update frequency is too sparse, the profile information will become outdated and insufficient to accurately serve the user's intent ranking. If the update frequency is too frequent, since data collection and profile calculation are performed directly on the mobile device, it will lead to a waste of system resources and problems such as battery life and power consumption. Figure 6D As shown, this embodiment provides a strategy for dynamically updating user profiles based on user mobile phone usage.

[0293] like Figure 6D As shown, the update strategy first obtains some system information and cumulative variables. The latter mainly refers to the cumulative number of times the gyroscope is stationary. The gyroscope activity is counted every 30 minutes. If there is activity in the past 30 minutes, it proves that the user is using the phone normally. At this time, the profile can be updated immediately (but you need to remember to clear the count).

[0294] If there has been no activity in the past 30 minutes, it means that the user has not touched the phone for a period of time. At this time, the inactivity count is incremented by 1 (recording how long the user has not touched the phone). Check whether the updated count exceeds the threshold (different thresholds correspond to normal activity time and late-night sleep time, with the former being smaller and the latter larger, so that the former updates more frequently and the latter does not need to be updated frequently). If it exceeds the threshold, the profile is forced to be updated once to avoid the profile not updating indefinitely (but remember to clear the accumulated count after the update).

[0295] It is evident that the aforementioned profile update strategy can not only keep up with users' normal mobile phone usage behavior in a timely manner, but also automatically update when the system is idle, thus dynamically adapting to user scenarios in real life.

[0296] In step 606, intent retrieval has been completed, thus obtaining the user's coarse-grained intent (such as schedule or address). At this point, it is only necessary to map the coarse-grained intent to fine-grained intent (which is essentially the service app). The mapping relationship is a fixed mapping determined in advance by the task product side. Then, based on the user preference score of the corresponding app according to the client-side interest profile, the intent ranking can be completed. Specifically, as follows... Figure 6E As shown.

[0297] Step 604: The mobile phone receives user input such as dragging and dropping operations generated across apps.

[0298] The pain point scenario of this invention is usually when users open multi-window mode on their mobile phones and display N apps on one screen at the same time. For ease of demonstration, a schematic diagram of a foldable screen phone is used for illustration.

[0299] First, such as Figure 2A As shown, when a user opens the multi-window mode on a foldable phone, there are two windows (two apps). On the left is the chat interface of the WeChat app, where the user has received a new contact information from a friend. On the right is the new contact interface of the Contacts app, where all slots are empty.

[0300] Next, as Figure 2A As shown, when a user long-presses the text "I am Xiao Wang, the housekeeper of Ziroom, 13430000111" on the left, this function relies on the phone's super drag and drop function and is an interaction method built into the OS system, which can "attach the text to the fingertip and prepare to start dragging and moving it".

[0301] At this point, the user triggered a drag-and-drop operation, dragging out the text "I am Xiao Wang, the housekeeper of Ziroom, 13430000111". This text is the beginning of the entire intent recognition process in this invention. Next, it will first go through the intent recall stage in step 505, such as... Figure 6E As shown, this embodiment will use a multi-path intent recall scheme to identify intent.

[0302] Step 605: The mobile phone responds to the operation input and performs multi-path intent recall on the device side to infer the user's potential multiple intents.

[0303] Through steps 601 and 602, the task data and task intent recognition model have been successfully constructed and trained (this embodiment of the invention requires two models: a small BERT model on the client side and a large 1B model on the client side). In the recall phase, the primary principle is to avoid overlooking any potential user intents. Theoretically, a multi-path recall design can be used, employing multiple sets of rule strategies and multiple task intent recognition models to ensure recall rate. However, to protect user privacy and security, the recall phase is usually required to be implemented offline on the client side, while a multi-path recall design increases system battery life and power consumption. Therefore, a scientific multi-path recall capability invocation scheme needs to be designed to simultaneously balance recall effectiveness and module performance. For example... Figure 6F The embodiment of the present invention shown provides a funnel-shaped call scheme for real-time automatic coordination of multi-path recall.

[0304] like Figure 6F As shown, the solution first introduces the three-way edge recall capabilities it relies on: rule-based strategy recall, BERT small model recall, and 1B large model recall. The rule-based strategy of the first recall loop is derived from the task data in step 101, which has been obtained through data mining of various intent categories. These are common phrase templates, such as address intents like "xxx I want to drive to xxx" or "Tomorrow's team building location is xxx". These phrase templates (i.e., rules) are pre-organized and written into the recall module for direct invocation and matching. If the user's input matches a rule of a certain category, the intent is considered matched. The BERT small model of the second recall loop and the 1B large model of the third recall loop have been prepared in advance through the model training process in step 502.

[0305] The following is the process of the entire multi-path recall capability invocation scheme. When the user triggers the recall process in step 504, the three recall capabilities will be activated in parallel. Since the computational load of these three is not on the same order of magnitude, it is almost certain that their completion order is as follows: rule-based strategies consume almost no time, possibly completing within 30ms; small models are slightly slower but still fast overall, possibly completing in about 120ms; large models are the slowest and consume the most system resources, possibly completing in 300ms. Based on the above facts, there will be three possible scenarios:

[0306] [Scenario 1] The smaller model completes earlier than the larger model (almost always), and the results are highly confident.

[0307] It is known that the overall capability of a smaller model is always weaker than that of a larger model, with one exception: when the scores of all N smaller models across N vertical categories fall within a "high confidence interval," the smaller model can be considered almost equivalent to the larger model. This interval is determined by the lower threshold T1 and the upper threshold T2, which can be predetermined and fixed according to the actual task requirements. In this embodiment, empirical values ​​of 0.15 and 0.85 are used. The physical meaning of this interval is: when the score of a smaller model for a certain vertical category is less than T1, it can be almost completely certain that the input content does not contain that intent; similarly, when it is greater than T2, it can be almost completely certain that the intent is contained.

[0308] Therefore, in this scenario, the small model recall is both fast and highly confident, allowing for the early interruption of the large model recall that is not yet complete. This effectively reduces the system resource consumption of the large model, and the entire recall process is also faster. In this case, the final recall result is "rule-based recall + small model recall".

[0309] [Scenario 2] The smaller model finishes earlier than the larger model (almost always), but the result is not very reliable.

[0310] As mentioned above, if the score of the small model in a certain vertical category does not hit the high confidence interval, then we need to wait for the recall result of the large model as usual. This is equivalent to letting the large model decide the results of "those vertical categories that the small model is not sure about". At this time, the final recall result is "rule recall + small model recall + large model recall".

[0311] [Scenario 3] The larger model is completed before the smaller model (almost impossible)

[0312] As mentioned above, although this situation is almost impossible, if it does occur, it is assumed that the reliability of the large model recall can completely cover that of the small model, and correspondingly, the small model recall that has not yet ended can be interrupted in advance. In this case, the final recall result is "rule-based recall + large model recall".

[0313] In summary, through step 505, a funnel-shaped invocation scheme with real-time automatic coordination of multi-path recall ensures both recall rate and, in most cases, the scientific and effective allocation of system resources, balancing recall effectiveness and module performance. At this point, the recall process concludes. During the user's drag-and-drop input in step 504, their potential intent (coarse-grained intent) has been inferred and can be sent to the next step 506 for scoring and ranking (fine-grained intent).

[0314] Step 506: On the mobile device, use offline interest profiling and other methods to score and rank multiple intentions.

[0315] The key question 3 in step 503, "How profiles affect intent ranking," has already fully described this, so it will not be repeated. At this point, the recall result (coarse-grained intent) from step 505 is known, and the process continues through step 503... Figure 7 After mapping intent (fine-grained intent), querying interest profiles on the device, and sorting by preference scores, the intent ranking process is complete. Next, simply display the results to the user on the mobile screen according to the design in step 507.

[0316] Step 507: Based on the sorting results, the mobile phone displays multiple intent service option buttons on the mobile phone.

[0317] For example, the intent service option button is the task control mentioned above.

[0318] Continuing with the same user input example, "I am Xiao Wang, the housekeeper of Ziroom, 13430000111," after the intent retrieval in step 505 and intent sorting in step 506, when the user finally drags the object to the "Create Contact" screen of the Contacts app on the right, multiple service buttons pop up based on the intent recognition results: Create Contact, Overwrite Contact, Create Contact and Make a Call, etc. Figure 6C As shown. At this point, the dragged object will have a semi-transparent animation effect, and the currently hovered service button will have a highlighted blinking animation effect.

[0319] Step 508: The mobile phone responds to the user's operation of dragging and releasing or clicking to specify a service button.

[0320] Continuing with the same user input example, "I am Xiao Wang, the housekeeper of Ziroom, 13430000111," after the intent service option buttons are displayed in step 507, if the user actually needs one of the services, they only need to release their finger above the corresponding hovering service button. Figure 2C As shown. At this point, the dragged object will have a fade-out animation effect added, and the selected button, other unselected buttons, and the origin of the dragged object will also display corresponding prompts or animations.

[0321] The prompt "Selected, executing..." indicates the execution of the next step 509, which is the parameter extraction stage.

[0322] Step 509: On the cloud side, extract parameters from the large model to drive the necessary parameters for downstream services.

[0323] Through step 502, the task parameter extraction model has been successfully trained. This embodiment of the invention selects a cloud-based 7B large model. A 7B large model was chosen because the purpose of the parameter extraction task is to extract key information from user input, requiring extremely high accuracy, which is crucial to the success of subsequent service execution. Therefore, if the model's parameter count is too small, the parameter extraction effect will be poor. Furthermore, if the parameter count is too large, such as a model with more than 13B, it will lead to slow model inference time. On the other hand, cloud-based deployment was chosen instead of direct integration onto the user's mobile device because models with more than 3B typically consume a large amount of system resources on mobile devices, while this problem does not exist on the cloud. Therefore, considering all factors, the cloud-based 7B large model solution was chosen, which can balance parameter extraction accuracy, inference time, and system resource consumption.

[0324] Continuing with the same user input example, "I am Xiao Wang, the housekeeper of Ziroom, 13430000111", if the user confirmed that they selected the "Create Contact" intent service in the previous step 508, the user input will be passed to the parameter-providing model. The reasoning result, as described in step 102, will be a dictionary, which is the necessary parameter used to drive the service, as shown in Table 6:

[0325]

[0326] Table 6. Inference results of the large-scale model with parameter extraction on sample input.

[0327] Step 510: The mobile phone receives the necessary parameters through the intent service, executes the corresponding service, and fulfills the user's needs.

[0328] After the parameter extraction step 509, the cloud-side parameter extraction model has correctly extracted the necessary driving parameters for the user-selected service. Next, it only needs to transmit these parameters to the service's API to start the service and fulfill the user's requirements. Specifically, as follows... Figure 6F As shown.

[0329] Thus, the entire intent recognition process is complete, from the user initiating the cross-app dragging behavior, to the on-device multi-path intent retrieval, then the on-device intent scoring and sorting, to the user releasing the finger to activate the cloud-side large model for parameter extraction, and finally the service successfully executing and displaying the result.

[0330] The task processing method provided in this embodiment can help users quickly and conveniently access the app services they need through a comprehensive intelligent system that organically combines key modules such as intent recall, intent sorting, and parameter extraction, while simultaneously covering both explicit and implicit intents. At the same time, the edge-cloud collaborative link design ensures the privacy and security of user input content.

[0331] It should be noted that the task processing method proposed in this embodiment has strong backward compatibility. For example, it can support complex cross-app operations, and therefore also simple non-cross-app operations; it can support text input, and therefore also image or file input; and it can support implicit intents, and therefore also simple explicit intents. Therefore, it has strong generalization ability and can be easily extended to various interaction scenarios, input scenarios, intent scenarios, etc.

[0332] It should be noted that each of the above method embodiments, or various possible implementations of each method embodiment, can be executed individually or in combination of any two or more. The specific implementation can be determined according to actual usage requirements, and this application embodiment does not impose any restrictions on this.

[0333] or,

[0334] It should be noted that the above-described method embodiments, or the various possible implementations of the method embodiments, can be executed individually, or, provided there are no contradictions, they can be combined with each other. The specific implementation can be determined according to actual usage requirements, and this application embodiment does not impose any restrictions on this.

[0335] The task processing method provided in this application can be executed by a task processing device. This application uses the execution of the task processing method by a task processing device as an example to illustrate the task processing device provided in this application.

[0336] Figure 7 This is a schematic diagram of a task processing device provided in an embodiment of this application. Figure 7 As shown, the task processing device 700 includes a receiving module 701 and a display module 702.

[0337] The receiving module 701 is used to receive the user's first input on the first information;

[0338] Display module 702 is used to respond to the first input and display the task processing result corresponding to the first task intent, wherein the first task intent is the task intent output by the task intent recognition model based on the first information; the task intent recognition model is trained based on the task intent label and the corresponding task intent description information.

[0339] In some embodiments of this application, the task intent description information includes at least one of the following: first task intent description information and second task intent description information; the first task intent description information is explicit intent description information, and the second task intent description information is implicit intent description information;

[0340] The first information includes at least one of the following: second information and third information; the second information is explicit intent description information, and the third information is implicit intent description information.

[0341] In some embodiments of this application, the display module 702 is further configured to:

[0342] Before displaying the task processing result corresponding to the first task intent, at least one task control corresponding to the first task intent is displayed, and each task control in the at least one task control indicates a task;

[0343] Receive a second input from the user to a first task control in the at least one task control;

[0344] The display module 702 is specifically used for:

[0345] In response to the second input, the task processing result corresponding to the first task intent is displayed.

[0346] In some embodiments of this application, the task intent recognition method of the task intent recognition model includes at least one of the following:

[0347] Methods for pre-setting task intent recall rules, methods for task intent recall in small neural network models, and methods for task intent recall in large neural network models;

[0348] The task intent recognition method of the task intent recognition model is determined based on at least one of a first confidence level, a second confidence level, and a third confidence level.

[0349] Wherein, the first confidence level is the confidence level of the task intent recalled using the preset intent recall rule, the second confidence level is the confidence level of the task intent recalled using the small neural network model intent recall method, and the third confidence level is the confidence level of the task intent recalled using the large neural network model intent recall method.

[0350] In some embodiments of this application, the task intent recognition model includes a semantic feature extraction module, an intent feature extraction module, and a task intent recognition module;

[0351] Combination Figure 7 ,like Figure 8 As shown, the device 700 further includes:

[0352] Processing module 703 is used to input the first information into the task intent recognition model, extract semantic features from the first information through the semantic extraction module, and output the first semantic feature information of the first information;

[0353] The intent feature extraction module extracts intent features from the first semantic feature information to obtain the intent feature information corresponding to the first information.

[0354] The task intent recognition module performs task intent recognition processing on the intent feature information to obtain M task intents and the score corresponding to each task intent, where M is a positive integer.

[0355] The task intent with the highest score among the M task intents is determined as the first task intent.

[0356] In some embodiments of this application, the processing module is further configured to:

[0357] Before inputting the first information into the task intent recognition model, at least one task intent label and at least one task intent description information corresponding to each task intent label are obtained.

[0358] Each task intent description is expanded to obtain at least one expanded task intent description.

[0359] The task recognition model is trained using the at least one task intent label and the at least one expanded task intent description information.

[0360] In some embodiments of this application, the task intent recognition module includes M task intent recognition sub-modules, each task intent recognition sub-module being used to recognize a task intent;

[0361] The processing module 703 is specifically used for:

[0362] Using the at least one task intent tag and the at least one expanded task intent description information, adjust the weight parameters of the intent feature extraction module and the weight parameters of the task intent recognition module;

[0363] The weight parameters of the corresponding task intent recognition submodule are adjusted using each task intent tag and its corresponding expanded task intent description information.

[0364] In some embodiments of this application, the display module 702 is specifically used for:

[0365] The application interface of the first application is displayed. The application interface includes the task processing result corresponding to the first task intent. The first application is the application that processes the task corresponding to the first task intent. The first information is information in the second application. The second application may be the same as or different from the first application.

[0366] In some embodiments of this application, the device 700 further includes:

[0367] Processing module 703 is used to obtain user profile data before displaying the application interface of the first application, wherein the user profile data represents the user's usage of the application.

[0368] Based on the user profile data, determine at least one application that the user prefers and the user's preference value for each of the at least one application;

[0369] Determine at least one third application from the at least one application that matches the task type of the task corresponding to the first task intent;

[0370] The application with the highest preference value among the at least one third application is identified as the first application.

[0371] In some embodiments of this application, the processing module 703 is specifically used for:

[0372] Data from the electronic device's gyroscope is collected at preset intervals.

[0373] User profile data is obtained based on the gyroscope data of the electronic device.

[0374] In some embodiments of this application, the processing module 703 is specifically used for:

[0375] If the gyroscope data collected in two consecutive transactions does not meet the first condition, then user profile data is obtained. The first condition is used to indicate that the gyroscope of the electronic device is stationary.

[0376] In some embodiments of this application, the processing module 703 is specifically used for:

[0377] If the gyroscope data collected in two consecutive transactions meet the first condition, the count value of the first counter is incremented by 1. The first condition is used to indicate that the gyroscope of the electronic device is stationary.

[0378] Based on the count value of the first counter and the time period in which the system time is located, user profile data is obtained;

[0379] The first counter is started before collecting gyroscope data from the electronic device, and the first counter is used to count the number of consecutive stationary periods of the gyroscope of the electronic device.

[0380] In some embodiments of this application, the processing module 703 is specifically used for:

[0381] If the count value of the first counter reaches a first value and the system time is within a first time period, then user profile data is acquired and the first counter is reset to zero; or,

[0382] If the count value of the first counter reaches the second value and the system time is in the second time period, then user profile data is obtained and the first counter is cleared.

[0383] Wherein, the second value is greater than the first value, and the second time period is later than the first time period.

[0384] In some embodiments of this application, the device 700 further includes:

[0385] Processing module 703 is used to input the first information and the task type of the task corresponding to the first task intent into the task parameter extraction model before displaying the task processing result corresponding to the first task intent, and extract the task parameters that match the task type from the first information.

[0386] Based on the task parameters, the task processing result is generated.

[0387] In some embodiments of this application, the task intent recognition model is deployed on a terminal, and the task parameter extraction model is deployed on a server that is communicatively connected to the terminal.

[0388] In some embodiments of this application, the processing module 703 is specifically used for:

[0389] The task parameters are invoked by the first application to process the task corresponding to the first task intent and obtain the task processing result. The first application is the application that processes the task corresponding to the first task intent.

[0390] In the task processing device provided in this application embodiment, a first input from a user regarding first information is received; in response to the first input, a task processing result corresponding to a first task intent is displayed. The first task intent is a task intent output by a task intent recognition model based on the first information; the task intent recognition model is trained based on task intent labels and corresponding task intent description information. Thus, even when the first information is not sufficiently clear, the user's task intent can be quickly and accurately identified through the task intent recognition model and the first information, thereby displaying the task processing result corresponding to the user's task intent, and improving the task processing efficiency of the electronic device.

[0391] The task processing device in this application embodiment can be an electronic device or a component within an electronic device, such as an integrated circuit or a chip. The electronic device can be a terminal or other devices besides a terminal. For example, the electronic device can be a mobile phone, tablet computer, laptop computer, PDA, in-vehicle electronic device, mobile internet device (MID), augmented reality (AR) / virtual reality (VR) device, robot, wearable device, ultra-mobile personal computer (UMPC), netbook, or personal digital assistant (PDA), etc. It can also be a server, network attached storage (NAS), personal computer (PC), television (TV), ATM, or self-service machine, etc. This application embodiment does not specifically limit the device.

[0392] The task processing device in this application embodiment can be a device with an operating system. The operating system can be Android, iOS, or other possible operating systems; this application embodiment does not specifically limit the specific operating system.

[0393] The task processing device provided in this application embodiment can achieve... Figures 1 to 4 The various processes implemented in the method implementation examples will not be described again here to avoid repetition.

[0394] Optionally, such as Figure 9 As shown, this application embodiment also provides an electronic device 900, including a processor 901 and a memory 902. The memory 902 stores a program or instructions that can run on the processor 901. When the program or instructions are executed by the processor 901, they implement the various steps of the above-described task processing method embodiment and can achieve the same technical effect. To avoid repetition, they will not be described again here.

[0395] It should be noted that the electronic devices in the embodiments of this application include the mobile electronic devices and non-mobile electronic devices described above.

[0396] Figure 10 A schematic diagram of the hardware structure of an electronic device to implement an embodiment of this application.

[0397] The electronic device 1000 includes, but is not limited to, the following components: radio frequency unit 1001, network module 1002, audio output unit 1003, input unit 1004, sensor 1005, display unit 1006, user input unit 1007, interface unit 1008, memory 1009, and processor 1010.

[0398] Those skilled in the art will understand that the electronic device 1000 may also include a power supply (such as a battery) for supplying power to various components. The power supply may be logically connected to the processor 1010 through a power management system, thereby enabling functions such as managing charging, discharging, and power consumption through the power management system. Figure 10 The electronic device structure shown does not constitute a limitation on the electronic device. The electronic device may include more or fewer components than shown, or combine certain components, or have different component arrangements, which will not be elaborated here.

[0399] The user input unit 1007 is used to receive the user's first input on the first information;

[0400] Display unit 1006 is used to respond to the first input and display the task processing result corresponding to the first task intent, wherein the first task intent is the task intent output by the task intent recognition model based on the first information; the task intent recognition model is trained based on the task intent label and the corresponding task intent description information.

[0401] In some embodiments of this application, the task intent description information includes at least one of the following: first task intent description information and second task intent description information; the first task intent description information is explicit intent description information, and the second task intent description information is implicit intent description information;

[0402] The first information includes at least one of the following: second information and third information; the second information is explicit intent description information, and the third information is implicit intent description information.

[0403] In some embodiments of this application, the display unit 1006 is further configured to:

[0404] Before displaying the task processing result corresponding to the first task intent, at least one task control corresponding to the first task intent is displayed, and each task control in the at least one task control indicates a task;

[0405] Receive a second input from the user to a first task control in the at least one task control;

[0406] The display unit 1006 is specifically used for:

[0407] In response to the second input, the task processing result corresponding to the first task intent is displayed.

[0408] In some embodiments of this application, the task intent recognition method of the task intent recognition model includes at least one of the following:

[0409] Methods for pre-setting task intent recall rules, methods for task intent recall in small neural network models, and methods for task intent recall in large neural network models;

[0410] The task intent recognition method of the task intent recognition model is determined based on at least one of a first confidence level, a second confidence level, and a third confidence level.

[0411] Wherein, the first confidence level is the confidence level of the task intent recalled using the preset intent recall rule, the second confidence level is the confidence level of the task intent recalled using the small neural network model intent recall method, and the third confidence level is the confidence level of the task intent recalled using the large neural network model intent recall method.

[0412] In some embodiments of this application, the task intent recognition model includes a semantic feature extraction module, an intent feature extraction module, and a task intent recognition module;

[0413] The processor 1010 is configured to input the first information into the task intent recognition model, extract semantic features from the first information through the semantic extraction module, and output the first semantic feature information of the first information.

[0414] The intent feature extraction module extracts intent features from the first semantic feature information to obtain the intent feature information corresponding to the first information.

[0415] The task intent recognition module performs task intent recognition processing on the intent feature information to obtain M task intents and the score corresponding to each task intent, where M is a positive integer.

[0416] The task intent with the highest score among the M task intents is determined as the first task intent.

[0417] In some embodiments of this application, the processor 1010 is further configured to:

[0418] Before inputting the first information into the task intent recognition model, at least one task intent label and at least one task intent description information corresponding to each task intent label are obtained.

[0419] Each task intent description is expanded to obtain at least one expanded task intent description.

[0420] The task recognition model is trained using the at least one task intent label and the at least one expanded task intent description information.

[0421] In some embodiments of this application, the task intent recognition module includes M task intent recognition sub-modules, each task intent recognition sub-module being used to recognize a task intent;

[0422] The processor 1010 is specifically used for:

[0423] Using the at least one task intent tag and the at least one expanded task intent description information, adjust the weight parameters of the intent feature extraction module and the weight parameters of the task intent recognition module;

[0424] The weight parameters of the corresponding task intent recognition submodule are adjusted using each task intent tag and its corresponding expanded task intent description information.

[0425] In some embodiments of this application, the display unit 1006 is specifically used for:

[0426] The application interface of the first application is displayed. The application interface includes the task processing result corresponding to the first task intent. The first application is the application that processes the task corresponding to the first task intent. The first information is information in the second application. The second application may be the same as or different from the first application.

[0427] In some embodiments of this application, the processor 1010 is configured to acquire user profile data before displaying the application interface of the first application, the user profile data representing the user's usage of the application;

[0428] Based on the user profile data, determine at least one application that the user prefers and the user's preference value for each of the at least one application;

[0429] Determine at least one third application from the at least one application that matches the task type of the task corresponding to the first task intent;

[0430] The application with the highest preference value among the at least one third application is identified as the first application.

[0431] In some embodiments of this application, the processor 1010 is specifically used for:

[0432] Data from the electronic device's gyroscope is collected at preset intervals.

[0433] User profile data is obtained based on the gyroscope data of the electronic device.

[0434] In some embodiments of this application, the processor 1010 is specifically used for:

[0435] If the gyroscope data collected in two consecutive transactions does not meet the first condition, then user profile data is obtained. The first condition is used to indicate that the gyroscope of the electronic device is stationary.

[0436] In some embodiments of this application, the processor 1010 is specifically used for:

[0437] If the gyroscope data collected in two consecutive transactions meet the first condition, the count value of the first counter is incremented by 1. The first condition is used to indicate that the gyroscope of the electronic device is stationary.

[0438] Based on the count value of the first counter and the time period in which the system time is located, user profile data is obtained;

[0439] The first counter is started before collecting gyroscope data from the electronic device, and the first counter is used to count the number of consecutive stationary periods of the gyroscope of the electronic device.

[0440] In some embodiments of this application, the processor 1010 is specifically used for:

[0441] If the count value of the first counter reaches a first value and the system time is within a first time period, then user profile data is acquired and the first counter is reset to zero; or,

[0442] If the count value of the first counter reaches the second value and the system time is in the second time period, then user profile data is obtained and the first counter is cleared.

[0443] Wherein, the second value is greater than the first value, and the second time period is later than the first time period.

[0444] In some embodiments of this application, the processor 1010 is used to input the first information and the task type of the task corresponding to the first task intent into a task parameter extraction model before displaying the task processing result corresponding to the first task intent, and to extract task parameters matching the task type from the first information.

[0445] Based on the task parameters, the task processing result is generated.

[0446] In some embodiments of this application, the task intent recognition model is deployed on a terminal, and the task parameter extraction model is deployed on a server that is communicatively connected to the terminal.

[0447] In some embodiments of this application, the processor 1010 is specifically used for:

[0448] The task parameters are invoked by the first application to process the task corresponding to the first task intent and obtain the task processing result. The first application is the application that processes the task corresponding to the first task intent.

[0449] In the electronic device provided in this application embodiment, a first input from a user regarding first information is received; in response to the first input, a task processing result corresponding to a first task intent is displayed. The first task intent is a task intent output by a task intent recognition model based on the first information; the task intent recognition model is trained based on task intent labels and corresponding task intent description information. Thus, even when the first information is not sufficiently clear, the user's task intent can be quickly and accurately identified through the task intent recognition model and the first information, thereby displaying the task processing result corresponding to the user's task intent, and improving the task processing efficiency of the electronic device.

[0450] It should be understood that, in this embodiment, the input unit 1004 may include a graphics processing unit (GPU) 10041 and a microphone 10042. The GPU 10041 processes image data of still images or videos obtained by an image capture device (such as a camera) in video capture mode or image capture mode. The display unit 1006 may include a display panel 10061, which may be configured in the form of a liquid crystal display, an organic light-emitting diode, etc. The user input unit 1007 includes at least one of a touch panel 10071 and other input devices 10072. The touch panel 10071 is also called a touch screen. The touch panel 10071 may include a touch detection device and a touch controller. Other input devices 10072 may include, but are not limited to, physical keyboards, function keys (such as volume control buttons, power buttons, etc.), trackballs, mice, joysticks, etc., which will not be described in detail here.

[0451] The memory 1009 can be used to store software programs and various data. The memory 1009 may primarily include a first storage area for storing programs or instructions and a second storage area for storing data. The first storage area may store the operating system, application programs or instructions required for at least one function (such as sound playback, image playback, etc.). Furthermore, the memory 1009 may include volatile memory or non-volatile memory, or both. The non-volatile memory may be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory can be random access memory (RAM), static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct memory bus RAM (DRRAM). The memory 1009 in this embodiment includes, but is not limited to, these and any other suitable types of memory.

[0452] The processor 1010 may include one or more processing units; optionally, the processor 1010 integrates an application processor and a modem processor, wherein the application processor mainly handles operations involving the operating system, user interface, and applications, and the modem processor mainly handles wireless communication signals, such as a baseband processor. It is understood that the aforementioned modem processor may also not be integrated into the processor 1010.

[0453] This application also provides a readable storage medium storing a program or instructions. When the program or instructions are executed by a processor, they implement the various processes of the above-described task processing method embodiments and achieve the same technical effects. To avoid repetition, they will not be described again here.

[0454] The processor is the processor in the electronic device described in the above embodiments. The readable storage medium includes computer-readable storage media, such as computer read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0455] This application embodiment also provides a chip, which includes a processor and a communication interface. The communication interface is coupled to the processor. The processor is used to run programs or instructions to implement the various processes of the above-described task processing method embodiments and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0456] It should be understood that the chip mentioned in the embodiments of this application may also be referred to as a system-on-a-chip, system chip, chip system, or system-on-a-chip, etc.

[0457] This application provides a computer program product, which is stored in a storage medium and executed by at least one processor to implement the various processes of the task processing method embodiments described above, and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0458] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one…" does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of this application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0459] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a computer software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0460] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.< / api> < / querys> < / querys> < / api> < / api> < / api> < / api> < / api> < / api> < / querys> < / querys> < / api> < / querys> < / api> < / querys> < / querys> < / api> < / api>

Claims

1. A task processing method, characterized in that, include: Receive the user's first input on the first information displayed on the first interface; In response to the first input, the task processing result corresponding to the first task intent is displayed, where the first task intent is the task intent output by the task intent recognition model based on the first information; The task intent recognition model is trained based on task intent labels and corresponding task intent description information.

2. The method according to claim 1, characterized in that, The task intent description information includes at least one of the following: first task intent description information and second task intent description information; the first task intent description information is explicit intent description information, and the second task intent description information is implicit intent description information; The first information includes at least one of the following: second information and third information; the second information is explicit intent description information, and the third information is implicit intent description information.

3. The method according to claim 1, characterized in that, Before displaying the task processing result corresponding to the first task intent, the method further includes: Display at least one task control corresponding to the first task intent, wherein each task control indicates a task; Receive a second input from the user to a first task control in the at least one task control; The display of the task processing result corresponding to the first task intent includes: In response to the second input, the task processing result corresponding to the first task intent is displayed; Wherein, the first task control is any one of the at least one task controls.

4. The method according to claim 1, characterized in that, The task intent recognition model includes at least one of the following methods for task intent recognition: Methods for pre-setting task intent recall rules, methods for task intent recall in small neural network models, and methods for task intent recall in large neural network models; The task intent recognition method of the task intent recognition model is determined based on at least one of a first confidence level, a second confidence level, and a third confidence level. Wherein, the first confidence level is the confidence level of the task intent recalled using the preset intent recall rule, the second confidence level is the confidence level of the task intent recalled using the small neural network model intent recall method, and the third confidence level is the confidence level of the task intent recalled using the large neural network model intent recall method.

5. The method according to claim 1, characterized in that, The task intent recognition model includes a semantic feature extraction module, an intent feature extraction module, and a task intent recognition module; The method further includes: The first information is input into the task intent recognition model, and the semantic extraction module extracts semantic features from the first information and outputs the first semantic feature information of the first information. The intent feature extraction module extracts intent features from the first semantic feature information to obtain the intent feature information corresponding to the first information. The task intent recognition module performs task intent recognition processing on the intent feature information to obtain M task intents and the score corresponding to each task intent, where M is a positive integer. The task intent with the highest score among the M task intents is determined as the first task intent.

6. The method according to claim 5, characterized in that, Before inputting the first information into the task intent recognition model, the method further includes: Obtain at least one task intent label and at least one task intent description information corresponding to each task intent label; Each task intent description is expanded to obtain at least one expanded task intent description. The task recognition model is trained using the at least one task intent label and the at least one expanded task intent description information.

7. The method according to claim 6, characterized in that, The task intent recognition module includes M task intent recognition sub-modules, each task intent recognition sub-module being used to recognize a task intent; The step of training the task recognition model using the at least one task intent label and the at least one expanded task intent description information includes: Using the at least one task intent tag and the at least one expanded task intent description information, adjust the weight parameters of the intent feature extraction module and the weight parameters of the task intent recognition module; The weight parameters of the corresponding task intent recognition submodule are adjusted using each task intent tag and its corresponding expanded task intent description information.

8. The method according to claim 1, characterized in that, The display of the task processing result corresponding to the first task intent includes: The application interface of the first application is displayed. The application interface includes the task processing result corresponding to the first task intent. The first application is the application that processes the task corresponding to the first task intent. The first information is information in the second application. The second application may be the same as or different from the first application.

9. The method according to claim 8, characterized in that, Before displaying the application interface of the first application, the method further includes: Acquire user profile data, which represents how users use the application; Based on the user profile data, determine at least one application that the user prefers and the user's preference value for each of the at least one application; Determine at least one third application from the at least one application that matches the task type of the task corresponding to the first task intent; The application with the highest preference value among the at least one third application is identified as the first application.

10. The method according to claim 9, characterized in that, The acquisition of user profile data includes: Data from the electronic device's gyroscope is collected at preset intervals. User profile data is obtained based on the gyroscope data of the electronic device.

11. The method according to claim 10, characterized in that, The acquisition of user profile data based on the gyroscope data of the electronic device includes: If the gyroscope data collected in two consecutive transactions does not meet the first condition, then user profile data is obtained. The first condition is used to indicate that the gyroscope of the electronic device is stationary.

12. The method according to claim 10, characterized in that, The acquisition of user profile data based on the gyroscope data of the electronic device includes: If the gyroscope data collected in two consecutive transactions meet the first condition, the count value of the first counter is incremented by 1. The first condition is used to indicate that the gyroscope of the electronic device is stationary. Based on the count value of the first counter and the time period in which the system time is located, user profile data is obtained; The first counter is started before collecting gyroscope data from the electronic device, and the first counter is used to count the number of consecutive stationary periods of the gyroscope of the electronic device.

13. The method according to claim 12, characterized in that, The process of obtaining user profile data based on the count value of the first counter and the time period in which the system time is located includes: If the count value of the first counter reaches a first value and the system time is within a first time period, then user profile data is acquired and the first counter is reset to zero; or, If the count value of the first counter reaches the second value and the system time is in the second time period, then user profile data is obtained and the first counter is cleared. Wherein, the second value is greater than the first value, and the second time period is later than the first time period.

14. The method according to claim 1, characterized in that, Before displaying the task processing result corresponding to the first task intent, the method further includes: Input the first information and the task type corresponding to the first task intent into the task parameter extraction model, and extract the task parameters that match the task type from the first information; Based on the task parameters, the task processing result is generated.

15. The method according to claim 14, characterized in that, The task intent recognition model is deployed on the terminal, and the task parameter extraction model is deployed on a server that communicates with the terminal.

16. The method according to claim 14 or 15, characterized in that, The step of generating the task processing result based on the task parameters includes: The task parameters are invoked by the first application to process the task corresponding to the first task intent and obtain the task processing result. The first application is the application that processes the task corresponding to the first task intent.

17. A task processing device, characterized in that, include: The receiving module is used to receive the user's first input on the first information; The display module is used to respond to the first input and display the task processing result corresponding to the first task intent, wherein the first task intent is the task intent output by the task intent recognition model based on the first information; The task intent recognition model is trained based on task intent labels and corresponding task intent description information.

18. The apparatus according to claim 17, characterized in that, The task intent description information includes at least one of the following: first task intent description information and second task intent description information; the first task intent description information is explicit intent description information, and the second task intent description information is implicit intent description information; The first information includes at least one of the following: second information and third information; the second information is explicit intent description information, and the third information is implicit intent description information.

19. An electronic device, characterized in that, It includes a processor and a memory, the memory storing a program or instructions that can run on the processor, the program or instructions being executed by the processor to implement the steps of the task processing method as described in any one of claims 1-16.

20. A readable storage medium, characterized in that, The readable storage medium stores a program or instructions that, when executed by a processor, implement the steps of the task processing method as described in any one of claims 1-16.

Images