Activity invitation method, device and equipment and computer readable storage medium

Abstract

The invention discloses an activity invitation method, device and equipment and a computer readable storage medium, and is applied to the technical field of computers, and the method comprises the steps: carrying out the first-stage activity invitation conversation with a target user through a pre-trained large language model, a voice recognition model and a text-to-voice model before introduction of activity details based on basic information; and when the stage anchor talk skill is detected, performing a second-stage activity invitation dialogue with the target user by using a pre-trained large language model, a voice recognition model and a text-to-voice model after introduction of activity details based on the target activity task. According to the method, accurate switching and dynamic adjustment of the invitation verbal skill are realized through staged cooperative work of double models, so that a large-model invitation agent (agent / agency) has a staged verbal skill adaptation capability similar to that of an artificial inviter, and the intelligent level and the conversion success rate of activity invitation are remarkably improved; and a technical support is provided for the online car-hailing platform to efficiently improve the driver activeness.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to an event invitation method, apparatus, device, and computer-readable storage medium. Background Technology

[0002] In the ride-hailing industry, inviting drivers to events is a core task for platforms to increase driver activity and promote special campaigns. This requires inviters to complete tasks such as identity introduction, information verification, and event promotion according to a set process, and to flexibly adjust their scripts based on driver feedback to improve conversion rates. Existing invitation models lack the ability to dynamically adapt their scripts and effectively highlight the advantages of events based on the dialogue content, resulting in a significant gap in invitation success rates compared to human inviters.

[0003] Therefore, the question is how to provide a large-scale invitation model with dynamic adaptability of the script to solve the problem of low intelligent conversion efficiency in event invitation scenarios. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide an event invitation method, apparatus, device and computer-readable storage medium, which solves the problem of poor dynamic adaptation capability of the script in the prior art, resulting in low intelligent conversion efficiency in event invitation scenarios.

[0005] To address the aforementioned technical problems, this invention provides an event invitation method, comprising:

[0006] Obtain basic information about the target users and the target activity tasks;

[0007] Based on the aforementioned basic information, a first invitation model, a speech recognition model, and a text-to-speech model are used to conduct a first-stage activity invitation dialogue with the target user; the first invitation model is a pre-trained large language model for introducing activity details;

[0008] When a stage anchor phrase is detected, a second stage activity invitation dialogue is conducted with the target user based on the target activity task using the second invitation model, the speech recognition model, and the text-to-speech model; the second invitation model is a pre-trained large language model that introduces the activity details.

[0009] Optionally, before engaging in the first phase of the activity invitation dialogue with the target user using the first invitation model, speech recognition model, and text-to-speech model, the method further includes:

[0010] Obtain the historical dialogue between the user and the inviter, and convert the historical dialogue into text data;

[0011] The text data is cleaned and segmented to obtain sub-dialogue data before the introduction of activity details;

[0012] The sub-dialogue data preceding the introduction of activity details is processed into first training data in a concatenated form; the concatenated form is a combination of the first system prompt and the first sub-dialogue record;

[0013] The first training data is used to train the first large language model, and the first large language model is fine-tuned using a fine-tuning algorithm to obtain the trained first invitation model.

[0014] Optionally, the sub-dialogue data preceding the introduction of activity details is processed into first training data in a concatenated form, including:

[0015] The first system prompt words are constructed based on the user's basic information and the sub-dialogue data before the introduction of activity details; the first system prompt words include at least the task description prompt words, the order setting step prompt words, the part-time / full-time inquiry prompt words, and the user information confirmation prompt words.

[0016] The first system prompt and the sub-dialogue data before the introduction of activity details are concatenated to obtain the first training data.

[0017] Optionally, the first training data is used to train a preset first large language model, and a fine-tuning algorithm is used to fine-tune the first large language model to obtain a trained first invitation model, including:

[0018] Instruction fine-tuning data is constructed based on the first training data; the instructions in the instruction fine-tuning data are the first system prompt words, historical dialogue data in the sub-dialogue data before introducing activity details, and the current user's question; the answers in the instruction fine-tuning data are the inviter's reply text in the sub-dialogue data before introducing activity details, based on the current user's question.

[0019] The first large language model is trained and fine-tuned based on the fine-tuning algorithm and the instruction fine-tuning data to obtain the trained first invitation model.

[0020] Optionally, before engaging in the second phase of the activity invitation dialogue with the target user using the second invitation model, the speech recognition model, and the text-to-speech model, the method further includes:

[0021] Obtain the historical dialogue between the user and the inviter, and convert the historical dialogue into text data;

[0022] The text data is cleaned and segmented to obtain sub-dialogue data after the activity details are introduced;

[0023] The sub-dialogue data following the introduction of activity details is processed into second training data in a concatenated form; the concatenation form is a combination of the second system prompt words and the second sub-dialogue records;

[0024] The second training data is used to train the preset second language model, and the second language model is fine-tuned using a fine-tuning algorithm to obtain the trained second invitation model.

[0025] Optionally, the sub-dialogue data following the introduction of activity details is processed into a second training data in a concatenated form, including:

[0026] The second system prompt words are constructed based on the sub-dialogue data following the introduction of activity details; the second system prompt words include at least task description prompt words, activity detail introduction prompt words, Q&A prompt words, and retention / promotion prompt words.

[0027] The second system prompt and the sub-dialogue data following the introduction of activity details are concatenated to obtain the second training data.

[0028] Optionally, the text data undergoes data cleaning and data segmentation processing, including:

[0029] The text data is cleaned using preset filtering rules and annotation methods to obtain cleaned text data.

[0030] Based on the stage anchor dialogue, the cleaned text data is segmented to obtain sub-dialogue data before the introduction of activity details and sub-dialogue data after the introduction of activity details.

[0031] The present invention also provides an event invitation device, comprising:

[0032] The acquisition module is used to acquire basic information about the target user and the target activity task;

[0033] The first dialogue module is used to conduct a first-stage activity invitation dialogue with the target user based on the basic information, using a first invitation model, a speech recognition model, and a text-to-speech model; the first invitation model is a pre-trained large language model before introducing activity details;

[0034] The second dialogue module is used to conduct a second-stage activity invitation dialogue with the target user based on the target activity task when a stage anchor phrase is detected, using a second invitation model, the speech recognition model, and the text-to-speech model; the second invitation model is a pre-trained large language model that introduces the activity details.

[0035] The present invention also provides an event invitation device, comprising:

[0036] Memory, used to store computer programs;

[0037] A processor for executing the computer program to implement the activity invitation method as described above.

[0038] The present invention also provides a computer-readable storage medium storing computer-executable instructions, which, when loaded and executed by a processor, implement the activity invitation method described above.

[0039] As can be seen, this invention obtains the basic information and target activity task of the target user; based on the basic information, it uses a first invitation model, a speech recognition model, and a text-to-speech model to conduct a first-stage activity invitation dialogue with the target user; the first invitation model is a pre-trained large language model before introducing activity details; when a stage anchor phrase is detected, based on the target activity task, a second invitation model, a speech recognition model, and a text-to-speech model are used to conduct a second-stage activity invitation dialogue with the target user; the second invitation model is a pre-trained large language model after introducing activity details. This invention achieves precise switching and dynamic adjustment of invitation phrases through the phased collaborative work of the two models, enabling the large model invitation agent (intelligent agent / proxy) to have phased phrase adaptation capabilities similar to human inviters, effectively catering to drivers' psychological concerns, significantly improving the intelligence level and conversion success rate of activity invitations; at the same time, it reduces the training costs and manpower investment of human inviters, ensures the compliance and standardization of the invitation process, and provides technical support for ride-hailing platforms to efficiently improve driver activity.

[0040] In addition, the present invention also provides an event invitation device, equipment, and computer-readable storage medium, which also have the above-mentioned beneficial effects. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0042] Figure 1 A flowchart of an activity invitation method provided in an embodiment of the present invention;

[0043] Figure 2 A flowchart illustrating an example processing method provided in an embodiment of the present invention;

[0044] Figure 3 Example diagram of the first training dataset preparation process provided in the embodiments of the present invention;

[0045] Figure 4 Example diagram of the second training dataset preparation process provided in this embodiment of the invention;

[0046] Figure 5 A flowchart illustrating a model training method provided in an embodiment of the present invention;

[0047] Figure 6 This is a schematic diagram of the structure of an activity invitation device provided in an embodiment of the present invention;

[0048] Figure 7 This is a schematic diagram of an activity invitation device provided in an embodiment of the present invention. Detailed Implementation

[0049] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0050] The activity invitation process follows a standard workflow, comprising a series of tasks to be completed sequentially. These include: introducing oneself and the purpose of the call; assisting drivers with order setup; inquiring whether the driver is part-time or full-time; verifying the driver's phone number and license plate number; and persuading and guiding the driver to participate in specific activities. Simultaneously, the inviter must adhere to a series of guidelines during the invitation process, including: highlighting the platform's advantages to pique the driver's interest when they decline; truthfully introducing the platform and related activities, avoiding fabricated rules; using polite and concise language; and absolutely refraining from abusive or derogatory language. The primary goal of the activity invitation process is to get as many drivers as possible to participate in the platform's specific activities, thereby increasing platform activity.

[0051] Due to its high business complexity, numerous behavioral norms, and the often low willingness of drivers to cooperate, event invitation work requires systematically trained personnel. Because of the high difficulty of event invitation tasks, the previous large-scale invitation model agent had a significantly lower success rate than real agents. This difference stems from the fact that event invitation tasks require understanding driver psychology, highlighting a specific advantage of the platform's event based on the driver's concerns, and persuading them. Real agents can adjust their messaging based on the current stage of the conversation. Therefore, the large-scale invitation model agent needs to possess the same communication skills to improve the success rate.

[0052] The core of this invention in solving the aforementioned problems lies in enabling the invitation model agent to change its speaking style according to the stage of the invitation task, making it more similar to the invitation style of a real human inviter, thereby achieving a more realistic invitation effect. This invention uses real human invitation dialogue data from ride-hailing drivers to further train and fine-tune the invitation model agent. The trained model possesses different speaking styles, allowing the invitation model agent to specifically change its speaking style according to the current stage of the invitation dialogue.

[0053] Please refer to the details. Figure 1 , Figure 1 A flowchart illustrating an activity invitation method provided in an embodiment of the present invention. The method may include:

[0054] S101: Obtain basic information about the target user and the target activity task.

[0055] The activity invitation method provided in this embodiment can be specifically applied to scenarios such as inviting ride-hailing drivers, truck drivers, food delivery drivers, and couriers. For example, in the ride-hailing industry, to increase the number of drivers participating in a specific activity, ride-hailing platform companies will call drivers who have already registered and verified on the platform to introduce the details of a specific activity launched by the platform (e.g., a new driver commission reduction activity, where the commission rate for drivers who purchase a membership is reduced to one percent, valid for one month), and then persuade the drivers to participate. This work is professionally called activity invitation, and the staff member making the calls to the drivers is called an inviter.

[0056] In this embodiment, the execution entity is an outbound call robot, which consists of an invitation model agent, a speech recognition model, and a text-to-speech model. The invitation model agent comprises a first invitation model and a second invitation model. This embodiment does not limit the speech recognition model; it can be a third-party ASR (Automatic Speech Recognition) tool, or it can be a TTS (Text-to-Speech) model (a technology that converts text on a digital interface into natural and fluent audio). Similarly, this embodiment does not limit the text-to-speech model; it can be a TTS tool, or it can be a deep speech model.

[0057] In this embodiment, the basic information of the target user and the target activity task are input into the outbound call robot, which can then complete the task of inviting the target user to a conversation. This embodiment does not limit the basic information; for example, it may include the user's name, authentication status, etc. This embodiment does not limit the target activity task; for example, it may include a new user commission reduction activity.

[0058] S102: Based on basic information, use the first invitation model, speech recognition model and text-to-speech model to conduct the first stage of activity invitation dialogue with the target user; the first invitation model is a pre-trained large language model before introducing the activity details.

[0059] In this embodiment, after inputting the target user's basic information, a dialogue can be initiated with the target user. Initially, an invitation model (i.e., the first invitation model) is used before introducing activity details. This embodiment does not specifically limit the large language model. For example, it can be a GPT series model (such as a generative pre-trained Transformer-4); or it can also be a Wenxin Yiyan model. It can be understood that the first invitation model in this embodiment can be understood as a pre-invitation process model, focusing on basic interactive tasks such as identity breaking, information verification, and process guidance, and is responsible for completing the dialogue opening and preliminary preparation.

[0060] S103: When the stage anchor dialogue is detected, the second invitation model, speech recognition model and text-to-speech model are used to conduct a second-stage activity invitation dialogue with the target user based on the target activity task; the second invitation model is a pre-trained large language model after introducing the activity details.

[0061] In this embodiment, when the stage anchor text is detected, the system switches to the "introducing activity details" invitation model (second invitation model) to conduct the activity invitation task. The stage anchor text in this embodiment can be the text output by the first invitation model during the initial dialogue with the target user. This embodiment does not specifically limit the stage anchor text. For example, it can be a keyword of the target activity task; or it can be the target activity task itself. It should be noted that in this step, based on the target activity task and basic information, the second invitation model, speech recognition model, and text-to-speech model can also be used to conduct a second-stage activity invitation dialogue with the target user, which can further enhance user goodwill. It can be understood that the second invitation model in this embodiment can be understood as an activity conversion model, primarily responsible for persuasive tasks such as demand insight, advantage matching, and objection resolution, focusing on increasing driver participation willingness.

[0062] This embodiment considers that before introducing the activity details, the inviter's focus is on helping drivers set up their order listeners, so the script is concise and accurate. However, once the activity introduction begins, the script becomes more enticing, urgent, and lengthy. When guiding someone through an app, conciseness and accuracy are paramount; lengthy explanations waste time. Conversely, when introducing activity details, a more lengthy explanation is preferable to fully convey the benefits of the activity. These are completely different script focuses and styles, and a single model cannot accommodate both styles simultaneously. Therefore, two invitation models with different styles / scripts need to be trained.

[0063] Furthermore, before conducting the first-stage activity invitation dialogue with the target user using the first invitation model, speech recognition model, and text-to-speech model, the process may further include: acquiring the user's historical dialogue with the inviter and converting the historical dialogue into text data; performing data cleaning and data segmentation on the text data to obtain sub-dialogue data before and after the introduction of activity details; processing the sub-dialogue data before the introduction of activity details into a first training data in a concatenated form; the concatenation form is a concatenation of the first system prompt word and the first sub-dialogue record; processing the sub-dialogue data after the introduction of activity details into a second training data in a concatenated form; the concatenation form is a concatenation of the second system prompt word and the second sub-dialogue record; training a preset first large language model using the first training data and fine-tuning the first large language model using a fine-tuning algorithm to obtain a trained first invitation model; training a preset second large language model using the second training data and fine-tuning the second large language model using a fine-tuning algorithm to obtain a trained second invitation model.

[0064] This embodiment uses historical, authentic human-invitation dialogue data, performing text conversion, data cleaning, and data segmentation to obtain the first training data and the second training data, respectively. The first training data is used to train the first large language model to obtain the first invitation model, and the second training data is used to train the second large language model to obtain the second invitation model. This embodiment does not specifically limit the implementation methods of text conversion, data cleaning, and data segmentation. This embodiment also does not limit the fine-tuning algorithm, the first large language model, or the second large language model. It is understood that the first and second large language models are the initial models before the first and second invitation models are trained, respectively.

[0065] Furthermore, the aforementioned data cleaning and data segmentation of text data may specifically include: using preset filtering rules and annotation methods to clean the text data to obtain cleaned text data; and segmenting the cleaned text data according to the stage anchor dialogue to obtain sub-dialogue data before the introduction of activity details and sub-dialogue data after the introduction of activity details.

[0066] This embodiment can use third-party ASR (speech recognition) tools to convert a large amount of real historical dialogues between inviters and drivers into text data (i.e., text data); using data cleaning methods such as rules and manual annotation, basic processing is performed on the text dialogue data, which may include: removing dialogues that were not answered / had 1-2 rounds of dialogue, dialogues that only involved inviters or drivers, and dialogues that were not successfully invited; each dialogue is divided into two stages according to the task: "before introducing the activity details" and "after introducing the activity details", thus processing a complete successfully invited dialogue into two sub-dialogues.

[0067] For example, you can refer to the following: Figure 2 , Figure 2 Here is a flowchart illustrating a processing method provided in this embodiment of the invention. (1) Classification and screening: Classify the conversations according to the inviter's name, the inviter's performance evaluation results, the driver's level, and the tags given after the call. Based on the above classification, screen the conversations with excellent scripts and high quality (Note: Driver level and post-call tags involve business knowledge, which will not be detailed here. The purpose of this step is to screen out the conversation records of excellent inviters and drivers with high willingness to cooperate based on business tags). (2) Data elimination: Eliminate conversations that were not connected or only had 1-4 rounds; eliminate the entire conversation of the inviter's follow-up call; eliminate incomplete conversations that did not complete the entire business process; eliminate... Only dialogues with the inviter or driver speaking; eliminate dialogues with dialects or ASR recognition errors; eliminate dialogues where invitations were not successful; (3) Cleaning and standardization: merge the continuous speaking content of the inviter or driver to ensure that the dialogue is an alternating dialogue format of one sentence from you and one sentence from me; rewrite the inviter's dialogue, rewriting the verbose and interjection-heavy sentences into fluent and concise sentences; (4) Splitting dialogues: based on the stage anchor phrase "Master, next I will introduce the xxx activity to you", split a complete dialogue into two sub-dialogues: "before introducing the details of the activity" and "after introducing the details of the activity".

[0068] Furthermore, the sub-dialogue data before the introduction of activity details is processed into first training data in a concatenated form. Specifically, this may include: constructing first system prompt words based on the user's basic information and the sub-dialogue data before the introduction of activity details; the first system prompt words include at least prompt words describing the task, prompt words for setting up the playlist, prompt words for inquiring about part-time or full-time positions, and prompt words for confirming user information; concatenating the first system prompt words with the sub-dialogue data before the introduction of activity details to obtain the first training data.

[0069] This embodiment describes the data preparation process for the first training data before introducing the activity details. The sub-dialogue data before introducing the activity details is processed into the form of `system_prompt` (the first system prompt word) and dialogue records. Here, `system_prompt` represents the system prompt word in the large language model domain, and is a variable that describes the task content to the large model. For details, please refer to... Figure 3 . Figure 3 Example diagram of the preparation process for the first training dataset provided in the embodiments of the present invention. (1) Query the basic information of the driver in the dialogue to be processed, which may include: name, gender, mobile phone number, license plate number, etc.; (2) Construct "system_prompt before introducing activity details", that is, the first system prompt word, which consists of several sub-prompts, including: task description prompt, order setting steps prompt, part-time / full-time inquiry prompt, driver information confirmation prompt, etc. Among them, the task description prompt refers to the part of the text describing the task content in the complete prompt, which is not the activity task, but the task that the large model has to complete as the inviter. (3) Concatenate the system_prompt and the "before introducing activity details" dialogue record (that is, the sub-dialogue data before introducing activity details) to form the first training data.

[0070] Furthermore, the sub-dialogue data after introducing the activity details is processed into a second training data in a concatenated form. Specifically, this may include: constructing the second system prompt words based on the sub-dialogue data after introducing the activity details; the second system prompt words include at least task description prompt words, activity detail introduction prompt words, Q&A prompt words, and retention / promotion prompt words; concatenating the second system prompt words and the sub-dialogue data after introducing the activity details to obtain the second training data.

[0071] This embodiment describes the data preparation process for the second training data after the activity details are introduced. The sub-dialogue data following the activity details are processed into the form of `system_prompt` (the second system prompt word) and dialogue records. Here, `system_prompt` represents the system prompt word in the large language model domain, and is a variable that describes the task content to the large model. For details, please refer to... Figure 4 . Figure 4 Example diagram of the preparation process for the second training dataset provided in the embodiments of the present invention. (1) Construct system_prompt (i.e., the second system prompt) after introducing the activity details, which consists of several sub-prompts, including: task description prompt, activity details introduction prompt, Q&A prompt, retention and facilitation prompt, etc.; (2) Concatenate system_prompt and the dialogue record after introducing the activity details (i.e., the sub-dialogue data after introducing the activity details) to form the second training data.

[0072] Furthermore, the first large language model is trained using the first training data, and then fine-tuned using a fine-tuning algorithm to obtain the trained first invitation model. Specifically, this may include: constructing instruction fine-tuning data based on the first training data; the instructions in the instruction fine-tuning data are the first system prompt words, historical dialogue data in the sub-dialogue data before introducing activity details, and the current user's question; the answers in the instruction fine-tuning data are the inviter's reply text corresponding to the current user's question in the sub-dialogue data before introducing activity details; and the first large language model is trained and fine-tuned using the fine-tuning algorithm and instruction fine-tuning data to obtain the trained first invitation model.

[0073] This embodiment describes the training process of the first invitation model. It should be noted that the training process of the second model can refer to the above process, only the training data differs. The training process of the first invitation model will be used as an example for explanation; please refer to [link / reference]. Figure 5 , Figure 5 Here is a flowchart of a model training method provided in this embodiment of the invention. (1) Use the training dialogue data before “introducing activity details” (i.e., the first training data mentioned above) to construct instruction fine-tuning data. That is, system_prompt (first system prompt word), historical dialogue data and the current user’s (e.g., the current driver’s) questions are used as instructions, and the inviter’s replies are used as answers. A 10-round dialogue can construct 10 instruction fine-tuning data. Among them, instruction fine-tuning is a commonly used training method in the field of large models, which will not be described in detail here. (2) Use the LoRa algorithm (a fine-tuning algorithm) to fine-tune the instructions and obtain the large model of the invitation dialogue before “introducing activity details” (i.e., the first invitation model). For example, the first large language model and the second large language model can be based on the qwen2.5-72b-instruct model. The LoRa algorithm is used to fine-tune the large model using the data before “introducing activity details” and after “introducing activity details”, respectively, so as to obtain two invitation models with different emphasis on the dialogue. Among them, qwen2.5-72b-instruct is an open source large language model.

[0074] The activity invitation method provided in this invention involves the following steps: S101, obtaining basic information about the target user and the target activity task; S102, based on the basic information, using a first invitation model, a speech recognition model, and a text-to-speech model to conduct a first-stage activity invitation dialogue with the target user; the first invitation model is a pre-trained large language model before introducing activity details; S103, when a stage anchor phrase is detected, using a second invitation model, a speech recognition model, and a text-to-speech model to conduct a second-stage activity invitation dialogue with the target user based on the target activity task; the second invitation model is a pre-trained large language model after introducing activity details. This invention achieves precise switching and dynamic adjustment of invitation phrases through the phased collaborative work of two models, enabling the large model invitation agent (intelligent agent / proxy) to possess phased phrase adaptation capabilities similar to human inviters, effectively catering to drivers' psychological concerns, significantly improving the intelligence level and conversion success rate of activity invitations; simultaneously reducing the training costs and manpower investment of human inviters, ensuring the compliance and standardization of the invitation process, and providing technical support for ride-hailing platforms to efficiently improve driver activity.

[0075] The activity invitation device provided in the embodiments of the present invention will be described below. The activity invitation device described below can be referred to in correspondence with the activity invitation method described above.

[0076] Please refer to the details. Figure 6 , Figure 6 A schematic diagram of an activity invitation device provided in an embodiment of the present invention may include:

[0077] Module 100 is used to acquire basic information about the target user and the target activity task.

[0078] The first dialogue module 200 is used to conduct a first-stage activity invitation dialogue with the target user based on the basic information, using a first invitation model, a speech recognition model, and a text-to-speech model; the first invitation model is a pre-trained large language model before introducing activity details;

[0079] The second dialogue module 300 is used to conduct a second-stage activity invitation dialogue with the target user based on the target activity task when a stage anchor dialogue is detected, using a second invitation model, the speech recognition model, and the text-to-speech model; the second invitation model is a pre-trained large language model that introduces the details of the activity.

[0080] Based on any of the above embodiments, the activity invitation device may further include:

[0081] The first historical dialogue acquisition module is used to acquire the historical dialogue between the user and the inviter before conducting the first stage of activity invitation dialogue with the target user using the first invitation model, speech recognition model and text-to-speech model, and convert the historical dialogue into text data.

[0082] The first text data processing module is used to perform data cleaning and data segmentation on the text data to obtain sub-dialogue data before introducing the activity details;

[0083] The first training data acquisition module is used to process the sub-dialogue data before the introduction of activity details into a first training data in a concatenated form; the concatenated form is a concatenation of the first system prompt words and the first sub-dialogue record;

[0084] The first large language model training module is used to train the preset first large language model using the first training data, and to fine-tune the first large language model using a fine-tuning algorithm to obtain the trained first invitation model.

[0085] Based on any of the above embodiments, the first training data acquisition module may include:

[0086] The first prompt word construction unit is used to construct the first system prompt word based on the user's basic information and the sub-dialogue data before the introduction of activity details; the first system prompt word includes at least a task description prompt word, a playlist setting step prompt word, an inquiry about part-time or full-time work prompt word, and a user information confirmation prompt word;

[0087] The first splicing unit is used to splice the first system prompt and the sub-dialogue data before the introduction of activity details to obtain the first training data.

[0088] Based on any of the above embodiments, the first large language model training module may include:

[0089] The instruction fine-tuning data construction unit is used to construct instruction fine-tuning data based on the first training data; the instructions in the instruction fine-tuning data are the first system prompt words, the historical dialogue data in the sub-dialogue data before the introduction of activity details, and the current user's question; the answers in the instruction fine-tuning data are the inviter's reply text in the sub-dialogue data before the introduction of activity details, based on the current user's question.

[0090] The model training unit is used to train and fine-tune the first large language model based on the fine-tuning algorithm and the instruction fine-tuning data to obtain the trained first invitation model.

[0091] Based on any of the above embodiments, the activity invitation device may further include:

[0092] The second historical dialogue acquisition module is used to acquire the historical dialogue between the user and the inviter before conducting the second stage of activity invitation dialogue with the target user using the second invitation model, the speech recognition model and the text-to-speech model, and convert the historical dialogue into text data.

[0093] The second text data processing module is used to perform data cleaning and data segmentation on the text data to obtain sub-dialogue data after introducing the activity details;

[0094] The second training data acquisition module is used to process the sub-dialogue data after introducing the activity details into a concatenated form of second training data; the concatenation form is a combination of the second system prompt words and the second sub-dialogue record.

[0095] The second training data is used to train the preset second language model, and the second language model is fine-tuned using a fine-tuning algorithm to obtain the trained second invitation model.

[0096] Based on any of the above embodiments, the second training data acquisition module may include:

[0097] The second prompt word construction unit is used to construct the second system prompt words based on the sub-dialogue data after the introduction of activity details; the second system prompt words include at least task description prompt words, activity details introduction prompt words, Q&A prompt words, and retention and facilitation prompt words.

[0098] The second splicing unit is used to splice the second system prompt and the sub-dialogue data after the introduction of activity details to obtain the second training data.

[0099] Based on any of the above embodiments, the first text data processing module or the second text data processing module may include:

[0100] The data processing unit is used to perform data cleaning on the text data using preset filtering rules and annotation methods to obtain cleaned text data.

[0101] The data segmentation unit is used to segment the cleaned text data according to the stage anchor dialogue to obtain sub-dialogue data before the introduction of activity details and sub-dialogue data after the introduction of activity details.

[0102] It should be noted that the order of the modules and units in the above-mentioned activity invitation device can be changed without affecting the logic.

[0103] The activity invitation device provided in this embodiment of the invention includes an acquisition module 100 for acquiring basic information of the target user and the target activity task; a first dialogue module 200 for conducting a first-stage activity invitation dialogue with the target user based on the basic information, using a first invitation model, a speech recognition model, and a text-to-speech model; the first invitation model is a pre-trained large language model before introducing activity details; and a second dialogue module 300 for conducting a second-stage activity invitation dialogue with the target user based on the target activity task when a stage anchor phrase is detected, using a second invitation model, the speech recognition model, and the text-to-speech model; the second invitation model is a pre-trained large language model after introducing activity details. This device achieves precise switching and dynamic adjustment of invitation scripts through the phased collaborative operation of two models. This enables the large-model invitation agent (intelligent agent / proxy) to have phased script adaptation capabilities similar to human inviters, effectively meeting drivers' psychological concerns and significantly improving the intelligence level and conversion success rate of event invitations. At the same time, it reduces the training costs and manpower investment of human inviters, ensures the compliance and standardization of the invitation process, and provides technical support for ride-hailing platforms to efficiently improve driver activity.

[0104] The activity invitation device provided in the embodiments of the present invention will be described below. The activity invitation device described below can be referred to in correspondence with the activity invitation method described above.

[0105] Please refer to Figure 7 , Figure 7 A schematic diagram of an activity invitation device provided in an embodiment of the present invention may include:

[0106] Memory 10 is used to store computer programs;

[0107] Processor 20 is used to execute computer programs to implement the above-described activity invitation method.

[0108] The memory 10, processor 20, and communication interface 31 all communicate with each other through the communication bus 32.

[0109] In this embodiment of the invention, the memory 10 is used to store one or more programs. The programs may include program code, which includes computer operation instructions. In this embodiment of the invention, the memory 10 may store programs for implementing the following functions:

[0110] Obtain basic information about the target users and the target activity tasks;

[0111] Based on basic information, the first invitation model, speech recognition model, and text-to-speech model are used to conduct the first stage of the activity invitation dialogue with the target user; the first invitation model is a pre-trained large language model before introducing the activity details;

[0112] When the stage anchor dialogue is detected, the second stage of the activity invitation dialogue is conducted with the target user based on the target activity task using the second invitation model, speech recognition model and text-to-speech model; the second invitation model is a pre-trained large language model that introduces the details of the activity.

[0113] In one possible implementation, the memory 10 may include a program storage area and a data storage area, wherein the program storage area may store the operating system and applications required for at least one function; and the data storage area may store data created during use.

[0114] Furthermore, memory 10 may include read-only memory and random access memory, providing instructions and data to the processor. A portion of the memory may also include NVRAM. The memory stores operating systems and operating instructions, executable modules, or data structures, or subsets thereof, or extended sets thereof, wherein the operating instructions may include various operating instructions for implementing various operations. The operating system may include various system programs for implementing various basic tasks and handling hardware-based tasks.

[0115] Processor 20 can be a central processing unit (CPU), an application-specific integrated circuit, a digital signal processor, a field-programmable gate array, or other programmable logic device. Processor 20 can be a microprocessor or any conventional processor. Processor 20 can call programs stored in memory 10.

[0116] Communication interface 31 can be an interface for the communication module, used to connect with other devices or systems.

[0117] Of course, it should be noted that, Figure 7 The structure shown does not constitute a limitation on the activity invitation device in the embodiments of the present invention. In practical applications, the activity invitation device may include more than Figure 7 More or fewer components as shown, or combinations of certain components.

[0118] The readable storage medium provided in the embodiments of the present invention is described below. The readable storage medium described below can be referred to in correspondence with the activity invitation method described above.

[0119] The present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the activity invitation method described above.

[0120] The computer-readable storage medium may include various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0121] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.

[0122] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0123] Finally, it should be noted that in this document, relationships such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0124] The above provides a detailed description of an activity invitation method, apparatus, device, and computer-readable storage medium provided by the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for inviting people to an event, characterized in that, include: Obtain basic information about the target users and the target activity tasks; Based on the aforementioned basic information, the first invitation model, speech recognition model, and text-to-speech model are used to conduct a first-stage activity invitation dialogue with the target user. The first invitation model is a pre-trained large language model that introduces the details of the event; When a stage anchor point is detected, a second stage activity invitation dialogue is conducted with the target user based on the target activity task using the second invitation model, the speech recognition model, and the text-to-speech model. The second invitation model is a pre-trained large language model that introduces the details of the event.

2. The activity invitation method according to claim 1, characterized in that, Before engaging in the first phase of the activity invitation dialogue with the target user using the first invitation model, speech recognition model, and text-to-speech model, the process also includes: Obtain the historical dialogue between the user and the inviter, and convert the historical dialogue into text data; The text data is cleaned and segmented to obtain sub-dialogue data before the introduction of activity details; The sub-dialogue data preceding the introduction of activity details is processed into first training data in a concatenated form; the concatenated form is a combination of the first system prompt and the first sub-dialogue record; The first training data is used to train the first large language model, and the first large language model is fine-tuned using a fine-tuning algorithm to obtain the trained first invitation model.

3. The activity invitation method according to claim 2, characterized in that, The sub-dialogue data preceding the introduction of activity details is processed into first training data in a concatenated form, including: The first system prompt words are constructed based on the user's basic information and the sub-dialogue data before the introduction of activity details; the first system prompt words include at least the task description prompt words, the order setting step prompt words, the part-time / full-time inquiry prompt words, and the user information confirmation prompt words. The first system prompt and the sub-dialogue data before the introduction of activity details are concatenated to obtain the first training data.

4. The activity invitation method according to claim 2, characterized in that, The first training data is used to train a preset first large language model, and a fine-tuning algorithm is used to fine-tune the first large language model to obtain a trained first invitation model, including: Instruction fine-tuning data is constructed based on the first training data; the instructions in the instruction fine-tuning data are the first system prompt words, historical dialogue data in the sub-dialogue data before introducing activity details, and the current user's question; the answers in the instruction fine-tuning data are the inviter's reply text in the sub-dialogue data before introducing activity details, based on the current user's question. The first large language model is trained and fine-tuned based on the fine-tuning algorithm and the instruction fine-tuning data to obtain the trained first invitation model.

5. The activity invitation method according to claim 1, characterized in that, Before engaging in the second phase of the activity invitation dialogue with the target user using the second invitation model, the speech recognition model, and the text-to-speech model, the process also includes: Obtain the historical dialogue between the user and the inviter, and convert the historical dialogue into text data; The text data is cleaned and segmented to obtain sub-dialogue data after the activity details are introduced; The sub-dialogue data following the introduction of activity details is processed into second training data in a concatenated form; the concatenation form is a combination of the second system prompt words and the second sub-dialogue records; The second training data is used to train the preset second language model, and the second language model is fine-tuned using a fine-tuning algorithm to obtain the trained second invitation model.

6. The activity invitation method according to claim 5, characterized in that, The sub-dialogue data following the introduction of activity details is processed into a second training data in a concatenated form, including: The second system prompt words are constructed based on the sub-dialogue data following the introduction of activity details; the second system prompt words include at least task description prompt words, activity detail introduction prompt words, Q&A prompt words, and retention / promotion prompt words. The second system prompt and the sub-dialogue data following the introduction of activity details are concatenated to obtain the second training data.

7. The activity invitation method according to any one of claims 2 and 5, characterized in that, The text data undergoes data cleaning and data segmentation, including: The text data is cleaned using preset filtering rules and annotation methods to obtain cleaned text data. Based on the stage anchor dialogue, the cleaned text data is segmented to obtain sub-dialogue data before the introduction of activity details and sub-dialogue data after the introduction of activity details.

8. An event invitation device, characterized in that, include: The acquisition module is used to acquire basic information about the target user and the target activity task; The first dialogue module is used to conduct a first-stage activity invitation dialogue with the target user based on the basic information, using a first invitation model, a speech recognition model, and a text-to-speech model. The first invitation model is a pre-trained large language model that introduces the details of the event; The second dialogue module is used to conduct a second-stage activity invitation dialogue with the target user based on the target activity task when a stage anchor point is detected, using the second invitation model, the speech recognition model, and the text-to-speech model. The second invitation model is a pre-trained large language model that introduces the details of the event.

9. An event invitation device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the activity invitation method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when loaded and executed by a processor, implement the activity invitation method as described in any one of claims 1 to 7.

Images