Artificial intelligence device dialog control method, apparatus and device, and medium

By obtaining user portrait information and inputting it into the dialogue generation model, the problem of lack of initiative and personalization of conversations between artificial intelligence devices is solved, and accurate and personalized dialogue with users is achieved, improving the user experience.

WO2024230570A1PCT designated stage expired Publication Date: 2024-11-14SHANGHAI SOULGATE TECH CO LTD

Patent Information

Application Number
PCT/CN2024/090719
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Priority Date
2023-05-10
Filing Date
2024-04-30
Publication Date
2024-11-14

AI Technical Summary

Technical Problem

The lack of initiative and personalization in existing AI devices in conversations with users leads to a degraded user experience.

Method used

By obtaining user portrait information, input it into the pre-established dialogue generation model, generating dialogue content corresponding to the user portrait, realizing active dialogue between artificial intelligence devices and users, and using dialogue scenario system model and multimodal generation model to improve dialogue Personalization.

Benefits of technology

It improves the initiative and personalization of dialogue between artificial intelligence devices, improves the user experience, and makes the dialogue between artificial intelligence devices and users more accurate and personalized.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN2024090719_14112024_PF_FP_ABST
    Figure CN2024090719_14112024_PF_FP_ABST
Patent Text Reader

Abstract

An artificial intelligence device dialog control method, apparatus and device, and a medium, relating to the field of artificial intelligence. The method comprises: acquiring information of a user portrait; inputting the information of the user portrait into a pre-established dialog generation model; and, upon obtaining the information of the user portrait, when the current time meets a preset requirement, outputting dialog content corresponding to the information of the user portrait to the user by means of the dialog generation model, so as to actively start a dialog with the user by means of the dialog content. Hence, in the method, when the current time meets the preset requirement, the dialog content is generated by means of the information of the user portrait so as to actively start a dialog with the user, thus achieving the initiative of an artificial intelligence device having a dialog with users; and in addition, the information of the user portrait at least comprises the name, the age and information posted by the user, so that the dialog between the artificial intelligence device and the user is more accurate and more personalized.
Need to check novelty before this filing date? Find Prior Art

Description

A method, device, equipment and medium for controlling dialogue between artificial intelligence devices

[0001] This application claims priority to the Chinese patent application filed with the China Patent Office on May 10, 2023, with application number 202310526411.3 and invention name “A method, device, equipment and medium for controlling dialogue between artificial intelligence devices”, the entire contents of which are incorporated by reference into this application. Technical Field

[0002] The present application relates to the field of artificial intelligence, and in particular to a method, apparatus, device, and medium for controlling conversations between artificial intelligence devices. Background Art

[0003] With the development of artificial intelligence (AI) technology, conversations between AI devices and users are becoming increasingly common. In existing intelligent conversations between AI devices and users, AI devices can only respond after the user sends a message, meaning they passively reply. Furthermore, because AI devices often rely on predefined response templates during conversations, the generated responses lack personalization, reducing the user experience.

[0004] It can be seen that how to improve the initiative and personalization of conversations with artificial intelligence devices, thereby enhancing the user conversation experience, is a technical problem that people in this field urgently need to solve.

[0005] Summary of the Invention

[0006] The purpose of this application is to provide a control method, device, equipment and medium for artificial intelligence device dialogue, which is used to improve the initiative and personalization of artificial intelligence device dialogue, thereby enhancing the user dialogue experience.

[0007] To solve the above technical problems, the present application provides a method for controlling a conversation between an artificial intelligence device, which is applied to an artificial intelligence device. The method includes:

[0008] Obtaining user portrait information; wherein the user portrait information includes at least name, age, and user posting information;

[0009] Inputting the user portrait information into a pre-established dialogue generation model;

[0010] After obtaining the information of the user portrait, if the current time meets the preset requirements, the dialogue content corresponding to the information of the user portrait is output to the user through the dialogue generation model, so as to actively initiate a dialogue with the user through the dialogue content.

[0011] Preferably, the dialogue generation model includes a dialogue scenario system model and / or a multimodal generation model; wherein, the dialogue scenario system model is established based on the information of the user attribute portrait in the information of the historical user portrait; and the multimodal generation model is established based on the information of the user behavior portrait in the information of the historical user portrait.

[0012] Preferably, after outputting the conversation content corresponding to the user portrait information to the user through the conversation generation model, the method further includes:

[0013] In the case of receiving the user's conversation, inputting the user portrait information into a pre-established Prompt-based generation model; wherein the Prompt-based generation model is established based on the user attribute portrait information in the historical user portrait information;

[0014] The conversation content corresponding to the user portrait information is output to the user through the Prompt-based generation model.

[0015] Preferably, establishing the dialogue scenario system model includes:

[0016] Get different levels of dialogue scenes;

[0017] According to the user attribute profile information in the historical user profile information, each of the conversation scenarios is configured with a plurality of conversation contents for engaging the user to form the conversation scenario system model;

[0018] Correspondingly, outputting the conversation content corresponding to the user portrait information to the user through the pre-established conversation scenario system model includes:

[0019] Extracting N pieces of conversation content that match the user portrait information from the pre-established conversation scenario system model based on the user attribute portrait information in the historical user portrait information;

[0020] Get the frequency of current active user contact;

[0021] When the frequency of the current active contact with the user is less than or equal to a threshold, a piece of the conversation content is randomly output from the N pieces of conversation content, and the conversation content is output to the user.

[0022] Preferably, training the multimodal generative model comprises:

[0023] Obtaining the content of posts issued by the user within a preset time period, comment data corresponding to each post, attribute information of the user who issued the post, and attribute information of the user who commented on the post; wherein the content of the post includes at least one of text data and image data;

[0024] In the case where the content of the post includes the text data and the image data, the image data in the post is encoded to obtain an image encoding token; the image encoding token, attribute information of the user who posted the post, the text data, and the comment data token are input into the multimodal generation model for training;

[0025] In the case where the content of the post includes the text data, the attribute information of the user who posted the post, the text data, and the comment data token are input into the multimodal generation model for training;

[0026] Correspondingly, outputting the conversation content corresponding to the user portrait information to the user through the pre-established multimodal generation model includes:

[0027] If the post passes the system's security review and is publicly visible, obtain the current frequency of active user contact;

[0028] When the frequency of the current active contact with the user is less than or equal to a threshold, the conversation content corresponding to the information of the user portrait is output to the user through the pre-established multimodal generation model.

[0029] Preferably, after encoding the image data in the post to obtain an image encoding token or before inputting the attribute information of the user who posted the post, the text data, and the comment data token into the multimodal generation model for training, the method further includes:

[0030] Cleaning the content of the post sent by the user and obtaining cleaned text data and cleaned image data;

[0031] Cutting the cleaned text data according to the size of the multimodal generation model, and obtaining the cut text data;

[0032] Processing the cleaned image data according to a preset number of image data corresponding to each post and obtaining the processed image data;

[0033] Filtering target text data and target image data that meet a first preset requirement from the intercepted text data and the processed image data;

[0034] Correspondingly, encoding the image data in the post to obtain an image encoding token, and inputting the image encoding token, attribute information of the user who posted the post, the text data, and the comment data token into the multimodal generation model for training includes:

[0035] After encoding the target image data in the post, the target image encoding token is obtained, and the target image encoding token, attribute information of the user who posted the post, the target text data, and the comment data token are input into the multimodal generation model for training;

[0036] The step of inputting the attribute information of the user who posted the post, the text data, and the comment data token into the multimodal generation model for training includes:

[0037] The attribute information of the user who posted the post, the target text data, and the comment data token are input into the multimodal generation model for training.

[0038] Preferably, it also includes:

[0039] During the training process of the multimodal generation model, the comment data token is controlled to participate in the model loss calculation of the multimodal generation model, and the content of the post sent by the user is controlled not to participate in the model loss calculation of the multimodal generation model.

[0040] In order to solve the above technical problems, the present application also provides a control device for artificial intelligence device dialogue, which is applied to the artificial intelligence device end, and the control device includes:

[0041] An acquisition module is used to acquire user portrait information; wherein the user portrait information includes at least name, age, and user posting information;

[0042] An input module, configured to input the user portrait information into a pre-established dialogue generation model;

[0043] The output module is used to output the conversation content corresponding to the user portrait information to the user through the conversation generation model after obtaining the user portrait information and when the current time meets the preset requirements, so as to actively initiate a conversation with the user through the conversation content.

[0044] In order to solve the above technical problems, the present application also provides a control device for communicating with an artificial intelligence device, including:

[0045] memory for storing computer programs;

[0046] A processor is used to implement the steps of the above-mentioned control method for artificial intelligence device dialogue when executing the computer program.

[0047] In order to solve the above technical problems, the present application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the control method for the above-mentioned artificial intelligence device dialogue are implemented.

[0048] The present application provides a method for controlling a conversation between an artificial intelligence device, which is applied to an artificial intelligence device. The method includes: obtaining user portrait information; inputting the user portrait information into a pre-established conversation generation model; starting from obtaining the user portrait information, when the current time meets the preset requirements, outputting conversation content corresponding to the user portrait information to the user through the conversation generation model, so as to proactively initiate a conversation with the user through the conversation content. It can be seen that in this method, when the current time meets the preset requirements, the conversation content is generated by the user portrait information to proactively initiate a conversation with the user, thereby realizing the initiative of the conversation between the artificial intelligence device and the user; and compared to the previous method in which the artificial intelligence device relies on a predefined reply template to reply, in the method provided by the present application, since the user portrait information includes at least the name, age, and user posting information, the method of generating conversation content based on the user portrait information makes the conversation between the artificial intelligence device and the user more accurate and personalized.

[0049] In addition, the present application also provides a control device for artificial intelligence device dialogue, a control device for artificial intelligence device dialogue, and a computer-readable storage medium, which have the same or corresponding technical features as the control method for artificial intelligence device dialogue mentioned above and have the same effects as above. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0051] FIG1 is a flow chart of a method for controlling an artificial intelligence device dialogue applied to an artificial intelligence device end according to an embodiment of the present application;

[0052] FIG2 is a schematic diagram of an active dialogue scenario system provided in an embodiment of the present application;

[0053] FIG3 is a schematic diagram of a multimodal combination method provided in an embodiment of the present application;

[0054] FIG4 is a flowchart of a method for generating conversation content that actively reaches out to users using a multimodal generation model, provided by an embodiment of the present application;

[0055] FIG5 is a structural diagram of a control device for communicating with an artificial intelligence device according to an embodiment of the present application;

[0056] FIG6 is a structural diagram of a control device for communicating with an artificial intelligence device provided in another embodiment of the present application. DETAILED DESCRIPTION

[0057] The following will be combined with the accompanying drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0058] The core of this application is to provide a control method, device, equipment and medium for artificial intelligence device dialogue, which is used to improve the initiative and personalization of artificial intelligence device dialogue, thereby enhancing the user dialogue experience.

[0059] In order to enable those skilled in the art to better understand the present application, the present application is further described in detail below in conjunction with the accompanying drawings and specific implementation methods. FIG1 is a flow chart of a control method for an artificial intelligence device dialogue applied to an artificial intelligence device end provided by an embodiment of the present application. As shown in FIG1 , the control method includes:

[0060] S10: Obtain user portrait information;

[0061] S11: Input the user portrait information into the pre-established dialogue generation model;

[0062] S12: After obtaining the user portrait information, if the current time meets the preset requirements, the dialogue content corresponding to the user portrait information is output to the user through the dialogue generation model, so as to actively initiate a dialogue with the user through the dialogue content.

[0063] In order to make the content of the generated conversation more personalized, in an embodiment of the present application, user portrait information is first obtained. The user portrait is the result of describing and analyzing the user's basic attributes, behaviors, interests and other information. Specifically, the user portrait information includes user attribute portrait information and user behavior portrait information. The user attribute portrait information includes at least name and age; the user behavior portrait information includes at least user posting information. There is no limitation on the collected user portrait information. In practice, in order to generate more accurate and personalized conversation content, more user portrait information can be collected. For example, in the collected user portrait information, the user attribute portrait information may include not only name and age, but also gender, hometown, address, emotional state, etc.

[0064] After obtaining the user portrait information, the user portrait information is input into a pre-established dialogue generation model. There is no limitation on the specific dialogue generation model, as long as it can generate dialogue content. Since the user portrait information contains information about the user attribute portrait and the user behavior portrait, a preferred embodiment is that the dialogue generation model includes a dialogue scenario system model and / or a multimodal generation model; wherein the dialogue scenario system model is established based on the user attribute portrait information in the historical user portrait information; and the multimodal generation model is established based on the user behavior portrait information in the historical user portrait information. After the user portrait information is input into the pre-established dialogue generation model, personalized dialogue content can be generated. It should be noted that the historical user portrait information is used to train the model. After obtaining the user portrait information, the user portrait information is judged. If the user portrait information only contains the user attribute portrait information, the dialogue scenario system model is used to generate dialogue content corresponding to the user portrait information; if the user portrait information only contains the user behavior portrait information, the multimodal generation model is used to generate dialogue content corresponding to the user portrait information; if the user portrait information contains both the user attribute portrait information and the user behavior portrait information, the dialogue scenario system model and the multimodal generation model can be used respectively to generate dialogue content corresponding to the user portrait information.

[0065] In the current process of dialogue between artificial intelligence devices and users, artificial intelligence devices only respond to dialogues initiated by users and do not actively contact users to respond, making the dialogue between artificial intelligence devices lack initiative and reducing the user's experience of dialogue with artificial intelligence devices. Therefore, in this embodiment, after the artificial intelligence device starts to obtain information about the user's profile, if the current time meets the preset requirements, the dialogue generation model outputs the dialogue content corresponding to the user's profile information to the user. It should be noted that the preset requirements here refer to the situation where the artificial intelligence device starts to obtain information about the user's profile and no dialogue is received from the user within the current time; or if the artificial intelligence device receives a dialogue from the user (which can be one or more rounds of dialogue) within the current time, the dialogue is interrupted. At this time, the dialogue content generated by the dialogue generation model can be used to actively contact the user. In practice, if active dialogue with the user is frequently triggered, the user's experience of dialogue with the artificial intelligence device may be reduced. Therefore, in practice, whether to continue active dialogue with the user can be determined based on the number of active contacts with the user.

[0066] This embodiment provides a method for controlling conversations with an artificial intelligence device, which is applied to an artificial intelligence device. The method includes: obtaining user portrait information; inputting the user portrait information into a pre-established conversation generation model; and after obtaining the user portrait information, if the current time meets preset requirements, outputting conversation content corresponding to the user portrait information to the user through the conversation generation model, so as to proactively initiate a conversation with the user through the conversation content. It can be seen that in this method, if the current time meets the preset requirements, the conversation content is generated based on the user portrait information to proactively initiate a conversation with the user, thereby achieving the initiative of the conversation between the artificial intelligence device and the user. Moreover, compared to the previous method in which the artificial intelligence device relies on a predefined reply template to reply, in the method provided by this application, since the user portrait information includes at least the name, age, and user posting information, the method of generating conversation content based on the user portrait information makes the conversation between the artificial intelligence device and the user more accurate and personalized.

[0067] In the above embodiment, the dialogue content output by the dialogue generation model actively triggers the dialogue with the user. After triggering the dialogue with the user, in order to make the dialogue between the artificial intelligence device and the user richer and more personalized, in practice, a preferred embodiment is to output the dialogue content corresponding to the user profile information to the user through the dialogue generation model, and further include:

[0068] When receiving a user's conversation, the user profile information is input into a pre-established Prompt-based generation model; wherein the Prompt-based generation model is established based on the user attribute profile information in the historical user profile information;

[0069] The prompt-based generation model outputs the conversation content corresponding to the user profile information to the user.

[0070] When controlling a Prompt-based generative model, in this embodiment, controlled features are added using a token feature approach. The resulting content after adding controlled features using the token feature approach is as follows: "Time: empty [SPEAKER1] Gender: empty, Age: empty, Attributes: empty [SEP] empty [SEP], Length: empty [SPEAKER2] Gender: male, Age: empty, Attributes: empty [SEP] empty [SEP], Length: short [START] [SPEAKER1] You're amazing [SYS] Very different, right..." The content before [START] is the token feature, and the content after [START] is the conversation content. The feature token and conversation content token are concatenated and incorporated into the model. During training, the feature token is not included in model loss calculations; its influence is eliminated through a loss mask. This feature addition approach ensures consistency between model pre-training and fine-tuning, and also allows for relatively uniform processing of subsequent feature expansion.

[0071] User profile-related attributes, content, and length can be attributed to session and user-dimensional features. Various feature tokens and text tokens are assembled together and put into the model for fine-tuning training.

[0072] In the infer stage, the attribute features of the user portrait can be passed to the model according to instructions to achieve a controlled effect.

[0073] In the method provided in this embodiment, natural language generation based on Prompt technology uses Prompt technology to build a generation model. This technology provides specific input to guide the model to generate output that meets specific needs, which provides a more efficient and flexible generation method for the dialogue system. During a normal conversation, the user's inherent attribute information, such as gender, age, hometown, emotional state, etc., is processed into specific feature token information through Prompt technology and input into the generation model together with the conversation content text token. The output is personalized reply content that meets the user's needs, thereby improving the accuracy and personalization of the conversation.

[0074] In order to proactively initiate a conversation with the user, this embodiment uses a pre-established conversation scenario system model to generate conversation content. When establishing the conversation scenario system model, a preferred implementation is that establishing the conversation scenario system model includes:

[0075] Get different levels of dialogue scenes;

[0076] According to the user attribute profile information in the historical user profile information, multiple conversation contents for engaging the user are configured for each conversation scenario to form a conversation scenario system model;

[0077] Correspondingly, outputting conversation content corresponding to the user profile information to the user through the pre-established conversation scenario system model includes:

[0078] Based on the user attribute profile information in the historical user profile information, N conversation contents that match the user profile information are extracted from the pre-established conversation scenario system model;

[0079] Get the frequency of current active user contact;

[0080] When the frequency of actively contacting the user is less than or equal to the threshold, a conversation content is randomly output from N conversation contents and the conversation content is output to the user.

[0081] Based on publicly available or de-identified user data, user profile features are mined and mapped to a proprietary conversation scenario system. N qualifying opening lines are extracted from this system. Within the constraints of natural frequency control, a random opening line is proactively sent to the user, providing more precise and proactive conversation services. Mined user profile features primarily include gender, city, occupation, hobbies, and specialties.

[0082] Figure 2 is a schematic diagram of an active dialogue scenario system provided by an embodiment of the present application. The active dialogue scenario system primarily includes six primary scenarios, such as User Specialties, User Hobbies, User Occupation, Holidays, Weather, and Daily Life, as shown in Figure 2, and 86 secondary scenarios, such as Sunny Day, Windy Day, and Hot Weather. Each sub-scenario is configured with nearly 100 specific, mined opening lines and greetings. Table 1 lists the opening lines and greetings corresponding to some of the active dialogue scenario systems.

[0083] Table 1 Opening greetings corresponding to some active dialogue scenario systems

[0084] For the same user, the AI ​​device's active contact and message sending function has a special frequency control mechanism to prevent excessive disturbance to the user: the same user will only be actively contacted by the AI ​​device once within N days.

[0085] In the method provided in this embodiment, the dialogue scenario system model established based on user attribute portrait features can provide users with more accurate and proactive dialogue services.

[0086] In the above embodiment, the conversation scenario system model established based on the user attribute profile features generates conversation content that actively reaches the user. In this embodiment, a multimodal generation model based on the user behavior profile features is also provided to generate conversation content that actively reaches the user. In implementation, training the multimodal generation model includes:

[0087] Obtaining the content of posts sent by users within a preset time period, comment data corresponding to each post, attribute information of users who sent the posts, and attribute information of users who commented on the posts; wherein the content of the posts includes at least one of text data and image data;

[0088] When the content of a post contains text data and image data, the image data in the post is encoded to obtain an image encoding token; the image encoding token, attribute information of the user who posted the post, text data, and comment data token are input into a multimodal generative model for training;

[0089] If the content of the post contains text data, the attribute information of the user who posted the post, the text data, and the comment data token are input into the multimodal generation model for training;

[0090] Correspondingly, outputting the conversation content corresponding to the user profile information to the user through the pre-established multimodal generation model includes:

[0091] Once the post has passed the system's security review and is visible to the public, obtain the frequency of active user contact.

[0092] When the frequency of currently actively contacting the user is less than or equal to a threshold, the conversation content corresponding to the user portrait information is output to the user through a pre-established multimodal generation model.

[0093] After encoding the image data in the post, obtaining the image encoding token or before inputting the attribute information, text data, and comment data token of the user who posted the post into the multimodal generation model for training, it also includes:

[0094] Clean the content of user posts and obtain cleaned text data and cleaned image data;

[0095] The cleaned text data is cut according to the size of the multimodal generative model, and the cut text data is obtained;

[0096] Processing the cleaned image data according to a preset number of image data corresponding to each post and obtaining the processed image data;

[0097] Filtering target text data and target image data that meet a first preset requirement from the intercepted text data and the processed image data;

[0098] Correspondingly, the image data in the post is encoded to obtain the image encoding token, and the image encoding token, the attribute information of the user who posted the post, the text data, and the comment data token are input into the multimodal generation model for training, including:

[0099] After encoding the target image data in the post, the target image encoding token is obtained. The target image encoding token, the attribute information of the user who posted the post, the target text data, and the comment data token are input into the multimodal generation model for training;

[0100] Inputting the user's attribute information, text data, and comment data tokens of the post into the multimodal generative model for training includes:

[0101] The attribute information of the user who posted the post, the target text data, and the comment data token are input into the multimodal generation model for training.

[0102] During the training process of the multimodal generation model, the comment data token is controlled to participate in the model loss calculation of the multimodal generation model, and the content of the posts sent by the user is controlled not to participate in the model loss calculation of the multimodal generation model.

[0103] User behavior primarily refers to the posting or commenting behavior of users within the software. Based on this behavior, the AI ​​device uses a multimodal generative model to autonomously generate comment messages to proactively reach and interact with users. The following describes the construction and training process of a multimodal generative model based on user behavior profiles, using specific examples.

[0104] (1) Data processing and construction

[0105] 1. Pull the desensitized postings and corresponding comment data (text and image type data) for the entire year on the platform and clean up abnormal data;

[0106] 2. The cleaned text data is truncated according to the maximum length of the model (512 / 1024). The image data is processed according to the maximum length of 4 images corresponding to each post. If there are less than 4 images, blank default images are used to supplement them. If there are more than 4 images, 4 images are randomly selected and put into the training data.

[0107] 3. Run the training data through the audit model, set the violation threshold, clean the illegal data, and improve data security;

[0108] 4. Mark the training data with the corresponding posting user attribute features and commenting user attribute features, such as the poster's gender, age, posting time, and the poster's city; the commenter's gender, age, and comment time;

[0109] 5. Expand the proportion of high-quality content data through interest entity model recognition, keyword library matching, or invalid data filtering.

[0110] (2) Training of multimodal generative models

[0111] The multimodal generation model used can be GPT3, Vision Transformer (ViT) model. The multimodal generation model used in this embodiment is the ViT model.

[0112] Figure 3 is a schematic diagram of a multimodal combination method provided by an embodiment of the present application. As shown in Figure 3, when the model is trained, the image encoding token and the post and comment text token are spliced ​​together and put into the model. During the training phase, the image encoding token does not participate in the model loss calculation, and the influence is eliminated by the loss mask. When using a multimodal generation model to generate conversation content, concat splicing is performed. The image information in the user's post is encoded by the ViT model and used as a unique token. It is placed in front of the text information encoding of the post through concat splicing as the input of the overall model, and the output of the model is the comment text information corresponding to the post.

[0113] FIG4 is a flow chart of a method for generating conversation content that actively reaches out to users using a multimodal generation model, provided in an embodiment of the present application. As shown in FIG4 , the method includes:

[0114] S13: Obtain user post content;

[0115] S14: The system reviews the content of the post;

[0116] S15: Determine whether the review is passed and visible; if so, proceed to step S16; if not, end;

[0117] S16: Obtain the frequency of actively reaching users;

[0118] S17: Determine whether the frequency of actively contacting the user meets the conditions; if so, proceed to step S18; if not, end;

[0119] S18: Call the multimodal generation model to generate comment replies and push the generated comments to the user;

[0120] S19: Update the number of times the user is actively reached.

[0121] In the method provided in this embodiment, the dialogue generation model established based on the user behavior profile features can provide users with more accurate and proactive dialogue services.

[0122] In the above embodiments, a method for controlling conversations between artificial intelligence devices is described in detail. This application also provides embodiments of a control apparatus for conversations between artificial intelligence devices and corresponding embodiments of the control apparatus for conversations between artificial intelligence devices. It should be noted that this application describes embodiments of the apparatus from two perspectives: one based on functional modules and the other based on hardware.

[0123] FIG5 is a structural diagram of a control device for communicating with an artificial intelligence device according to an embodiment of the present application. Based on the functional modules of this embodiment, the control device includes:

[0124] The acquisition module 10 is used to obtain user portrait information; wherein the user portrait information includes at least name, age, and user posting information;

[0125] Input module 11, used to input user portrait information into a pre-established dialogue generation model;

[0126] The output module 12 is used to output the conversation content corresponding to the user portrait information to the user through the conversation generation model after obtaining the user portrait information and when the current time meets the preset requirements, so as to actively initiate a conversation with the user through the conversation content.

[0127] Since the embodiments of the apparatus part correspond to the embodiments of the method part, the embodiments of the apparatus part refer to the description of the embodiments of the method part and are not described here in detail. The invention has the same beneficial effects as the control method for communicating with an artificial intelligence device mentioned above.

[0128] FIG6 is a structural diagram of a control device for artificial intelligence device dialogue provided by another embodiment of the present application. This embodiment is based on the hardware perspective. As shown in FIG6 , the control device for artificial intelligence device dialogue includes:

[0129] Memory 20, for storing computer programs;

[0130] The processor 21 is configured to implement the steps of the method for controlling the dialogue between artificial intelligence devices as described in the above embodiment when executing a computer program.

[0131] Among them, the processor 21 may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 21 can be implemented in at least one hardware form of a digital signal processor (DSP), a field programmable gate array (FPGA), and a programmable logic array (PLA). The processor 21 may also include a main processor and a coprocessor. The main processor is a processor for processing data in the awake state, also known as a CPU; the coprocessor is a low-power processor for processing data in the standby state. In some embodiments, the processor 21 may be integrated with a GPU, which is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor 21 may also include an artificial intelligence (AI) processor, which is used to process computing operations related to machine learning.

[0132] The memory 20 may include one or more computer-readable storage media, which may be non-transitory. The memory 20 may also include high-speed random access memory, and non-volatile memory, such as one or more disk storage devices, flash memory storage devices. In this embodiment, the memory 20 is at least used to store the following computer program 201, wherein, after the computer program is loaded and executed by the processor 21, it can implement the relevant steps of the control method for the dialogue of the artificial intelligence device disclosed in any of the aforementioned embodiments. In addition, the resources stored in the memory 20 may also include an operating system 202 and data 203, etc., and the storage method may be temporary storage or permanent storage. Among them, the operating system 202 may include Windows, Unix, Linux, etc. The data 203 may include but is not limited to the data involved in the control method for the dialogue of the artificial intelligence device mentioned above.

[0133] In some embodiments, the control device for the artificial intelligence device dialogue may also include a display screen 22, an input and output interface 23, a communication interface 24, a power supply 25 and a communication bus 26.

[0134] Those skilled in the art will understand that the structure shown in FIG6 does not constitute a limitation on the control device for the dialogue between artificial intelligence devices, and may include more or fewer components than shown in the figure.

[0135] The control device for artificial intelligence device dialogue provided in the embodiment of the present application includes a memory and a processor. When the processor executes the program stored in the memory, it can implement the following method: the control method for artificial intelligence device dialogue, the effect is the same as above.

[0136] Finally, the present application also provides an embodiment corresponding to a computer-readable storage medium. The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps described in the above method embodiment.

[0137] It is understandable that if the method in the above embodiment is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and executes all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0138] The computer-readable storage medium provided in this application includes the above-mentioned control method for dialogue with artificial intelligence devices, and the effect is the same as above.

[0139] The above is a detailed introduction to the control method, device, equipment and medium for the dialogue with an artificial intelligence device provided by the present application. The various embodiments in the specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same and similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description. It should be pointed out that for ordinary technicians in this technical field, without departing from the principles of this application, several improvements and modifications can be made to the present application, and these improvements and modifications also fall within the scope of protection of the claims of this application.

[0140] It should also be noted that, in this specification, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variants thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus comprising the element.

Claims

1. A method for controlling a conversation between an artificial intelligence device, characterized in that: Applied to the artificial intelligence device side, the control method includes: Obtaining user portrait information; wherein the user portrait information at least includes name, age, and user posting information; Inputting the user portrait information into a pre-established dialogue generation model; After obtaining the information of the user portrait, when the current time meets the preset requirements, the dialogue content corresponding to the information of the user portrait is output to the user through the dialogue generation model, so as to actively initiate a dialogue with the user through the dialogue content.

2. The method for controlling the dialogue of an artificial intelligence device according to claim 1, characterized in that: The dialogue generation model includes a dialogue scenario system model and / or a multimodal generation model; Among them, the dialogue scenario system model is established based on the information of user attribute portrait in the information of historical user portrait; and the multimodal generation model is established based on the information of user behavior portrait in the information of historical user portrait.

3. The method for controlling the dialogue of an artificial intelligence device according to claim 2, characterized in that: After outputting the dialogue content corresponding to the information of the user portrait to the user through the dialogue generation model, the method further includes: In the case of receiving the user's conversation, inputting the user portrait information into a pre-established Prompt-based generation model; wherein the Prompt-based generation model is established according to the user attribute portrait information in the historical user portrait information; The conversation content corresponding to the information of the user portrait is output to the user through the Prompt-based generation model.

4. The method for controlling the dialogue of an artificial intelligence device according to claim 2 or 3, characterized in that: Establishing the dialogue scenario system model includes: Get different levels of dialogue scenes; According to the information of the user attribute portrait in the information of the historical user portrait, each of the conversation scenes is configured with a plurality of conversation contents for contacting the user to form the conversation scene system model; Correspondingly, the pre-established dialogue scenario system model outputs the user portrait The information corresponding to the conversation content to the user includes: Extracting N pieces of conversation content matching the information of the user portrait from the pre-established conversation scenario system model according to the information of the user attribute portrait in the information of the historical user portrait; Get the frequency of actively reaching users at present; When the frequency of the current active contact with the user is less than or equal to a threshold, one of the conversation contents is randomly output from the N conversation contents, and the conversation content is output to the user.

5. The method for controlling the dialogue of an artificial intelligence device according to claim 2, characterized in that: Training the multimodal generative model comprises: Obtaining the content of posts sent by the user within a preset time period, comment data corresponding to each post, attribute information of the user who sent the post, and attribute information of the user who commented on the post; wherein the content of the post includes at least one of text data and image data; In the case where the content of the post includes the text data and the image data, the image data in the post is encoded to obtain an image encoding token; the image encoding token, the attribute information of the user who issued the post, the text data, and the comment data token are input into the multimodal generation model for training; In the case where the content of the post includes the text data, the attribute information of the user who posted the post, the text data, and the comment data token are input into the multimodal generation model for training; Correspondingly, outputting the conversation content corresponding to the information of the user portrait to the user through the pre-established multimodal generation model includes: When the post passes the security review of the system and is visible to the public, obtain the current frequency of actively reaching the user; When the current frequency of actively contacting the user is less than or equal to a threshold, the conversation content corresponding to the information of the user portrait is output to the user through the pre-established multimodal generation model.

6. The method for controlling the dialogue of an artificial intelligence device according to claim 5, characterized in that: After encoding the image data in the post to obtain the image encoding token or before inputting the attribute information of the user who posted the post, the text data, and the comment data token into the multimodal generation model for training, it also includes: Cleaning the content of the post sent by the user and obtaining cleaned text data and cleaned image data; Cutting the cleaned text data according to the size of the multimodal generation model, and obtaining the cut text data; Processing the cleaned image data according to a preset number of image data corresponding to each post and obtaining the processed image data; Filtering target text data and target image data that meet a first preset requirement from the intercepted text data and the processed image data; Correspondingly, the step of encoding the image data in the post to obtain an image encoding token, and inputting the image encoding token, the attribute information of the user who posted the post, the text data, and the comment data token into the multimodal generation model for training includes: After encoding the target image data in the post, the target image encoding token is obtained, and the target image encoding token, the attribute information of the user who issued the post, the target text data, and the comment data token are input into the multimodal generation model for training; The step of inputting the attribute information of the user who posts the post, the text data, and the comment data token into the multimodal generation model for training includes: The attribute information of the user who posts the post, the target text data, and the comment data token are input into the multimodal generation model for training.

7. The method for controlling the dialogue of an artificial intelligence device according to claim 6, characterized in that: Also includes: During the training process of the multimodal generation model, the comment data token is controlled to participate in the model loss calculation of the multimodal generation model, and the content of the post sent by the user is controlled not to participate in the model loss calculation of the multimodal generation model.

8. A control device for artificial intelligence device dialogue, characterized in that: Applied to an artificial intelligence device, the control device includes: The acquisition module is used to obtain user portrait information; wherein the user portrait information is At least include name, age, and user posting information; An input module, used to input the information of the user portrait into a pre-established dialogue generation model; The output module is used to output the dialogue content corresponding to the user portrait information to the user through the dialogue generation model after obtaining the user portrait information and when the current time meets the preset requirements, so as to actively initiate a dialogue with the user through the dialogue content.

9. A control device for artificial intelligence device dialogue, characterized in that: include: Memory for storing computer programs; A processor, configured to implement the steps of the method for controlling a dialogue between artificial intelligence devices as claimed in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by the processor, the steps of the method for controlling the dialogue of an artificial intelligence device as described in any one of claims 1 to 7 are implemented.

Citation Information

Patent Citations

  • Home service robot cloud multimode dialogue method, device and system

    CN109658928A

  • Online self-service dialogue method and device, computer equipment and computer readable medium

    CN112306326A

  • Active dialogue method and device based on behavior analysis, equipment and storage medium

    CN115527527A

  • Artificial intelligence equipment dialogue control method and device, equipment and medium

    CN116628153A

  • Digital content security and communications system using artificial intelligence (AI) based machine learning and predictive analysis

    US20200293587A1

Cited By

  • Data generation method and device, storage medium and computing equipment

    CN119250041A

  • Heat supply secondary network personalized regulation and control method based on large model and multi-modal portrait

    CN120355158A

  • Dialogue reply method and device based on multiple modes, electronic equipment and medium

    CN120596658A

  • Instruction generation method and device

    CN120723304A

  • Emotion-sharing-driven multi-mode cognitive large model anthropomorphic interaction optimization method

    CN122020539A