An intelligent outbound call method and apparatus

Abstract

The application discloses an intelligent outbound call method and device, and relates to the technical field of computers. A specific implementation of the method comprises the following steps: determining a finite state machine conversation flow template corresponding to an outbound call task; performing the outbound call task based on the finite state machine conversation flow template to make an outbound call to a user; and in response to receiving response information input by the user for the outbound call task, searching a pre-configured conversation data set according to the response information, and generating reply information based on the search result. According to the embodiment, more high-quality conversation data is expanded based on the finite state machine conversation flow template, and a conversation data search enhancement technology is introduced, so that the most relevant conversation data can be searched and utilized in real time in the identification process, thereby improving the identification accuracy and adaptability.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to an intelligent outbound calling method and apparatus. Background Technology

[0002] Current outbound calling technologies all require converting users' voice content into text, then using natural language understanding technology to perform intent recognition on the text content. Based on the intent recognition results, users can choose to respond with pre-configured fixed scripts, or use a generative model to generate a response and then convert it into voice for interaction with the user. In existing outbound calling systems, call quality largely depends on the accuracy of automatic speech recognition technology. It is prone to recognition errors when encountering unfamiliar expressions and has poor adaptability. Summary of the Invention

[0003] In view of this, embodiments of the present invention provide an intelligent outbound calling method and apparatus, which expands more high-quality dialogue data based on a finite state machine dialogue flow template, and also introduces dialogue data retrieval enhancement technology, which can retrieve and utilize the most relevant dialogue data in real time during the recognition process, thereby improving recognition accuracy and adaptability.

[0004] To achieve the above objectives, according to one aspect of the present invention, an intelligent outbound calling method is provided, comprising:

[0005] Determine the finite state machine dialogue flow template corresponding to the outbound call task;

[0006] Execute outbound calling tasks based on a finite state machine dialogue flow template to make outbound calls to users;

[0007] In response to receiving the user's input for the outbound call task, the system retrieves a pre-configured dialogue dataset based on the response information and generates a reply based on the retrieval results.

[0008] Optionally, before determining the finite state machine dialogue flow template corresponding to the outbound call task, the following steps are also included:

[0009] Configure a finite state machine dialogue flow template; the finite state machine dialogue flow template includes the dialogue flow's task information, template identification information, and node information;

[0010] Historical dialogue data is generated based on a finite state machine dialogue flow template and acquired historical task data.

[0011] Historical dialogue data is modified according to a predefined rewriting strategy, a dialogue dataset is generated based on the modified historical dialogue data, and the dialogue dataset is stored in the database.

[0012] Optionally, the finite state machine dialogue flow template includes a persona-generating agent, which generates historical dialogue data based on the finite state machine dialogue flow template and acquired historical task data, including:

[0013] Initialize the persona generation agent based on historical task data, and generate the user persona based on the initialized persona generation agent.

[0014] Construct user response agents and customer service response agents based on user personas and node information in the finite state machine dialogue flow template;

[0015] Historical dialogue data is generated based on user-responding AI agents and customer service-responding AI agents.

[0016] Optionally, the rewriting strategy includes a dialect rewriting strategy, which modifies historical dialogue data according to a predefined rewriting strategy, including:

[0017] The first pre-trained language model rewrites historical dialogue data into dialect versions.

[0018] Optionally, the rewriting strategy includes a typo rewriting strategy, which modifies historical dialogue data according to a predefined rewriting strategy, including:

[0019] The historical dialogue data is segmented and stop words are removed to obtain the segmentation results;

[0020] Based on the similarity of pinyin or the similarity of character shape, obtain a set of candidate words for the word segmentation results;

[0021] The word segmentation results are replaced according to the preset replacement quantity threshold and the candidate word set, and multiple candidate sentences are generated;

[0022] The sentence perplexity of multiple candidate sentences is determined and ranked, and the modified historical dialogue data is determined based on the ranking results and a pre-trained second-largest language model.

[0023] Optionally, a pre-configured dialogue dataset is retrieved based on the response information, including:

[0024] Convert the voice message in the response information into a text message;

[0025] Based on text messages and preset semantic similarity and pinyin similarity strategies, the dialogue dataset is retrieved.

[0026] Optionally, the method further includes:

[0027] In response to receiving a user's rejection message for the outbound call task or reaching the end node in the finite state machine dialogue flow template, a preset callback interface is invoked to return the outbound call result based on the callback interface.

[0028] According to a second aspect of the present invention, an intelligent outbound calling device is provided, comprising:

[0029] The determination module is used to determine the finite state machine dialogue flow template corresponding to the outbound call task;

[0030] The execution module is used to perform outbound calling tasks based on a finite state machine dialogue flow template to make outbound calls to users;

[0031] The response module is used to respond to the response information received from the user for the outbound call task, retrieve the pre-configured dialogue dataset based on the response information, and generate reply information based on the retrieval results.

[0032] According to a third aspect of the present invention, an electronic device is provided, comprising:

[0033] One or more processors;

[0034] Memory, used to store one or more programs.

[0035] When one or more programs are executed by one or more processors, the one or more processors implement the methods of any of the above embodiments.

[0036] According to a fourth aspect of the present invention, a computer-readable medium is provided having a computer program stored thereon, which, when executed by a processor, implements the method of any of the above embodiments.

[0037] According to a fifth aspect of the present invention, a computer program product is provided, including a computer program that, when executed by a processor, implements the method of any of the above embodiments.

[0038] One embodiment of the above invention has the following advantages or beneficial effects: by determining the finite state machine dialogue flow template corresponding to the outbound call task; executing the outbound call task to make an outbound call to the user based on the finite state machine dialogue flow template; in response to receiving the response information input by the user for the outbound call task, retrieving the pre-configured dialogue dataset according to the response information, and generating reply information based on the retrieval results; this embodiment expands more high-quality dialogue data based on the finite state machine dialogue flow template, and also introduces dialogue data retrieval enhancement technology, which can retrieve and utilize the most relevant dialogue data in real time during the recognition process, thereby improving the recognition accuracy and adaptability. Attached Figure Description

[0039] The accompanying drawings are provided to better understand the invention and are not intended to unduly limit the scope of the invention. Wherein:

[0040] Figure 1 This is a schematic diagram of the main flow of the intelligent outbound calling method according to an embodiment of the present invention;

[0041] Figure 2 This is a schematic diagram of the main flow of an intelligent outbound calling method according to a preferred embodiment of the present invention;

[0042] Figure 3 This is a schematic diagram of the main modules of the intelligent outbound calling device according to an embodiment of the present invention;

[0043] Figure 4 This is an exemplary system architecture diagram in which embodiments of the present invention can be applied;

[0044] Figure 5 This is a schematic diagram of the structure of a computer system suitable for implementing terminal devices or servers of the present invention. Detailed Implementation

[0045] The following description, in conjunction with the accompanying drawings, illustrates exemplary embodiments of the present invention, including various details to aid understanding. These details should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the invention. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0046] It should be noted that the acquisition, storage, and application of personal information involved in the embodiments of the present invention comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0047] Current outbound calling technologies all require converting users' voice content into text, then using natural language understanding technology to perform intent recognition on the text content. Based on the intent recognition results, users can choose to respond with pre-configured fixed scripts, or use a generative model to generate a response and then convert it into voice for interaction with the user. In existing outbound calling systems, call quality largely depends on the accuracy of automatic speech recognition technology. It is prone to recognition errors when encountering unfamiliar expressions and has poor adaptability.

[0048] In view of this, according to one aspect of the present invention, an intelligent outbound calling method is provided.

[0049] Figure 1 This is a schematic diagram of the main flow of the intelligent outbound calling method according to an embodiment of the present invention. Figure 1 As shown, the intelligent outbound calling method according to an embodiment of the present invention includes the following steps S101 to S103.

[0050] Step S101: Determine the finite state machine dialogue flow template corresponding to the outbound call task.

[0051] Outbound calling tasks refer to the system automatically dialing user numbers and engaging in pre-defined voice interactions to achieve specific business objectives, such as customer follow-up, order confirmation, marketing promotion, or information verification. Different types of outbound calling tasks correspond to different dialogue flows to ensure that communication with users conforms to business logic, improving interaction efficiency and user experience. Finite State Machine (FSM) dialogue flow templates are a modeling method for the possible interactive states and state transition rules during outbound calling. An FSM dialogue flow template consists of multiple state nodes, each representing the current dialogue state, such as "initial greeting," "user confirmation," "user rejection," "supplementary information," or "end of dialogue." State transitions depend on user input or system judgment, ensuring the controllability and logic of the entire outbound calling process.

[0052] Determining the finite state machine (FSM) dialogue flow template corresponding to an outbound call task can be based on predefined matching of business rules. Upon receiving an outbound call task request, the appropriate FSM dialogue flow template can be directly matched according to the task type (such as order confirmation, satisfaction survey, collection reminder, etc.). For example, an order confirmation task might use an FSM structure of "greeting → order details verification → user confirmation / rejection → end," while a collection reminder task might include multiple states such as "greeting → bill reminder → payment intention confirmation → payment guidance / delay negotiation → end." Alternatively, it can be based on dynamically configured template generation. The FSM dialogue flow template can be dynamically adjusted according to specific parameters of the outbound call task, such as target user group, dialogue history, and business priority. For example, for high-value customers, the FSM can contain richer branch logic, such as "providing personalized offers → evaluating user response → adjusting recommendation schemes," while a simpler path can be used for ordinary users to improve outbound call efficiency. The above methods can be used in combination. When the business is clear and the process is stable, predefined matching can be used to improve efficiency; when flexible adjustments or optimizations to the dialogue experience are needed, dynamic configuration can be used to adapt to the needs of different scenarios.

[0053] Step S102: Execute outbound call tasks based on the finite state machine dialogue flow template to make outbound calls to users.

[0054] When executing outbound calls, the system automates voice interaction with users based on a selected finite state machine (FSM) dialogue flow template, ensuring the call process aligns with predetermined business objectives and interaction logic. The template defines the dialogue states that may be involved in the entire outbound call process, the different user response methods, and the corresponding state transition rules. This allows the system to dynamically adjust the interaction flow, improving communication efficiency and user experience. Specifically, it can execute according to a preset fixed state machine flow. After initiating the outbound call task, it progresses step-by-step according to the preset state sequence of the FSM. This approach is suitable for scenarios with clear business logic and relatively standardized user responses, such as order confirmation and payment reminders. Due to the fixed flow, it can execute efficiently, reducing unnecessary branch processing and improving outbound call efficiency. Alternatively, it can dynamically adjust based on real-time user feedback. When executing the FSM dialogue flow, it not only proceeds according to the preset flow but also incorporates factors such as user voice input, pause duration, and emotion analysis to adjust the state machine's execution path in real time. This approach is more suitable for scenarios requiring higher flexibility, such as customer care and after-sales follow-up, making the outbound call process more natural and human-centered.

[0055] Step S103: In response to receiving the user's response information for the outbound call task, retrieve the pre-configured dialogue dataset based on the response information, and generate reply information based on the retrieval results.

[0056] During outbound call execution, user response information is processed to generate appropriate reply information. For this purpose, this embodiment pre-configures a dialogue dataset, which contains various possible user responses and their corresponding optimal response schemes. This dataset can be constructed based on historical outbound call interaction data, business rule settings, and through large-scale model generation and optimization, thereby improving the comprehensiveness and effectiveness of the data. The dialogue dataset covers general dialogue patterns as well as industry-specific interaction logic, such as common user questions in order confirmation scenarios and standard customer service answers.

[0057] Upon receiving a user's response, the dialogue dataset can be searched using semantic matching to find the most relevant reply. For example, using a Bidirectional Generative Embedding (BGE) vector model, the user's response is converted into a semantic vector, and its cosine similarity is calculated with the standard response vectors stored in the dataset to rank and retrieve the most relevant responses. This method effectively handles user inputs with different expressions but similar semantics; for instance, "How do I do this?" and "How should I do this?" can be matched with the same customer service answer. Alternatively, keyword or rule-based matching can be used to segment the user's response and trigger specific replies based on preset business rules or keywords. For example, in a debt collection scenario, if the user's reply contains "paid," a "payment confirmation" reply can be directly invoked without complex semantic calculations, thus speeding up the response.

[0058] After retrieving candidate responses, the final response information is generated based on specific business needs. For simple and clear matching results, such as fixed responses triggered by keywords, the matched text can be used directly as the reply. For multiple candidate responses retrieved through semantic matching, further methods such as contextual analysis, business strategy weighting, or large-scale model-assisted decision-making can be used to select the response most suitable for the current conversation state. For example, when a user expresses a question, the conversation history can be used to determine whether additional explanations are needed, or whether the response can be simplified to improve user comprehension. In addition, in certain scenarios, such as when a user expresses refusal to complete a task, reassuring and guiding responses can be generated according to business needs to salvage the conversation as much as possible and maximize the success rate of outbound calls.

[0059] This invention provides an embodiment that determines a finite state machine dialogue flow template corresponding to an outbound call task; executes the outbound call task based on the finite state machine dialogue flow template to make an outbound call to the user; in response to receiving the user's response information for the outbound call task, retrieves a pre-configured dialogue dataset based on the response information, and generates reply information based on the retrieval results. This embodiment expands more high-quality dialogue data based on the finite state machine dialogue flow template and also introduces dialogue data retrieval enhancement technology, which can retrieve and utilize the most relevant dialogue data in real time during the recognition process, thereby improving recognition accuracy and adaptability.

[0060] Optionally, before determining the finite state machine dialogue flow template corresponding to the outbound call task, the process further includes: configuring the finite state machine dialogue flow template; the finite state machine dialogue flow template includes the task information, template identification information, and node information of the dialogue flow; generating historical dialogue data based on the finite state machine dialogue flow template and the acquired historical task data; modifying the historical dialogue data according to a predefined rewriting strategy, generating a dialogue dataset based on the modified historical dialogue data, and storing the dialogue dataset in the database.

[0061] This embodiment configures a finite state machine dialogue flow template to enable outbound call tasks to execute according to predetermined dialogue logic. The finite state machine dialogue flow template defines the overall framework of the outbound call task, including task information, template identification information, and node information, etc. Specifically, the task information describes the applicable business scenario for the dialogue flow, such as customer service pre-sales consultation, order confirmation, and collection reminders; the template identification information uniquely identifies the dialogue flow template for accurate matching during outbound call task execution; and the node information defines the critical path of the dialogue, including possible user input types, customer service response strategies, and transition conditions between different dialogue states, etc.

[0062] After configuration, historical dialogue data can be generated based on the finite state machine dialogue flow template and acquired historical task data. Historical task data originates from past outbound call interaction records, including actual user responses, customer service response strategies, and final task completion status. By analyzing this historical data and mapping it to the current dialogue flow template, a representative historical dialogue dataset can be constructed, thereby improving the system's adaptability to real user interactions. To further optimize the quality of historical dialogue data, it can be adjusted according to predefined rewriting strategies. Rewriting strategies can be aimed at optimizing language style, such as adjusting formal expressions to more natural, colloquial versions to improve user experience; or they can be rewritten for data augmentation, such as introducing dialect versions of user responses or generating different variations based on typos to enhance the robustness of the dialogue data. Through these rewriting methods, the generated dialogue data can be made more diverse, thereby improving the outbound call system's adaptability in complex scenarios. Based on the rewritten historical dialogue data, a high-quality dialogue dataset can be further constructed. This dataset not only contains dialogue samples of different task types but can also optimize the data structure and labels according to business needs, making it more suitable for automated processing of outbound call tasks. The processed dialogue dataset is stored in a database so that the outbound calling system can efficiently retrieve and match it during actual task execution.

[0063] Besides building dialogue datasets based on historical task data, large-scale model generation can also supplement dialogue data. For example, based on existing dialogue flow structures, large models can simulate different user roles and engage in dialogue with customer service agents, thereby generating a large amount of dialogue data that conforms to business logic. This approach can compensate for insufficient historical data, especially in new business scenarios, where new dialogue datasets can be quickly built through strategy setting and data-driven approaches.

[0064] Optionally, the finite state machine dialogue flow template includes a persona generation agent, which generates historical dialogue data based on the finite state machine dialogue flow template and acquired historical task data. This includes: initializing the persona generation agent based on the historical task data; generating a user persona based on the initialized persona generation agent; constructing a user response agent and a customer service response agent based on the user persona and node information in the finite state machine dialogue flow template; and generating historical dialogue data based on the user response agent and the customer service response agent.

[0065] The persona generation agent is used to simulate different types of user characteristics, enabling outbound calling tasks to be more accurately tailored to different user groups. The persona generation agent is initialized based on historical task data. The initialization process may include analyzing historical interaction records, extracting user behavior patterns, language styles, preference characteristics, and other information, and storing this information in a structured manner for use in subsequent dialogue data generation. After initialization, the persona generation agent can be used to generate user personas with different characteristics, including dimensions such as age, occupation, region, interests, and purchasing habits. This makes the simulated user responses more consistent with the expression habits of specific groups. For example, for elderly users, a slower, more formal response can be generated, while for younger users, a more conversational and social expression can be used.

[0066] Based on the generated user personas and node information in the finite state machine dialogue flow template, user response agents and customer service response agents can be constructed. The user response agent generates diverse responses that match the specific user identity based on the characteristics of the user persona and the sub-intent information of the task flow nodes, ensuring the authenticity and diversity of user feedback at each node. The customer service response agent matches appropriate dialogue strategies based on the user's answer and intent, generating reasonable customer service responses to ensure the dialogue progresses according to the preset process. The generation of the user response agent can be combined with existing example data and fine-tuned using a large model to make the responses more realistic and as concise and conversational as possible. Based on the user response agent and the customer service response agent, complete historical dialogue data can be constructed, covering the interaction situations of different task types, different user personas, and different dialogue branches. This historical dialogue data can not only be used to train the outbound call system, improving its adaptability in real tasks, but also serve as a knowledge base during outbound call task execution, improving the system's response quality.

[0067] Besides methods that build persona-generating agents based on historical task data, persona templates can also be defined directly based on existing business rules and expert knowledge. For example, when there is insufficient historical data, industry experts can define the characteristics of different types of user groups and directly generate corresponding dialogue data based on these characteristics. This method can quickly build dialogue datasets suitable for specific business scenarios in the initial stage.

[0068] Optionally, the rewriting strategy includes a dialect rewriting strategy, which modifies historical dialogue data according to a predefined rewriting strategy, including: rewriting historical dialogue data into a dialect version based on a pre-trained first language model.

[0069] When rewriting historical dialogue data, a dialect rewriting strategy can be introduced to adapt to the language habits of users in different regions, making the dialogue content of outbound calls more relevant to the target user group. Specifically, based on a pre-trained primary language model, historical dialogue data can be rewritten into dialect versions of the corresponding regions. This primary language model can be fine-tuned through large-scale dialect corpora, enabling it to understand and generate different dialects. During the rewriting process, the model can generate dialogue versions with grammar, vocabulary, and tone that better match the target dialect habits based on standard Mandarin text. For example, the Mandarin "Hello, how can I help you?" can be rewritten as the Cantonese version "Hello, how can I help you?" or the Sichuanese version "Hello, what can I help you?". The dialect rewriting strategy can dynamically adjust the target dialect by combining information such as the user's geographical location, historical interaction records, and device language settings. Furthermore, dialect rewriting is not limited to simple text translation; it can also be based on local cultural habits and expressions to make the rewritten dialogue more natural.

[0070] In addition, a method based on dialect dictionary and rule matching can also be used for rewriting. According to the pre-constructed dialect vocabulary database and sentence pattern conversion rules, by searching and replacing key words or phrases in Mandarin, they are converted into corresponding dialect expressions. For example, through dictionary rules, "chi fan" (having a meal) can be converted into "shi fan" in Cantonese, or "shi yan" in Hokkien. This rule-based method can be combined with statistical learning methods to improve accuracy and achieve more stable rewriting effects in specific fields such as customer service and marketing.

[0071] Optionally, the rewriting strategy includes a typo rewriting strategy, and modifying historical conversation data according to a pre-defined rewriting strategy, comprising: performing word segmentation processing on historical conversation data and removing stop words to obtain word segmentation results; obtaining a candidate word set of the word segmentation results based on Pinyin similarity or glyph similarity; replacing the word segmentation results according to a preset replacement number threshold and the candidate word set, and generating a plurality of candidate sentences; determining sentence perplexity of the plurality of candidate sentences and sorting the candidate sentences, so as to determine the modified historical conversation data based on a sorting result and a pre-trained second large language model.

[0072] When rewriting historical conversation data, a typo rewriting strategy can be introduced to enhance the robustness of the conversation data, simulate possible spelling errors made by users when inputting, and improve the adaptability of the system in real scenarios. Specifically, word segmentation processing is performed on historical conversation data, stop words are removed, and word segmentation results of core content are retained; based on Pinyin similarity or glyph similarity, a candidate word set is generated for each word segmentation result. For example, for the word "yin hang" (bank), "yin hang" (wrong character with same Pinyin) can be generated based on Pinyin similarity, and "yin hang" (wrong character with similar glyph) can be generated based on glyph similarity. After obtaining the candidate word set, according to the preset replacement number threshold, each sentence can replace at most two words, and a plurality of rewritten candidate sentences are generated based on the candidate word set. The sentence perplexity of these candidate sentences is calculated and sorted, and the top five sentences with the lowest perplexity are selected to ensure the readability and rationality of the generated sentences. Based on the pre-trained second large language model, the rationality of the candidate sentences is further evaluated, and the modified historical conversation data is finally determined, so that the rewritten sentences conform to grammatical specifications without changing the original meaning. In addition to the rewriting method based on Pinyin similarity or glyph similarity, a statistical learning method based on a large-scale misspelling corpus can also be used for typo rewriting, which learns common misspelling patterns by analyzing a large amount of user input data and constructs corresponding error-correct mapping relationships.

[0073] Optionally, retrieving a pre-configured dialogue data set according to response information, comprising: converting a voice message in the response information into a text message; retrieving the dialogue data set based on the text message, a preset semantic similarity strategy and a Pinyin similarity strategy.

[0074] When retrieving a pre-configured dialogue dataset, user responses can be standardized by converting speech to text using speech recognition technology. This converted text is then used as input for the search, and a pre-defined semantic similarity strategy and a pinyin similarity strategy are combined to search the dataset. In semantic similarity retrieval, a vector model is used to convert the text message into a semantic vector, and the cosine similarity between this vector and pre-stored dialogue text vectors in the database is calculated. Based on the similarity score, the matching results are sorted, and candidate dialogues with similarities exceeding a set threshold are selected. In pinyin similarity retrieval, candidate text and input text are converted into pinyin sequences, and the length of their longest common subsequence (LCS) is calculated. The ratio of this length to the pinyin length of the input text serves as a similarity metric, selecting dialogue data that meets the threshold. Finally, by combining the retrieval results of semantic and pinyin similarity, the most relevant historical dialogue data is obtained for generating response information. Alternatively, edit distance can be used to calculate text similarity, suitable for scenarios with spelling errors or minor input errors, improving the accuracy and efficiency of the response.

[0075] Optionally, the method further includes: in response to receiving a user's rejection message for the outbound call task or reaching the end node in the finite state machine dialogue flow template, invoking a preset callback interface to return the outbound call result based on the callback interface.

[0076] During the execution of an outbound call task, if the user provides explicit rejection information, or the dialogue flow reaches the end node in the finite state machine dialogue flow template, the outbound call task termination process is triggered. At this time, the outbound call result is returned to the upper-layer business system or relevant modules by calling a preset callback interface for subsequent processing. The returned information from the callback interface can include the call status (such as success, failure, user rejection), key dialogue information (such as the user's final feedback, dialogue path), and the overall execution status of the outbound call task, etc., enabling the execution status of the outbound call task to be tracked in real time and supporting further data analysis or business decisions. Alternatively, a message queue approach can be used to push the outbound call result to a designated business processing queue in the form of an asynchronous message, enabling subsequent execution of corresponding business logic based on this information, such as updating user status, adjusting outbound call strategies, or performing subsequent customer relationship maintenance.

[0077] Figure 2 This is a schematic diagram of the main flow of an intelligent outbound calling method according to a preferred embodiment of the present invention. Figure 2As shown, this embodiment adds a finite state machine dialogue flow template to the outbound call system. The template contains basic task information, template ID, and node information for this dialogue flow. The dialogue flow includes multiple branches and nodes throughout the entire process from the start to the end of the dialogue. Directed edges between nodes indicate that the dialogue state can flow from the originating node to the end node. Each node contains a global task description, a necessary explanation for the current node, a clarification instance, and a selection instance. The clarification intent is used to handle all other intents not within the sub-intent. All dialogue flow nodes share a large model input splicing template, which can be simply represented as: [Task Explanation] + [Sub-intent Description] + [Specific Task Custom Information] + [Example Dialogue Record] + [Sub-intent Instance] + [Clarification Instance] + [Current Dialogue Record] + [User Input in This Round] + [User Input Retrieval Result].

[0078] Based on the dialogue flow, initiate persona-based dialogue data generation. Obtain a batch of untraceable task data (with anonymization requirements) from the business system, and initialize the persona generation agent based on the format of the task-specific custom information in the dialogue flow. For each round of dialogue generation, generate user personas with different task attributes from the persona generation agent. Based on the dialogue flow, construct a user response agent and a customer service response agent. The user response agent generates multiple responses based on the persona and the sub-intent information of the task flow nodes, covering different sub-intents of the current node. The customer service response agent generates corresponding responses based on the user's answer and intent. The user response agent's responses use a given example as a reference, requiring responses to be as concise and conversational as possible. Generation begins from the initial node of the dialogue flow, traversing the nodes of the dialogue flow until reaching the end node corresponding to the dialogue flow branch.

[0079] Based on the dialogue generation results from the above steps, the single-turn generation results are rewritten using data augmentation to address dialects and typos, and then saved to the dialogue database. Specifically, the dialect rewriting scheme involves rewriting the user's response into a dialect version using a large model. The typo rewriting scheme involves segmenting the user's response into words, obtaining a set of candidate words based on similar pinyin or at most two-character changes after removing stop words, and replacing the original words with the selected words. Each sentence can have a maximum of two words replaced at a time. The generated set of selected sentences is sorted by sentence perplexity, and the top five sentences with the lowest perplexity are selected. The large model then judges the rationality of the candidate sentences. The rewritten results from both schemes are stored in the ES database, with the scenario identifiers being the template ID, the status of the current dialogue flow node, the semantic vector of the text, and the pinyin of the text, facilitating subsequent retrieval.

[0080] When the outbound calling system receives an outbound call task request, the system determines the outbound call dialogue flow template based on the outbound call type and creates a unique Session ID to maintain the global dialogue state, including dialogue history, current node, update time, response status, etc.

[0081] When a user's phone rings, the system redirects to the voice interaction module. If the user doesn't answer or hangs up, a callback interface is invoked to return the result. If the user inputs a voice response, the ASR (Automatic Speech Recognition) module converts the voice message into a text message. The response text message is used for mixed retrieval of dialogue data, including semantic similarity and pinyin similarity. Semantic similarity is achieved by vectorizing the text using a BGE vector model to obtain semantic vectors. The semantic vectors are ranked and recalled by calculating the cosine similarity between the semantic vectors and the corresponding text vectors in the database. Pinyin similarity is measured by comparing the longest common subsequence (LCS) of the candidate text's pinyin with the pinyin of the response text. Both recall metrics are evenly distributed across a recall threshold; only responses exceeding the threshold are recalled. The session record is retrieved based on the SessionId. The system determines whether the voice was interrupted based on the current session's response status, update time, and voice event. If interrupted, the current session is merged with the previous session; otherwise, a separate response is given. The large model inputs text, and the large model responds with the intent selection and reply result. The reply result is sent back to the voice interaction module, and the current conversation is updated according to the intent selection result. It is then determined whether the conversation stream has reached its end point. If it has, a hang-up command is issued and the callback interface is called to return the result.

[0082] This embodiment enhances and expands dialogue data based on a large model, utilizing advanced large language model technology to generate more high-quality dialogue data. This enriches and improves the training dataset of the ASR system, enhancing its recognition accuracy and robustness. Since dialects and speech anomalies (such as stuttering, background noise, etc.) negatively impact ASR performance, this embodiment rewrites the dialogue data to better adapt the ASR system to these complexities. A dialogue data retrieval enhancement mechanism is also introduced. By using the ASR recognition results as input and employing a hybrid retrieval method combining Chinese Pinyin character-level and semantic representation, the most relevant dialogue data can be retrieved and utilized in real-time during the recognition process.

[0083] According to a second aspect of the present invention, an intelligent outbound calling device is provided.

[0084] Figure 3 This is a schematic diagram of the main modules of an intelligent outbound calling device according to an embodiment of the present invention. Figure 3 As shown, the intelligent outbound calling device 300 includes:

[0085] Module 301 is used to determine the finite state machine dialogue flow template corresponding to the outbound call task;

[0086] Execution module 302 is used to execute outbound call tasks based on a finite state machine dialogue flow template to make outbound calls to users;

[0087] The response module 303 is used to respond to the response information received from the user for the outbound call task, retrieve the pre-configured dialogue dataset based on the response information, and generate reply information based on the retrieval results.

[0088] Optionally, the intelligent outbound calling device 300 also includes a configuration module, which is used for:

[0089] Configure a finite state machine dialogue flow template; the finite state machine dialogue flow template includes the dialogue flow's task information, template identification information, and node information;

[0090] Historical dialogue data is generated based on a finite state machine dialogue flow template and acquired historical task data.

[0091] Historical dialogue data is modified according to a predefined rewriting strategy, a dialogue dataset is generated based on the modified historical dialogue data, and the dialogue dataset is stored in the database.

[0092] Optionally, the finite state machine dialogue flow template includes a persona-generating intelligent agent, and the configuration module is also used for:

[0093] Initialize the persona generation agent based on historical task data, and generate the user persona based on the initialized persona generation agent.

[0094] Construct user response agents and customer service response agents based on user personas and node information in the finite state machine dialogue flow template;

[0095] Historical dialogue data is generated based on user-responding AI agents and customer service-responding AI agents.

[0096] Optionally, the rewriting strategy includes a dialect rewriting strategy, and the configuration module is also used for:

[0097] The first pre-trained language model rewrites historical dialogue data into dialect versions.

[0098] Optionally, the rewriting strategy includes a typo rewriting strategy, and the configuration module is also used for:

[0099] The historical dialogue data is segmented and stop words are removed to obtain the segmentation results;

[0100] Based on the similarity of pinyin or the similarity of character shape, obtain a set of candidate words for the word segmentation results;

[0101] The word segmentation results are replaced according to a preset replacement quantity threshold and a candidate word set, and multiple candidate sentences are generated.

[0102] The sentence perplexity of multiple candidate sentences is determined and ranked, and the modified historical dialogue data is determined based on the ranking results and a pre-trained second-largest language model.

[0103] Optionally, the response module 303 is also used for:

[0104] Convert the voice message in the response information into a text message;

[0105] Based on text messages and preset semantic similarity and pinyin similarity strategies, the dialogue dataset is retrieved.

[0106] Optionally, the intelligent outbound calling device 300 also includes a calling module, which is used for:

[0107] In response to receiving a user's rejection message for the outbound call task or reaching the end node in the finite state machine dialogue flow template, a preset callback interface is invoked to return the outbound call result based on the callback interface.

[0108] It should be noted that the specific implementation details of the intelligent outbound calling device in this invention have been described in detail in the intelligent outbound calling method above, so the details will not be repeated here.

[0109] According to a third aspect of the present invention, an electronic device is provided, comprising: one or more processors; and a storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method provided in the first aspect of the present invention.

[0110] According to a fourth aspect of the present invention, a computer-readable medium is provided having a computer program stored thereon, which, when executed by a processor, implements the method provided in the first aspect of the present invention.

[0111] According to a fifth aspect of the present invention, a computer program product is provided, including a computer program that, when executed by a processor, implements the method provided in the first aspect of the present invention.

[0112] Figure 4 An exemplary system architecture 400 is shown that can be applied to the intelligent outbound calling method or intelligent outbound calling device of the present invention.

[0113] like Figure 4As shown, system architecture 400 may include terminal devices 401, 402, and 403, a network 404, and a server 405. Network 404 serves as the medium for providing communication links between terminal devices 401, 402, and 403 and server 405. Network 404 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0114] Users can use terminal devices 401, 402, and 403 to interact with server 405 via network 404 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 401, 402, and 403, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social media platform software, etc. (for example only).

[0115] Terminal devices 401, 402, and 403 can be various electronic devices with displays that support web browsing, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0116] Server 405 can be a server that provides various services, such as a backend management server that supports shopping websites browsed by users using terminal devices 401, 402, and 403 (for example only). The backend management server can analyze and process data such as received outbound call task requests, and feed back the processing results (such as outbound call results - for example only) to the terminal devices.

[0117] It should be noted that the intelligent outbound calling method provided in the embodiments of the present invention is generally run by server 405, and correspondingly, the intelligent outbound calling device is generally set in server 405.

[0118] It should be understood that Figure 4 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0119] The following is for reference. Figure 5 It shows a schematic diagram of the structure of a computer system 500 suitable for implementing a terminal device of the present invention. Figure 5 The terminal device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the present invention.

[0120] like Figure 5As shown, the computer system 500 includes a central processing unit (CPU) 501, which can perform various appropriate actions and processes based on programs stored in read-only memory (ROM) 502 or programs loaded from storage section 508 into random access memory (RAM) 503. The RAM 503 also stores various programs and data required for the operation of the system 500. The CPU 501, ROM 502, and RAM 503 are interconnected via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0121] The following components are connected to I / O interface 505: an input section 506 including a keyboard, mouse, etc.; an output section 507 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 508 including a hard disk, etc.; and a communication section 509 including a network interface card such as a LAN card, modem, etc. The communication section 509 performs communication processing via a network such as the Internet. Drive 510 is also connected to I / O interface 505 as needed. Removable media 511, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., are installed on drive 510 as needed so that computer programs read from them can be installed into storage section 508 as needed.

[0122] In particular, according to the embodiments disclosed in this invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 509, and / or installed from removable medium 511. When the computer program is run by the central processing unit (CPU) 501, it performs the functions defined above in the system of this invention.

[0123] It should be noted that the computer-readable medium shown in this invention can be a computer-readable signal medium, a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this invention, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media can also be any computer-readable medium other than computer-readable storage media, which can send, propagate, or transmit programs for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0124] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more operable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually operate substantially in parallel, and they may sometimes operate in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0125] The modules described in the embodiments of the present invention can be implemented in software or hardware. The described modules can also be housed in a processor; for example, a processor can be described as including a determining module, an execution module, and a response module. The names of these modules do not necessarily limit the module itself; for example, the determining module can also be described as "a module for determining the finite state machine dialogue flow template corresponding to an outbound call task."

[0126] In another aspect, the present invention also provides a computer-readable medium, which may be included in the device described in the above embodiments; or it may exist independently and not assembled into the device. The computer-readable medium carries one or more programs that, when executed by the device, cause the device to include: determining a finite state machine dialogue flow template corresponding to an outbound call task; executing the outbound call task to make an outbound call to a user based on the finite state machine dialogue flow template; and, in response to receiving response information input by the user for the outbound call task, retrieving a pre-configured dialogue dataset based on the response information and generating reply information based on the retrieval results.

[0127] The computer program product provided in this embodiment of the invention includes a computer program that, when executed by a processor, implements the intelligent outbound calling method in this embodiment of the invention.

[0128] The technical solution of the present invention has the following advantages or beneficial effects: by determining the finite state machine dialogue flow template corresponding to the outbound call task; executing the outbound call task to make an outbound call to the user based on the finite state machine dialogue flow template; in response to receiving the response information input by the user for the outbound call task, retrieving the pre-configured dialogue dataset according to the response information, and generating reply information based on the retrieval results; this embodiment expands more high-quality dialogue data based on the finite state machine dialogue flow template, and also introduces dialogue data retrieval enhancement technology, which can retrieve and utilize the most relevant dialogue data in real time during the recognition process, thereby improving the recognition accuracy and adaptability.

[0129] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can occur depending on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

[0130] It should be noted that the acquisition, storage, and application of user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

Claims

1. A smart outbound calling method, characterized in that, include: Determine the finite state machine dialogue flow template corresponding to the outbound call task; The outbound call task is executed based on the finite state machine dialogue flow template to make outbound calls to users; In response to receiving the user's response information for the outbound call task, the system retrieves a pre-configured dialogue dataset based on the response information and generates reply information based on the retrieval results.

2. The method according to claim 1, characterized in that, Before determining the finite state machine dialogue flow template corresponding to the outbound call task, the following steps are also included: Configure a finite state machine dialogue flow template; the finite state machine dialogue flow template includes the dialogue flow's task information, template identification information, and node information; Historical dialogue data is generated based on the finite state machine dialogue flow template and the acquired historical task data. The historical dialogue data is modified according to a predefined rewriting strategy, a dialogue dataset is generated based on the modified historical dialogue data, and the dialogue dataset is stored in a database.

3. The method according to claim 2, characterized in that, The finite state machine dialogue flow template includes a persona-generating agent, which generates historical dialogue data based on the finite state machine dialogue flow template and acquired historical task data, including: The persona generation agent is initialized based on the historical task data, and the user persona is generated based on the initialized persona generation agent. Construct a user response agent and a customer service response agent based on the user persona and the node information in the finite state machine dialogue flow template. Historical dialogue data is generated based on the user response AI and the customer service response AI.

4. The method according to claim 2, characterized in that, The rewriting strategy includes a dialect rewriting strategy, which modifies the historical dialogue data according to a predefined rewriting strategy, including: The historical dialogue data is rewritten into dialect versions based on a pre-trained first language model.

5. The method according to claim 2, characterized in that, The rewriting strategy includes a typo rewriting strategy, which modifies the historical dialogue data according to a predefined rewriting strategy, including: The historical dialogue data is segmented and stop words are removed to obtain the segmentation results; Based on the similarity of pinyin or the similarity of character shape, obtain a set of candidate words for the word segmentation results; The word segmentation results are replaced according to a preset replacement quantity threshold and the candidate word set, and multiple candidate sentences are generated. The sentence perplexity of the multiple candidate sentences is determined and ranked, and the modified historical dialogue data is determined based on the ranking results and a pre-trained second large language model.

6. The method according to claim 1, characterized in that, Retrieving a pre-configured dialogue dataset based on the response information includes: Convert the voice message in the response information into a text message; Based on the text messages and preset semantic similarity and pinyin similarity strategies, the dialogue dataset is retrieved.

7. The method according to claim 1, characterized in that, The method further includes: In response to receiving a user's rejection message for the outbound call task or reaching the end node in the finite state machine dialogue flow template, a preset callback interface is invoked to return the outbound call result based on the callback interface.

8. An intelligent outbound calling device, characterized in that, include: The determination module is used to determine the finite state machine dialogue flow template corresponding to the outbound call task; The execution module is used to execute the outbound call task based on the finite state machine dialogue flow template to make outbound calls to users; The response module is used to respond to the response information received from the user for the outbound call task, retrieve a pre-configured dialogue dataset based on the response information, and generate reply information based on the retrieval results.

9. An electronic device, characterized in that, include: One or more processors; Storage device for storing one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1-7.

10. A computer-readable medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1-7.

11. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1-7.

Images