Authentication information detection method, system, equipment and medium

Abstract

The invention relates to an authentication information detection method, system and device and a medium. The method comprises the steps of obtaining an identity label of a target user; determining missing authentication information of the target user based on the identity label; determining a task target based on the missing authentication information; inputting the task target and the missing authentication information into a large language model, generating guide information for guiding the operation of a target user, and starting the dialogue; sending the guide information to the target user, and receiving reply information sent by the target user; and detecting whether the missing authentication information is complemented according to the reply information, and ending the dialogue after detecting that the missing authentication information is complemented. According to the method, the current operation state of the user can be judged more accurately and quickly, so that the guiding process is dynamically adjusted.

Description

Technical Field

[0001] The present invention relates to the field of intelligent dialogue technology, and in particular to a method, system, device and medium for detecting authentication information. Background Art

[0002] When inviting users on the freight invitation invitation platform, the staff will guide the target users to complete a series of operations through direct communication to complete the authentication information, such as confirming information, uploading materials or solving questions, etc. This requires the staff to have high language guidance skills and adaptability, which undoubtedly increases the invitation cost.

[0003] Traditional methods, generally based on the "NLU (Natural Language Understanding) + DM (Dialogue Management) + NLG (Natural Language Generation)" architecture, can conduct human-computer dialogue, but lack the ability to adapt to complex scenarios and dynamic tasks, and cannot effectively replace manual work, which makes the overall cost of authentication-related work high and inefficient. Therefore, it is necessary to provide a method, system, device and medium for detecting authentication information. Summary of the invention

[0004] In view of the shortcomings of the prior art described above, the purpose of the present invention is to provide a method, system, device and medium for detecting authentication information, which improves the problem that the prior art cannot take into account both accuracy and efficiency when guiding users to complete the authentication information in the face of high-complexity task scenarios.

[0005] To achieve the above-mentioned purpose and other related purposes, the present invention provides a method for detecting authentication information, including: obtaining an identity identifier of a target user; determining the missing authentication information of the target user based on the identity identifier; determining a task objective based on the missing authentication information; inputting the task objective and the missing authentication information into a large language model, generating guidance information for guiding the target user's operation and starting this conversation; sending the guidance information to the target user, and receiving a reply message sent by the target user; detecting whether the missing authentication information is completed based on the reply information, and ending the conversation after detecting that the missing authentication information is completed.

[0006] In one embodiment of the present invention, the type of this conversation is a voice call, and sending the guide information to the target user and receiving the reply information sent by the target user includes: performing text-to-speech processing on the guide information to convert the guide information into voice guide information; sending the voice guide information to the target user and receiving voice-type reply information sent by the target user; performing voice recognition on the voice-type reply information to generate text-type reply information.

[0007] In one embodiment of the present invention, the voice-type reply information is subjected to voice recognition to generate text-type reply information, including: performing voice recognition on the voice-type reply information to obtain initial text-type reply information; performing semantic checking on the initial text-type reply information to generate final text-type reply information.

[0008] In one embodiment of the present invention, detecting whether the missing authentication information is completed based on the reply information, and ending the conversation after detecting that the missing authentication information is completed, includes: performing semantic analysis on the reply information to determine whether there is an action to complete the missing authentication information in the content of the reply information: if so, inputting the completion status of the missing authentication information and the reply information into the large language model again, generating end-of-conversation information, and sending the end-of-conversation information to the target user to end the conversation; if not, inputting the reply information, the task target and the missing authentication information into the large language model again, generating new guidance information based on the reply information, and sending the new guidance information to the target user.

[0009] In one embodiment of the present invention, the large language model is obtained by training based on a pre-acquired training set and a corresponding label set, and the process of obtaining each training sample and the corresponding label in the training set is: based on the pre-acquired conversation information between the inviter and the user; wherein each conversation information carries the identity of the inviter and the identity of the user and the corresponding conversation content; data cleaning is performed on the conversation information to obtain cleaned conversation information; the cleaned conversation information is standardized to obtain standardized conversation information; from the standardized conversation information, according to the user's identity, the user's missing authentication information is searched and identified from the user library, and the corresponding task goal is determined based on the missing authentication information; the conversation content of the user's identity is selected from the standardized conversation information, and the conversation content of the user's identity and the missing authentication information and the corresponding task goal are used as training samples together, and the conversation content of the corresponding inviter's identity is used as a label.

[0010] In one embodiment of the present invention, the training process of the large language model includes: inputting the training set into the pre-trained large language model to obtain a prediction guidance information set; calculating the difference between the prediction guidance set and the label set; based on the difference, optimizing all parameters of the large language model according to the full parameter fine-tuning method to obtain a trained large language model.

[0011] In one embodiment of the present invention, the training process of the large language model includes: inputting the training set into the pre-trained large language model to obtain a prediction guidance information set; calculating the difference between the prediction guidance set and the label set; based on the difference, performing incremental pre-training on the large language model according to the full parameter fine-tuning method to obtain an initially trained large language model; performing instruction fine-tuning on the initially trained large language model, and optimizing the local parameters of the initially trained large language model based on the LoRA algorithm to obtain a trained large language model.

[0012] In one embodiment of the present invention, a system for detecting authentication information is also provided, the system comprising: an identity acquisition module for acquiring the identity of a target user; a missing authentication information determination module for determining the missing authentication information of the target user based on the identity; a target determination module for determining the task target of this conversation based on the missing authentication information; a guidance information generation module for inputting the task target and the missing authentication information into a large language model to generate guidance information for guiding the target user's operation; an information feedback module for sending the guidance information to the target user and receiving reply information sent by the target user; a missing information detection module for detecting whether the missing authentication information is completed based on the reply information, and terminating the conversation after detecting that the missing authentication information is completed.

[0013] In one embodiment of the present invention, an electronic device is also provided, including: one or more processors; a storage device for storing one or more programs, when the one or more programs are executed by the one or more processors, the electronic device implements any of the authentication information detection methods described above.

[0014] In one embodiment of the present invention, a computer-readable storage medium is further provided, on which a computer program is stored. When the computer program is executed by a processor of a computer, the computer executes any of the above-mentioned authentication information detection methods.

[0015] As described above, a method, system, device and medium for detecting authentication information of the present invention have the following beneficial effects: obtaining the identity of the target user and determining the missing authentication information of the target user based on the identity, and accurately identifying the incomplete authentication process through the identity, dynamically generating the task target, thereby ensuring the accuracy of the subsequent guidance process. According to the task target and the missing authentication information of the user, a large language model is used to generate personalized natural language guidance information and send it to the user to guide the user to complete the authentication operation. The reply information fed back by the user is received in real time, and the reply information is analyzed to confirm whether the authentication information has been completed, and the conversation is ended after the information is detected to be completed. Through this real-time feedback mechanism, the current operating status of the user can be judged more accurately and quickly to dynamically adjust the guidance process. The present invention improves the problem that the prior art cannot take into account both accuracy and efficiency when guiding users to complete authentication information in high-complexity task scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A schematic diagram of a flow chart of a method for detecting authentication information provided by an embodiment of the present invention;

[0017] Figure 2 A schematic diagram of a process for data cleaning and standardization provided by an embodiment of the present invention;

[0018] Figure 3 A schematic diagram of a process for training a large language model provided by an embodiment of the present invention;

[0019] Figure 4 Shown is a structural block diagram of a system for detecting authentication information provided by an embodiment of the present invention;

[0020] Figure 5 Shown is a structural schematic diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0021] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict.

[0022] It should be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present invention, and thus the drawings only show components related to the present invention rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component may be changed arbitrarily, and the component layout may also be more complicated.

[0023] In the following description, numerous details are discussed to provide a more thorough explanation of the embodiments of the present invention. However, it is obvious to those skilled in the art that the embodiments of the present invention can be implemented without these specific details. In other embodiments, well-known structures and devices are shown in the form of block diagrams rather than in detail to avoid making the embodiments of the present invention difficult to understand.

[0024] The inventor found that in the field of online car-hailing, in order to increase the number of registered drivers, online car-hailing platform companies will call drivers who have provided basic information such as name and mobile phone number on the platform to remind and help drivers complete subsequent data upload and certification. The professional name of this work is "driver invitation", and the staff who call the driver is called "inviter". Driver invitation work has a standard workflow, including a series of tasks that need to be completed in sequence, mainly including: introducing yourself and the intention of the call, confirming whether the driver meets the registration conditions (whether the license plate is local, whether the age of the car or driving experience meets the requirements, etc.), introducing relevant information of the platform, reminding the driver to upload necessary documents such as driver's license or vehicle registration, helping the driver to upload information or answer related questions on site, and asking whether it is possible to send text messages to invite drivers to participate in platform activities. At the same time, the "inviter" needs to follow a series of specifications during the invitation process, including: introducing the advantages of the platform to the driver when the driver refuses to arouse the driver's interest; truthfully introducing the platform and related activities, and not fabricating non-existent fixed things; polite and concise language; absolutely no insults, derogatory words, etc.

[0025] The main goal of the "driver invitation" work is to allow as many drivers as possible who have provided basic information to complete the subsequent registration and certification steps and officially become drivers of the online car-hailing platform. The "driver invitation" work has high business complexity, more behavioral norms, and drivers are less willing to cooperate most of the time. Therefore, this work must be done by invitation personnel who have undergone systematic training. In the past, the conversation robot based on the 'NLU+DM+NLG' architecture could not complete it. This also led to a high cost for driver invitation work. Due to the rapid development of LLM (large-scale language model), the conversation model represented by openAI's gpt-4o is no different from a real person in terms of conversation ability. Therefore, it is possible to use a large language model similar to GPT-4o to complete the driver invitation task. However, since the current mainstream large model is general-purpose, it cannot be simply used to replace manual invitations.

[0026] In view of the above situation, the present invention provides a method for detecting authentication information, which obtains the identity of the target user and determines the missing authentication information of the target user based on the identity. The identity can accurately identify the incomplete authentication process and dynamically generate the task target, so as to ensure the accuracy of the subsequent guidance process. According to the task target and the missing authentication information of the user, a large language model is used to generate personalized natural language guidance information and send it to the user to guide the user to complete the authentication operation. The reply information fed back by the user is received in real time, and the reply information is analyzed to confirm whether the authentication information has been completed, and the conversation is ended after the information is detected to be completed. Through this real-time feedback mechanism, the current operation status of the user can be judged more accurately and quickly to dynamically adjust the guidance process. The large language model is further trained and fine-tuned by combining real artificial invitation dialogue data to make it usable. The present invention improves the problem that the prior art cannot take into account both accuracy and efficiency when guiding users to complete authentication information in high-complexity task scenarios.

[0027] See also Figure 1 , the authentication information detection method includes the following steps:

[0028] S1. Obtain the identity of the target user.

[0029] The detection method described in the present invention can be applied to any invitation platform, including but not limited to inviting couriers, inviting takeaways, inviting truck drivers, inviting online car-hailing drivers, etc. For the convenience of description, the following is an example of inviting online car-hailing drivers, but it should be understood that other invitation methods are still within the scope of protection of the present invention. When a user intends to join an invitation platform, a part of the information will be pre-registered on the invitation platform. This information is the user's identity information, where the identity includes but is not limited to contact numbers, names or other unique identifiers such as user IDs. When it is necessary to invite a user, the invitation system will screen out relevant users who have not completed authentication from the user database, and obtain detailed authentication information of the user by querying their identity.

[0030] S2. Determine the missing authentication information of the target user based on the identity identifier.

[0031] The user database pre-stores the user's identity and corresponding authentication information. By calling the target user's identity, the platform can query the authentication status of the target user, including completed authentication tasks, unfinished tasks, and rejected authentication information. Specifically, since the invitation platform can involve a variety of different businesses, the business type to which the user belongs can be determined based on the user's identity, and the predefined authentication task list of the business type can be loaded, wherein the authentication task list records all the authentication information and specific requirements that need to be completed for the business type. For example, for online car-hailing users, the authentication information includes but is not limited to uploading ID cards, uploading driver's licenses, etc. After obtaining the actual authentication status of the user from the user database through the identity, the current actual authentication status of the target user is compared with the task list, and the missing authentication information of the target user can be determined. Among them, the missing authentication information includes but is not limited to information that the user has not filled in, information that is rejected after the user uploads, etc. For rejected information, the reasons for rejection (such as blurred photos, etc.) will also be recorded.

[0032] S3. Determine a task objective based on the missing authentication information.

[0033] According to the unfinished authentication information of the target user, the task goal of this invitation is determined. It can be understood that if there are more than one unfinished authentication information, the unfinished authentication information can be parsed item by item and the corresponding sub-task goals can be generated. Specifically, the user's authentication status can be determined by the user's identity identification, and the authentication task can be compared with the task list one by one to determine the completion status of the current authentication task. For authentication tasks with a status of "unfinished", the task is marked as a new sub-task goal. For authentication tasks with a status of "rejected", it is necessary to generate a correction sub-task goal in combination with the rejection reason. Exemplarily, the rejected task is: driver's license upload (reason: blurred photo), then the generated correction task sub-goal is: remind the user to re-take the above clear driver's license photo. Based on the preset task priority, all new sub-task goals and correction sub-task goals are sorted to generate a complete task goal.

[0034] S4. Input the task objective and the missing authentication information into a large language model, generate guidance information for guiding the target user to operate, and start this dialogue.

[0035] The task objectives and missing authentication information are structured to generate a format suitable for the large language model, and input into the trained large language model to obtain guidance information for guiding the target user's operation and start this conversation. In addition, in order to improve the accuracy of the model, the background information of the target user can also be input into the large language model at the same time, wherein the user's background information includes but is not limited to name, gender, mobile phone number, current year, month, day, whether it is close to a festival, the city where the user is located, etc. The large language model analyzes the task objectives and the current authentication progress, and combines the relevant policies of the city where the user is located, as well as festival-related policies, etc., to generate guidance information in natural language form, and start this conversation with the user. The large language model can be any model that can realize intelligent dialogue, including but not limited to GPT4, GPT4O, qwen2.5-72b-instruct, etc. Preferably, the large language model is a qwen2.5-72b-instruct model so that multi-round dialogue scenarios can be better handled.

[0036] S5. Send the guide information to the target user, and receive reply information sent by the target user.

[0037] The types of conversations in the present invention include but are not limited to voice calls, SMS calls, APP push, etc., as long as the generated guidance information can be sent to the user, it is not limited here. After the large language model generates the guidance information, the platform formats the guidance information generated by the large language model, generates a format that adapts to the information transmission channel, and sends it to the relevant target users. To ensure successful transmission, the platform will monitor the sending status of the information in real time, and when it monitors that the information fails to be sent, for SMS calls and APP push, a resend mechanism can be used to send it again. For voice calls, a redial mechanism or manual intervention can be used to ensure that the information is delivered as much as possible. After the invitation platform sends the guidance information to the user, it will receive the reply information fed back by the user within a preset time and parse it to confirm the user's task completion status.

[0038] In an embodiment of the present invention, the type of the current conversation is a voice call, and sending the guide information to the target user and receiving a reply message sent by the target user includes:

[0039] Performing text-to-speech processing on the guidance information to convert the guidance information into voice guidance information;

[0040] Sending the voice guidance information to the target user, and receiving the voice reply information sent by the target user;

[0041] The voice-type reply information is subjected to voice recognition to generate text-type reply information.

[0042] In order to improve the efficiency of completing the authentication task, in this embodiment, the type of dialogue is a voice call. For voice calls, since the large language model cannot directly parse the voice content, the user replies received by the large language model and the guidance information generated are in text form, so it is necessary to first convert the guidance information generated by the large language model into voice guidance information through the text-to-speech (TTS) technology, and generate an audio file adapted to the telephone voice call accordingly. After the voice conversion is completed, the outbound robot will be automatically triggered to initiate a call with the target user. After the call is established, the invitation platform automatically plays the aforementioned generated voice guidance information through the outbound robot to guide the user to complete the authentication information. For example, the voice guidance information sent by the outbound robot is "Hello, you have not uploaded your driver's license yet, please upload a clear photo of your driver's license on the "My Information" page". After the user answers the voice guidance information, he can interact with the invitation platform through voice reply, for example, the user replies "I have uploaded my driver's license". After receiving the voice reply information from the outbound call robot, the invitation platform converts the voice content into text through automatic speech recognition (ASR) technology to generate a text reply information. By analyzing the content of the text reply information, it determines whether the user has completed the relevant authentication information.

[0043] In one embodiment of the present invention, the step of performing speech recognition on the voice-type reply information to generate text-type reply information includes:

[0044] Performing speech recognition on the voice-type reply information to obtain initial text-type reply information;

[0045] A semantic check is performed on the initial text-type reply information to generate final text-type reply information.

[0046] After the invitation platform receives the user's voice reply information through the outbound call robot, it performs voice recognition on it, realizes voice-to-text conversion, and generates the initial text reply information. Since recognition errors, spelling errors, etc. may occur during voice recognition, it is also necessary to perform semantic checks on the generated initial text reply information to obtain the final text reply information to improve the accuracy of text information conversion. Among them, semantic checks include but are not limited to spelling checks, grammar checks, semantic integrity supplements, etc., as long as they can improve the accuracy of text information.

[0047] S6. Detecting whether the missing authentication information is complete according to the reply information, and ending the current conversation after detecting that the missing authentication information is complete.

[0048] After receiving the target user's reply, the invitation platform uses natural language processing technology to parse the reply content and confirm whether the user has completed the current authentication task. If the user's current authentication task has been completed, based on the large language model, a message to end the conversation is generated and sent to the user. If the user has not completed the authentication task, a new guidance message is generated to continue to guide the user to complete the subsequent operations.

[0049] In one embodiment of the present invention, detecting whether the authentication information is complete according to the reply information, and ending the current conversation after detecting that the authentication information is complete, includes:

[0050] Perform semantic analysis on the reply information to determine whether there is any action to complete the missing authentication information in the reply information:

[0051] If so, the completion status of the missing authentication information and the reply information are input into the large language model again, and a conversation end message is generated, and the conversation end message is sent to the target user to end the conversation;

[0052] If not, the reply information, the task target and the missing authentication information are input into the large language model again, new guidance information is generated based on the reply information, and the new guidance information is sent to the target user.

[0053] After receiving the reply information from the user, the invitation platform performs semantic analysis on the reply information based on natural language processing technology to analyze whether there is an action to complete the authentication information in the user's reply. If the user's reply information contains the action of completing the authentication information, the user's reply information and the completion of the authentication information are input into the large language model again to generate the end of the dialogue information (such as "Thank you for your cooperation! Your driver's license has been uploaded successfully, this call is over, goodbye!"), and sent to the user through voice call, SMS or APP push, etc., to end the dialogue. On the contrary, if the user's reply information does not contain the completion action of the authentication information, the user's current reply information, task objectives, and authentication progress and other authentication information are re-entered into the large language model, so that the large language model generates new guidance information based on the user's current reply and sends it to the user again. Exemplarily, if the user replies "I don't know how to upload", the guidance information generated by the large language model can be "Please click 'Upload driver's license' on the 'My Information' page and select a clear driver's license photo to submit. If you need help, please contact the platform customer service." Through this interactive dialogue with the user, the user is guided to complete the authentication information, and finally the dialogue ends after the user completes it.

[0054] In one embodiment of the present invention, the large language model is obtained by training based on a pre-acquired training set and a corresponding label set. The acquisition process of each training sample and the corresponding label in the training set is:

[0055] Based on pre-acquiring conversation information between the inviter and the user, wherein each conversation information carries the identity identifier of the inviter and the identity identifier of the user and the corresponding conversation content;

[0056] Performing data cleaning on the conversation information to obtain cleaned conversation information;

[0057] Performing standardization processing on the cleaned dialogue information to obtain standardized dialogue information;

[0058] Searching and identifying the missing authentication information of the user from the user database according to the user's identity identifier in the standardized dialogue information, and determining the corresponding task target based on the missing authentication information;

[0059] The conversation content of the user's identity is selected from the standardized conversation information, the conversation content of the user's identity and the missing authentication information and the corresponding task objectives are used as training samples, and the conversation content of the corresponding inviter's identity is used as a label.

[0060] After obtaining a complete conversation message between the inviter and the user, data cleaning is first performed, wherein the types of data cleaning include but are not limited to conversation screening, format optimization, etc. Exemplarily, since the conversation message carries the identity identifiers of the inviter and the user, the inviter's performance evaluation results, etc., it can be first determined whether the conversation message is available based on the inviter's performance evaluation results. If the evaluation result is lower than the preset evaluation standard, the conversation message is unavailable, deleted from the training set, and a new conversation message is selected. Then it is determined whether the conversation message meets the following judgment criteria. If it does, the conversation message is removed and replaced with another conversation message for processing. The judgment criteria are: no connection or only 1-4 rounds of conversation; the inviter's return visit record (not the first invitation); the conversation is incomplete and the complete business process has not been completed; the conversation only contains the inviter or the user's unilateral speech; there are a lot of dialects or ASR recognition errors in the conversation. After the above data cleaning, the cleaned conversation information is obtained, and the cleaned conversation information is standardized to obtain standardized conversation information. Among them, the standardization method includes but is not limited to the merging optimization of the conversation content and the rewriting of sentences. For example, the contents of the inviter's or user's consecutive speeches are merged to ensure that the dialogue content is presented in an alternating form of "you speak, I speak", and the inviter's dialogue is rewritten to rewrite the verbose sentences and sentences with many modal particles into smooth and concise sentences.

[0061] According to the user's identity in the standardized conversation information, the user's authentication status is retrieved from the user database, where the authentication status includes the user's completed authentication information and incomplete authentication information, and the incomplete authentication information is identified. According to the method determined by the aforementioned task objectives, the task objectives are generated based on these incomplete authentication information. The conversation content related to the user's identity is extracted from the standardized conversation information, and the user's conversation content is integrated with the missing authentication information and the task objectives to form a complete training sample. The conversation content of the inviter in the corresponding standardized conversation information is used as the label of the training sample for training the large language model.

[0062] Specifically, see Figure 2 For the acquired conversation information, high-quality calls are screened out according to the inviter level, clue level, post-call tags, etc., and conversations with fewer conversation rounds than the preset threshold are eliminated. The above processing is processed by rules and scripts to preliminarily screen out conversation information that meets the requirements. The conversation content is further optimized by manually cleaning and standardizing the initially screened conversation information, and the final conversation data is used as the training sample of the model for model training.

[0063] In one embodiment of the present invention, the training process of the large language model includes:

[0064] Inputting the training set into a pre-trained large language model to obtain a prediction guidance information set;

[0065] Calculating the difference between the prediction guide set and the label set;

[0066] Based on the difference, all parameters of the large language model are optimized according to the full parameter fine-tuning method to obtain a trained large language model.

[0067] The cleaned and standardized training set is input into the pre-trained large language model. For each training sample, the user's identity, the historical conversation of this call, and the task goal are input into the large language model as a training sample, so that the model generates relevant guidance information based on the user identity, conversation history, and task goal. To guide the user to complete the authentication task. The difference between the generated guidance information and the corresponding true label is calculated by the loss function, wherein the loss function includes but is not limited to cross entropy loss, mean square error loss, KL divergence, etc. According to the difference, the full parameter fine-tuning method is used to adjust the various parameters of the large language model layer by layer through back propagation. When the difference is less than the preset difference threshold or the training reaches the preset number of rounds, the trained large language model is finally obtained. By fine-tuning all parameters of the large language model, for the complex reasoning scenario of multi-round dialogues such as the present invention, the model can have a strong understanding ability, so that more accurate guidance information can be generated according to the user's reply.

[0068] In one embodiment of the present invention, the training process of the large language model includes:

[0069] Inputting the training set into a pre-trained large language model to obtain a prediction guidance information set;

[0070] Calculating the difference between the prediction guide set and the label set;

[0071] Based on the difference, the large language model is incrementally pre-trained according to the full parameter fine-tuning method to obtain the initially trained large language model;

[0072] The initially trained large language model is fine-tuned according to instructions, and local parameters of the initially trained large language model are optimized based on the LoRA algorithm to obtain a trained large language model.

[0073] The aforementioned training set after data cleaning and standardization is input into the pre-trained large language model. For each training sample: the identity identifier, historical dialogue information and task target of the dialogue user are input into the large language model so that the model generates corresponding guidance information, thereby guiding the user to perform a specific authentication operation. The difference between the guidance information generated by the model and the corresponding true label is calculated by the loss function, wherein the loss function includes but is not limited to cross entropy loss, mean square error loss, KL divergence, etc. According to the calculated difference, all parameters of the pre-trained large language model are adjusted by the back propagation algorithm, and the large language model is iteratively trained until the difference is less than the preset difference threshold or the number of iterative training reaches the preset number threshold, then the initially trained large language model is obtained. In order to further improve the recognition ability of the model, the present invention also locally fine-tunes the parameters of the model through instruction fine-tuning and LoRA algorithm. Specifically, for each training sample, the user's dialogue history, task target, user identifier and corresponding task instruction are input into the initially trained large language model again so that the model learns how to generate output that meets the requirements according to the task instruction. Among them, the task instruction is a description used to guide the model to complete a specific task, so that the model can understand the task information and generate response information that meets the requirements. The prediction guidance set output by the model and the corresponding true label set are recalculated, and based on the calculation results, the local parameters of the model are optimized and updated through the LoRA algorithm to obtain the final trained large language model. By fine-tuning the instructions and updating the local parameters of the model with the LoRA algorithm, the robustness of the model in specific task scenarios is effectively improved, resource consumption is reduced, and the generalization ability of the model is improved.

[0074] Specifically, see Figure 3First, the pre-trained large language model (such as the qwen2.5-72b-instruct model) is incrementally pre-trained through the algorithm's general dialogue data (i.e., the dialogue record between the driver and the inviter) to obtain an initially trained large language model. Then, by constructing instruction fine-tuning data, the instruction fine-tuning data is sent to the initially trained large language model, and the instruction fine-tuning is performed again to obtain a trained large language model. Among them, the present invention modularizes task-related information into several sub-prompts (prompts) by constructing a system-level prompt (system_prompt) suitable for instruction fine-tuning, including a description of the task prompt, a registration rule prompt, a platform advantage prompt, a driver information prompt, a driver authentication progress prompt, and other dynamic information prompts. Combined with the general dialogue data, the system_prompt, historical dialogues, and the current user's questions are used as input instructions, and the inviter's reply is used as a reference answer to construct an instruction fine-tuning data set. Specifically, a 10-round conversation can be split into 10 instruction fine-tuning data to fine-tune the initially trained large language model, so that the trained model can generate more targeted guidance information.

[0075] See also Figure 4 , the authentication information detection system 100 includes: an identity acquisition module 110, a missing authentication information determination module 120, a target determination module 130, a guide information generation module 140, an information feedback module 150 and a missing information detection module 160. The above-mentioned identity acquisition module 110 is used to obtain the identity of the target user. The missing authentication information determination module 120 is used to determine the missing authentication information of the target user based on the identity. The target determination module 130 is used to determine the task target of this dialogue based on the authentication information. The guide information generation module 140 is used to input the task target and the missing authentication information into the large language model to generate guidance information for guiding the target user to operate. The information feedback module 150 is used to send the missing guidance information to the target user and receive the reply information sent by the target user. The missing information detection module 160 detects whether the missing authentication information is completed according to the reply information, and ends the current dialogue after detecting that the missing authentication information is completed.

[0076] For the specific definition of the authentication information detection system, please refer to the definition of the authentication information detection method above, which will not be repeated here. Each module in the above authentication information detection system can be implemented in whole or in part by software, hardware and a combination thereof. The above modules can be embedded in or independent of the processor in the computer device in hardware format, or stored in the memory in the computer device in software format, so that the processor can call the operations corresponding to the above modules.

[0077] It should be noted that, in order to highlight the innovative part of the present invention, the present embodiment does not introduce modules that are not closely related to solving the technical problem proposed by the present invention, but this does not mean that there are no other modules in the present embodiment.

[0078] See also Figure 5 The electronic device 1 may include a memory 12, a processor 13 and a bus, and may also include a computer program stored in the memory 12 and executable on the processor 13, such as a detection program for authentication information.

[0079] Among them, the memory 12 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, mobile hard disk, multimedia card, card-type memory (for example: SD or DX memory, etc.), magnetic memory, disk, optical disk, etc. In some embodiments, the memory 12 can be an internal storage unit of the electronic device 1, such as a mobile hard disk of the electronic device 1. In other embodiments, the memory 12 can also be an external storage device of the electronic device 1, such as a plug-in mobile hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the electronic device 1. Further, the memory 12 can also include both an internal storage unit of the electronic device 1 and an external storage device. The memory 12 can not only be used to store application software and various types of data installed in the electronic device 1, such as the code for detecting authentication information, etc., but also can be used to temporarily store data that has been output or is to be output.

[0080] In some embodiments, the processor 13 may be composed of an integrated circuit, for example, a single packaged integrated circuit, or a plurality of packaged integrated circuits with the same or different functions, including one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and combinations of various control chips. The processor 13 is the control core (Control Unit) of the electronic device 1, and uses various interfaces and lines to connect various components of the entire electronic device 1, and executes or executes programs or modules (such as authentication information detection programs, etc.) stored in the memory 12, and calls data stored in the memory 12 to execute various functions of the electronic device 1 and process data.

[0081] The processor 13 executes the operating system and various installed applications of the electronic device 1. The processor 13 executes the applications to implement the steps in the above authentication information detection method.

[0082] Exemplarily, the computer program may be divided into one or more modules, which are stored in the memory 12 and executed by the processor 13 to complete the present application. The one or more modules may be a series of computer program instruction segments capable of completing specific functions, which are used to describe the execution process of the computer program in the electronic device 1. For example, the computer program may be divided into an identity acquisition module 110, a missing authentication information determination module 120, a target determination module 130, a guidance information generation module 140, an information feedback module 150, and a missing information detection module 160.

[0083] The above-mentioned integrated unit implemented in the form of a software function module can be stored in a computer-readable storage medium, and the computer-readable storage medium can be non-volatile or volatile. The above-mentioned software function module is stored in a storage medium, including several instructions for enabling a computer device (which can be a personal computer, a computer device, or a network device, etc.) or a processor to perform part of the functions of the authentication information detection method described in each embodiment of the present application.

[0084] In summary, the present invention discloses a method, system, device and medium for detecting authentication information, obtains the identity of the target user and determines the missing authentication information of the target user based on the identity, and can accurately identify the unfinished authentication process through the identity, dynamically generate the task target, and thus ensure the accuracy of the subsequent guidance process. According to the task target and the missing authentication information of the user, a large language model is used to generate personalized natural language guidance information and send it to the user to guide the user to complete the authentication operation. The reply information fed back by the user is received in real time, and the reply information is analyzed to confirm whether the authentication information has been completed, and the conversation is ended after the information is detected to be completed. Through this real-time feedback mechanism, the current operation status of the user can be judged more accurately and quickly to dynamically adjust the guidance process. In this way, the invitation platform does not need to train and employ a large number of inviters to perform user authentication guidance work, thereby saving costs and reducing management complexity. In addition, the large voice model can strictly abide by the established business processes and dialogue specifications, reduce complaints caused by emotional fluctuations and uneven work capabilities during the real-person invitation process, and greatly improve the user experience. The present invention improves the problem that the prior art cannot take into account both accuracy and efficiency when guiding users to complete authentication information in highly complex task scenarios. Therefore, the present invention effectively overcomes various shortcomings in the prior art and has high industrial utilization value.

[0085] The above embodiments are merely illustrative of the principles and effects of the present invention, and are not intended to limit the present invention. Anyone familiar with the art may modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by a person of ordinary skill in the art without departing from the spirit and technical concept disclosed by the present invention shall still be covered by the claims of the present invention.

Claims

1. A method for detecting authentication information, characterized in that: Applied to the invitation platform, the method includes: Get the target user's identity. Determining missing authentication information of the target user based on the identity identifier; Determining a task objective based on the missing authentication information; Inputting the task goal and the missing authentication information into a large language model, generating guidance information for guiding the target user to operate and starting this dialogue; Sending the guide information to the target user, and receiving a reply message sent by the target user; The missing authentication information is detected to be complete according to the reply information, and the current conversation is ended after it is detected that the missing authentication information is complete.

2. The method for detecting authentication information according to claim 1, characterized in that: The type of the current conversation is a voice call, and sending the guide information to the target user and receiving a reply message sent by the target user includes: Performing text-to-speech processing on the guidance information to convert the guidance information into voice guidance information; Sending the voice guidance information to the target user, and receiving the voice reply information sent by the target user; The voice-type reply information is subjected to voice recognition to generate text-type reply information.

3. The method for detecting authentication information according to claim 2, characterized in that: The step of performing speech recognition on the voice-type reply information to generate text-type reply information includes: Performing speech recognition on the voice-type reply information to obtain initial text-type reply information; A semantic check is performed on the initial text-type reply information to generate final text-type reply information.

4. The method for detecting authentication information according to claim 1, characterized in that: The detecting whether the missing authentication information is completed according to the reply information, and ending the current conversation after detecting that the missing authentication information is completed, includes: Perform semantic analysis on the reply information to determine whether there is any action to complete the missing authentication information in the reply information: If so, the completion status of the missing authentication information and the reply information are input into the large language model again, and a conversation end message is generated, and the conversation end message is sent to the target user to end the conversation; If not, the reply information, the task target and the missing authentication information are input into the large language model again, new guidance information is generated based on the reply information, and the new guidance information is sent to the target user.

5. The method for detecting authentication information according to claim 1, characterized in that: The large language model is obtained by training based on a pre-acquired training set and a corresponding label set. The acquisition process of each training sample and the corresponding label in the training set is: Based on pre-acquiring conversation information between the inviter and the user, wherein each conversation information carries the identity identifier of the inviter and the identity identifier of the user and the corresponding conversation content; Performing data cleaning on the conversation information to obtain cleaned conversation information; Performing standardization processing on the cleaned dialogue information to obtain standardized dialogue information; Searching and identifying the missing authentication information of the user from the user database according to the user's identity identifier in the standardized conversation information, and determining the corresponding task target based on the missing authentication information; The conversation content of the user's identity is selected from the standardized conversation information, the conversation content of the user's identity and the missing authentication information and the corresponding task objectives are used as training samples, and the conversation content of the corresponding inviter's identity is used as a label.

6. The method for detecting authentication information according to claim 5, characterized in that: The training process of the large language model includes: Inputting the training set into a pre-trained large language model to obtain a prediction guidance information set; Calculating the difference between the prediction guide set and the label set; Based on the difference, all parameters of the large language model are optimized according to the full parameter fine-tuning method to obtain a trained large language model.

7. The method for detecting authentication information according to claim 5, characterized in that: The training process of the large language model includes: Inputting the training set into a pre-trained large language model to obtain a prediction guidance information set; Calculating the difference between the prediction guide set and the label set; Based on the difference, the large language model is incrementally pre-trained according to the full parameter fine-tuning method to obtain the initially trained large language model; The initially trained large language model is fine-tuned according to instructions, and local parameters of the initially trained large language model are optimized based on the LoRA algorithm to obtain a trained large language model.

8. A system for detecting authentication information, characterized in that: The system comprises: An identity acquisition module, used to acquire the identity of a target user; A missing authentication information determination module, used to determine the missing authentication information of the target user based on the identity identifier; A target determination module, used to determine the task target of this dialogue based on the missing authentication information; A guidance information generation module, used for inputting the task target and the missing authentication information into a large language model to generate guidance information for guiding the target user to operate; An information feedback module, used to send the guidance information to the target user and receive reply information sent by the target user; The missing information detection module detects whether the missing authentication information is completed according to the reply information, and ends the current conversation after detecting that the missing authentication information is completed.

9. An electronic device, characterized in that: The electronic device comprises: one or more processors; A storage device for storing one or more programs, which, when executed by the one or more processors, enables the electronic device to implement the authentication information detection method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the computer program is executed by a processor of a computer, the computer is caused to execute the method for detecting authentication information as claimed in any one of claims 1 to 7.

Images