Information processing method and device, computer equipment and storage medium

Abstract

The invention discloses an information processing method and device, computer equipment and a storage medium, and relates to the technical field of artificial intelligence. The method is applied to a server and comprises the steps that the server receives first request information sent by first electronic equipment; determining a generative large language model corresponding to the first user identity identifier from a plurality of generative large language models as a target model; inputting the first request information into a target model to obtain first feedback information corresponding to the first request information; and sending the first feedback information to the first electronic equipment. Therefore, the target model learns the preference and speaking habit of the user corresponding to the first user identity identifier, that is, the target model is obtained by performing targeted and customized training on the user corresponding to the first user identity identifier. Therefore, the first feedback information for the first request information can be more accurately generated by utilizing the target model, namely, the accuracy of information feedback is improved, and meanwhile, the feedback information is more personalized.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to an information processing method, apparatus, computer equipment and storage medium. Background Art

[0002] A large language model (LLM) is a deep learning model trained with a large amount of text data that can generate natural language text or understand the meaning of language text. Large language models can handle a variety of natural language tasks, such as text classification, question answering, and dialogue, and are an important path to artificial intelligence.

[0003] However, in the related art, it is not possible to process the request information of some users in a more personalized manner according to different users, which results in inaccurate feedback on the request information of different users. Summary of the invention

[0004] The present application proposes an information processing method, apparatus, computer equipment and storage medium to improve the accuracy of information feedback.

[0005] In a first aspect, an embodiment of the present application provides an information processing method, which is applied to a server, and the method includes: receiving a first request information sent by a first electronic device, the first request information carrying a first user identity of a current user of the first electronic device; determining a generative large language model corresponding to the first user identity from multiple generative large language models as a target model, multiple generative large language models corresponding one-to-one to multiple preset user identities, the generative large language model is used to identify the intent of the request information and output feedback information corresponding to the intention identification result, the target model is pre-trained based on a first training sample set associated with the first user identity, and the first training sample set is generated based on context information corresponding to the first user identity in at least one application installed on the first electronic device; inputting the first request information into the target model to obtain first feedback information corresponding to the first request information; and sending the first feedback information to the first electronic device.

[0006] In a second aspect, an embodiment of the present application provides an information processing method, which is applied to a first electronic device, and the method includes: sending a first request message to a server, the first request message carrying a first user identity of a current user of the first electronic device; receiving first feedback information sent by the server based on the first request message, the first feedback information being a generative large language model corresponding to the first user identity determined by the server from multiple generative large language models as a target model; and inputting the first request message into the target model to obtain first feedback information corresponding to the first request message, wherein the multiple generative large language models correspond one-to-one to multiple preset user identities, the generative large language model is used to perform intent recognition on the request information, and output feedback information corresponding to the intent recognition result, the target model is pre-trained based on a first training sample set associated with the first user identity, and the first training sample set is generated based on context information corresponding to the first user identity in at least one application installed on the first electronic device; and outputting the first feedback information.

[0007] In a third aspect, an embodiment of the present application provides an information processing device, which is applied to a server, and the device includes: an information receiving module, which is used to receive a first request information sent by a first electronic device, and the first request information carries a first user identity of the current user of the first electronic device; a model determination module, which is used to determine a generative large language model corresponding to the first user identity from multiple generative large language models as a target model, and multiple generative large language models correspond one-to-one to multiple preset user identities. The generative large language model is used to identify the intent of the request information and output feedback information corresponding to the intention recognition result. The target model is pre-trained based on a first training sample set associated with the first user identity, and the first training sample set is generated based on context information corresponding to the first user identity in at least one application installed on the first electronic device; a feedback information acquisition module, which is used to input the first request information into the target model to obtain first feedback information corresponding to the first request information; an information feedback module, which is used to send the first feedback information to the first electronic device.

[0008] In a fourth aspect, an embodiment of the present application provides an information processing device, which is applied to a first electronic device, and the device includes: an information sending module, which is used to send a first request message to a server, and the first request message carries a first user identity of the current user of the first electronic device; an information receiving module, which is used to receive first feedback information sent by the server based on the first request message, and the first feedback information is a generative large language model corresponding to the first user identity determined by the server from multiple generative large language models as a target model; and the first request message is input into the target model to obtain first feedback information corresponding to the first request message, and multiple generative large language models correspond one-to-one to multiple preset user identities, and the generative large language model is used to perform intent recognition on the request information and output feedback information corresponding to the intent recognition result, and the target model is pre-trained based on a first training sample set associated with the first user identity, and the first training sample set is generated based on context information corresponding to the first user identity in at least one application installed on the first electronic device; an information output module, which is used to output the first feedback information.

[0009] In a fifth aspect, an embodiment of the present application provides a computer device, comprising: one or more processors; a memory; one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs are configured to execute the above-mentioned method.

[0010] In a sixth aspect, an embodiment of the present application provides a computer-readable storage medium, in which a program code is stored. The program code can be called by a processor to execute the above method.

[0011] In the solution provided by the present application, the server receives a first request message sent by a first electronic device, wherein the first request message carries a first user identity of the current user of the first electronic device; determines a generative large language model corresponding to the first user identity from a plurality of generative large language models as a target model, wherein the plurality of generative large language models correspond one-to-one to a plurality of preset user identities, and the generative large language model is used to perform intent recognition on the request message and output feedback information corresponding to the intent recognition result; inputs the first request message into the target model to obtain first feedback information corresponding to the first request message; and sends the first feedback information to the first electronic device. Since the target model is pre-trained based on a first training sample set associated with the first user identity, and the first training sample set is generated based on context information corresponding to the first user identity in at least one application installed on the first electronic device. In this way, the target model can learn the preferences and speaking habits of the user corresponding to the first user identity, which is equivalent to the target model being trained in a targeted and customized manner for the user corresponding to the first user identity. Therefore, by using the target model, the first feedback information of the first request message corresponding to the first user identity can be generated more accurately, that is, the accuracy of the information feedback is improved, and the feedback information is also made more personalized. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0013] Figure 1 A schematic diagram of the system architecture of an information processing system 10 provided in an embodiment of the present application is shown.

[0014] Figure 2 A flow chart of an information processing method provided in an embodiment of the present application is shown.

[0015] Figure 3 A flow chart of an information processing method provided by another embodiment of the present application is shown.

[0016] Figure 4 A flowchart of an information processing method provided by yet another embodiment of the present application is shown.

[0017] Figure 5 A flowchart of an information processing method provided in yet another embodiment of the present application is shown.

[0018] Figure 6It is a block diagram of an information processing device provided according to an embodiment of the present application.

[0019] Figure 7 It is a block diagram of an information processing device provided according to another embodiment of the present application.

[0020] Figure 8 It is a block diagram of a computer device according to an embodiment of the present application for executing an information processing method according to an embodiment of the present application.

[0021] Fig. 9 It is a storage unit of an embodiment of the present application for storing or carrying a program code for implementing an information processing method according to an embodiment of the present application. DETAILED DESCRIPTION

[0022] In order to enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0023] It should be noted that in some processes described in the specification, claims and the above-mentioned drawings of the present application, multiple operations that appear in a specific order are included, and these operations may not be executed or executed in parallel in the order in which they appear in this article. The sequence numbers of operations such as S110, S120, etc. are only used to distinguish different operations, and the sequence numbers themselves do not represent any execution order. In addition, these processes may include more or less operations, and these operations may be executed in sequence or in parallel. And, the terms "first", "second", etc. in the specification, claims and the above-mentioned drawings of the present application are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described here can be implemented in an order other than those illustrated or described here. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or server that includes a series of steps or sub-modules is not necessarily limited to those steps or sub-modules explicitly listed, but may include other steps or sub-modules that are not explicitly listed or that are inherent to these processes, methods, products, or devices.

[0024] The inventor proposes an information processing method, device, computer equipment and storage medium. The information processing method provided by the embodiment of the present application is described in detail below.

[0025] See also Figure 1 , Figure 1 This is a schematic diagram of a system architecture of an information processing system 10 provided in an embodiment of the present application. The information processing system 10 at least includes a server 101 and a first electronic device 102 .

[0026] The server 101 may be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, network acceleration services (Content Delivery Network, CDN), and big data and artificial intelligence platforms. The first electronic device 102 includes but is not limited to smart phones, tablet computers, laptop computers, desktop computers, smart watches, e-book readers, MP3 (Moving Picture Experts Group Audio Layer III, Moving Picture Experts Group Audio Layer 3) players, MP4 (Moving Picture Experts Group Audio Layer IV, Moving Picture Experts Group Audio Layer 4) players, and smart speakers.

[0027] It can be understood that the server 101 and the first electronic device 102 can be directly or indirectly connected via wired or wireless communication.

[0028] In some embodiments, the server 101 can receive a first request message sent by the first electronic device 102, where the first request message carries a first user identity of the current user of the first electronic device 102; determine a generative large language model corresponding to the first user identity from multiple generative large language models as a target model, where multiple generative large language models correspond one-to-one to multiple preset user identities, and the generative large language model is used to identify the intent of the request message and output feedback information corresponding to the intent identification result, where the target model is pre-trained based on a first training sample set associated with the first user identity, and the first training sample set is generated based on context information corresponding to the first user identity in at least one application installed on the first electronic device; input the first request message into the target model to obtain first feedback information corresponding to the first request message; and send the first feedback information to the first electronic device 102.

[0029] Please refer to Figure 2 , Figure 2 A flowchart of an information processing method provided in an embodiment of the present application is applied to a server. Figure 2The information processing method provided in the embodiment of the present application is described in detail. The information processing method may include the following steps:

[0030] Step S210: receiving a first request message sent by a first electronic device, wherein the first request message carries a first user identity of a current user of the first electronic device.

[0031] Step S220: Determine a generative large language model corresponding to the first user identity from multiple generative large language models as a target model, multiple generative large language models correspond one-to-one to multiple preset user identity identifiers, the generative large language model is used to identify the intent of the request information and output feedback information corresponding to the intention identification result, the target model is pre-trained based on a first training sample set associated with the first user identity identifier, and the first training sample set is generated based on context information corresponding to the first user identity identifier in at least one application installed on the first electronic device.

[0032] In this embodiment, the first request information at least includes the request content input by the current user of the first electronic device, and the input methods include but are not limited to text input, voice input, and visual input; it can be understood that in order to enable the server to distinguish the identity of the requester corresponding to the received request information, the first electronic device will send the first user identity identifier of the current user of the first electronic device to the server when sending the first request information.

[0033] Among them, the first user identity identifier is a unique identifier that can represent the identity of the user. The identifier can be at least one of the user's name, ID number, email account, mobile phone number and social account, and this embodiment does not limit this.

[0034] It can be understood that the generative large language model is used to identify the intent of the request information and output feedback information corresponding to the intention identification result; in other words, the generative large language model can be regarded as a robot that communicates with the user and can understand and respond to the request information input by the user. The above-mentioned multiple generative large language models correspond one-to-one to multiple preset user identities, that is, each generative large language model is pre-trained for a preset user identity. Specifically, the context information of the preset user identity in at least one application in the electronic device used by the preset user identity is obtained, and a training sample set is generated based on the context information; the large language model is fine-tuned and supervised based on the training sample set. That is to say, each generative large language model can learn the preferences and behaviors of its corresponding preset user identity, which is equivalent to creating a digital twin version of the owner.

[0035] Based on this, after obtaining the first user identity carried by the first request information, the generative large language model corresponding to the first user identity can be determined from multiple generative large language models as the target model. Among them, the target model is pre-trained based on the first training sample set associated with the first user identity, and the first training sample set is generated based on the context information corresponding to the first user identity in at least one application installed on the first electronic device. At this time, the obtained target model can be understood as the server-side, digital twin version of the owner corresponding to the first user identity.

[0036] The above application is at least one of instant messaging applications, shopping applications, game applications, email applications, music applications, navigation applications, office applications, etc. Context information can be understood as information about the interaction between a user and other users or devices in any application. For example, historical chat information of a target user corresponding to a first user identity in an instant messaging application and a corresponding user corresponding to a second user identity can be used as context information corresponding to the first identity in the instant messaging application.

[0037] In some embodiments, the first training sample set includes multiple training samples, each of which includes a first interactive text and a first reply text input by a target user for the first interactive text, and the target user is the user corresponding to the first user identity. The target model is trained in the following manner: the first interactive text is input into the pre-trained first initial model to obtain a second reply text predicted for the first interactive text; based on the degree of difference between the first reply text and the second reply text, a first loss value is determined; according to the first loss value, the first initial model is iteratively trained until the first target condition is met to obtain the target model. At this time, the target model is frozen, and the target model is set as an inference model. Freezing the target model can be understood as saving the current model network parameters of the target model, and setting the target model as an inference model can be understood as the target model can be applied to identify the intent in the request information at this time, and output feedback information corresponding to the intent recognition result.

[0038] Among them, the first target condition can be: detecting that the first loss value is less than a preset value, detecting that the first loss value no longer changes, or the number of training times reaches a preset number, etc. It can be understood that after the first initial model is iteratively trained for multiple training cycles according to the first training sample set, wherein each training cycle includes multiple iterative trainings, the parameters in the first initial model are continuously optimized, so that the above-mentioned detection of the first loss value becomes smaller and smaller, and finally becomes a fixed value, or is less than the above preset value. At this time, it means that the first initial model has converged; of course, it can also be determined that the first initial model has converged after the number of training times reaches the preset number. At this time, the converged first initial model can be used as the above-mentioned target model. Among them, the preset value and the preset number of times are both pre-set, and their values ​​can also be adjusted according to different application scenarios. This embodiment does not limit this.

[0039] It can be understood that the first initial model is a language model that has been pre-trained on a large scale and has a large volume, such as an LLM model. In this way, fine-tuning the pre-trained first initial model using a small amount of first training samples can not only improve the speed of model training, but also enable the first initial model after fine-tuning the model parameters to more specifically identify the first request information corresponding to the target user. In addition, the training process of the target model can be executed in the server mentioned in this application or in other devices, and this application does not limit this.

[0040] In some embodiments, in the fine-tuning training phase of the target model, after the target model converges, the accuracy of the target model can be tested using a test sample set, that is, the accuracy of the feedback data output by the target model for all test samples in the test sample set is statistically calculated; and the accuracy is fed back to the first electronic device. If a confirmation instruction based on the accuracy feedback is received from the first electronic device, it indicates that the target model has reached the accuracy requirement of the first electronic device, and the model network parameters of the target model can be saved at this time; if a continuing training instruction based on the accuracy feedback is received from the first electronic device, it indicates that the target model has not reached the accuracy requirement of the first electronic device, and the server will continue to use the negative samples that the target model failed to accurately predict in the previous training process to continue model training until the accuracy fed back to the first electronic device reaches the accuracy requirement of the first electronic device. Among them, the accuracy requirement of the first electronic device can be a preset accuracy rate value, for example, 90%; or it can be an accuracy value temporarily determined by the user, which is not limited in this embodiment. The trained target model can be regarded as a personal digital version of the Chat Generative Pre-trained Transformer (ChatGPT).

[0041] Step S230: input the first request information into the target model to obtain first feedback information corresponding to the first request information.

[0042] Step S240: sending the first feedback information to the first electronic device.

[0043] Furthermore, after screening out the target model customized for the target user, the first request information sent by the target user through the first electronic device can be input into the target model, and the target model performs intent recognition on the first request information and outputs corresponding feedback information based on the intent recognition result, that is, outputs first feedback information corresponding to the first request information; and then the server sends the first feedback information to the first electronic device.

[0044] In some embodiments, the first feedback information may only include feedback data, which includes at least one of text feedback data, image feedback data, audio feedback data, and video feedback data. For example, the first request information is the text data of "What is passion fruit", and the first feedback information output by the target model may be the text data of "Passiflora edulis Sims, a vine plant of the genus Passiflora in the family Passifloraceae. The stem is glabrous; the leaves are papery and glabrous on both sides; the stipules are linear-lanceolate; the flowers are fragrant and white, the sepals are oblong, the petals are lanceolate, and the ovary is obovate; the fruit is ovoid, the peel is hard, the flowering period is from April to June, and the fruiting period is from July to April of the following year." For another example, the first request information is the text data of "What does passion fruit look like", and the first feedback information output by the target model may be an image data of a passion fruit.

[0045] In other embodiments, the first feedback information may include a start instruction for a target function in a target application in the first electronic device, and feedback data for the target function in the target application, the feedback data including at least one of text feedback data, image feedback data, and audio feedback data. Based on this, the feedback data and the start instruction are sent to the first electronic device, and the start instruction is used to instruct the first electronic device to start the target function in the target application and output feedback data using the target function. Among them, the start instruction for the target function in the target application can be a start instruction generated after the target model recognizes the intention of the first request information and determines the operation corresponding to the reply to the intention; the start instruction and the feedback data can be generated at the same time, or one after the other, and this embodiment does not limit this.

[0046] Exemplarily, the target application is an email application, the target function is an email reply function, and the feedback data is the email reply content generated for the received target email. That is to say, after the server sends the first feedback information to the first target device, the first target device can start the email reply function in the email application, and generate a target reply email for the target email based on the feedback data, and send the target reply email to the electronic device corresponding to the target email. It can be understood that since the target model has already learned the preferences, behaviors and speaking style of the target user corresponding to the first user identity in advance, the generated email reply content is also generated by imitating the target user's email reply tone and reply method. In this way, the target model can be regarded as a personal digital assistant for the target user, which effectively and efficiently helps the target user reply emails; greatly improving the target user's email reply efficiency and work efficiency.

[0047] Similarly, the target model can also help target users read emails, reply to text messages, search for information, and generate and manage schedules, thereby greatly improving user work efficiency and saving time and energy.

[0048] Furthermore, the first request information mentioned above can be actively input by the user; it can also be generated by triggering an application in the first electronic device. For example, after receiving an email, the email application will generate a corresponding first request information to the server based on the received email. Correspondingly, the server can respond to the first request information and generate a corresponding first feedback information to the first electronic device. In other words, the target model can be used to realize functions such as automatic reply to emails, reply to text messages, or manage schedules without the need for active triggering by the user. In other words, the target model on the server side can be used to complete daily chores on behalf of the target user, so that the user can have multiple clones to improve their work efficiency and save time and energy.

[0049] Optionally, the target application can also be a smart home application. In this case, the target function is a control function, and the feedback data is text data such as "a certain home device is about to be controlled." That is to say, the server can identify through the target model that the first request information is to control other devices, for example, to control devices such as smart TVs or smart access control systems. Correspondingly, the server can send feedback data and startup instructions to the first electronic device. The first electronic device can send the startup instructions to the corresponding other devices through the target application to control the other devices to perform startup operations related to the startup instructions, for example, controlling the smart TV to switch channels, turn up or down the volume; for example, controlling the smart access control system to open the door. At the same time, the feedback data can also be displayed through the target application to prompt the target user that the device is about to be controlled.

[0050] Optionally, the target application can also be a sales management application, and correspondingly, the target functions can be sales process statistics functions, sales data statistics and sales analysis functions. After sending feedback data to the first electronic device, the server helps the target user to realize the company's sales management, for example, automating sales processes, counting sales data, providing sales analysis, automating advertising delivery, processing data and advertising analysis and other functions.

[0051] In addition, the target model in the server can also help enterprises achieve digital transformation. For example, it can help enterprises achieve customer relationship management, such as recording customer information, tracking customer behavior, and providing personalized services. This embodiment will not be described one by one.

[0052] The above-mentioned LLM model can realize data interaction with applications in electronic devices through the plug-in function disclosed by ChatGPT.

[0053] In this embodiment, since the target model for processing the first request information is pre-trained based on the first training sample set associated with the first user identity, and the first training sample set is generated based on the context information corresponding to the first user identity in at least one application installed on the first electronic device. It can be seen that the target model has learned the preferences and speaking habits of the user corresponding to the first user identity, which is equivalent to the target model being trained in a targeted and customized manner for the user corresponding to the first user identity. Therefore, using the target model, the first feedback information of the first request information corresponding to the first user identity can be generated more accurately, that is, the accuracy of the information feedback is improved, and the feedback information is also more personalized. The target model can be regarded as a personal digital assistant of the user corresponding to the first user identity. When the server receives the request information corresponding to the first user identity, it can simulate the speaking style and personality of the user corresponding to the first user identity to generate feedback information; and then it can replace the user corresponding to the first user identity to handle daily chats, reply to text messages, reply to emails, order takeout, and work summaries, etc., so that the user can have an extra avatar to improve their work efficiency and save time and energy.

[0054] Please refer to Figure 3 , Figure 3 A flowchart of an information processing method provided in another embodiment of the present application is applied to a server. Figure 3 The information processing method provided in the embodiment of the present application is described in detail. The information processing method may include the following steps:

[0055] Step S310: receiving a first request message sent by a first electronic device, wherein the first request message carries a first user identity of a current user of the first electronic device.

[0056] Step S320: Determine a generative large language model corresponding to the first user identity from multiple generative large language models as a target model, multiple generative large language models correspond one-to-one to multiple preset user identity identifiers, the generative large language model is used to identify the intent of the request information and output feedback information corresponding to the intention identification result, the target model is pre-trained based on a first training sample set associated with the first user identity identifier, and the first training sample set is generated based on context information corresponding to the first user identity identifier in at least one application installed on the first electronic device.

[0057] In this embodiment, the specific implementation of step S310 to step S320 can refer to the content of the above-mentioned embodiment, which will not be repeated here.

[0058] Step S330: If the first request information carries a target image, determine an image recognition model corresponding to the first user identity from a plurality of pre-trained image recognition models as a target recognition model.

[0059] It can be understood that the first request information can include a target image, that is, the target user corresponding to the first user identity can input the target image; since the input form of the target model is required to be in text form, the target image cannot be directly input into the target model for intent recognition, and the target image needs to be first input into the image recognition model for image recognition. Based on this, the image recognition model can be pre-trained to ensure that when the request information includes the target image, the content contained in the target image can be recognized in combination with the image recognition model.

[0060] The above-mentioned multiple image recognition models are pre-trained for multiple preset user identity identifiers, and the multiple image recognition models correspond one-to-one to the multiple preset user identity identifiers; that is, an image recognition model is specifically trained for the preset user corresponding to each preset user identity identifier, and the obtained target recognition model can be understood as an image recognition model specifically trained for the target user corresponding to the first user identity identifier.

[0061] Optionally, the object recognition model is trained by:

[0062] First, an image sample set corresponding to the first user identity is obtained. The image sample set is generated based on images in the album associated with the first user identity. Each sample image in the image sample set carries first label information, and the first label information is used to characterize the image content contained in the sample image.

[0063] Secondly, each sample image is input into the second initial model after pre-training to obtain the predicted second label information for each sample image, and the second label information is used to characterize the image content contained in the sample image identified by the second initial model after pre-training. Among them, the second initial model after pre-training can be a text-image pre-training model (Constrastive Language-Image Pre-training, CLIP), and the CLIP model generally adopts contrast learning in the pre-training stage. Among them, since the image sample set involves the personal privacy of the target user corresponding to the first user identity, the image sample set is used to perform privacy fine-tuning training on the pre-trained CLIP model, that is, other devices except the first electronic device are prohibited from viewing or obtaining the image sample set from the server.

[0064] Finally, based on the difference between the first label information and the second label information, a second loss value is determined; according to the second loss value, the second initial model is iteratively trained until the second target condition is met, thereby obtaining a target recognition model. After the training of the target recognition model is completed, the target recognition model can be multimodally aligned with the target model after fine-tuning training, that is, the recognition result output by the target recognition model is used as the input of the target model.

[0065] Among them, the second target condition can be: the second loss value detected is less than the preset value, the second loss value detected no longer changes, or the number of training times reaches the preset number of times, etc. It can be understood that after the second initial model is iteratively trained for multiple training cycles according to the image sample set, wherein each training cycle includes multiple iterative trainings, the parameters in the second initial model are continuously optimized, so that the above-mentioned second loss value detected becomes smaller and smaller, and finally becomes a fixed value, or is less than the above preset value. At this time, it means that the second initial model has converged; of course, it can also be determined that the second initial model has converged after the number of training times reaches the preset number of times. At this time, the converged second initial model can be used as the above-mentioned target model. Among them, the preset value and the preset number of times are pre-set, and their values ​​can also be adjusted according to different application scenarios. This embodiment does not limit this.

[0066] It can be understood that the second initial model is a large-scale image recognition model that has undergone large-scale pre-training, such as the CLIP model. In this way, fine-tuning the pre-trained first initial model using a small number of image sample sets can not only improve the speed of model training, but also enable the second initial model after fine-tuning the model parameters to more specifically identify the image content contained in the target image input by the target user. In addition, the training process of the target recognition model can be executed in the server mentioned in this application, or in other devices, and this application does not limit this.

[0067] Step S340: input the target image into the target recognition model to obtain a first recognition result corresponding to the target image, wherein the first recognition result is text-type information and is used to characterize the content contained in the target image.

[0068] Furthermore, the target image can be input into the target recognition model for image recognition, and the content contained in the target image can be identified as the first recognition result corresponding to the target image. The first recognition result is text type information. For example, if the target image contains an apple, after the target image is input into the target recognition model, the output first recognition result can be information of the text type "A photo of the apple". In addition, since the image sample set used to train the target recognition model is generated based on the images in the album associated with the first user identity, and the album generally contains daily screenshots of the target user corresponding to the first user identity, images taken by the user himself, and favorite images saved from the network; in this way, the fine-tuned trained target recognition model can more accurately identify the common objects or scene images of the target user, that is, improve the recognition accuracy of the target recognition model for the target image input by the target user; that is, improve the accuracy of the first recognition result, and then improve the accuracy of the first feedback information output based on the target model based on the first recognition result.

[0069] Step S350: input the first recognition result into the target model to obtain first feedback information corresponding to the first request information.

[0070] In other embodiments, if the first request information carries target audio, the server can use automatic speech recognition (ASR) technology to convert the target audio into corresponding target text; then input the target text into the target model to obtain first feedback information corresponding to the first request information.

[0071] Step S360: Send the first feedback information to the first electronic device.

[0072] In this embodiment, the specific implementation of step S350 to step S360 can refer to the content of the above-mentioned embodiment, which will not be repeated here.

[0073] In this embodiment, an image sample set is generated through images in an album associated with a first user identity identifier, and an image recognition model corresponding to the first user identity identifier is trained based on the image sample set; since the album generally contains daily screenshots of the target user corresponding to the first user identity identifier, images taken by the user himself, and favorite images saved from the Internet; in this way, the target recognition model after fine-tuning training can more accurately recognize objects or scene images that are common to the target user, that is, the recognition accuracy of the target recognition model for the target image input by the target user is improved, that is, the accuracy of the first recognition result is improved, and then the accuracy of the first feedback information output based on the target model based on the first recognition result can be improved.

[0074] Please refer to Figure 4 , Figure 4 A flowchart of an information processing method provided in another embodiment of the present application is applied to a first electronic device. Figure 4 The information processing method provided in the embodiment of the present application is described in detail. The information processing method may include the following steps:

[0075] Step S410: Sending a first request message to a server, wherein the first request message carries a first user identity of a current user of the first electronic device.

[0076] In some embodiments, the first request information is generated based on the operation information input by the user. The input operation information may be text information, image information, or audio information, for example, the first request information may be text information or audio information of "Please help me order fish-flavored pork takeaway from a Chinese restaurant near my company".

[0077] Step S420: receiving first feedback information sent by the server based on the first request information, wherein the first feedback information is that the server determines a generative large language model corresponding to the first user identity from multiple generative large language models as a target model; and inputs the first request information into the target model to obtain first feedback information corresponding to the first request information, wherein the multiple generative large language models correspond one-to-one to multiple preset user identity identifiers, and the generative large language model is used to perform intent recognition on the request information and output feedback information corresponding to the intent recognition result, and the target model is pre-trained based on a first training sample set associated with the first user identity identifier, and the first training sample set is generated based on context information corresponding to the first user identity identifier in at least one application installed on the first electronic device.

[0078] The specific implementation of how the first feedback information is generated by the target model in the server can refer to the content of the above embodiment, which will not be repeated here. Based on this, after the server generates the first feedback information, the first electronic device can receive the first feedback information.

[0079] Step S430: output the first feedback information.

[0080] The first feedback information may be outputted in a manner of being displayed on a current display interface of the first electronic device, or may be outputted via audio, which is not limited in this embodiment.

[0081] In some embodiments, after step S430, context information corresponding to the first user identity in at least one target application within the target duration can be obtained at each target duration, as target context information, and the target application is an application installed by the first electronic device with an information sharing function in an enabled state. Each application in the first electronic device can be provided with a status switch control corresponding to the information sharing function, based on which the target user can turn on or off the state of the information sharing function of each application based on the status switch control. The application representation with the information sharing function in the enabled state is an application from which the target user can extract the context information of the application by default, and therefore, context information corresponding to the first user identity in at least one target application can be obtained at each target duration.

[0082] Furthermore, the target context information is sent to the server so that the server updates the target model based on the target context information. That is, the first electronic device will regularly upload the target context information to update the target model to improve the accuracy of the target model's recognition of the target user's request information and the accuracy of the feedback information; that is, the prediction effect and prediction accuracy of the target model are improved through the new target context information.

[0083] In addition, when the first electronic device sends the context information corresponding to the first user identity in at least one target application for the first time, the user's identity and attributes, as well as navigation habit routes, learning of addresses such as home and company, and other related information are also sent to the server. Correspondingly, the server can associate these related information with the first user identity so that the target model can determine the static information attributes of the user corresponding to the first user identity based on these related information.

[0084] In some embodiments, after step S430, the user may also input the second request information based on the first feedback information. For example, the first request information may be "help me order the fish-flavored shredded pork takeaway from the Chinese restaurant near the company", and the first feedback information is "can I order the fish-flavored shredded pork takeaway from the restaurant named xxx?" At this time, the aforementioned second request information is "yes". Based on this, the server can also input the second request information into the target model, and the corresponding target model outputs the second feedback data, and the second feedback data can include the start instruction for the target function in the target application in the first electronic device, and the feedback data for the target function in the target application; for example, the target function can be the takeaway ordering function, and the feedback data can be the text information of "ordered". In this way, the target model can be used to replace the user to complete multiple continuous and complex tasks. Of course, it can not only help users order takeaways, but also help users make phone calls, reply to WeChat, navigate and other interactive tasks, and this embodiment does not limit this.

[0085] In this embodiment, since the target model for processing the first request information sent by the first electronic device is pre-trained based on the first training sample set associated with the first user identity, and the first training sample set is generated based on the context information corresponding to the first user identity in at least one application installed on the first electronic device. It can be seen that the target model has learned the preferences and speaking habits of the user corresponding to the first user identity, which is equivalent to the target model being trained in a targeted and customized manner for the user corresponding to the first user identity. Therefore, using the target model, the first feedback information of the first request information corresponding to the first user identity can be generated more accurately, that is, the accuracy of the information feedback is improved, and the feedback information is also more personalized. The target model can be regarded as a personal digital assistant of the user corresponding to the first user identity. When the server receives the request information corresponding to the first user identity, it can simulate the speaking style and personality of the user corresponding to the first user identity to generate feedback information; and then it can replace the user corresponding to the first user identity to handle daily chats, reply to text messages, reply to emails, order takeout, and work summaries, etc., so that the user can have an extra avatar, improve their work efficiency, and save time and energy.

[0086] Please refer to Figure 5 , Figure 5 A flowchart of an information processing method provided in another embodiment of the present application is applied to a first electronic device. Figure 5 The information processing method provided in the embodiment of the present application is described in detail. The information processing method may include the following steps:

[0087] Step S510: if interaction information sent by the second electronic device is received through the currently running application, and it is detected that the proxy feedback function of the currently running application is in an enabled state, the first request information is generated based on the interaction information.

[0088] In this embodiment, the first request information may not be generated based on the information input by the target user corresponding to the first user identity identifier, but may be generated based on the interactive information sent by the second electronic device received by the application currently running in the first electronic device. For example, the application may be an email application or a text message application, that is, when the email application or the text message application receives the target email or target text message sent by the second electronic device to the first electronic device, it can generate the first request information based on the target email or target text message.

[0089] It is understandable that whether or not to implement automatic reply of SMS or email is generally determined according to user needs. Therefore, a proxy function control with a proxy feedback function can be pre-set for each application, and the target user can turn on or off the proxy feedback function based on the proxy function control. When the proxy feedback function is turned on, the first request information is generated according to the aforementioned interaction information. In other words, when the waiting feedback function is turned on, it is equivalent to the target user turning on the target model's automatic processing authority for the application, realizing personal digital management of the application, and the target model acts as an agent for the target user to process the first request information generated by the application.

[0090] In other embodiments, if the interactive information sent by the second electronic device is received through the currently running application, it is detected that the proxy feedback function of the currently running application is turned on and the target field is not included in the interactive information, the first request information is generated based on the interactive information. Among them, the target field can be pre-set, for example, it can be a field that obviously contains advertising information, or a commonly used field containing spam or spam text messages. In other words, the first request information is not generated for any interactive information sent by the second electronic device, but some interactive information that does not need to be replied will be filtered out, and only the corresponding first request information will be generated for the necessary information, so that the target model can perform proxy processing. In this way, the personal information and property security of the target user are guaranteed to a certain extent, and at the same time, the consumption of computing resources of the server for processing unimportant request information is reduced.

[0091] Step S520: Send a first request message to the server, where the first request message carries a first user identity of the current user of the first electronic device.

[0092] Step S530: receiving first feedback information sent by the server based on the first request information, wherein the first feedback information is that the server determines a generative large language model corresponding to the first user identity from multiple generative large language models as a target model; and inputs the first request information into the target model to obtain first feedback information corresponding to the first request information, wherein the multiple generative large language models correspond one-to-one to multiple preset user identity identifiers, and the generative large language model is used to perform intent recognition on the request information and output feedback information corresponding to the intent recognition result, and the target model is pre-trained based on a first training sample set associated with the first user identity identifier, and the first training sample set is generated based on context information corresponding to the first user identity identifier in at least one application installed on the first electronic device.

[0093] Step S540: output the first feedback information.

[0094] In this embodiment, the specific implementation of step S520 to step S540 can refer to the content of the above-mentioned embodiment, which will not be repeated here.

[0095] In other embodiments, if the interactive information sent by the second electronic device is received through the currently running application, and it is detected that the proxy feedback function of the currently running application is in a closed state, a first prompt message is generated based on the interactive information, and the first prompt message is displayed, and the first prompt message is used for the user to view the interactive information. It can be understood that when the proxy feedback function is in a closed state, it represents that the target user does not want to use the target model deployed by the server to automatically process on his own; at this time, the first prompt message can be generated and displayed to prompt the user to view the interactive information sent by the second electronic device. For example, the first prompt message is displayed in the form of a pop-up window to prompt that the target email sent by the second electronic device is currently received in the email application.

[0096] In this embodiment, the first electronic device will only automatically process applications with the proxy feedback function turned on, realize personal digital management of the application, and the target model will act as a proxy for the target user to process the first request information generated by the application. At this time, the target model is equivalent to the digital twin user of the user corresponding to the first user identity for the application with the proxy feedback function turned on, and can replace the user corresponding to the first user identity to complete the interactive tasks of this part of the application.

[0097] Please refer to Figure 6, which shows a structural block diagram of an information processing device 600 provided in an embodiment of the present application, which is applied to a server. The device 600 may include: an information receiving module 610, a model determination module 620, a feedback information acquisition module 630 and an information feedback module 640.

[0098] The information receiving module 610 is used to receive first request information sent by a first electronic device, where the first request information carries a first user identity of a current user of the first electronic device.

[0099] The model determination module 620 is used to determine a generative large language model corresponding to the first user identity from multiple generative large language models as a target model. Multiple generative large language models correspond one-to-one to multiple preset user identity identifiers. The generative large language model is used to identify the intent of the request information and output feedback information corresponding to the intent identification result. The target model is pre-trained based on a first training sample set associated with the first user identity identifier, and the first training sample set is generated based on context information corresponding to the first user identity identifier in at least one application installed on the first electronic device.

[0100] The feedback information acquisition module 630 is used to input the first request information into the target model to obtain first feedback information corresponding to the first request information.

[0101] The information feedback module 640 is used to send the first feedback information to the first electronic device.

[0102] In some embodiments, the first training sample set includes multiple training samples, each of which includes a first interactive text and a first reply text input by a target user for the first interactive text, and the target user is the user corresponding to the first user identity. The information processing device 600 may also include: a first model training module. The model training module can be specifically used to: input the first interactive text into the pre-trained first initial model to obtain a second reply text predicted for the first interactive text; determine a first loss value based on the degree of difference between the first reply text and the second reply text; and iteratively train the first initial model according to the first loss value until the first target condition is met to obtain the target model.

[0103] In some embodiments, the information processing device 600 may further include: a recognition model determination module and an image recognition module. The recognition model determination module may be used to determine an image recognition model corresponding to the first user identity from a plurality of pre-trained image recognition models as a target recognition model before inputting the first request information into the target model to obtain first feedback information corresponding to the first request information, if the first request information carries a target image. The image recognition module may be used to input the target image into the target recognition model to obtain a first recognition result corresponding to the target image, wherein the first recognition result is text-type information, and the first recognition result is used to characterize the content contained in the target image.

[0104] In this manner, the feedback information acquisition module 630 may be specifically configured to input the first recognition result into the target model to obtain first feedback information corresponding to the first request information.

[0105] In this manner, the information processing device 600 may further include: a second model training module. The second model training module may be specifically used to: obtain an image sample set corresponding to the first user identity identifier, the image sample set is generated based on images in the album associated with the first user identity identifier, each sample image in the image sample set carries first label information, the first label information is used to characterize the image content contained in the sample image; input each sample image into the pre-trained second initial model to obtain predicted second label information for each sample image, the second label information is used to characterize the image content contained in the sample image identified by the pre-trained second initial model; determine a second loss value based on the degree of difference between the first label information and the second label information; iteratively train the second initial model according to the second loss value until the second target condition is met to obtain the target recognition model.

[0106] In some embodiments, the first feedback information includes a start instruction for a target function in a target application in the first electronic device, and feedback data for the target function in the target application, wherein the feedback data includes at least one of text feedback data, image feedback data, and audio feedback data. The information feedback module 640 may be configured to send the feedback data and the start instruction to the first electronic device, wherein the start instruction is configured to instruct the first electronic device to start the target function in the target application and output the feedback data using the target function.

[0107] Please refer to Figure 7, which shows a structural block diagram of an information processing device 700 provided by another embodiment of the present application, which is applied to a first electronic device. The device 700 may include: an information sending module 710, an information receiving module 720 and an information output module 730.

[0108] The information sending module 710 is used to send a first request message to a server, where the first request message carries a first user identity of a current user of the first electronic device.

[0109] The information receiving module 720 is used to receive the first feedback information sent by the server based on the first request information, where the first feedback information is the server determining the generative large language model corresponding to the first user identity from multiple generative large language models as the target model; and inputting the first request information into the target model to obtain the first feedback information corresponding to the first request information, where the multiple generative large language models correspond one-to-one to multiple preset user identity identifiers, and the generative large language model is used to perform intent recognition on the request information and output feedback information corresponding to the intent recognition result, where the target model is pre-trained based on a first training sample set associated with the first user identity identifier, and the first training sample set is generated based on context information corresponding to the first user identity identifier in at least one application installed on the first electronic device.

[0110] The information output module 730 is used to output the first feedback information.

[0111] In some embodiments, the information processing device 700 may further include: a request information generation module and a prompt information generation module. The request information generation module may be used to generate the first request information based on the interaction information if the interaction information sent by the second electronic device is received through the currently running application and the proxy feedback function of the currently running application is detected to be in the turned-on state before the first request information is sent to the server. The prompt information generation module may be used to generate the first prompt information based on the interaction information if the interaction information sent by the second electronic device is received through the currently running application and the proxy feedback function of the currently running application is detected to be in the turned-off state, and the first prompt information is displayed, and the first prompt information is used for the user to view the interaction information.

[0112] In this manner, the request information generation module can also be specifically used for: if interaction information sent by a second electronic device is received through a currently running application, it is detected that the proxy feedback function of the currently running application is turned on and the interaction information does not contain a target field, then the first request information is generated based on the interaction information.

[0113] In some other embodiments, the information processing device 700 may further include: an operation receiving module and a request information generating module. The operation receiving module may be used to receive operation information input by a user before sending the first request information to the server. The request information generating module may be used to generate the first request information based on the operation information.

[0114] In some embodiments, the information processing device 700 may further include: a context information acquisition module and a context information sending module. The context information acquisition module may be used to obtain, after the first feedback information is output, context information corresponding to the first user identity in at least one target application within the target time period at every target time period, as target context information, wherein the target application is an application installed by the first electronic device with an information sharing function turned on. The context information sending module may be used to send the target context information to the server so that the server updates the target model based on the target context information.

[0115] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices and modules can refer to the corresponding processes in the aforementioned method embodiments, and will not be repeated here.

[0116] In several embodiments provided in the present application, the coupling between modules may be electrical, mechanical or other forms of coupling.

[0117] In addition, each functional module in each embodiment of the present application can be integrated into a processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The above integrated modules can be implemented in the form of hardware or software functional modules.

[0118] In summary, the server receives a first request message sent by a first electronic device, wherein the first request message carries a first user identity of the current user of the first electronic device; determines a generative large language model corresponding to the first user identity from a plurality of generative large language models as a target model, wherein the plurality of generative large language models correspond one-to-one to a plurality of preset user identities, and the generative large language model is used to perform intent recognition on the request message, and output feedback information corresponding to the intent recognition result; inputs the first request message into the target model to obtain first feedback information corresponding to the first request message; and sends the first feedback information to the first electronic device. Since the target model is pre-trained based on a first training sample set associated with the first user identity, and the first training sample set is generated based on context information corresponding to the first user identity in at least one application installed on the first electronic device. In this way, the target model can learn the preferences and speaking habits of the user corresponding to the first user identity, which is equivalent to the target model being trained in a targeted and customized manner for the user corresponding to the first user identity. Therefore, using the target model, the first feedback information of the first request message corresponding to the first user identity can be generated more accurately, that is, the accuracy of the information feedback is improved, and the feedback information is also made more personalized.

[0119] The following will be combined Figure 8 A computer device provided by the present application is described.

[0120] Reference Figure 8 , Figure 8The structural block diagram of a computer device 800 provided in an embodiment of the present application is shown, and the above method provided in an embodiment of the present application can be executed by the computer device 800. The computer device can be an electronic terminal with a data processing function, and the electronic terminal includes but is not limited to a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart watch, an e-book reader, an MP3 (Moving Picture Experts Group Audio Layer III, Moving Picture Experts Compression Standard Audio Layer 3) player, an MP4 (Moving Picture Experts Group Audio Layer IV, Moving Picture Experts Compression Standard Audio Layer 4) player, and smart home devices, etc.; of course, the computer device can also be a server, and the server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, network acceleration services (Content Delivery Network, CDN), and basic cloud computing services such as big data and artificial intelligence platforms.

[0121] The computer device 800 in the embodiment of the present application may include one or more of the following components: a processor 801, a memory 802, and one or more applications, wherein the one or more applications may be stored in the memory 802 and configured to be executed by one or more processors 801, and the one or more programs are configured to execute the method as described in the aforementioned method embodiment.

[0122] The processor 801 may include one or more processing cores. The processor 801 uses various interfaces and lines to connect various parts of the entire computer device 800, and executes various functions and processes data of the computer device 800 by running or executing instructions, programs, code sets or instruction sets stored in the memory 802, and calling data stored in the memory 802. Optionally, the processor 801 can be implemented in at least one hardware form of digital signal processing (Digital Signal Processing, DSP), field programmable gate array (Field-Programmable Gate Array, FPGA), and programmable logic array (Programmable Logic Array, PLA). The processor 801 can integrate one or a combination of a central processing unit (Central Processing Unit, CPU), a graphics processing unit (Graphics Processing Unit, GPU) and a modem. Among them, the CPU mainly processes the operating system, user interface and application programs; the GPU is responsible for rendering and drawing display content; and the modem is used to process wireless communications. It can be understood that the above-mentioned modem can also be integrated into the processor 801 and implemented separately through a communication chip.

[0123] The memory 802 may include a random access memory (RAM) or a read-only memory (ROM). The memory 802 may be used to store instructions, programs, codes, code sets or instruction sets. The memory 802 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for implementing at least one function (such as a touch function, a sound playback function, an image playback function, etc.), instructions for implementing the following various method embodiments, etc. The data storage area may also store data (such as the various corresponding relationships described above) created by the computer device 800 during use.

[0124] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices and modules can refer to the corresponding processes in the aforementioned method embodiments, and will not be repeated here.

[0125] In several embodiments provided in the present application, the coupling or direct coupling or communication connection between the modules shown or discussed may be an indirect coupling or communication connection through some interfaces, devices or modules, which may be electrical, mechanical or other forms.

[0126] In addition, each functional module in each embodiment of the present application can be integrated into a processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The above integrated modules can be implemented in the form of hardware or software functional modules.

[0127] Please refer to Fig. 9 , which shows a structural block diagram of a computer-readable storage medium provided in an embodiment of the present application. The computer-readable medium 900 stores program codes, which can be called by a processor to execute the method described in the above method embodiment.

[0128] The computer readable storage medium 900 may be an electronic memory such as a flash memory, an EEPROM (electrically erasable programmable read-only memory), an EPROM, a hard disk, or a ROM. Optionally, the computer readable storage medium 900 includes a non-transitory computer-readable storage medium. The computer readable storage medium 900 has storage space for program code 910 that performs any method step of the above method. These program codes can be read from or written to one or more computer program products. The program code 910 can be compressed, for example, in an appropriate form.

[0129] In some embodiments, a computer program product or computer program is provided, the computer program product or computer program includes computer instructions, the computer instructions are stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the electronic device performs the steps in the above-mentioned method embodiments.

[0130] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. An information processing method, characterized in that: Applied to a server, the method comprises: Receiving a first request message sent by a first electronic device, wherein the first request message carries a first user identity of a current user of the first electronic device; Determining a generative large language model corresponding to the first user identity from a plurality of generative large language models as a target model, wherein the plurality of generative large language models correspond one-to-one to a plurality of preset user identity identifiers, the generative large language model is used to perform intent recognition on the request information, and output feedback information corresponding to the intent recognition result, wherein the target model is pre-trained based on a first training sample set associated with the first user identity identifier, and the first training sample set is generated based on context information corresponding to the first user identity identifier in at least one application installed on the first electronic device; Inputting the first request information into the target model to obtain first feedback information corresponding to the first request information; The first feedback information is sent to the first electronic device.

2. The method according to claim 1, characterized in that The first training sample set includes a plurality of training samples, each of the training samples includes a first interactive text and a first reply text input by a target user for the first interactive text, and the target user is a user corresponding to the first user identity identifier; The target model is trained in the following way: Inputting the first interactive text into a pre-trained first initial model to obtain a second reply text predicted for the first interactive text; Determining a first loss value based on the degree of difference between the first reply text and the second reply text; According to the first loss value, the first initial model is iteratively trained until a first target condition is met, thereby obtaining the target model.

3. The method according to claim 1, characterized in that Before inputting the first request information into the target model to obtain first feedback information corresponding to the first request information, the method further includes: If the first request information carries a target image, determine an image recognition model corresponding to the first user identity from a plurality of pre-trained image recognition models as a target recognition model; Inputting the target image into the target recognition model to obtain a first recognition result corresponding to the target image, wherein the first recognition result is text-type information and is used to characterize the content contained in the target image; The step of inputting the first request information into the target model to obtain first feedback information corresponding to the first request information includes: The first recognition result is input into the target model to obtain first feedback information corresponding to the first request information.

4. The method according to claim 3, characterized in that The target recognition model is trained in the following way: Acquire an image sample set corresponding to the first user identity, where the image sample set is generated based on images in an album associated with the first user identity, and each sample image in the image sample set carries first label information, where the first label information is used to characterize image content contained in the sample image; Inputting each sample image into the pre-trained second initial model to obtain predicted second label information for each sample image, wherein the second label information is used to characterize the image content contained in the sample image identified by the pre-trained second initial model; Determining a second loss value based on a degree of difference between the first label information and the second label information; According to the second loss value, the second initial model is iteratively trained until the second target condition is met, thereby obtaining the target recognition model.

5. The method according to any one of claims 1 to 4, characterized in that: The first feedback information includes a start instruction for a target function in a target application in the first electronic device, and feedback data for the target function in the target application, wherein the feedback data includes at least one of text feedback data, image feedback data, and audio feedback data, and sending the first feedback information to the first electronic device includes: The feedback data and the start-up instruction are sent to the first electronic device, where the start-up instruction is used to instruct the first electronic device to start the target function in the target application and output the feedback data using the target function.

6. An information processing method, characterized in that: Applied to a first electronic device, the method includes: Sending a first request message to a server, wherein the first request message carries a first user identity of a current user of the first electronic device; Receiving first feedback information sent by the server based on the first request information, the first feedback information is a generative large language model corresponding to the first user identity identifier determined by the server from multiple generative large language models as a target model; and inputting the first request information into the target model to obtain first feedback information corresponding to the first request information, multiple generative large language models correspond one-to-one to multiple preset user identity identifiers, the generative large language model is used to perform intent recognition on the request information, and output feedback information corresponding to the intent recognition result, the target model is pre-trained based on a first training sample set associated with the first user identity identifier, and the first training sample set is generated based on context information corresponding to the first user identity identifier in at least one application installed on the first electronic device; The first feedback information is output.

7. The method according to claim 6, characterized in that Before sending the first request information to the server, the method further includes: If the interaction information sent by the second electronic device is received through the currently running application, and it is detected that the proxy feedback function of the currently running application is turned on, generating the first request information based on the interaction information; If interaction information sent by a second electronic device is received through a currently running application, and it is detected that the agent feedback function of the currently running application is in a closed state, first prompt information is generated based on the interaction information and displayed, and the first prompt information is used for the user to view the interaction information.

8. The method according to claim 7, characterized in that If the interaction information sent by the second electronic device is received through the currently running application, and it is detected that the proxy feedback function of the currently running application is turned on, generating the first request information based on the interaction information includes: If interaction information sent by the second electronic device is received through the currently running application, it is detected that the proxy feedback function of the currently running application is turned on and the interaction information does not include a target field, the first request information is generated based on the interaction information.

9. The method according to claim 6, characterized in that Before sending the first request information to the server, the method further includes: Receive operation information input by the user; Based on the operation information, the first request information is generated.

10. The method according to any one of claims 6 to 9, characterized in that: After outputting the first feedback information, the method further includes: At each target duration, obtaining context information corresponding to the first user identity in at least one target application within the target duration as target context information, wherein the target application is an application installed by the first electronic device and has an information sharing function enabled; The target context information is sent to the server, so that the server updates the target model based on the target context information.

11. An information processing device, characterized in that: Applied to a server, the device comprises: An information receiving module, configured to receive a first request message sent by a first electronic device, wherein the first request message carries a first user identity of a current user of the first electronic device; a model determination module, used to determine a generative large language model corresponding to the first user identity from a plurality of generative large language models as a target model, wherein the plurality of generative large language models correspond one-to-one to a plurality of preset user identity identifiers, the generative large language model is used to perform intent recognition on the request information, and output feedback information corresponding to the intent recognition result, the target model is pre-trained based on a first training sample set associated with the first user identity identifier, and the first training sample set is generated based on context information corresponding to the first user identity identifier in at least one application installed on the first electronic device; a feedback information acquisition module, configured to input the first request information into the target model to obtain first feedback information corresponding to the first request information; An information feedback module is used to send the first feedback information to the first electronic device.

12. An information processing device, characterized in that: Applied to a first electronic device, the device comprises: An information sending module, used for sending a first request message to a server, wherein the first request message carries a first user identity of a current user of the first electronic device; An information receiving module is used to receive first feedback information sent by the server based on the first request information, wherein the first feedback information is a generative large language model corresponding to the first user identity identifier determined by the server from multiple generative large language models as a target model; and the first request information is input into the target model to obtain first feedback information corresponding to the first request information, wherein the multiple generative large language models correspond one-to-one to multiple preset user identity identifiers, the generative large language model is used to perform intent recognition on the request information, and output feedback information corresponding to the intent recognition result, wherein the target model is pre-trained based on a first training sample set associated with the first user identity identifier, and the first training sample set is generated based on context information corresponding to the first user identity identifier in at least one application installed on the first electronic device; An information output module is used to output the first feedback information.

13. A computer device, characterized in that: include: one or more processors; Memory; One or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs are configured to execute the method according to any one of claims 1 to 10.

14. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program codes, which can be called by a processor to execute the method according to any one of claims 1 to 10.

Images