An invitation method, apparatus, device, and computer-readable storage medium

Abstract

This invention discloses an invitation method, apparatus, device, and computer-readable storage medium, applied in the field of computer technology. The method includes: retrieving the historical invitation process with the highest matching degree from a historical invitation dialogue database based on the current dialogue context; constructing dynamic invitation prompts based on the historical invitation process with the highest matching degree and the current invitation stage; obtaining the target invitation content using an invitation model based on the dynamic invitation prompts and the current dialogue context; and making the invitation based on the target invitation content. This invention identifies the historical invitation process with the highest matching degree, then guides the user along the invitation path within that process. The dynamic invitation prompts precisely locate the current invitation stage of the dialogue. As the dialogue progresses, this current dialogue context is dynamically updated in the prompts, thereby accurately constraining the next action of the large model, improving the accuracy of responses and increasing the invitation success rate.

Description

Technical Field

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

[0002] In AI-powered outbound call scenarios in industries such as ride-hailing, food delivery, and insurance, the system needs to persuade drivers / customers to visit offline stores or complete registration within a short period. Taking the "driver to store" invitation as an example, the traditional process typically includes six fixed steps: ① confirm vehicle type → ② introduce platform advantages → ③ inquire about location → ④ inform store address → ⑤ initiate invitation → ⑥ confirm time. To improve intelligence, existing technologies pre-write the complete process (e.g., "confirm vehicle type → introduce advantages → inquire about location → inform store → initiate invitation → confirm time"), and a large model generates the response. While this eliminates the need for step-by-step scripting, the process sequence is locked within a fixed framework, lacking interpretability and process controllability, resulting in low invitation accuracy.

[0003] It is evident that improving the accuracy of invitations is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide an invitation method, apparatus, device and computer-readable storage medium, which solves the technical problem of low invitation accuracy in the prior art.

[0005] To address the aforementioned technical problems, the present invention provides an invitation method, comprising: Based on the current dialogue context, retrieve the historical invitation process with the highest matching degree from the historical invitation dialogue database; Based on the historical invitation process with the highest matching degree and the current invitation stage, dynamic invitation prompts are constructed; Based on the dynamic invitation prompt and the current dialogue context, the invitation model is used to obtain the target invitation content, and an invitation is made based on the target invitation content.

[0006] Optionally, after obtaining the target invitation content using the invitation model based on the dynamic invitation prompt and the current dialogue context, and making an invitation based on the target invitation content, the method further includes: Obtain the actual response content of the user based on the target invitation content; Determine the semantic similarity between the actual response content and the expected response of the historical invitation process with the highest matching degree in the current invitation stage; Determine whether the semantic similarity is lower than a preset deviation threshold; When the similarity is lower than the preset deviation threshold, the step of retrieving the historical invitation process with the highest matching degree is re-executed.

[0007] Optionally, based on the current dialogue context, retrieve the historical invitation process with the highest matching degree from the historical invitation dialogue database, including: Based on the current dialogue context, completed invitation steps, and a preset number of recent invitation dialogue contents, the historical invitation processes in the historical invitation dialogue database are retrieved to determine the historical invitation process with the highest matching degree.

[0008] Optionally, based on the current dialogue context, completed invitation steps, and a preset number of recent invitation dialogue contents, the historical invitation processes in the historical invitation dialogue database are retrieved to determine the historical invitation process with the highest matching degree, including: A retrieval query is constructed based on the current dialogue context, the completed invitation steps, and the preset number of recent invitation dialogue contents, and the retrieval query is encoded to obtain a retrieval vector; Based on the retrieval vector, a similarity search is performed in the vector database corresponding to the historical invitation dialogue database to recall the first candidate invitation process set; Based on the completed invitation steps, the second candidate invitation process set is recalled through an inverted index; The invitation processes in the first candidate invitation process set and the second candidate invitation process set are merged and weighted, and the invitation process ranked first is determined as the historical invitation process with the highest matching degree.

[0009] Optionally, based on the historical invitation process with the highest matching degree and the current invitation stage, dynamic invitation prompts are constructed, including: Obtain a predefined instruction template, which contains variable placeholders for filling in invitation dialogue information; The step sequence, information slots, and dialogue text from the historical invitation process with the highest matching degree are filled into the corresponding variable placeholders in the instruction template to obtain the dynamic invitation prompt.

[0010] Optionally, before retrieving the most relevant historical invitation process from the historical invitation dialogue database based on the current dialogue context, the following steps are also included: The invitation voice is converted into a set of invitation text dialogues using automatic speech recognition technology; Each invitation text dialogue in the set of invitation text dialogues is structured into a target structured path; wherein, the target structured path includes the sequence of execution steps, the set key information slots, the complete invitation dialogue text, and the final result of the invitation session; All target structured paths are stored in the historical invitation dialogue database, and an inverted index is built for the sequence of executed steps; wherein, the inverted index is an index that includes the correspondence between the sequence of executed steps and the structured paths.

[0011] The present invention also provides an invitation device, comprising: The module for determining the historical invitation process with the highest matching degree is used to retrieve the historical invitation process with the highest matching degree from the historical invitation dialogue database based on the current dialogue context. The dynamic invitation prompt word construction module is used to construct dynamic invitation prompt words based on the historical invitation process with the highest matching degree and the current invitation stage; The invitation module is used to obtain the target invitation content based on the dynamic invitation prompt and the current dialogue context using the invitation model, and to make an invitation based on the target invitation content.

[0012] Optionally, the aforementioned invitation device further includes: The actual response content determination module is used to obtain the actual response content of the user based on the target invitation content; The semantic similarity determination module is used to determine the semantic similarity between the actual response content and the expected response of the historical invitation process with the highest matching degree in the current invitation stage; The judgment module is used to determine whether the semantic similarity is lower than a preset deviation threshold; The module for re-determining the historical invitation process with the highest matching degree is used to re-execute the step of retrieving the historical invitation process with the highest matching degree when the similarity is lower than the preset deviation threshold.

[0013] The present invention also provides an invitation device, comprising: Memory, used to store computer programs; A processor for executing the computer program to implement the steps of the above-described invitation method.

[0014] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described invitation method.

[0015] The present invention also provides a computer program product, including a computer program / instructions, which, when executed by a processor, implement the steps of the above-described invitation method.

[0016] As can be seen, this invention retrieves the historical invitation process with the highest matching degree from the historical invitation dialogue database based on the current dialogue context; constructs dynamic invitation prompts based on the historical invitation process with the highest matching degree and the current invitation stage; obtains the target invitation content using the invitation model based on the dynamic invitation prompts and the current dialogue context, and then makes the invitation based on the target invitation content. The beneficial effects of this invention are: compared with directly inputting and determining all response content, this invention determines a historical invitation process with the highest matching degree, and then guides the user along the invitation path in the historical invitation process with the highest matching degree. Moreover, the dynamic invitation prompts accurately locate the invitation stage of the current dialogue. As the dialogue progresses, this "current invitation stage" will be dynamically updated in the prompts, thereby accurately constraining the next action of the large language model, thus improving the accuracy of the response and the invitation success rate.

[0017] In addition, the present invention also provides an invitation device, apparatus, and computer-readable storage medium, which also have the above-mentioned beneficial effects. Attached Figure Description

[0018] 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.

[0019] Figure 1 A flowchart of an invitation method provided in an embodiment of the present invention; Figure 2 A flowchart illustrating an invitation method provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of an invitation device provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of an invitation device provided in an embodiment of the present invention. Detailed Implementation

[0020] 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.

[0021] Please refer to Figure 1 , Figure 1A flowchart illustrating an invitation method provided in an embodiment of the present invention. The method may include: S101, based on the current dialogue context, retrieve the historical invitation process with the highest matching degree from the historical invitation dialogue database.

[0022] Each step in this embodiment can be executed by a designated electronic device, which can be a server, a portable terminal, or other form. This electronic device contains memory modules, the specific number of which is not limited. In this embodiment, the current dialogue context refers to the multi-turn dialogue records between the user and the invitation system under real-time monitoring. The historical invitation dialogue database in this embodiment is a saved complete conversation chain of a manually invited user (invitation steps + corresponding script). This embodiment determines the matching degree between the current dialogue context and the invitation path sequence in the historical invitation dialogue database, thus using the historical invitation path with the highest matching degree as the historical invitation process. In this embodiment, the already traversed path (current dialogue context) is used as the Query (question), and the Top-K (first K) complete chains are retrieved, that is, based on the current dialogue context, the most matching subsequent invitation process is retrieved in real-time. This embodiment decomposes the retrieved remaining links into a Step-Agent queue, which sequentially drives TTS (Text-to-Speech) / ASR (Automatic Speech Recognition). (TTS is responsible for speaking the intended message in a professional and clear voice. ASR is responsible for listening to and understanding what the invitee said.) The core purpose of designing the Step-Agent is to transform the "static path" into an "executable and interruptible unit." A retrieved path is a "dead" text record. The execution engine needs to turn it into a "living," step-by-step executable task, and after each step, it can check external feedback (user response) to decide whether to continue to the next step or replan. Therefore, the Step-Agent is a crucial processing node in this process. Its core role is that each Step-Agent knows how to handle the step it is responsible for (generating the script, invoking TTS / ASR, understanding the user's intent). The "user intent" it outputs is the sole basis for the deviation monitor to determine whether to continue the current path. Different Step-Agent subclasses can be easily defined for different types of steps (such as inquiry, confirmation, and sales) to achieve more granular control. From the retrieved Top-K candidate links, select the optimal link (the historical invitation process with the highest matching degree), and decompose the remaining steps that have not yet been executed into a Step-Agent queue, and drive the TTS / ASR to execute in sequence.

[0023] It should be further explained that, based on any of the above embodiments, S101, which retrieves the historical invitation process with the highest matching degree from the historical invitation dialogue database based on the current dialogue context, may include: searching the historical invitation process in the historical invitation dialogue database based on the current dialogue context, completed invitation steps, and a preset number of recent invitation dialogue contents to determine the historical invitation process with the highest matching degree. In this embodiment, the completed step sequence (i.e., the completed invitation steps) and the dialogue text of the most recent rounds can be extracted and combined into a search query to retrieve the historical invitation process in the historical invitation dialogue database. In this embodiment, the completed invitation steps are standardized step sequences that have been executed and automatically identified and extracted from the current dialogue history; the dialogue text of the most recent rounds (a preset number of recent invitation dialogue contents) in this embodiment are the original dialogue records of the most recent interactions between the user and the system, which are unrefined natural language. Combining the two in this embodiment is equivalent to using both the "logical framework" and the "specific context" as two keys to find the most similar script in the historical invitation dialogue database. This ensures that the system can both follow the correct business processes and generate flexible responses that fit the details of the current conversation.

[0024] It should be further explained that, based on any of the above embodiments, the above-mentioned retrieval of historical invitation processes from the historical invitation dialogue database based on the current dialogue context, completed invitation steps, and a preset number of recent invitation dialogue contents to determine the historical invitation process with the highest matching degree may include: S1011: Construct a retrieval query based on the current dialogue context, completed invitation steps, and a preset number of recent invitation dialogue contents, and encode the retrieval query to obtain a retrieval vector.

[0025] This embodiment can perform hybrid retrieval based on retrieval vectors. Hybrid retrieval includes both semantic and step-by-step retrieval, specifically a vector-inverted hybrid index. The vector is responsible for semantic similarity, and the inverted index is responsible for step sequence matching. In this embodiment, the current query is not a single-turn user statement, but a composite query condition composed of system state (Seq_done) and dialogue context (Turns). It describes "what is happening" from both the dimensions of "business process" and "dialogue context," thereby enabling a more comprehensive and accurate retrieval of the most relevant requests and processes from the historical invitation dialogue database. The completed invitation steps used in this embodiment ensure the overall direction is correct, and the preset number of recent invitation dialogue contents ensures the optimal response strategy. Only by combining the two can the most appropriate historical invitation process in terms of strategy and context be retrieved from massive historical data.

[0026] S1012, based on the retrieval vector, perform a similarity search in the vector database corresponding to the historical invitation dialogue database to recall the first candidate invitation process set.

[0027] This embodiment can encode the retrieval query Q into a vector, perform a similarity search in the vector database corresponding to the historical invitation dialogue database, and recall the top-K historical paths that are semantically most similar.

[0028] S1013, based on the completed invitation steps, recall the second candidate invitation process set through the inverted index.

[0029] This embodiment can retrieve the Top-K historical paths matching the completed step sequence through an inverted index, forming a second candidate invitation process set. In this embodiment, the inverted index is like the final index of a book, telling you which keyword appears on which pages (e.g., "apple → pages 5, 18, 92"), instead of you flipping through pages to find "apple". Forward index: finds content (keywords) by page number (document). Inverted index: finds page number (document) by content (keyword). That is, given a keyword (completed invitation steps), it directly returns all documents containing it (the second candidate invitation process set). In this embodiment, the inverted index can complement vector retrieval (finding semantically similar elements), forming a dual guarantee of "precise step matching + fuzzy semantic matching," ensuring that the retrieved "invitation processes" are both fast and accurate.

[0030] S1014, perform a weighted fusion sort on the invitation processes in the first candidate invitation process set and the second candidate invitation process set, and determine the invitation process ranked first as the historical invitation process with the highest matching degree.

[0031] This embodiment can filter invitation processes based on the first and second candidate invitation process sets: retaining only the top-scoring results and eliminating low-scoring or invalid candidates. It extracts key "next step suggestions" and "recommended scripts." The results are packaged into clear, structured instructions (such as JSON (Lightweight Data Interchange Format)) to directly guide what to say and do next. This embodiment can combine semantic and step-by-step matching scores to weight and rank the recall results, selecting the unique and optimal historical successful path as the "script" for the current dialogue (i.e., the historical invitation process with the highest matching degree). This embodiment employs a hybrid retrieval mechanism of "semantic recall + step-by-step recall," whose core advantage lies in significantly improving the accuracy, robustness, and business alignment of retrieval results through dual verification and complementarity. It is understandable that the core purpose of hybrid recall is to combine the advantages of different retrieval methods, leveraging their strengths and compensating for their weaknesses, to achieve a more accurate and robust retrieval effect. Using any single method may have shortcomings, while a hybrid strategy ensures that the system can find the most suitable "script" under any circumstances. The purpose of hybrid recall is to dynamically and intelligently match an "optimal historical template" for each conversation within milliseconds, thereby greatly improving the success rate and efficiency of invitations, while ensuring the stability and professionalism of service quality.

[0032] It should be further noted that, based on any of the above embodiments, before retrieving the historical invitation process with the highest matching degree from the historical invitation dialogue database based on the current dialogue context, the process may further include: The invitation voice is converted into a set of invitation text dialogues using automatic speech recognition technology. Each invitation text dialogue in the set is then structured into a target structured path. The target structured path includes a sequence of execution steps, designated key information slots, the complete invitation dialogue text, and the final result of the invitation session. All target structured paths are stored in a historical invitation dialogue database, and an inverted index is created for the sequence of execution steps. The inverted index is an index that includes the correspondence between the sequence of execution steps and the structured path. In this embodiment, the inverted index represents the correspondence between the sequence of execution steps and the historical invitation process. This embodiment collects tens of thousands of complete, successful and failed human agent recordings over several months and transcribes these recordings into a set of text dialogues D={d1, d2, ...} using automatic speech recognition (ASR) technology. Each dialogue d... i Structured into a standardized path i Path i It contains four key parts: Seq i : The sequence of steps to be executed (e.g., [ask about car model, ask about region, invite someone]). Slot iKey information slots collected (e.g., {vehicle type: heavy truck, city: Shanghai}). Dialog i : The complete dialogue text between the user (the invitee) and the inviter. Result i : The final result of the session (success / failure). All structured paths... i Stored in a vector database and assigned to a step sequence Seq i An inverted index is built for fast retrieval. The historical invitation dialogue database in this embodiment not only stores the historical invitation process (the sequence of executed steps), but also includes the complete dialogue text, the final result of the session, and key information slots, thereby improving the accuracy of subsequent construction of dynamic invitation prompts.

[0033] S102, Based on the historical invitation process with the highest matching degree and the current invitation stage, construct dynamic invitation prompts.

[0034] This embodiment dynamically populates information such as dialogue, steps, and slots from the most closely matched historical invitation process into a preset instruction template, forming a powerful dynamic invitation prompt. This instruction explicitly instructs the large language model: what role to play, what steps to follow, what tone to reference, and what content to generate. This embodiment can format the retrieved complete link (the most closely matched historical invitation process) into a long text instruction in real time, including a) the full text of the historical dialogue, b) the steps already taken, c) the current slot, and d) mandatory / optional skip rules; adaptive length truncation and placeholder mechanisms ensure that the maximum context of the large language model is never exceeded. This embodiment does not retrain or fine-tune the large language model; it only uses a variable system-prompt to allow the model to generate the next round of responses under the constraint of "remaining steps in the current link"; it utilizes the large model's in-context learning capabilities to reproduce manual strategies in one go. The core of the "dynamic invitation prompt" in this embodiment lies in its real-time construction "based on the most similar complete historical link." Its "dynamic" nature is primarily reflected in two aspects: Dynamic content changes: After each user speaks, the system re-searches to find the most matching complete dialogue link. This means that the content of each generated prompt may be different, entirely depending on which historical path is most applicable at the moment. It is not a fixed template. Real-time state updates: The prompt not only includes the steps but also precisely locates the current step in this link of the dialogue (e.g., "Step 1 completed, now step 2 must be executed"). As the dialogue progresses, this "current step" is dynamically updated in the prompt, thus precisely constraining the next action of the larger model.

[0035] It should be further explained that, based on any of the above embodiments, S102, which constructs a dynamic invitation prompt based on the historical invitation process with the highest matching degree and the current invitation stage, may include: obtaining a predefined instruction template, which contains variable placeholders for filling invitation dialogue information; filling the step sequence, information slots, and dialogue text of the historical invitation process with the highest matching degree into the corresponding variable placeholders in the instruction template to obtain a dynamic invitation prompt. This embodiment dynamically fills the dialogue, steps, slots, and other information from the retrieved historical invitation process with the highest matching degree into a preset instruction template, forming a powerful dynamic invitation prompt. This invention creates a "contextualized" expert guidance manual, rather than a "static" operation list, which fundamentally solves the shortcomings of large language models in professional tasks, such as lack of specific business knowledge and real-time context.

[0036] S103: Based on the dynamic invitation prompts and the current dialogue context, the invitation model is used to obtain the target invitation content, and the invitation is made based on the target invitation content.

[0037] In this embodiment, dynamically generated invitation prompts and the current dialogue context are input into a large language model (i.e., the invitation model in this application can be a large language model). Under the strict constraints of the aforementioned dynamic invitation prompts, the invitation model generates the next sentence, which is both strategic and naturally fluent. It should be noted that sensitive words can be filtered for each step of the dialogue; a manually preset "no skipping" whitelist is allowed to ensure compliance: sensitive words are automatically filtered for each sentence; and skipping is prevented: key steps are forced to be executed sequentially and cannot be skipped.

[0038] It should be further noted that, based on any of the above embodiments, after S103 obtains the target invitation content using the invitation model based on the dynamic invitation prompt and the current dialogue context, and makes the invitation based on the target invitation content, it may further include: S104, Obtain the actual response content of the user based on the target invitation content; S105, Determine the semantic similarity between the actual response content and the expected response of the historical invitation process with the highest matching degree in the current invitation stage; S106, Determine whether the semantic similarity is lower than the preset deviation threshold; S107, when the similarity is lower than the preset deviation threshold, the step of retrieving the historical invitation process with the highest matching degree is re-executed.

[0039] This embodiment can calculate the similarity between the driver's response and the locked link in real time, and trigger a re-search if it is lower than the threshold. This embodiment can monitor whether the actual response content of the user's subsequent response deviates from the expected response in the selected "historical invitation process with the highest matching degree". If the deviation exceeds the deviation threshold (e.g., semantic similarity <0.75), a re-search is immediately triggered (returning to S101), and the optimal path (historical invitation process with the highest matching degree) is reselected according to the latest dialogue state to achieve self-correction and ensure that the dialogue does not go astray. The invitation process in this embodiment can specifically include: Call start: The Path-Retriever uses an empty path search (e.g., using "the purpose of the call" (e.g., to book a follow-up class) and "driver information" (e.g., a new driver) to search. For example, directly look for "successful script process for booking a follow-up class for a new driver" to start), and lock the link L=[a,b,c,d,e,f]. a,b,c,d,e,f are the nodes of the link, which are nodes (or steps). They are connected to form a complete link (L). Just like the series of steps "greeting - self-introduction - asking about needs" constitutes a complete script. Upon completion of each step, the engine advances to the next node in the chain and generates a script. Existing technologies begin with a mechanical response, such as "Hello?" to a user's question. This invention begins with a proactive opening, such as saying, "Hello, I'm an advisor from XX platform, here to help you…", directly starting the sales pitch and making it more targeted. If the driver's response causes a deviation, the system immediately re-searches using the already traversed path and the latest response to obtain a new chain L', and continues execution. The session ends upon completion of the chain, successful invitation, or the driver hanging up. Existing technologies operate on a "you ask, I answer" basis, easily leading to digressions. This invention uses a complete script (because it utilizes the most relevant historical invitation process), gradually guiding the conversation and preventing deviations from the invitation topic.

[0040] An invitation method provided by this invention may include: S101, retrieving the historical invitation process with the highest matching degree from a historical invitation dialogue database based on the current dialogue context; S102, constructing a dynamic invitation prompt based on the historical invitation process with the highest matching degree and the current invitation stage; S103, obtaining the target invitation content using an invitation model based on the dynamic invitation prompt and the current dialogue context, and making the invitation based on the target invitation content. This invention identifies a historical invitation process with the highest matching degree, then guides the user along the invitation path within that process. The dynamic invitation prompt accurately locates the current invitation stage of the dialogue. As the dialogue progresses, this "current invitation node" is dynamically updated in the prompt, thereby precisely constraining the next action of the large model, improving the accuracy of responses and increasing the invitation success rate.

[0041] For a clearer understanding of this invention, please refer to the following details. Figure 2 , Figure 2A flowchart illustrating an invitation method provided in an embodiment of the present invention may specifically include: S201 stores complete historical invitation conversations in a structured manner according to "step sequence + full dialogue text + slot + result", forming a Path-Level dialogue library (historical invitation dialogue database).

[0042] S202, during the call, uses the sequence of steps already taken and the most recent conversation as queries in real time to retrieve one or more complete links (historical invitation processes with the highest matching degree).

[0043] In this embodiment, the sequence of invitation steps already performed and the recent dialogue context can be used as search criteria to retrieve information from the historical invitation dialogue database. The retrieval in this embodiment can employ a hybrid "vector + inverted index" recall method, and the n-gram matching weight of the step sequence is ≥0.3.

[0044] S203 concatenates the retrieved links into dynamic invitation prompts in real time and feeds them into a large language model that has not been retrained.

[0045] The dynamic invitation prompts in this embodiment may include the full text of dialogues from historical success cases; a list of completed steps; information on currently filled slots; and explicit rules indicating whether the large language model can skip directly to the next step or must roll back.

[0046] S204, using a large language model to generate the next round of responses under the constraints of the dynamic invitation prompt.

[0047] S205, If the user's behavior deviates from the current link, repeat steps S202-S204 to update the remaining links.

[0048] In this embodiment, deviation monitoring can trigger link re-retrieval in real time through a semantic similarity threshold, and the threshold can be dynamically adjusted according to the business scenario.

[0049] This invention identifies the most relevant historical invitation process from successfully recorded conversations based on the current dialogue. This process is then transformed into a mandatory step-by-step instruction. A large language model generates specific responses under this instruction. Success rate improved: in real-world testing, the success rate for driver invitations increased from 23% to 42%. Deployment is possible solely based on existing human recordings, reducing the iteration cycle from weeks to days. High interpretability: any skipped step / rewind can be traced back to the specific human intervention, facilitating compliance auditing. General and scalable: simply replacing the historical invitation dialogue database allows migration to other invitation and customer service scenarios.

[0050] The invitation device provided in the embodiments of the present invention will be described below. The invitation device described below can be referred to in correspondence with the invitation method described above.

[0051] Please refer to the details. Figure 3 , Figure 3 A schematic diagram of an invitation device provided in an embodiment of the present invention may include: The module 100 for determining the historical invitation process with the highest matching degree is used to retrieve the historical invitation process with the highest matching degree from the historical invitation dialogue database based on the current dialogue context. The dynamic invitation prompt word construction module 200 is used to construct dynamic invitation prompt words based on the historical invitation process with the highest matching degree and the current invitation stage; The invitation module 300 is used to obtain the target invitation content based on the dynamic invitation prompt and the current dialogue context using the invitation model, and to make an invitation based on the target invitation content.

[0052] Furthermore, based on any of the above embodiments, the invitation device further includes: The actual response content determination module is used to obtain the actual response content of the user based on the target invitation content; The semantic similarity determination module is used to determine the semantic similarity between the actual response content and the expected response of the historical invitation process with the highest matching degree in the current invitation stage; The judgment module is used to determine whether the semantic similarity is lower than a preset deviation threshold; The module for re-determining the historical invitation process with the highest matching degree is used to re-execute the step of retrieving the historical invitation process with the highest matching degree when the similarity is lower than the preset deviation threshold.

[0053] Furthermore, based on any of the above embodiments, the historical invitation process determination module 100 with the highest matching degree may include: The unit for determining the historical invitation process with the highest matching degree is used to search the historical invitation process in the historical invitation dialogue database based on the current dialogue context, the completed invitation steps and a preset number of recent invitation dialogue contents, and determine the historical invitation process with the highest matching degree.

[0054] Furthermore, based on any of the above embodiments, the historical invitation process determination unit with the highest matching degree may include: The retrieval vector determination subunit is used to construct a retrieval query based on the current dialogue context, the completed invitation steps, and the preset number of recent invitation dialogue contents, and to encode the retrieval query to obtain a retrieval vector; The first candidate invitation process set determination subunit is used to perform a similarity search in the vector database corresponding to the historical invitation dialogue database based on the retrieval vector, and recall the first candidate invitation process set. The second candidate invitation process set determination subunit is used to recall the second candidate invitation process set based on the completed invitation steps through an inverted index. The highest matching historical invitation process determination subunit is used to perform a weighted fusion sorting of the invitation processes in the first candidate invitation process set and the second candidate invitation process set, and determine the invitation process ranked first as the highest matching historical invitation process.

[0055] Furthermore, based on any of the above embodiments, the dynamic invitation prompt word construction module 200 may include: The instruction template acquisition unit is used to acquire a predefined instruction template, which includes variable placeholders for filling in invitation dialogue information. The dynamic invitation prompt word determination unit is used to fill the step sequence, information slots and dialogue text of the historical invitation process with the highest matching degree into the corresponding variable placeholders in the instruction template to obtain the dynamic invitation prompt word.

[0056] Furthermore, based on any of the above embodiments, the invitation device may further include: The module for determining the set of invitation text dialogues is used to convert invitation voice into a set of invitation text dialogues through automatic speech recognition technology; The target structured path determination module is used to structure each invitation text dialogue in the invitation text dialogue set into a target structured path; wherein, the target structured path includes the sequence of execution steps, the set key information slots, the complete invitation dialogue text, and the final result of the invitation session; The inverted index determination module is used to store all target structured paths into the historical invitation dialogue database and to build an inverted index for the sequence of executed steps; wherein, the inverted index is an index that includes the correspondence between the sequence of executed steps and the structured path.

[0057] It should be noted that the order of the modules and units in the aforementioned invitation device can be changed without affecting the logic.

[0058] An invitation device provided in this invention may include: a historical invitation process determination module 100, used to retrieve the historical invitation process with the highest matching degree from a historical invitation dialogue database based on the current dialogue context; a dynamic invitation prompt word construction module 200, used to construct a dynamic invitation prompt word based on the historical invitation process with the highest matching degree and the current invitation stage; and an invitation module 300, used to obtain the target invitation content using an invitation model based on the dynamic invitation prompt word and the current dialogue context, and to make an invitation based on the target invitation content. This invention determines a historical invitation process with the highest matching degree, then guides the user along the invitation path within that process. The dynamic invitation prompt word accurately locates the current invitation stage of the dialogue. As the dialogue progresses, this "current invitation node" is dynamically updated in the prompt word, thereby precisely constraining the next action of the large model, improving the accuracy of responses and increasing the invitation success rate.

[0059] The following describes an invitation device provided by an embodiment of the present invention. The invitation device described below can be referred to in correspondence with the invitation method described above.

[0060] Please refer to Figure 4 , Figure 4 A schematic diagram of an invitation device provided in an embodiment of the present invention may include: Memory 10 is used to store computer programs; Processor 20 is used to execute computer programs to implement the invitation method described above.

[0061] The memory 10, processor 20, and communication interface 30 all communicate with each other through the communication bus 40.

[0062] 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: Based on the current dialogue context, retrieve the historical invitation process with the highest matching degree from the historical invitation dialogue database; Based on the historical invitation process with the highest matching degree and the current invitation stage, dynamic invitation prompts are constructed; Based on dynamic invitation prompts and the current dialogue context, the invitation model is used to obtain the target invitation content, and the invitation is made based on the target invitation content.

[0063] 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.

[0064] 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.

[0065] 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.

[0066] The communication interface 30 can be an interface for the communication module, used to connect with other devices or systems.

[0067] Of course, it should be noted that, Figure 4 The structure shown does not constitute a limitation on the invitation device in the embodiments of the present invention. In practical applications, the invitation device may include more than Figure 4 More or fewer components as shown, or combinations of certain components.

[0068] The following describes the computer-readable storage medium provided in the embodiments of the present invention. The computer-readable storage medium described below can be referred to in correspondence with the invitation method described above.

[0069] 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 invitation method described above.

[0070] 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.

[0071] 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.

[0072] 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.

[0073] 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.

[0074] The foregoing has provided a detailed description of the 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 the 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. An invitation method, characterized in that, include: Based on the current dialogue context, retrieve the historical invitation process with the highest matching degree from the historical invitation dialogue database; Based on the historical invitation process with the highest matching degree and the current invitation stage, dynamic invitation prompts are constructed; Based on the dynamic invitation prompt and the current dialogue context, the invitation model is used to obtain the target invitation content, and an invitation is made based on the target invitation content.

2. The invitation method according to claim 1, characterized in that, After obtaining the target invitation content using the invitation model based on the dynamic invitation prompt and the current dialogue context, and then making an invitation based on the target invitation content, the process further includes: Obtain the actual response content of the user based on the target invitation content; Determine the semantic similarity between the actual response content and the expected response of the historical invitation process with the highest matching degree in the current invitation stage; Determine whether the semantic similarity is lower than a preset deviation threshold; When the similarity is lower than the preset deviation threshold, the step of retrieving the historical invitation process with the highest matching degree is re-executed.

3. The invitation method according to claim 1, characterized in that, Based on the current dialogue context, retrieve the historical invitation process with the highest matching degree from the historical invitation dialogue database, including: Based on the current dialogue context, completed invitation steps, and a preset number of recent invitation dialogue contents, the historical invitation processes in the historical invitation dialogue database are retrieved to determine the historical invitation process with the highest matching degree.

4. The invitation method according to claim 3, characterized in that, Based on the current dialogue context, completed invitation steps, and a preset number of recent invitation dialogues, the historical invitation processes in the historical invitation dialogue database are retrieved to determine the historical invitation process with the highest matching degree, including: A retrieval query is constructed based on the current dialogue context, the completed invitation steps, and the preset number of recent invitation dialogue contents, and the retrieval query is encoded to obtain a retrieval vector; Based on the retrieval vector, a similarity search is performed in the vector database corresponding to the historical invitation dialogue database to recall the first candidate invitation process set; Based on the completed invitation steps, the second candidate invitation process set is recalled through an inverted index; The invitation processes in the first candidate invitation process set and the second candidate invitation process set are merged and weighted, and the invitation process ranked first is determined as the historical invitation process with the highest matching degree.

5. The invitation method according to any one of claims 1 to 4, characterized in that, Based on the historical invitation process with the highest matching degree and the current invitation stage, dynamic invitation prompts are constructed, including: Obtain a predefined instruction template, which contains variable placeholders for filling in invitation dialogue information; The step sequence, information slots, and dialogue text from the historical invitation process with the highest matching degree are filled into the corresponding variable placeholders in the instruction template to obtain the dynamic invitation prompt.

6. The invitation method according to claim 1, characterized in that, Before retrieving the most relevant historical invitation process from the historical invitation dialogue database based on the current dialogue context, the process also includes: The invitation voice is converted into a set of invitation text dialogues using automatic speech recognition technology; Each invitation text dialogue in the set of invitation text dialogues is structured into a target structured path; wherein, the target structured path includes the sequence of execution steps, the set key information slots, the complete invitation dialogue text, and the final result of the invitation session; All target structured paths are stored in the historical invitation dialogue database, and an inverted index is built for the sequence of executed steps; wherein, the inverted index is an index that includes the correspondence between the sequence of executed steps and the structured paths.

7. An invitation device, characterized in that, include: The module for determining the historical invitation process with the highest matching degree is used to retrieve the historical invitation process with the highest matching degree from the historical invitation dialogue database based on the current dialogue context. The dynamic invitation prompt word construction module is used to construct dynamic invitation prompt words based on the historical invitation process with the highest matching degree and the current invitation stage; The invitation module is used to obtain the target invitation content based on the dynamic invitation prompt and the current dialogue context using the invitation model, and to make an invitation based on the target invitation content.

8. The invitation device according to claim 7, characterized in that, Also includes: The actual response content determination module is used to obtain the actual response content of the user based on the target invitation content; The semantic similarity determination module is used to determine the semantic similarity between the actual response content and the expected response of the historical invitation process with the highest matching degree in the current invitation stage; The judgment module is used to determine whether the semantic similarity is lower than a preset deviation threshold; The module for re-determining the historical invitation process with the highest matching degree is used to re-execute the step of retrieving the historical invitation process with the highest matching degree when the similarity is lower than the preset deviation threshold.

9. An invitation device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the steps of the invitation method as described in any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the invitation method as described in any one of claims 1 to 6.

Images