Message group sending method and device based on large model, equipment and storage medium

Abstract

The invention provides a message group sending method and device based on a large model, equipment and a storage medium, relates to the technical field of computers, in particular to the technical fields of artificial intelligence, deep learning and large models, and can be applied to application scenes such as intelligent document editing and intelligent assistants. According to the specific implementation scheme, the method comprises the following steps: acquiring historical interaction data between a sender and a plurality of receivers; determining relation role tags, portrait feature tags and message sending opportunities of the multiple receivers according to the historical interaction data; respectively generating message contents corresponding to each receiver in the plurality of receivers according to the message theme selected by the sender and the relation role tags and the portrait feature tags of the plurality of receivers by using a large model; and sending the message to the plurality of receivers according to the message content and the message sending opportunity. According to the scheme disclosed by the invention, personalized group sending messages can be generated and sent, so that the accuracy of the messages and the user experience are improved.

Description

Technical Field

[0001] The present disclosure relates to the field of computer technology, and in particular to the fields of artificial intelligence, deep learning, and large-model intelligent assistant technology. Background Art

[0002] The content of group messages is usually repetitive, and the same content is difficult to get careful reading and positive feedback from the recipient. And if only the recipient's name or nickname is used in the message, and the copy lacks personalized content related to the recipient, it is difficult to show respect and care for the recipient. Summary of the invention

[0003] The present invention provides a large model-based group messaging method, device, equipment and storage medium.

[0004] According to one aspect of the present disclosure, a method for group messaging based on a large model is provided, comprising:

[0005] Get historical interaction data between a sender and multiple receivers;

[0006] Determine the relationship role labels, portrait feature labels and message sending timing of multiple receivers based on historical interaction data; the relationship role labels are used to characterize the classification results of the identity role of the receiver in the relationship between the sender and the receiver;

[0007] Using the large model, based on the message subject selected by the sender, the relationship role labels and the portrait feature labels of the multiple receivers, the message content corresponding to each of the multiple receivers is generated respectively; and

[0008] Send messages to multiple recipients based on the content and timing of the message.

[0009] According to another aspect of the present disclosure, a large model-based group messaging device is provided, comprising:

[0010] A data acquisition module, used to acquire historical interaction data between a sender and multiple receivers;

[0011] An analysis module, used to determine the relationship role labels, portrait feature labels and message sending timing of multiple receivers based on historical interaction data; wherein the relationship role label is used to characterize the classification result of the identity role of the receiver in the relationship between the sender and the receiver;

[0012] A content generation module, for generating message content corresponding to each of the multiple receivers respectively based on the message subject selected by the sender, the relationship role labels and the portrait feature labels of the multiple receivers by using the large model; and

[0013] The sending module is used to send messages to multiple recipients according to the message content and the message sending timing.

[0014] According to another aspect of the present disclosure, there is provided an electronic device, comprising:

[0015] at least one processor; and

[0016] a memory communicatively connected to the at least one processor; wherein,

[0017] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute any method in the embodiments of the present disclosure.

[0018] According to another aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to enable the computer to execute any method according to the embodiments of the present disclosure.

[0019] According to another aspect of the present disclosure, a computer program product is provided, including a computer program, which implements any method according to the embodiments of the present disclosure when executed by a processor.

[0020] The present invention can generate and send personalized group messages, thereby improving the accuracy of messages and user experience.

[0021] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present disclosure.

[0023] Figure 1 is an application scenario diagram provided according to an embodiment of the present disclosure;

[0024] Figure 2 It is a flowchart of a message group sending method based on a large model provided according to an embodiment of the present disclosure;

[0025] Figure 3 Schematic diagram of a method for sending group messages based on a large model according to another embodiment of the present disclosure

[0026] Figure 4 It is a structural diagram of a message group sending device based on a large model provided according to an embodiment of the present disclosure;

[0027] Figure 5is a block diagram of an electronic device for implementing an embodiment of the present disclosure. DETAILED DESCRIPTION

[0028] The following is a description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope of the present disclosure. Similarly, for the sake of clarity and conciseness, the description of well-known functions and structures is omitted in the following description.

[0029] The present disclosure proposes a method for group messaging based on a large model, which can be applied to a data processing device. For example, the device can be deployed in a terminal or a server or other processing device to implement group messaging. Figure 1 The application scenarios shown are as follows: Figure 1 As shown, the application scenario may include a terminal 110 and an analysis server 120. Taking the case where the device to which the method is applied is deployed in the analysis server 120 as an example, the user may send a request to use the group messaging method to the analysis server 120 through the terminal 110, and the analysis server 120 may obtain the historical interaction data between the multiple recipients and the sender in the request from the terminal 110 or a third-party data storage server. After the analysis server 120 uses the group messaging method to analyze the historical interaction data, it may use the large model to generate personalized message content for each recipient, and feed the generated message content back to the terminal 110. The large model may also run on any third-party server 130. The terminal 110 may receive and display the message content, and group message to multiple recipients at a selected time.

[0030] The terminal 110 is connected to the analysis server 120 via a wireless network or a wired network. Optionally, the terminal 110 is a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart watch, a vehicle-mounted terminal, etc., but is not limited thereto. The terminal 110 installs and runs an application that supports group messaging. For example, chat software, information software, communication software, or email software.

[0031] The analysis server 120 and the third-party server can be independent servers, or server clusters or distributed systems, or cloud servers that provide basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, as well as big data and artificial intelligence platforms.

[0032] Figure 2 The following is a flow chart of a method for group messaging based on a large model according to an embodiment of the present disclosure. Figure 2As shown, the method comprises at least the following steps:

[0033] S201. Obtain historical interaction data between a sender and multiple receivers.

[0034] In the disclosed embodiment, both the sender and the receiver refer to messages that need to be sent to a group. When the sender generates a request for sending a group message on the terminal, it is necessary to select multiple communication objects as the recipients of the message. The communication objects can be directly selected from the address book, the friend list of the software, the contact list of the email, or manually input. When the sender's group message request is received, the historical interaction data between the sender and the multiple recipients are obtained according to the multiple recipients selected by the sender to send the group message.

[0035] Historical interaction data includes, but is not limited to, historical message records, interaction data, and social network information. Among them, historical message records may include text data, image data, emoticon data, reposted links, etc. Social network information may include information published by the recipient in social software, such as status, tweets, personal signatures, etc. Interaction data may include interactive operations such as quoting, replying, forwarding, commenting, liking, etc. between the sender and the receiver in a two-person conversation or a group conversation on a certain conversation message, or may be interactive operations such as replying, commenting, liking, etc. on a certain social network message.

[0036] The scope of acquisition of historical interaction data can be selected and limited by the sender and authorized by the sender. In addition, in order to further protect personal privacy, when acquiring historical interaction data, the recipient's personal information, such as social account information, email address, mobile phone number, personal account ID (Identity Document, identity identification number), etc., can be erased, and each recipient can be given a new virtual identifier, such as receiver 1, receiver 2, receiver 3..., and this virtual identifier is associated with the corresponding receiver's historical interaction data, rather than with any receiver's actual unique identifier. The mapping relationship table between the virtual identifier and the receiver can be retained in the sender's terminal device, so that the historical interaction data obtained by the analysis server cannot be associated with the receiver's actual identity.

[0037] It should be noted that in the technical solution disclosed in this disclosure, historical interaction data is limited to historical data between the sender and the receiver, as well as the social network information disclosed by the receiver, and does not involve the acquisition of other private information of the sender or the receiver. The acquisition, storage and application of personal information are in compliance with the relevant laws and regulations and do not violate public order and good customs.

[0038] S202: Determine the relationship role labels, portrait feature labels and message sending timing of multiple receivers based on the historical interaction data, wherein the relationship role label is used to represent the classification result of the identity role of the receiver in the relationship between the sender and the receiver.

[0039] In the disclosed embodiment, the identity or role of the recipient in the relationship between the recipient and the sender determines how the sender should address the recipient when communicating with the recipient, and what tone and writing style can better reflect respect for the recipient. Therefore, the relationship role label is used to determine the way to address the recipient in the message content on the one hand, and on the other hand, it is also used to influence the tone and style of the message content. For example, a more casual and informal tone can be used for close friends, which may include slang or nicknames. For colleagues / superiors, a formal and professional tone can be used to avoid overly personal content. For customers / partners, politeness and professionalism can be emphasized, which may include recommendations or promotional content.

[0040] The following example lists possible relationship role labels for several types of relationships:

[0041] 1. Social Relationships:

[0042] Close friends, casual friends, classmates, family members, partners, neighbors

[0043] 2. Working relationship:

[0044] Colleagues, superiors, subordinates, customers, partners, suppliers

[0045] 3. Professional Relationship:

[0046] Mentor / Teacher, Student, Consultant / Advisor, Coach, Expert

[0047] 4. Business relations:

[0048] Investors, business partners, competitors, supply chain members

[0049] 5. Interests and hobbies:

[0050] Mutual enthusiasts, community members, sports teammates

[0051] 6. Social Media Relationships:

[0052] Fans, followers, netizens

[0053] 7. Customer Relationship:

[0054] Potential customers, existing customers, VIP customers, after-sales customers

[0055] The relationship role label can be one or more. When there are multiple labels, any two labels should not belong to the same relationship type.

[0056] Portrait feature tags are used to influence the specific content of messages, so that group messages have personalized content for different recipients, and at the same time, make the message content interesting to the recipients.

[0057] The timing of message sending can be understood as the best time to send a message. If a message is sent at this time, when the message reaches the recipient, it is more likely to receive the recipient's timely attention and feedback.

[0058] S203. Using the large model, according to the message subject selected by the sender, the relationship role labels and the portrait feature labels of the multiple receivers, the message content corresponding to each of the multiple receivers is generated respectively.

[0059] In the disclosed embodiment, when initiating a group message request, the sender can edit and input the message subject, or select from preset subjects or templates. For example, holiday greetings, daily care, work notification, project progress report, process transmission, homework assignment, event time notification, commercial promotion, etc. Natural language generation technology can be used to input the message subject, relationship role labels and portrait feature labels of multiple recipients into the big model, so as to use the big model to generate personalized message content that meets the characteristics of each recipient.

[0060] It should be noted that the relationship role label, portrait feature label and message sending time are all obtained by analyzing each recipient individually. Therefore, the message content and message sending time are personalized. However, if some recipients have less or no historical interaction data, the general relationship type can be derived as the relationship role label based on the recipient's nickname, note name, group name, gender and other general information, such as the above-mentioned working relationship, teacher-student relationship, business relationship, etc. For the portrait feature label, the lack of this label does not lead to the inability to generate the message content, but it reduces the personalization of the message content. You can use a more common portrait feature label in a certain type of relationship.

[0061] S204: Send the message to multiple recipients according to the message content and the message sending timing.

[0062] In the disclosed embodiment, after the message content is generated, the corresponding message content can be sent to the recipients according to the message sending timing of each recipient.

[0063] It should be noted that if some recipients have little or no historical interaction data, and it is difficult to analyze the timing of message sending, this option can be omitted, which will not cause the message content to fail to be sent, but only reduce the degree of personalization in the timing of message sending. In the default state, the message can be sent immediately.

[0064] According to the solution of the embodiment of the present disclosure, personalized group messages can be generated and sent, thereby improving the recipient's acceptance of the message and the message reach.

[0065] In a possible implementation, S202 determines the relationship role tags, portrait feature tags, and message sending timings of multiple recipients based on historical interaction data, further comprising the steps of:

[0066] S2021. Determine relationship role labels and portrait feature labels for multiple recipients based on historical interaction data.

[0067] S2022. Determine the time characteristics of the communication behaviors of multiple recipients based on the time information of multiple historical messages included in the historical interaction data.

[0068] S2023. Determine the timing of sending messages to multiple recipients based on the message subject selected by the sender and / or the time characteristics of the communication behaviors of multiple recipients.

[0069] In the disclosed embodiments, a classification algorithm may be used to determine the relationship role labels, portrait feature labels, and message sending timing of multiple recipients. Common classification algorithms include decision trees, random forests, support vector machines, logistic regression, or deep learning models. Taking a classification model using deep learning technology as an example, a training sample may be created, and the classification model may be trained using feature data contained in the historical interaction data of the training sample and the relationship role labels and portrait feature labels calibrated in advance. The model is adjusted and the optimal combination of parameters is selected to improve the prediction accuracy of the model. Extracting feature data from historical interaction data that is meaningful for classifying relationship role labels and portrait feature labels is a key step in ensuring that the classification model can accurately predict relationship role labels and portrait feature labels. The following are some examples of feature types and specific features that may be used:

[0070] 1. Basic interactive features

[0071] Interaction frequency: The number of times two parties interact with each other over a period of time. A high frequency of interactions usually indicates a close relationship (such as close friends or colleagues).

[0072] Response time: The average time it takes for the other person to reply during each interaction. Quick responses may indicate a stronger relationship.

[0073] Interaction length: The duration of an interaction (such as the length of a phone call or number of chat messages). Longer interactions may indicate a deeper relationship.

[0074] Interaction method: The channel through which the interaction takes place, such as email, phone, social media, instant messaging, etc. Some methods may represent a more formal or informal relationship.

[0075] 2. Content Features

[0076] Interaction topics: Topic modeling, such as the Latent Dirichlet Allocation (LDA) topic model, is used to analyze the main topics of historical interaction content. Different relationship roles may discuss different types of topics. For example, "colleagues" may discuss work, while "friends" may discuss more about personal life.

[0077] Language features: Analyze language style and word usage through Natural Language Processing (NLP) technology:

[0078] Tone: whether formal or informal language is used, whether there are honorifics, exclamation marks, abbreviations, etc.

[0079] Title: The way of addressing the other party in the message, such as "Dear XX", "Mr. / Ms. XX", "Hi, XX", etc.

[0080] Sentiment analysis: Analyze the emotional tendency of the interactive content, whether it is positive, negative or neutral. Interactions with more positive emotions usually indicate more friendly and close relationships.

[0081] 3. Time Characteristics

[0082] Interaction time distribution:

[0083] Temporal regularity: whether the interaction is regular (such as interaction at a fixed time every week or every day). Some relationship roles (such as colleagues) may interact during working hours, while friends may interact during non-working hours.

[0084] Interaction interval: The average time interval between two interactions. Shorter intervals generally mean more frequent interactions, which may indicate a closer relationship.

[0085] Seasonality: Are interactions concentrated during certain time periods (such as holidays, birthdays, anniversaries, etc.) For example, family and close friends may interact more during these times.

[0086] 4. Interaction Direction

[0087] Bidirectionality of interaction: Analyze whether the interaction is one-way (e.g. one party sends a lot of messages, the other party responds less) or two-way (both parties respond frequently). Bidirectional interactions usually indicate an equal and close relationship.

[0088] Ratio of initiators: Who initiates the interaction more often, and whether one party always takes the initiative and the other party responds passively. Clients or supervisors may be more passive receivers, while friends or colleagues may be more equal.

[0089] 5. Social Network Characteristics

[0090] Social Network Analysis: By analyzing the structure of the social network with the sender, more information can be extracted:

[0091] Number of mutual friends or contacts: If two people have a large number of mutual friends, it may indicate a close relationship.

[0092] Position in a social network: The "centrality" metric in a social network can help determine how important someone is in the overall network. People who interact more with central nodes are likely to be close ties.

[0093] 6. Deep Features of Interactive Content

[0094] Lexical richness: If a rich and varied vocabulary is used in the interaction, it may indicate the depth of the interaction and the importance of the relationship.

[0095] Specificity of expression: Highly specific content of interactions, such as detailed discussions of personal matters or projects, often indicates a deeper relationship.

[0096] Emotional diversity: Do both parties express a variety of emotions (joy, concern, anger, etc.) in their interactions? Richer emotional interactions usually mean a closer relationship.

[0097] VII. Geographical location characteristics

[0098] Geographic distance: Do the two parties live in the same or close geographic location. A closer distance may mean more offline interactions and a closer relationship.

[0099] Number of offline meetings: If the interaction record includes offline meeting information, the frequency of offline meetings is an important feature.

[0100] 8. Social Role Characteristics

[0101] Organizational hierarchy: If the interaction is within a company or team, consider the other person’s position and level in the organization. Interaction patterns between superiors, colleagues, and subordinates are often different.

[0102] Social media interaction characteristics: On social media platforms, analyze interactive behaviors such as "likes", "comments", and "shares". Frequent interactions may indicate a close relationship.

[0103] Through the above-mentioned exemplary multi-dimensional features and other potential features, complex and diverse features can be obtained. The classification model can more accurately predict the relationship role labels and portrait feature labels of the recipients, thereby achieving the goal of personalized group messaging.

[0104] In step S2022, the above time features can be reused when analyzing the timing of message sending. The following are several specific methods for analyzing and optimizing the timing of message sending by using time features:

[0105] 1. Regularity of interaction time

[0106] Analyze recipients’ historical interaction times: Determine the time range in which each recipient prefers to interact by analyzing the time periods in which they responded to messages in past interactions. For example, some recipients may respond more during work hours, while others may be active in the evenings or on weekends.

[0107] Method: Extract the specific time points (hours, days, weeks, months) when interactions occurred from historical interaction data to generate a time distribution. Through statistics and visualization (such as histograms and heat maps), we can find the time periods with high-frequency interactions.

[0108] Application: Send messages during these high-frequency periods to increase the chances of your messages being seen and responded to.

[0109] 2. Interaction interval

[0110] Calculate the best time between messages: By observing the time interval between two interactions, find the appropriate time window for re-contact. For example, if the recipient replies to a message every 3 days on average, then send a new message 3 days after the last interaction.

[0111] Method: By analyzing the historical interaction data of each recipient, the average interval between two interactions is calculated, and the appropriate sending cycle is determined. Based on this time interval, messages are sent at the appropriate time point to avoid sending messages too frequently or too long, thereby reducing message interruptions and maintaining the recipient's engagement.

[0112] 3. Rhythmic sending

[0113] Discover regular interaction patterns: Analyze whether there are periodic interaction patterns, such as fixed time nodes such as weekly, monthly, and quarterly. For example, some recipients may have a higher probability of interaction on a certain day of the week (such as Friday afternoon). This rhythm can serve as an important reference for sending messages.

[0114] Method: Use time series analysis tools, such as autocorrelation analysis, to discover the periodic trends of interactions and identify patterns of regular interactions. If it is found that the recipient has regular interaction behavior within a certain period of time, you can align the message sending with its interaction cycle and send messages during its active period.

[0115] 4. Time rules and working hours

[0116] Take advantage of the recipient's work time or free time: Different types of recipients have different levels of activity at different times. For example, business people may be more likely to check emails during work hours, while home users may be more likely to receive messages during non-work hours. Understanding these patterns can help optimize the time of message delivery.

[0117] Methodology: Determine how active a recipient is on weekdays or weekends by comparing their interaction time with their likely work hours. This can be further segmented by combining information such as industry and occupation. Infer the recipient's best work hours based on their industry, occupation, and historical interaction time, ensuring that business recipients are sent emails during work hours, or that casual users are sent messages during non-work hours.

[0118] 5. Interactive seasonal patterns

[0119] Take advantage of special dates or seasonal events: Some engagement may be concentrated around specific dates or holidays, such as the recipient’s birthday, important holidays, or specific industry events. By sending messages during these times, you can increase the relevance and open rates of your messages.

[0120] Method: Analyze past interaction data to see if there are more interactions during specific seasons or dates, such as holidays, anniversaries, etc. Send personalized greetings or relevant messages during these special periods. For example, send customized blessing messages on customers’ birthdays, or send relevant information on the eve of industry conferences.

[0121] 6. Sending based on active window

[0122] Track the recipient’s active time window: For some social media platforms or instant messaging tools, you can monitor the recipient’s online status in real time and send messages when the recipient is in an active time window. This method can ensure that the recipient is in a high interactive state when the message is sent.

[0123] Method: Obtain the real-time activity status of the recipient through the API (Application Programming Interface) or monitoring tools provided by the platform, or infer future activity time through historical activity records. Send messages when the recipient is online or recently active to increase the chance of being viewed and responded to immediately.

[0124] In addition to the above specific operation methods, you can also use machine learning to build time series prediction models, such as ARIMA (Autoregressive Integrated Moving Average), LSTM (Long Short-Term Memory) and other time series prediction models to model the receiver's historical interactive behavior and predict the most suitable time to send messages in the future. Machine learning can comprehensively consider various time characteristics and automatically find the best time to send.

[0125] Use time series models to predict historical interaction time points and infer possible active time periods in the future. Input variables can be the aforementioned interaction frequency, time interval, periodicity, etc. Send messages during the predicted high-activity period to increase the chances of being viewed and responded to.

[0126] For the timing of sending group messages, the urgency and appropriateness of the message subject selected by the sender also need to be considered. For example, messages expressing care on specific dates such as holidays, anniversaries, birthdays, etc. must be sent on the same day, but the specific time of the day can be determined in combination with the time characteristics of the receiver's communication behavior. For message subjects with high urgency, such as work notifications and instruction transmission, the messages should generally be sent immediately, regardless of the receiver's behavioral habits.

[0127] According to the solution of the embodiment of the present disclosure, the active time window of each recipient can be determined according to the individual differences of the recipients, and the timing of message sending can be personalized by utilizing the unique interaction habits of each recipient.

[0128] In a possible implementation, S2021 determines the relationship role labels and portrait feature labels of multiple recipients based on the historical interaction data, further comprising the steps of:

[0129] S2021a. Determine identity characteristics of multiple recipients based on the address information and topic information contained in the historical interaction data.

[0130] S2021b. Determine relationship role labels based on identity features.

[0131] S2021c. Determine the portrait feature label based on the behavior pattern information and topic information contained in the historical interaction data.

[0132] In the disclosed embodiment, for historical interaction data, text records including emails, social media conversations, instant messages, etc. are obtained, and the text is segmented, and then named entity recognition is performed to extract address and topic information from the text, such as names, titles, locations, companies, etc. Address information can reflect the relationship between the sender and the receiver, especially the address in formal or informal occasions. The address information extraction method may include:

[0133] 1. Keyword matching: keyword matching is performed through a predefined list of address words (such as "Mr.", "Ms.", "Dear", etc.).

[0134] 2. Context analysis: Use the context to analyze the specific forms of address, such as "Dear Mr. Li" and "Dear friend" represent different relationships.

[0135] 3. Language model: With the help of pre-trained language models, such as BERT (Bidirectional Encoder Representations from Transformers) and GPT (Generative Pre-trained Transformer), the implicit relationship of the address is understood in context.

[0136] The classification of address information is not limited to:

[0137] 1. Formal titles: such as "Dear", "Manager Li", "Professor Wang", etc., may indicate that the recipient is a client, superior or elder.

[0138] 2. Informal titles: such as "Dear", "Old Friend", "Xiao Li", etc., usually indicate that the recipient is a friend or close contact.

[0139] Different relationship roles usually involve different topics. Topic information can be obtained by analyzing keywords, semantic features, etc. in the interactive content. The methods for extracting topic information can include:

[0140] 1.LDA topic model: Use the LDA model to extract multiple potential topics from historical interactive content and analyze topic distribution, such as "project progress", "weekend party", "company activities", etc.

[0141] 2. Keyword extraction: Extract high-frequency keywords from interactive content through TF-IDF (Term Frequency-Inverse Document Frequency) or TextRank algorithm.

[0142] 3. Sentiment analysis: Use sentiment analysis to understand the tone and emotional tendency of the interaction and determine whether the topic is more formal (such as work-related) or more casual (such as life, entertainment).

[0143] The categories of topic information include but are not limited to:

[0144] 1. Work / business related: discussing topics such as projects, contracts, collaborations, etc., which may indicate that the relationship role is a client, colleague, or partner.

[0145] 2. Personal life related: discussing topics such as hobbies, holiday plans, family, etc., which may indicate that the relationship roles are friends or family members.

[0146] 3. Professional field discussions: technical and academic topics, which may indicate that the relationship role is a mentor, colleague or industry expert.

[0147] The identity of the recipient can be constructed through the address information and topic information. Specific methods may include:

[0148] 1. Address information features: Combine the form of address (formal or informal) and the person being addressed (position, seniority, etc.) to determine the other person’s identity (such as a boss, client, or friend).

[0149] 2. Topic information characteristics: Based on the topic classification results, the main direction of the interaction content can be determined. More interactions on work-related topics may mean that the recipient is a client or colleague, while more interactions on life-related topics may mean that the recipient is a friend or family member.

[0150] 3. Cross-analysis: Combine the address information with the topic information for cross-analysis. For example, "formal address + work topic" may indicate that the other party is a client or a supervisor, while "informal address + personal topic" may indicate that the other party is a friend. For example:

[0151] Title: Manager Wang, Topic: Project progress → Identity: Work partner (client or supervisor)

[0152] Name: Xiao Zhang, Topic: Weekend party → Identity: Friend

[0153] Title: Dear, Topic: Family gathering → Identity: Close friend or family member

[0154] In step S2021b, the classification model is used to further determine the relationship role label of the recipient according to the identity features. The role label can be subdivided into multiple categories, such as friends, colleagues, customers, bosses, family members, etc. The classification model can map the identity features obtained according to different address information and topic information to the corresponding relationship role label.

[0155] In step S2021c, the address information can be removed from the historical interaction data preprocessed in step S2021a, the topic information can be retained, and the behavior pattern information and emotional tendency information of the recipient can be further analyzed. The image feature label used to describe the personalized characteristics of the recipient can be determined through the topic information, behavior pattern information, emotional tendency information, etc. The image feature label can cover the following aspects:

[0156] Interest tags: reflect the recipient's interests in themes, activities, products or topics.

[0157] Behavioral pattern labels: reflect the recipient’s interaction frequency, active time period, information feedback speed, etc.

[0158] Emotional tendency label: reflects the receiver's emotional expression in the interaction, whether it is more inclined to positive, negative or neutral emotions.

[0159] According to the scheme of the embodiment of the present disclosure, the relationship role label of the recipient is determined by the address information and topic information, which can provide a basis for personalized group messaging, ensuring that each recipient receives a personalized message that matches his or her identity and interests; further analysis of historical interaction data is performed to derive the recipient's portrait feature label to help design personalized message content, so that group messaging has higher relevance and appeal to different recipients, thereby improving the effectiveness of the message.

[0160] In a possible implementation, S2021c determines the portrait feature label according to the behavior pattern information and topic information included in the historical interaction data, including at least one of the following methods:

[0161] According to the topic information included in the historical interaction data, a first tag for representing the interests and hobbies of the recipient is determined. The first tag is used to influence the topic type included in the message content.

[0162] According to the proprietary language content included in the historical interaction data, a second tag for characterizing the social group attribute of the recipient is determined. The second tag is used to influence the type of terms included in the message content.

[0163] According to the interaction frequency contained in the historical interaction data, a third tag for characterizing the depth of the relationship between the sender and the receiver is determined. The third tag is used to influence the emotionality of the message content.

[0164] According to the length information of the plurality of historical messages included in the historical interaction data, a fourth tag for characterizing the length of the message content of the receiver is determined. The fourth tag is used to influence the length of the message content.

[0165] According to the feedback speed included in the historical interaction data, a fifth tag for characterizing the receiver's interaction enthusiasm is determined. The fifth tag is used to influence the degree of interactivity of the message content.

[0166] According to the cultural background information contained in the historical interaction data, a sixth tag for characterizing the cultural background of the receiver is determined. The sixth tag is used to influence the topic scope of the message content.

[0167] In the disclosed embodiment, for the first tag, the extracted main topic information is mapped with various interest categories to determine the interest and hobby tags of the recipient. For example, if the topic of "travel" appears frequently, it is marked as "travel enthusiast". According to the interest and hobby tags of the recipient, the topic type in the group message is adjusted in a targeted manner to push content that suits their interests. For example, the latest technology information, product release information, etc. are added to the message content generated for "technology enthusiasts". Health and sports-related content are added to the message content generated for sports enthusiasts. Recommended music, concert information or music services are added to the message content for music lovers.

[0168] For the second label, the unique language style, expression or vocabulary used by different groups in the interaction can be identified to generate a second label for characterizing the social group attributes of the recipient, thereby deciding whether to use the exclusive language or content of a specific group. Specific methods may include:

[0169] Analysis of language style classification: Through NLP technology, the language style of text content is classified to identify the language characteristics unique to different groups. Young groups usually contain a lot of Internet terms, emoticons, slang and abbreviations. Middle-aged groups usually use more formal and stable language, with more complete sentences, less slang and emoticons, and the content tends to be practical.

[0170] Keyword and symbol analysis: Use TF-IDF or word frequency analysis techniques to identify words, phrases, or symbols commonly used by specific groups in the interaction data. For example, popular words, abbreviations, and emoticons used by young groups can be used as iconic language features.

[0171] The extracted language content is matched with the predefined proprietary language vocabulary to determine whether the recipient belongs to a specific social group, such as the young group, the middle-aged group, the elderly group, the working group, the student group, etc. Based on the matching results, the corresponding social group attribute label is generated.

[0172] Young group label: If a large number of Internet terms, abbreviations or emoticons appear in the recipient's historical interaction content, a "young group" label can be generated.

[0173] Middle-aged group label: If the recipient's language style in interaction is more formal and the vocabulary used is more traditional and stable, a "middle-aged group" label is generated.

[0174] For the third label, the total number of interactions between the sender and the receiver (the frequency of sending and receiving messages) can be counted, and the average response time and reply interval of the message can be calculated. The receivers are divided into different levels of relationship depth according to the interaction frequency. For example: a high interaction frequency indicates a deep relationship. A medium interaction frequency indicates an average relationship. A low interaction frequency indicates a shallow relationship. Generate the corresponding relationship depth label based on the interaction frequency data. Assign a relationship depth label to each receiver to reflect the degree of relationship between the sender and the receiver.

[0175] For the fourth label, historical messages can be obtained from the historical interaction data, and the length of each message can be counted, including the number of characters, words, sentences, voice duration, etc. These statistics can be used to describe the level of detail of the message. Then the average message length of each recipient is calculated. If the length of the recipient's historical messages is generally short, it indicates that they may prefer concise information. If the length of the recipient's historical messages is generally long, it indicates that they may prefer detailed information. Therefore, according to the statistical results of the recipient's message length, it is classified into the corresponding message length label. For recipients with shorter historical message lengths, a "concise" label is generated. For recipients with longer historical message lengths, a "detailed" label is generated.

[0176] For the fifth label, the feedback speed after sending the message (such as response time, number of clicks, time interval between replies, etc.) can be analyzed from the interaction data with the receiver. Calculate the average response time of the receiver to each message. The faster the feedback, the more active the interaction. Record whether the receiver frequently participates in the interaction, such as reply, click, expression reaction, etc. If the receiver responds quickly in most interactions, it means that he tends to actively interact and has a high degree of interaction enthusiasm. If the receiver's response time is long, or there are few responses, it means that the interaction enthusiasm is low. Analyze the data of each receiver and assign the corresponding positivity label. For example: Recipients whose average response time is less than a certain threshold are marked as "positive feedback". Recipients whose average response time is greater than a certain threshold are marked as "negative feedback".

[0177] For the sixth tag, the sixth tag may be determined according to the cultural background information in the historical interaction data, and the tag may be used to influence the topic scope of the message content. The specific steps may include:

[0178] Collect data on the recipient's geographic location, the language the recipient uses or the language style in historical interactions, and the recipient's discussion of religion or holidays in previous interactions (such as holiday greetings).

[0179] Predict the cultural background of the recipient based on the geographic location information. For example, if the recipient's interactive information shows that he is located in Europe, Japan, the United States, etc., his cultural background can be inferred based on these geographic locations.

[0180] Analyze the language style and intonation used by the recipient based on language and expression, and identify the cultural background. For example, the use of American English, British English, Japanese, etc. are associated with different cultural backgrounds.

[0181] Analyze whether the recipient mentions specific holidays in the interaction based on religious background, infer their religious belief background, and mark whether sensitive topics should be avoided.

[0182] Assign appropriate cultural background labels to each recipient based on geographic location, language, and religious background. Label examples:

[0183] Country labels: "United States", "Japan", "France".

[0184] Language tags: "American English", "French".

[0185] Religious labels: certain religion, no religion.

[0186] According to the solution of the embodiment of the present disclosure, through multi-dimensional interest feature tags, the content of each message is highly personalized in terms of language, tone, form, topic and expression, ensuring that the content is uniquely targeted and attractive to each recipient.

[0187] In a possible implementation, S203 uses the big model to generate message content corresponding to each of the multiple receivers according to the message subject selected by the sender, the relationship role tags and the portrait feature tags of the multiple receivers, and further includes the steps of:

[0188] S2031. Determine a first prompt word according to the message subject selected by the sender.

[0189] In the disclosed embodiment, the message subject selected by the sender is obtained as the main purpose or core content of the group message, such as new product release, event invitation, holiday greetings, etc. Key information can be extracted from the message subject as the first prompt word. For example:

[0190] A theme: New product release, first prompt word: New product release, product introduction, latest features

[0191] B theme: event invitation, first prompt word: event invitation, participation opportunity, exclusive experience

[0192] Theme C: Mid-Autumn Festival Blessings, the first prompt words: Mid-Autumn Festival, Happy Holidays, Blessings.

[0193] S2032: Determine a second prompt word according to the relationship role labels of the multiple recipients.

[0194] The second prompt word is used to influence the tone and style of the message content. The relationship role label itself can be used as part of the second prompt word, or a guiding or requiring prompt word with corresponding tone and style can be given according to the relationship role label, for example:

[0195] A Recipient’s relationship role label: Friend, Second prompt word: The tone can be relaxed and friendly, and the style can be casual.

[0196] B Receiver’s relationship role label: customer, second prompt: A formal, polite, and respectful tone needs to be used.

[0197] C Recipient’s relationship role label: superior. Second prompt: The other party is my superior. The message content must be generated using a formal, professional, and respectful tone and style.

[0198] S2033. Determine a third prompt word according to the image feature label.

[0199] The third prompt word is used to influence at least one of the topic type, word type, emotional level, message content length, interactivity level or topic scope of the message content. The portrait feature label itself can be used as part of the third prompt word, or the portrait feature label can be expanded to give a clearer prompt. For example:

[0200] A recipient’s portrait feature tags: technology enthusiast, active in interaction, likes short content, then the third prompt words may include: technology topics, short, and highly interactive.

[0201] B. Characteristic labels of recipient’s portrait: middle-aged group, conservative, British cultural background. The third prompt words may include: practical topics in life, formal terms, and British cultural sensitivity.

[0202] C. The portrait characteristic tags of the recipient: young people, music lovers, not active in interaction. The third prompt words can include: music topics, relaxed tone, and short messages.

[0203] The portrait feature tag includes at least one of the first tag, the second tag, the third tag, the fourth tag, the fifth tag, and the sixth tag. The first tag is used to influence the topic type included in the message content.

[0204] The impact of the second tag on the message content can be divided into:

[0205] Use more casual, internet-friendly expressions in messages for younger audiences, for example, by adding emoticons, internet buzzwords, or more interactive language.

[0206] For the middle-aged group, the message content is more stable, using a more formal and traditional language style, the content is close to the practicality of life, and avoids using too much slang and informal expressions.

[0207] The impact of the third tag on the message content can be divided into:

[0208] 1. Deep relationship receiver:

[0209] Add personal, emotional touches to your messages. For example, mention a past shared experience, a personal feeling, or a joke in the conversation.

[0210] Create open-ended questions or interactive content that encourages the recipient to continue interacting. For example, "How did the topic we discussed last go?" or "I'm curious what you think about this new project?"

[0211] 2. General relationship recipients:

[0212] Keep your message neutral and objective, and avoid overly emotional or personal language. For example, "Here's our latest product update, we hope it helps."

[0213] Be concise and direct: Provide direct information, avoid lengthy content and complex interactions, and mainly use one-way transmission.

[0214] Include a simple survey or request for feedback, but don’t get too in-depth or personal. For example, “Please review the attachment and let me know if it meets your needs.”

[0215] 3. Shallow relationship receiver:

[0216] Use formal, standard expressions and avoid overly personal language or emotional content.

[0217] Messages are primarily informational and do not require specific interaction or feedback. For example, “Here is our new product catalog, please check it out.”

[0218] The impact of the fourth label on the message content can be divided into:

[0219] 1. Simple receiver:

[0220] Create short, focused messages for concise recipients. Avoid lengthy descriptions and get straight to the point. For example: "Update: Our product is now live. Click here to learn more."

[0221] 2. Detailed receiver:

[0222] Produce detailed, informative messages for verbose recipients. Include more context, explanations, and details. Example: "We are excited to introduce our newest product to you. Below is a detailed feature description, user reviews, and usage guide for the product."

[0223] The impact of the fifth tag on the message content can be divided into:

[0224] 1. Positive Feedback Recipient:

[0225] Increase interactive content: For recipients of positive feedback, generate more interactive message content, such as:

[0226] Provide more recommendations, product suggestions, or Q&A content to encourage continued interaction.

[0227] Add open-ended questions or personalize interactions like “What do you think of our new features? We welcome your feedback!”

[0228] Frequent push: You can moderately increase the push frequency of messages and ensure that the content remains fresh.

[0229] 2. Negative feedback recipients:

[0230] Reduce duplicate content: Avoid sending similar content, reduce the frequency of push notifications, and avoid interruptions. For example: "We noticed that you were not interested last time, and the content recommended this time is more in line with your needs."

[0231] Adjust the tone and content: Use a more neutral and gentle tone, more concise content, and topics that are closer to the recipient's potential interests. For example, for content that is not of interest, provide new options, such as "We have new products that may better meet your needs."

[0232] The impact of the sixth tag on the message content may include:

[0233] 1. Geographical location and language influence:

[0234] Localize message content: Based on the recipient's country tag, ensure that the message content is adapted to the local language and culture. For example:

[0235] Use local languages ​​(such as French, German, etc.) or cultural customs to express holiday greetings.

[0236] Avoid using expressions that could cause cultural misunderstandings. For example, avoid greeting people on a specific holiday that is considered normal in some cultures but a topic that should be avoided in other cultures.

[0237] 2. Influence of religious background:

[0238] Neutral topics or religious customization: For recipients with religious backgrounds, avoid touching on sensitive religious topics, especially among non-religious recipients. Customized content can be provided for recipients with religious backgrounds.

[0239] 3. Influence of social habits and customs:

[0242] Cultural style adaptation: adjust the tone and intonation according to the cultural habits of different countries. For example:

[0243] Use a more formal, polite tone when speaking to recipients from a Japanese cultural background.

[0244] Use a more open and direct communication style with recipients from an American cultural background.

[0245] S2034: For each of the multiple receivers, the corresponding first prompt word, second prompt word and third prompt word are combined to obtain a personalized prompt word.

[0246] For each recipient, the first prompt word, the second prompt word, and the third prompt word corresponding to the recipient are combined to obtain personalized prompt words that can reflect the specific characteristics of each recipient, which are used as input text of the large model.

[0247] S2035. Input the personalized prompt words into the large model to obtain the message content corresponding to each recipient.

[0248] The combined personalized prompt words are input into a large language model (ie, a large model), and the final personalized message is obtained according to the output result of the large model.

[0249] Any large model in the relevant technology can be used. Commonly used large language models include Wenxinyiyan, GPT-3 / GPT-4, ChatGPT, BERT, XLNet, BLOOM, etc.

[0250] According to the scheme of the embodiment of the present disclosure, by combining the subject selected by the sender, the relationship role label and the portrait feature label of the receiver, personalized prompt words are generated, and a large language model is input to generate message content, which can greatly improve the personalization level of group messages and the sense of participation of the receiver.

[0251] In a possible implementation, S2034 combines the corresponding first prompt word, second prompt word, and third prompt word for each of the multiple receivers to obtain a personalized prompt word, further comprising the steps of:

[0252] Content preference parameters of a plurality of recipients are determined from interaction frequencies included in the historical interaction data.

[0253] For the first receiver whose content tendency parameter reaches a threshold, a recent conversation topic is determined from the historical interaction data of the first receiver.

[0254] Use a recent conversation topic as the fourth prompt.

[0255] The first prompt word, the second prompt word, the third prompt word and the fourth prompt word corresponding to the first receiver are combined to obtain a personalized prompt word.

[0256] In the disclosed embodiment, the interaction frequency analyzed when determining the relationship depth can be used. When the interaction frequency of a certain receiver reaches a preset threshold, for example, more than 3 interactions in a week, content related to the receiver's recent interactions can be generated, such as unfinished tasks last time, topics discussed previously, etc. The specific operation is to determine the recent conversation topic from the historical interaction data of such first receiver, and add the topic as the fourth prompt word to the personalized prompt word, so that the message content output by the large model includes the recent conversation topic.

[0257] According to the solution of the embodiment of the present disclosure, the receiver will feel that the sender has a high degree of attention to their interaction, especially to continue the previous unfinished matters or continue the discussed topics, which enhances the coherence and personalized experience of the interaction.

[0258] In a possible implementation, S203 uses the big model to generate message content corresponding to each of the multiple receivers according to the message subject selected by the sender, the relationship role tags and the portrait feature tags of the multiple receivers, and further includes the steps of:

[0259] The time impact parameters of the message are determined based on the message subject and the timing of message sending selected by the sender.

[0260] Using the large model, the message content corresponding to each of the multiple receivers is generated according to the message topic selected by the sender, the relationship role labels of multiple receivers, the portrait feature labels and the time impact parameters.

[0261] In the disclosed embodiment, the introduction of time influencing parameters can make group messaging more closely aligned with the recipient's daily rhythm. Specifically, the message content can be adapted to the recipient's state, and the recipient's mood and energy state vary in different time periods. For example, the morning is suitable for receiving encouraging content, while the evening is more suitable for receiving relaxing content that helps the recipient relax. This type of time-adapted content can improve the receptivity of information. The relevance of messages can be improved: in emergencies, more direct expressions can quickly convey key information, while in daily time periods, the content can be more interactive or emotional, avoiding information that is too urgent and blunt. Whether it is a regular notification, reminder, holiday greeting, or emergency, the generated content can be automatically adjusted according to the time of sending, improving the flexibility and adaptability of the message.

[0262] According to the solution of the embodiment of the present disclosure, the tone and content style are adjusted according to the time, so that the recipient can feel a more intimate interactive experience, and the probability of paying attention to and responding to the message is increased.

[0263] In a possible implementation, S202 determines the relationship role tags, portrait feature tags, and message sending timings of multiple recipients based on historical interaction data, further comprising the steps of:

[0264] Extract feedback data of multiple recipients for group messages from historical interaction data.

[0265] Based on historical interaction data and feedback data, determine the relationship role labels, portrait feature labels and message sending timing of multiple recipients.

[0266] In the disclosed embodiments, when it is not the first time to send a group message, feedback data of previous group messages can be obtained from historical interaction data, and the receiver's response to previous group messages can be further analyzed through each receiver's feedback data on the group messages, such as reading rate, click rate, reply status, etc. For example, whether the receiver responds to the topic contained in the message content. Whether the reply to the group message is praiseworthy. The timing of message sending can also be optimized based on the feedback performance in different time periods in the historical data. For example, analyze the time period in which the receiver has a higher open rate in a certain time period, or a time period in which the reply enthusiasm is higher. These time periods can be set as priority sending times to increase the reading rate and interaction rate of the message.

[0267] Through machine learning or rule engines, a model is established based on the collected historical interaction data and feedback data to automatically update the recipient's relationship role label, portrait feature label and optimal sending time.

[0268] According to the solution of the embodiment of the present disclosure, by comprehensively considering the feedback from the receiver and using the reinforcement learning algorithm to optimize the generation and sending strategies, the accuracy of the classification model is continuously optimized.

[0269] In a possible implementation, a flow chart of a method for sending group messages based on a large model is provided. Figure 3 As shown, the method comprises the following steps:

[0270] 1. Data collection and processing:

[0271] Through APIs and data stream processing frameworks (such as Kafka and Spark), the interaction data between users and recipients is collected in real time.

[0272] After cleaning and preprocessing, the data is stored in a distributed database for subsequent analysis.

[0273] 2. Relationship analysis and role classification:

[0274] A deep learning model (such as Transformer) is used to analyze the interaction history between users and receivers and generate relationship labels.

[0275] 3. Personalized content generation:

[0276] The message generation module selects an appropriate template or completely generates personalized content based on the recipient's relationship role, historical interaction data, etc.

[0277] The module supports multi-language processing to ensure that contextual message content is generated in different cultural backgrounds.

[0278] 4. Intelligent scheduling and sending:

[0279] Based on the chat communication behavior data between the sender and the receiver, the best time to send is selected to avoid network congestion during peak hours.

[0280] Supports parallel sending via multiple channels (SMS, email, social platforms).

[0281] 5. Feedback collection and optimization:

[0282] The system collects feedback data by monitoring the recipient's behavior (such as clicks, replies, reading time, etc.).

[0283] Use reinforcement learning algorithms to optimize generation and sending strategies to form a self-evolving system.

[0284] The interaction relationship and interaction process between the modules of the system include:

[0285] 1. The user portrait management module collects data and transmits it to the relationship analysis module.

[0286] 2. The relationship analysis module generates relationship tags and transmits them to the personalized content generation module.

[0287] 3. The personalized content generation module creates personalized messages and hands them over to the intelligent scheduling module.

[0288] 4. The intelligent scheduling module sends a message and transmits the feedback data to the feedback optimization module.

[0289] 5. The feedback optimization module adjusts the strategy of the user portrait management module based on the feedback data.

[0290] According to the solution of the embodiment of the present disclosure, personalized messages can be generated and sent automatically without manual intervention, which greatly reduces the complexity of operation and time cost. Through machine learning algorithms, the system can automatically adjust and optimize the message content to make it more in line with the needs of the sender. It is not only suitable for group sending of holiday greetings, but also can be applied to various business scenarios, such as customer relationship management, marketing promotion, customer care, etc.

[0291] Figure 4 Schematic diagram of a large model-based message group sending device according to an embodiment of the present disclosure. Figure 4 As shown, the device at least includes:

[0292] The data acquisition module 401 is used to acquire historical interaction data between a sender and multiple receivers.

[0293] The analysis module 402 is used to determine the relationship role labels, portrait feature labels and message sending timing of multiple receivers based on the historical interaction data. The relationship role label is used to represent the classification result of the identity role of the receiver in the relationship between the sender and the receiver.

[0294] The content generation module 403 is used to generate message content corresponding to each of the multiple receivers using the large model according to the message subject selected by the sender, the relationship role labels and the portrait feature labels of the multiple receivers. And

[0295] The sending module 404 is used to send a message to multiple recipients according to the message content and the message sending timing.

[0296] In a possible implementation, the analysis module 402 includes:

[0297] The label determination submodule is used to determine the relationship role labels and portrait feature labels of multiple recipients based on historical interaction data.

[0298] The information determination submodule is used to determine the time characteristics of the communication behaviors of multiple receivers based on the time information of multiple historical messages contained in the historical interaction data.

[0299] The timing determination submodule is used to determine the message sending timing for multiple receivers according to the message subject selected by the sender and / or the time characteristics of the communication behaviors of multiple receivers.

[0300] In a possible implementation, the label determination submodule is used to:

[0301] The identity characteristics of multiple recipients are determined based on the address information and topic information contained in the historical interaction data.

[0302] Determine relationship role labels based on identity characteristics.

[0303] Determine the portrait feature labels based on the behavioral pattern information and topic information contained in the historical interaction data.

[0304] In a possible implementation manner, the tag determination submodule is further configured to perform at least one of the following manners:

[0305] According to the topic information included in the historical interaction data, a first tag for representing the interests and hobbies of the recipient is determined. The first tag is used to influence the topic type included in the message content.

[0306] According to the proprietary language content included in the historical interaction data, a second tag for characterizing the social group attribute of the recipient is determined. The second tag is used to influence the type of terms included in the message content.

[0307] According to the interaction frequency contained in the historical interaction data, a third tag for characterizing the depth of the relationship between the sender and the receiver is determined. The third tag is used to influence the emotionality of the message content.

[0308] According to the length information of the plurality of historical messages included in the historical interaction data, a fourth tag for characterizing the length of the message content of the receiver is determined. The fourth tag is used to influence the length of the message content.

[0309] According to the feedback speed included in the historical interaction data, a fifth tag for characterizing the receiver's interaction enthusiasm is determined. The fifth tag is used to influence the degree of interactivity of the message content.

[0310] According to the cultural background information contained in the historical interaction data, a sixth tag for characterizing the cultural background of the receiver is determined. The sixth tag is used to influence the topic scope of the message content.

[0311] In a possible implementation, the content generation module 403 is used to:

[0312] The first prompt word is determined according to the message subject selected by the sender.

[0313] A second prompt word is determined according to the relationship role labels of the multiple recipients.

[0314] Determine the third prompt word based on the image feature label.

[0315] Among them, the second prompt word is used to influence the tone and style of the message content, and the third prompt word is used to influence at least one of the topic type, word type, emotional level, message content length, interactivity level or topic scope of the message content.

[0316] For each of the multiple receivers, the corresponding first prompt word, second prompt word and third prompt word are combined to obtain a personalized prompt word.

[0317] The personalized prompt words are input into the large model to obtain the message content corresponding to each recipient.

[0318] In a possible implementation, the content generation module 403 is further configured to:

[0319] Content preference parameters of a plurality of recipients are determined from interaction frequencies included in the historical interaction data.

[0320] For the first receiver whose content tendency parameter reaches a threshold, a recent conversation topic is determined from the historical interaction data of the first receiver.

[0321] Use a recent conversation topic as the fourth prompt.

[0322] The first prompt word, the second prompt word, the third prompt word and the fourth prompt word corresponding to the first receiver are combined to obtain a personalized prompt word.

[0323] In a possible implementation, the content generation module 403 is further configured to:

[0324] The time impact parameters of the message are determined based on the message subject and the timing of message sending selected by the sender.

[0325] Using the large model, the message content corresponding to each of the multiple receivers is generated according to the message topic selected by the sender, the relationship role labels of multiple receivers, the portrait feature labels and the time impact parameters.

[0326] In a possible implementation, the analysis module 402 is further configured to:

[0327] Extract feedback data of multiple recipients for group messages from historical interaction data.

[0328] Based on historical interaction data and feedback data, determine the relationship role labels, portrait feature labels and message sending timing of multiple recipients.

[0329] For the description of specific functions and examples of each module and submodule of the device in the embodiment of the present disclosure, reference can be made to the relevant description of the corresponding steps in the above method embodiment, which will not be repeated here.

[0330] In the technical solution disclosed herein, the acquisition, storage and application of user personal information involved are in compliance with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0331] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.

[0332] Figure 5 A schematic block diagram of an example electronic device 500 that can be used to implement an embodiment of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or required herein.

[0333] like Figure 5 As shown, the device 500 includes a computing unit 501, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 502 or a computer program loaded from a storage unit 508 into a random access memory (RAM) 503. In the RAM 503, various programs and data required for the operation of the device 500 can also be stored. The computing unit 501, the ROM 502, and the RAM 503 are connected to each other via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0334] A number of components in the device 500 are connected to the I / O interface 505, including: an input unit 506, such as a keyboard, a mouse, etc.; an output unit 507, such as various types of displays, speakers, etc.; a storage unit 508, such as a disk, an optical disk, etc.; and a communication unit 509, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 509 allows the device 500 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0335] The computing unit 501 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 501 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 501 performs the various methods and processes described above, such as a message grouping method based on a large model. For example, in some embodiments, a message grouping method based on a large model may be implemented as a computer software program, which is tangibly included in a machine-readable medium, such as a storage unit 508. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 500 via the ROM 502 and / or the communication unit 509. When the computer program is loaded into the RAM 503 and executed by the computing unit 501, one or more steps of the message grouping method based on the large model described above may be performed. Alternatively, in other embodiments, the computing unit 501 may be configured to execute the large model-based group messaging method in any other appropriate manner (for example, by means of firmware).

[0336] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0337] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0338] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0339] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0340] The systems and techniques described herein may be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), and the Internet.

[0341] A computer system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The relationship of client and server is generated by computer programs running on respective computers and having a client-server relationship with each other. The server may be a cloud server, a server of a distributed system, or a server combined with a blockchain.

[0342] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this document does not limit this.

[0343] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the principles of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A message group sending method based on a large model, comprising: Get historical interaction data between a sender and multiple receivers; Determine the relationship role labels, portrait feature labels and message sending timing of the multiple recipients according to the historical interaction data; wherein the relationship role labels are used to characterize the classification results of the identity roles of the recipients in the relationship between the sender and the recipient; Using the large model, according to the message subject selected by the sender, the relationship role labels and the portrait feature labels of the multiple receivers, respectively generate message content corresponding to each of the multiple receivers; and Sending messages to the multiple recipients according to the message content and the message sending timing.

2. The method according to claim 1, wherein: Determining, based on the historical interaction data, relationship role tags, portrait feature tags, and message sending timings of the multiple recipients, including: Determining relationship role labels and portrait feature labels of the multiple recipients according to the historical interaction data; Determining time characteristics of communication behaviors of the multiple recipients according to time information of multiple historical messages included in the historical interaction data; The message sending timings of the multiple recipients are determined according to the message subject selected by the sender and / or the time characteristics of the communication behaviors of the multiple recipients.

3. The method according to claim 2, wherein: Determining, based on the historical interaction data, relationship role labels and portrait feature labels of the multiple recipients, including: Determining identity characteristics of the multiple recipients based on the address information and topic information included in the historical interaction data; Determining a relationship role label according to the identity characteristics; Determine the portrait feature label based on the behavior pattern information and topic information contained in the historical interaction data.

4. The method according to claim 3, wherein: Determining the portrait feature label according to the behavior pattern information and topic information contained in the historical interaction data includes at least one of the following methods: Determine a first tag for characterizing the interests of the recipient based on the topic information contained in the historical interaction data; the first tag is used to influence the topic type contained in the message content; Determining a second tag for characterizing the social group attribute of the recipient based on the proprietary language content contained in the historical interaction data; the second tag is used to influence the type of terms contained in the message content; Determining, based on the interaction frequency included in the historical interaction data, a third tag for characterizing the depth of the relationship between the sender and the receiver; the third tag is used to influence the degree of emotionality of the message content; Determining, according to the length information of the plurality of historical messages included in the historical interaction data, a fourth tag for characterizing the length of the message content of the receiver; the fourth tag is used to affect the length of the message content; Determining a fifth tag for characterizing the receiver's interactive enthusiasm according to the feedback speed included in the historical interactive data; the fifth tag is used to influence the degree of interactivity of the message content; A sixth tag for characterizing the cultural background of the recipient is determined according to the cultural background information included in the historical interaction data; the sixth tag is used to influence the topic scope of the message content.

5. The method according to claim 1 or 4, wherein: Using the large model, according to the message subject selected by the sender, the relationship role labels and the portrait feature labels of the multiple receivers, the message content corresponding to each of the multiple receivers is generated respectively, including: Determining a first prompt word according to the message subject selected by the sender; Determining a second prompt word according to the relationship role labels of the multiple recipients; Determining a third prompt word according to the image feature label; The second prompt word is used to influence the tone and style of the message content, and the third prompt word is used to influence at least one of the topic type, word type, emotional level, message content length, interactivity or topic scope of the message content; For each of the multiple receivers, combining the corresponding first prompt word, the second prompt word and the third prompt word to obtain a personalized prompt word; The personalized prompt words are input into the large model to obtain the message content corresponding to each recipient.

6. The method according to claim 5, wherein: For each of the multiple receivers, combining the corresponding first prompt word, the second prompt word, and the third prompt word to obtain a personalized prompt word includes: determining content preference parameters of the plurality of recipients from the interaction frequencies included in the historical interaction data; For the first receiver whose content tendency parameter reaches the threshold, determining a recent conversation topic from the historical interaction data of the first receiver; Using the recent conversation topic as the fourth prompt word; The first prompt word, the second prompt word, the third prompt word and the fourth prompt word corresponding to the first recipient are combined to obtain a personalized prompt word.

7. The method according to claim 5, wherein: Using the large model, according to the message subject selected by the sender, the relationship role labels and the portrait feature labels of the multiple receivers, the message content corresponding to each of the multiple receivers is generated respectively, including: Determining a time impact parameter of the message according to the message subject selected by the sender and the message sending timing; By using the large model, the message content corresponding to each of the multiple receivers is generated according to the message subject selected by the sender, the relationship role labels of the multiple receivers, the portrait feature labels and the time impact parameters.

8. The method according to claim 1, determining the relationship role labels, portrait feature labels and message sending timing of the multiple recipients according to the historical interaction data, comprises: Extracting feedback data of the plurality of recipients for the group message from the historical interaction data; The relationship role labels, portrait feature labels and message sending timings of the multiple recipients are determined based on the historical interaction data and the feedback data.

9. A message group sending device based on a large model, comprising: A data acquisition module, used to acquire historical interaction data between a sender and multiple receivers; An analysis module, for determining the relationship role labels, portrait feature labels and message sending timing of the multiple recipients according to the historical interaction data; wherein the relationship role labels are used to characterize the classification results of the identity roles of the recipients in the relationship between the sender and the recipient; a content generation module, configured to generate message content corresponding to each of the multiple receivers by using a large model according to the message subject selected by the sender, the relationship role labels and the portrait feature labels of the multiple receivers; and The sending module is used to send the message to the multiple recipients according to the message content and the message sending timing.

10. The device according to claim 9, wherein: The analysis module comprises: A label determination submodule, used to determine the relationship role labels and portrait feature labels of the multiple recipients based on the historical interaction data; An information determination submodule, for determining the time characteristics of the communication behaviors of the plurality of receivers according to the time information of the plurality of historical messages included in the historical interaction data; The timing determination submodule is used to determine the message sending timing of the multiple receivers according to the message subject selected by the sender and / or the time characteristics of the communication behaviors of the multiple receivers.

11. The device according to claim 10, wherein: The label determination submodule is used for: Determining identity characteristics of the multiple recipients based on the address information and topic information included in the historical interaction data; Determining a relationship role label according to the identity characteristics; Determine the portrait feature label based on the behavior pattern information and topic information contained in the historical interaction data.

12. The device according to claim 11, wherein The tag determination submodule is further configured to execute at least one of the following methods: Determine, based on the topic information included in the historical interaction data, a first tag for characterizing the interests and hobbies of the recipient; the first tag is used to influence the topic type included in the message content; Determining a second tag for characterizing the social group attribute of the recipient based on the proprietary language content contained in the historical interaction data; the second tag is used to influence the type of terms contained in the message content; Determining, based on the interaction frequency included in the historical interaction data, a third tag for characterizing the depth of the relationship between the sender and the receiver; the third tag is used to influence the degree of emotionality of the message content; Determining, according to the length information of the plurality of historical messages included in the historical interaction data, a fourth tag for characterizing the length of the message content of the receiver; the fourth tag is used to affect the length of the message content; Determining a fifth tag for characterizing the receiver's interactive enthusiasm according to the feedback speed included in the historical interactive data; the fifth tag is used to influence the degree of interactivity of the message content; A sixth tag for characterizing the cultural background of the recipient is determined according to the cultural background information included in the historical interaction data; the sixth tag is used to influence the topic scope of the message content.

13. The device according to claim 9 or 12, wherein: The content generation module is used for: Determining a first prompt word according to the message subject selected by the sender; Determining a second prompt word according to the relationship role labels of the multiple recipients; Determining a third prompt word according to the image feature label; The second prompt word is used to influence the tone and style of the message content, and the third prompt word is used to influence at least one of the topic type, word type, emotional level, message content length, interactivity or topic scope of the message content; For each of the multiple receivers, combining the corresponding first prompt word, the second prompt word and the third prompt word to obtain a personalized prompt word; The personalized prompt words are input into the large model to obtain the message content corresponding to each recipient.

14. The device according to claim 13, wherein: The content generation module is also used for: determining content preference parameters of the plurality of recipients from the interaction frequencies included in the historical interaction data; For the first receiver whose content tendency parameter reaches the threshold, determining a recent conversation topic from the historical interaction data of the first receiver; Using the recent conversation topic as the fourth prompt word; The first prompt word, the second prompt word, the third prompt word and the fourth prompt word corresponding to the first recipient are combined to obtain a personalized prompt word.

15. The device according to claim 13, wherein: The content generation module is also used for: Determining a time-influencing parameter of the message according to the message subject selected by the sender and the timing of sending the message; By using the large model, the message content corresponding to each of the multiple receivers is generated according to the message subject selected by the sender, the relationship role labels of the multiple receivers, the portrait feature labels and the time impact parameters.

16. The device according to claim 9, wherein the analysis module is further configured to: Extracting feedback data of the plurality of recipients for the group message from the historical interaction data; The relationship role labels, portrait feature labels and message sending timings of the multiple recipients are determined based on the historical interaction data and the feedback data.

17. An electronic device comprising: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 8.

18. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to cause the computer to execute the method according to any one of claims 1-8.

19. A computer program product, comprising a computer program, which, when executed by a processor, implements the method according to any one of claims 1 to 8.

Images