Talk skill generation method and system, product, equipment and storage medium

Abstract

The invention discloses a verbal skill generation method and system, a product, equipment and a storage medium, which break through the limitation that a traditional method only depends on single input by acquiring multi-dimensional input information covering conversation context, user personalized characteristics and business stages, and provide comprehensive data support for dynamic adjustment. A first prompt word is generated based on the information, a plurality of adaptive basic strategies are accurately obtained from a strategy verbal strategy library containing a mapping relation between strategy definition and customer characteristics, a composite strategy fitting a current user and a scene is formed through integration, and mechanical application of a single strategy is avoided; then, a second cue word is generated in combination with a composite strategy and multi-dimensional input information, strategy constraints and personalized requirements are fused, finally, a large language model is input to generate a natural language verbal skill, it is ensured that the verbal skill strictly follows a customized strategy, the unique features and the service stage of a user are responded, the conversation context is matched, and the current situation that replies are all the same is changed; and dynamic adjustment which varies from time to time is realized.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence natural language processing technology, and in particular to a speech generation method, system, product, device and storage medium. Background Technology

[0002] In the field of artificial intelligence natural language processing technology, especially in business dialogue scenarios such as intelligent customer service and intelligent marketing, existing technologies generally adopt retrieval-based static dialogue recommendation systems. These systems build a pre-built static dialogue database and, relying on keyword matching or simple intent recognition technology, retrieve fixed dialogues from the database that are most relevant to the user's current input as responses.

[0003] However, such systems suffer from significant rigidity and insufficient adaptability. Their core operational logic is limited to mechanical information matching, failing to establish a deep understanding mechanism for multi-dimensional information in the dialogue scenario. Specifically, they cannot effectively capture and integrate contextual information, accurately identify users' personalized characteristics, or adapt to the current business stage. This results in system outputs with a fixed format and content, exhibiting a monotonous nature and failing to adjust to the differences in user characteristics, the dynamic changes in the dialogue process, or different stages of business progress. In actual business dialogues, whether facing users with different needs, preferences, and spending power, or at different business stages such as initial contact, objection handling, or final facilitation, the system can only return uniform pre-set responses, failing to achieve dynamic communication that varies from person to person and from time to time. This severely impacts the adaptability and effectiveness of the dialogue, making it difficult to meet the personalized and precise communication needs of business scenarios. Summary of the Invention

[0004] To address the aforementioned issues, this application provides a script generation method, system, product, device, and storage medium.

[0005] The embodiments of this application disclose the following technical solutions:

[0006] The first aspect of this application provides a script generation method, including:

[0007] Obtain multi-dimensional input information, and generate a first prompt word based on the multi-dimensional input information;

[0008] Based on the first prompt word, multiple basic strategies are obtained from the strategy script library, and the multiple basic strategies are integrated to obtain a composite strategy. The strategy script library includes several basic strategies that represent the correspondence between strategy definitions and applicable customer characteristics.

[0009] Based on the composite strategy and the multi-dimensional input information, a second prompt word is generated;

[0010] The second prompt word is input into the large language model to generate natural language speech that conforms to the composite strategy.

[0011] In one possible implementation, the integration of the multiple basic strategies to obtain a composite strategy includes:

[0012] Based on historical objection records, current business stage, and user profile characteristics from multi-dimensional input information, determine the logical priority and combination relationship of each basic strategy;

[0013] Based on the logical priority and combination relationship of each basic strategy, redundant information between the multiple basic strategies is removed, connection information between the multiple basic strategies is added, and a composite strategy is obtained to represent the execution order and logical relationship.

[0014] In one possible implementation, the basic strategies in the strategy dialogue library are structured entries, which at least include strategy encoding, strategy definition, applicable conditions, and customer feature mapping fields. These structured entries are obtained through manual input or by machine learning models from historical successful dialogues.

[0015] In one possible implementation, the multi-dimensional input information includes at least user profiles, current dialogue status, customer insight information, objection summaries, and current business stage information; wherein, the user profile includes price sensitivity, decision-making attributes, and service preference characteristics; the objection summary includes historical objection records and the latest objection content; and the customer insight information includes the user's service usage preferences and demand tendencies.

[0016] In one possible implementation, the method further includes:

[0017] Record the business outcome data after the agent adopts the natural language dialogue;

[0018] The input information, composite strategies, and generated scripts corresponding to dialogues that achieve positive business results are stored as success case triples.

[0019] Based on the success case triples, the large language model is adjusted; and / or, the success case triples are transformed into basic strategies and updated to the strategy discourse library.

[0020] In one possible implementation, the business outcome data includes transaction completion status, objection resolution effectiveness, user satisfaction, and agent adoption status; the adjustment of the large language model based on the success case triples includes:

[0021] The large language model is adjusted by using the triples corresponding to successful business results, which include multi-dimensional input information, composite strategies, and natural language phrasing.

[0022] This application provides a script generation system in embodiment two, including:

[0023] The acquisition unit is used to acquire multi-dimensional input information and generate a first prompt word based on the multi-dimensional input information;

[0024] An integration unit is used to retrieve multiple basic strategies from a strategy script library based on the first prompt word, and integrate the multiple basic strategies to obtain a composite strategy. The strategy script library includes several basic strategies that represent the correspondence between strategy definitions and applicable customer characteristics.

[0025] The first generation unit is used to generate a second prompt word based on the composite strategy and the multi-dimensional input information;

[0026] The second generation unit is used to input the second prompt word into the large language model and generate natural language speech that conforms to the composite strategy.

[0027] A third aspect of this application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, it implements the speech generation method as described in the first aspect above.

[0028] A fourth aspect of this application provides a computer program product, which, when run on a computer, executes the speech generation method described in the first aspect above.

[0029] A fifth aspect of this application provides a computer-readable storage medium storing instructions that, when executed on a terminal device, cause the terminal device to perform the speech generation method described in the first aspect above.

[0030] Compared with the prior art, this application has the following beneficial effects:

[0031] First, multi-dimensional input information is acquired, breaking the limitations of relying solely on single keywords for information acquisition. This provides a comprehensive basis for subsequent dynamic adjustments. Then, based on this multi-dimensional information, a first prompt word is generated. Multiple suitable basic strategies are precisely retrieved from a strategy script library containing strategy definitions and customer characteristic mapping relationships. By integrating these strategies, a composite strategy that fits the current user state and business scenario is formed, avoiding the mechanical application of a single fixed strategy. Subsequently, a second prompt word is generated by combining the composite strategy with multi-dimensional input information, deeply integrating strategy constraints and personalized needs into the prompt word. Finally, the input is fed into a large language model to generate natural language scripts, ensuring that the scripts strictly follow the composite strategy adapted to user characteristics and business stage, while also responding to the context of the dialogue. This completely eliminates the dilemma of monotonous responses and achieves dynamic adjustments that vary from person to person and time to time. Attached Figure Description

[0032] To more clearly illustrate the technical solutions in the embodiments of this application 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 some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0033] Figure 1 A flowchart illustrating a script generation method provided in this application embodiment;

[0034] Figure 2 This is a structural diagram of a script generation system provided in an embodiment of this application. Detailed Implementation

[0035] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0036] To facilitate understanding of the technical solutions provided in the embodiments of this application, the background technology involved in the embodiments of this application will be described below.

[0037] As mentioned earlier, the system can currently only perform fixed matching from a pre-stored static script library, resulting in three serious shortcomings: First, it is rigid and lacks adaptability. The system cannot understand the dialogue context, user personal characteristics, and business stage, leading to monotonous responses and an inability to dynamically adjust to different people and times. Second, it lacks strategic thinking and cannot combine responses. The pre-stored scripts are isolated points, and when users raise multiple or complex questions, the system cannot integrate different strategies to generate a unified and comprehensive response. Third, it suffers from maintenance bottlenecks and a lack of creativity. The script library relies entirely on manual operation and updates, which is costly, slow, and the system itself lacks creative capabilities, resulting in mechanical and unconvincing responses.

[0038] To address the aforementioned issues, this application's embodiments comprehensively analyze multi-dimensional input information to extract abstract communication strategy instructions. Based on these instructions, specific natural language responses are then generated. These two steps are both independent and closely linked, ensuring precise matching between strategy planning and script expression. In the strategy formation stage, instead of simply retrieving a single strategy from the knowledge base, it deeply integrates complex user states, including historical objections, current sales stage, and user profiles. Multiple basic strategies are precisely retrieved from the knowledge base and creatively combined to form a new, highly customized composite strategy that perfectly meets user needs, completely overcoming the limitations of traditional simple retrieval. In the script generation stage, the established strategy plan is strictly followed. Prompt words clearly define the generation boundaries, ensuring the generation process always revolves around the strategy requirements, avoiding deviations caused by improvisation, and ensuring the final script complies with business compliance standards, closely adheres to core objectives, and maintains a high degree of consistency with the strategy. To support the implementation of the above process, this application adopts a structured strategy script knowledge base design. Each piece of information in the knowledge base not only contains the final script template, but also covers key contents such as strategy coding, strategy definition, applicable conditions and customer feature mapping, so that this information can be effectively identified, understood and flexibly combined, providing a solid foundation for the formation of accurate strategies and the generation of personalized scripts.

[0039] It should be noted that the script generation method, system, product, device, and medium provided in this application can be applied to the field of artificial intelligence natural language processing technology, particularly involving the cross-disciplinary fields of large language models, intelligent customer service, intelligent marketing, and human-computer dialogue systems. Specifically, this application is applied to business dialogue scenarios such as intelligent customer service and telemarketing, realizing the automatic generation of personalized scripts driven by strategies. The above are merely examples and do not limit the application areas of the script generation method, system, product, device, and medium provided in this application. Furthermore, the embodiments of this application may not limit the executing entity of script generation. For example, the script generation method in the embodiments of this application can be applied to data processing devices such as terminal devices or servers. The terminal device can be an electronic device such as a computer or a personal digital assistant (PDA). The server can be a standalone server, a cloud server, or a cluster server composed of multiple servers.

[0040] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0041] The following embodiment illustrates a script generation method provided in this application. See also... Figure 1 ,Should Figure 1 A flowchart of a script generation method provided in this application embodiment, the method including:

[0042] S101. Obtain multi-dimensional input information and generate a first prompt word based on the multi-dimensional input information.

[0043] The multi-dimensional input information includes at least user profiles, current dialogue status, customer insight information, objection summaries, and current business stage information.

[0044] User profile information is extracted by connecting to the customer management system and business transaction records, covering core characteristics such as name, current sales stage, and price sensitivity. The current sales stage clearly defines the core objective of the conversation as closing the deal, directly guiding the strategy to focus on achieving the transaction; price sensitivity characteristics determine that the strategy should emphasize value presentation to alleviate the user's price concerns; and name information provides the foundation for personalized expression in subsequent communication.

[0045] Customer insights are derived from historical dialogue records, user service usage patterns, and business interaction data mining, including service preferences and decision-making attributes. Service preferences clearly identify users' interests in value-added services such as car washes and chauffeur services, providing specific direction for strategy matching; the attribute of independent decision-makers indicates that communication should emphasize core value and autonomy, avoiding over-guidance.

[0046] The objection summary information is generated by capturing the user's current expression through the real-time dialogue monitoring module, and then performing intent recognition and objection classification using natural language processing technology. It includes the objection type and the original expression. This information directly identifies the core conflict in the current dialogue: the user's resistance to the WeChat communication channel and confusion about the contact person, thus clearly defining the scope for strategic recall.

[0047] For example, the user profile could be name (Miao xx), current sales stage (tail-end closing), and price sensitivity (yes). Customer insights could include preferences for services such as car washes and chauffeur services; and the user being an independent decision-maker. The objection summary could be the latest objection being "communication channel objection," with the user's original statement: "It's inconvenient. There are too many WeChat messages, and they're all spam; I can't see them at all. I don't know who you are; there are already more than a dozen on Pacific Channel."

[0048] In one possible implementation, after all the original input information is collected, the system performs standardization processing, classifies and integrates the scattered data according to dimensions such as profile insights and objections, extracts key information and removes invalid content, forming a standardized dataset that can be parsed and correlated by machines, laying the data foundation for the generation of the first prompt word.

[0049] The first prompt word is generated based on the preprocessed multi-dimensional input information. The core purpose is to convey clear task instructions, complete user status and business scenario constraints to the strategy generation engine, guide the engine to accurately recall the matching basic strategy from the structured strategy script library, and reserve adjustment space for subsequent strategy integration.

[0050] In one possible implementation, prompt word construction revolves around five core modules. First, the role is clearly defined: the engine is positioned as a senior insurance sales strategy expert, focusing on handling customer objections in the final closing stage, ensuring the engine matches business scenario needs from a professional perspective. Second, a scenario description is provided, integrating the user's current final closing stage and the objection scenario of the communication channel, allowing the engine to clearly grasp the dialogue context. Next, core user characteristics are extracted, presenting key attributes such as price sensitivity, independent decision-making, and service preferences, clarifying the core basis for strategy adaptation. Then, the core demands are clarified, accurately conveying the core contradiction of user resistance to WeChat communication and confusion regarding the contact person, helping the engine focus on addressing key issues. Finally, task instructions and output requirements are set. The instruction engine, combining user characteristics and the current scenario, recalls suitable basic strategies from the strategy script library and can make reasonable fine-tuning based on user characteristics; simultaneously, the output content is standardized to include strategy code, strategy definition, and adaptation instructions, clarifying the matching logic between the strategy and user characteristics and business scenario, facilitating the efficient integration of subsequent composite strategies.

[0051] In one possible implementation, generating a first prompt word based on the multi-dimensional input information includes:

[0052] First, the preprocessed multi-dimensional input information is hierarchically integrated, logically organized into business scenario dimension, user status dimension, core objection dimension, and feature adaptation dimension. The business scenario dimension focuses on the transaction-oriented goal of facilitating the end-user experience; the user status dimension integrates the identity of independent decision-makers and price sensitivity attributes; the core objection dimension extracts the core contradictions of objections from communication channels; and the feature adaptation dimension relates to service preferences such as car wash and chauffeur services. The integrated standardized information is then transformed into a text format conforming to the semantic understanding habits of a large language model and input into the strategy retrieval module of a large language model (such as qwen2.5-72b) with strong logical reasoning capabilities. This module has a dual core function: firstly, based on the field matching relationship between the input information and the structured strategy discourse library, it accurately retrieves the appropriate basic strategy; secondly, relying on the contextual association and logical combination capabilities of the large language model, it provides semantic parsing and reasoning support for the generation of subsequent composite strategies, ensuring that the strategy combination both conforms to rule constraints and meets the personalized needs of users. The key reason for choosing this type of large language model is that it can deeply understand the multi-dimensional constraints in complex business scenarios, accurately capture the adaptation logic of strategies and user characteristics, and meet the flexible reasoning capabilities required for the generation of composite strategies.

[0053] The core prompts must comprehensively cover the task boundaries, the user profile, operational instructions, and output specifications to ensure that the strategy generation engine has clear objectives and constraints. - Example of a core prompt: "You are a senior sales expert. The current user is in the 'tail-end closing' stage and has raised a 'communication channel objection': [User's original statement]. This user is price-sensitive. Please recall the most suitable response strategy from the strategy library, and you can fine-tune it according to its characteristics. Output the strategy code and a brief description."

[0054] S102. Based on the first prompt word, obtain multiple basic strategies from the strategy script library, and integrate the multiple basic strategies to obtain a composite strategy.

[0055] The strategy script library includes several basic strategies that represent the correspondence between strategy definitions and applicable customer characteristics, ensuring the professionalism, adaptability, and efficiency of strategy planning and script generation. This strategy script library contains several basic strategies, each clearly representing the correspondence between the strategy definition and applicable customer characteristics, achieving a precise link between strategies and user needs.

[0056] Each basic strategy is stored in the form of a structured entry. The core fields of the entry cover the strategy code, strategy definition, applicable conditions, and customer feature mapping. Regardless of whether the initial construction method is manual input or generated by machine learning models from historical successful dialogues, a unified structured format is maintained to ensure that the system can efficiently retrieve, analyze, and combine each basic strategy.

[0057] Taking the scenario of addressing objections in communication channels as an example, the basic strategies pre-set in the strategy script library include: Strategy A is a value decomposition strategy related to price objections. Its strategy is defined as allocating the price to a daily basis and associating it with the value of the core guarantee. The corresponding applicable customer characteristics are price-sensitive users, especially suitable for users in the final sales closing stage who have price concerns and focus on the practicality of the guarantee; Strategy B is a scenario-based description strategy related to rights objections. Its strategy is defined as combining abstract rights with users' specific life scenarios to stimulate demand. The corresponding applicable customer characteristics are users with clear service preferences who value practical use value, especially suitable for independent decision-makers and users who need value-added services such as car wash and chauffeur services.

[0058] These basic strategies do not exist in isolation. Their structured design and clear correspondence with customer characteristics enable the strategy generation engine to accurately recall multiple suitable basic strategies based on user profiles, customer insights, and business stages from multi-dimensional input information. This provides a solid foundation for the creative integration of subsequent composite strategies and effectively supports the generation of personalized and strategic communication.

[0059] In one possible implementation, the integration of the multiple strategies to obtain a composite strategy includes:

[0060] Based on historical objection records, current business stage, and user profile characteristics from multi-dimensional input information, and combined with the applicable conditions and customer feature mapping rules corresponding to each basic strategy in the structured strategy script library, the logical priority and combination relationship of each basic strategy are determined. Among them, the ranking of logical priorities is based on the urgency of the user's core demands, the goal orientation of the current business stage (such as prioritizing strategies that focus on transaction completion in the final closing stage), and the compatibility between user profile characteristics and basic strategies (such as prioritizing value decomposition strategies for price-sensitive users, and prioritizing scenario-based descriptive strategies that highlight the value of autonomous choice for independent decision-makers). The determination of combination relationships needs to be based on the full coverage of the user's multi-dimensional needs, ensuring that multiple basic strategies respectively address the user's explicit objections and implicit demands (such as simultaneously responding to price concerns and service value doubts), forming a complementary and synergistic strategy matrix.

[0061] Based on the logical priority and combination relationship of the various basic strategies, the core content of multiple basic strategies is first semantically analyzed to remove redundant information that is repeated between strategies (such as merging and simplifying common value expressions involved in different basic strategies). Then, according to the natural logic of the communication scenario and the path of business goal advancement, connecting information between strategies is added (such as when transitioning from a value decomposition strategy to a scenario-based description strategy, supplementing with connecting logic such as combining your daily usage needs after clarifying the cost advantage). Finally, a composite strategy with a clear execution order (such as being arranged from high to low priority) and clear logical connections (such as first alleviating core concerns and then strengthening added value, first responding to explicit objections and then exploring potential needs) is formed. This composite strategy needs to fully present the execution order, functional division and connecting logic of each basic strategy through a structured expression to ensure that the subsequent script generation stage can be accurately implemented.

[0062] In one possible implementation, the integration of the multiple strategies to obtain a composite strategy includes:

[0063] Based on historical objection records, current business stage, and user profile characteristics (including price sensitivity, decision-making attributes, and service preferences) from multi-dimensional input information, and combined with the applicable conditions and customer feature mapping rules of each basic strategy in the structured strategy script library, the logical priority and combination relationship of each basic strategy are determined. The logical priority is ranked according to the urgency of the user's core demands and the goal orientation of the business stage. The customer feature mapping rules need to accurately match the applicable conditions of the basic strategies and user profile characteristics to ensure that the priority ranking is in line with user needs and business progress logic.

[0064] For scenarios involving single-dimensional or multi-dimensional objections from users, the basic strategies with the highest logical priority are used as core response strategies, while the remaining basic strategies serve as auxiliary and supplementary strategies, constructing a core-auxiliary combination framework. The core strategy focuses on directly responding to the user's core demands and eliminating communication barriers, while the auxiliary strategies enhance communication effectiveness, supplement value support, and promote the achievement of business goals around the core strategy, ensuring that the strategy combination not only directly addresses key contradictions but also forms a complete response logic chain.

[0065] The execution content of each basic strategy is creatively integrated. Redundant information in repeated expressions between strategies is identified and eliminated through semantic analysis. At the same time, the connection logic between strategies is supplemented according to the natural logic of the communication scenario and the advancement path of business objectives. This allows multiple basic strategies to be organically linked in the order of core to auxiliary, forming a composite strategy with a clear execution order and coherent logical hierarchy, avoiding the awkwardness of strategy splicing.

[0066] Based on user profile characteristics and the adaptability requirements of the current business stage, the execution details of each basic strategy in the composite strategy are fine-tuned; the specific expression of value decomposition is strengthened in combination with user price sensitivity, the independent decision-maker attributes are adapted to highlight the space for autonomous choice, the service preferences are aligned with the contextual presentation of corresponding value-added services, and the strategy focus is optimized in close alignment with the goal orientation of the current business stage, so as to ensure that the composite strategy not only complies with the rule constraints of the structured strategy script library, but also accurately matches the personalized needs of users, and always promotes efficient progress around the business objectives.

[0067] For example, when independent decision-makers who are price-sensitive and prefer chauffeur services at the final closing stage raise objections to communication channels, first determine Strategy A (price objection - value breakdown) as the core strategy (adapting to price sensitivity and responding to implicit core demands), and Strategy B (rights objection - scenario-based description) as the auxiliary strategy (adapting to service preferences and independent decision-making attributes, driving the transaction). Eliminate redundant value statements in the two strategies, supplement with the high cost-effectiveness of core guarantees, and connect with the logic of your daily usage needs; fine-tune Strategy A to highlight the irreplaceable nature of core guarantees such as third-party liability insurance, and fine-tune Strategy B to focus on chauffeur service scenarios such as working overtime late or gathering with friends, ultimately forming a composite strategy that first alleviates price concerns through value breakdown, and then strengthens service value through scenario-based description.

[0068] S103. Based on the composite strategy and the multi-dimensional input information, generate a second prompt word.

[0069] The abstract strategies generated upstream are deeply integrated with specific user information, serving as input for the script generation module. The strategy scheme represents the planned instructions on "what to say and how to say it," such as first decomposing the value proposition, then describing it in a contextualized way. User information includes personalized data such as user profiles, current conversation status, and historical preferences, such as user "Miao Qianlan," price sensitivity, and being in the closing stage. This ensures that the generated scripts not only conform to the strategy logic but also fit the specific user's characteristics and context, achieving personalized expression guided by the strategy.

[0070] Example of core prompt: "Please strictly follow the following strategy to generate a response to user 'Miao Qianlan'. Strategy: [The above-generated strategy scheme]. Requirements: Friendly and professional tone, and tailored to her price sensitivity." The phrase "Please strictly follow the following strategy" serves as a command to ensure the large model adheres to the established strategy, limiting its flexibility and ensuring business compliance and alignment with objectives. "Strategy: [The above-generated strategy scheme]" embeds the composite strategy output from upstream as a core instruction, making the generation process controllable and explainable. "Tailored to user 'Miao Qianlan'...Friendly and professional tone, and tailored to her price sensitivity" incorporates specific user identity and characteristics within the strategy framework, specifying the expression style to make the response more human and targeted.

[0071] In one possible implementation, generating the second prompt word based on the composite strategy and the multi-dimensional input information includes: decomposing the composite strategy into step-by-step execution instructions that can be recognized by a large language model, using the execution order, logical relationship, and core requirements of the composite strategy as core constraints; extracting user-personalized features (including price sensitivity, decision-making attributes, and service preferences), current business stage, dialogue scenario, and core objection demands from the multi-dimensional input information, and converting them into adaptability parameters in the prompt word; and explicitly embedding business compliance requirements, goal-oriented standards, and strategy consistency constraint clauses into the second prompt word. Limit the generation boundaries of the large language model to avoid deviating from business strategies; add tone adaptation instructions, which are dynamically adjusted according to user profile characteristics and dialogue scenarios, including gentle value expressions for price-sensitive users, emphasis on the autonomy of independent decision-makers, and scenario-based description requirements for service-oriented users, while matching the professional standards of business dialogue scenarios; construct a structured prompt word framework, which includes role definitions, defining the large language model as a strategy executor, composite strategy execution instructions, user adaptation parameters, constraint clauses, and tone requirements, to ensure that the large language model generates natural language speech that meets the requirements according to clear logic.

[0072] S104. Input the second prompt word into the large language model to generate natural language speech that conforms to the composite strategy.

[0073] By leveraging the natural language generation capabilities of large language models, composite strategies are transformed into specific persuasive language that aligns with users' personalized needs, meets business objectives, and possesses strong communication appeal. This achieves precise conversion of strategic intent into concrete expression, completely resolving the core pain points of existing technologies, such as mechanical language, lack of specificity, and logical coherence.

[0074] For example, the natural language dialogue could be: "Ms. Miao, I completely understand your price considerations. Let's do the math. This comprehensive coverage actually costs less than two yuan per day, covering the biggest risks on the road. For example, if you accidentally bump into a luxury car, the high-value third-party liability insurance can prevent you from suffering huge financial losses. This is truly worry-free. (Complete Strategy A here) In addition, you mentioned the service. For ladies like you who often drive independently, our chauffeur service is particularly useful. Whether you're working late or having a small gathering with friends, you always have a safe guarantee. This peace of mind is invaluable. (Complete Strategy B here)."

[0075] This generated sales pitch is a perfect instance of composite strategy C (strategy A + strategy B): Strategy A (value decomposition) transforms the abstract premium price into concrete, perceptible daily value and risk hedging through expressions such as "less than two yuan a day" and "avoiding huge economic losses," directly addressing the user's "price sensitivity." Strategy B (scenario-based description) embeds the "designated driver service" benefit into the user's actual life through highly scenario-based and personalized descriptions such as "ladies like you who often drive independently," "working late," and "gathering with friends," evoking emotional resonance and awareness of the need. The speech uses "In addition, you mentioned services" for a natural transition, demonstrating the organic logical connection between multiple strategy points and forming a coherent and hierarchical persuasive whole.

[0076] In one possible implementation, a feedback learning step is also included: recording business outcome data after the agent adopts the natural language dialogue; storing the input information, composite strategy, and generated dialogue corresponding to the dialogue that achieves a positive business outcome as success case triples; performing supervised fine-tuning on the large language model used for strategy integration and / or dialogue generation based on the success case triples; and / or converting the success case triples into new basic strategy entries and updating them to the strategy dialogue library.

[0077] In one possible implementation, the business outcome data includes transaction completion status, objection resolution effectiveness, and user satisfaction rating; the supervised fine-tuning of the large language model based on the success case triples includes: using the success case triples corresponding to positive business outcomes as positive samples to fine-tune the large language model; and, optionally, using the triples corresponding to negative business outcomes as negative samples to participate in the fine-tuning process.

[0078] In one possible implementation, before converting the success case triplet into a new basic strategy entry and updating it to the strategy script library, the method further includes: desensitizing the user input information in the success case triplet; and manually or automatically reviewing the compliance and validity of the new strategy definition and applicable conditions extracted from the triplet.

[0079] This means continuously optimizing model performance and the quality of the strategy script library through feedback from business practices, ensuring that the practicality and adaptability of the methods are constantly improved with the use cases.

[0080] First, the business results data after the agents adopted the natural language script are recorded. This recording process is real-time and comprehensive, covering the complete business feedback after the script is used, providing real and effective data support for subsequent optimization and preventing the optimization direction from deviating from the actual business needs.

[0081] Next, the input information, combined strategies, and generated scripts corresponding to dialogues that achieve positive business results are stored as success case triplets. The input information encompasses comprehensive data including multi-dimensional user characteristics, business stages, objections, and background; the combined strategies include core content such as logical priority relationships and execution points; and the generated scripts are the specific expression of the strategy's implementation. These three elements together form a complete success case triplet that preserves the entire successful process from user needs to strategy formulation and script implementation, providing structured, high-quality material for subsequent reuse and learning.

[0082] Based on the aforementioned success case triples, the large language model is adjusted. By injecting high-quality successful experiences into the model, it continuously learns successful strategy planning logic and conversational expression styles, gradually improving the accuracy and adaptability of strategy generation. Alternatively, the success case triples are transformed into basic strategies and updated to the strategy and conversational library. Key information such as core response logic, applicable conditions, and customer characteristic mapping relationships from the successful cases are extracted to form standardized basic strategy items. This enriches the coverage scenarios and adaptation dimensions of the strategy and conversational library, allowing for more high-quality basic strategies to be called upon in subsequent strategy generation.

[0083] The transaction completion status directly reflects the effectiveness of the sales pitch in driving business objectives and is a core indicator for judging whether business results are positive; the objection resolution effect reflects the sales pitch's ability to respond to users' core demands, which is determined by observing whether users no longer insist on the original objection or generate new related objections; user satisfaction provides feedback from the user's subjective experience perspective, reflecting the sales pitch's communicative affinity and persuasiveness; agent adoption rate reflects the sales pitch's practicality and ease of operation, with a high agent adoption rate indicating that the sales pitch fits the needs of actual work scenarios.

[0084] The adjustment of the large language model based on success case triples involves using triples corresponding to successful business results—containing multi-dimensional input information, composite strategies, and natural language phrases—as positive samples and inputting them into the optimization training phase of the large language model. Through continuous feeding of positive samples, the model deeply learns the adaptation logic between multi-dimensional input information and composite strategies, as well as the conversion rules between composite strategies and natural language phrases. This gradually optimizes the model's strategy generation algorithm and phrase expression logic, making the composite strategies generated by the model more aligned with user needs and business goals, and the generated natural language phrases more personalized, targeted, and effective in communication, thus achieving continuous iterative upgrades to the model's capabilities.

[0085] In summary, this application has the following beneficial effects:

[0086] This technology completely breaks through the core limitation of existing technologies that mechanically retrieve static dialogue scripts, upgrading the system's core capabilities from simple information matching to in-depth strategy planning and creative execution. Existing technologies can only retrieve fixed responses from a pre-set dialogue script library based on user intent, essentially "information transfer"; while this application, through the collaborative operation of strategy generation and dialogue script generation, can analyze the user's multi-dimensional state, plan communication logic, and create suitable dialogue scripts like a senior expert, achieving a leapfrog improvement in the intelligence level of the dialogue system, enabling intelligent dialogue to shift from passive response to proactive decision-making.

[0087] Compared to the rigid and generic responses of existing technologies, this application boasts exceptional dynamic adaptability. The system can capture real-time state changes throughout the conversation, combining detailed user profiles (such as price sensitivity, service preferences, and decision-making attributes) with the current business stage (such as initial contact or closing the deal) to dynamically adjust communication strategies and wording. Whether the user is price-sensitive or service-oriented, whether in the objection handling or sales promotion stage, they can receive targeted communication solutions, truly achieving personalized interaction for each individual and completely resolving the pain point of existing technologies using a single script for everyone.

[0088] Existing technologies, when faced with multiple cross-domain objections from users (such as both excessively high prices and dissatisfaction with service), can only retrieve single-phrase responses and cannot form comprehensive solutions. This application, however, leverages powerful strategy integration capabilities to creatively combine multiple basic strategies, generating logically clear and comprehensive solutions that address multiple concerns. For example, when users simultaneously have resistance to communication channels and price concerns, the system can integrate value decomposition and scenario-based service presentation strategies, alleviating price anxieties while enhancing service value, effectively addressing complex communication scenarios that existing technologies struggle to handle.

[0089] Automation has completed the entire process from user information analysis and strategy planning to script generation, completely eliminating the reliance on manual writing and maintenance of massive amounts of static scripts. Customer service and marketing personnel no longer need to spend a lot of time searching for or brainstorming scripts; they can directly use the personalized and strategic responses generated by the system, significantly improving response speed and work efficiency. At the same time, the generated scripts strictly adhere to business strategies and compliance requirements, balancing logical coherence and persuasiveness. Compared to manually written generic scripts, they offer superior communication quality and conversion rates, achieving simultaneous optimization of efficiency and quality.

[0090] Unlike existing static, read-only script libraries, this system, on the one hand, can distill scattered human experience into standardized strategy assets, transforming them into reusable strategy atoms, thus reducing the manual costs and delays in daily script creation. On the other hand, by recording agents' adoption of scripts and subsequent business results (such as transaction status and objection resolution effectiveness), successful cases are formed into an input-strategy-script triplet. This provides high-quality positive samples for model fine-tuning and can also be added to the strategy script library after anonymization and review. This closed loop of practice feedback, model optimization, and knowledge base iteration allows the system to continuously learn from changes in business scenarios and high-quality communication experiences, with its adaptability and intelligence constantly improving with use.

[0091] The above are some specific implementations of the script generation method provided in the embodiments of this application. Based on this, this application also provides a corresponding script generation system. The system provided in the embodiments of this application will be described below from the perspective of functional modularization. Figure 2 This is a structural diagram of a script generation system provided in an embodiment of this application.

[0092] The system includes:

[0093] The acquisition unit 110 is used to acquire multi-dimensional input information and generate a first prompt word based on the multi-dimensional input information.

[0094] Integration unit 111 is used to obtain multiple basic strategies from the strategy script library based on the first prompt word, and integrate the multiple basic strategies to obtain a composite strategy. The strategy script library includes several basic strategies that represent the correspondence between strategy definitions and applicable customer characteristics.

[0095] The first generation unit 112 is used to generate a second prompt word based on the composite strategy and the multi-dimensional input information;

[0096] The second generation unit 113 is used to input the second prompt word into the large language model to generate natural language speech that conforms to the composite strategy.

[0097] This application also provides corresponding devices and computer storage media for implementing the script generation scheme provided in this application.

[0098] The device includes a memory and a processor. The memory is used to store instructions or code, and the processor is used to execute the instructions or code to cause the device to perform the speech generation method described in any embodiment of this application.

[0099] The computer storage medium stores code, and when the code is run, the device running the code implements the speech generation method described in any embodiment of this application.

[0100] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems or apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple, and relevant parts can be referred to the method section.

[0101] It should be understood that in this application, "at least one" refers to one or more items, and "more" refers to two or more items. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one" or similar expressions refer to any combination of these items, including any combination of singular or plural items. For example, "at least one" of a, b, or c can represent: a, b, c, a and b, a and c, b and c, or a and b and c, where a, b, and c can be single or multiple.

[0102] It should be understood that the terms center, longitudinal, transverse, up, down, front, back, left, right, vertical, horizontal, top, bottom, inside, outside, etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application.

[0103] It should be noted that, unless otherwise explicitly specified and limited, the terms installation, connection, and linking should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections or electrical connections; they can refer to direct connections or indirect connections through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0104] It should also be noted that, in this document, relational terms 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 thereof 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 a process, method, article, or apparatus. Without further limitations, an element defined by the statement "comprising a..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0105] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0106] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for generating dialogue scripts, characterized in that, include: Obtain multi-dimensional input information, and generate a first prompt word based on the multi-dimensional input information; Based on the first prompt word, multiple basic strategies are obtained from the strategy script library, and the multiple basic strategies are integrated to obtain a composite strategy. The strategy script library includes several basic strategies that represent the correspondence between strategy definitions and applicable customer characteristics. Based on the composite strategy and the multi-dimensional input information, a second prompt word is generated; The second prompt word is input into the large language model to generate natural language speech that conforms to the composite strategy.

2. The method according to claim 1, characterized in that, The integration of the multiple basic strategies to obtain a composite strategy includes: Based on historical objection records, current business stage, and user profile characteristics from multi-dimensional input information, determine the logical priority and combination relationship of each basic strategy; Based on the logical priority and combination relationship of each basic strategy, redundant information between the multiple basic strategies is removed, connection information between the multiple basic strategies is added, and a composite strategy is obtained to represent the execution order and logical relationship.

3. The method according to claim 1, characterized in that, The basic strategies in the strategy dialogue library are structured entries. Each structured entry contains at least a strategy code, strategy definition, applicable conditions, and customer feature mapping fields. The structured entries are obtained through manual input or by machine learning models from historical successful dialogues.

4. The method according to claim 1, characterized in that, The multi-dimensional input information includes at least user profiles, current dialogue status, customer insight information, objection summary, and current business stage information; wherein, the user profile includes price sensitivity, decision-making attributes, and service preference characteristics; the objection summary includes historical objection records and the latest objection content; and the customer insight information includes the user's service usage preferences and demand tendencies.

5. The method according to claim 1, characterized in that, The method further includes: Record the business outcome data after the agent adopts the natural language dialogue; The input information, composite strategies, and generated scripts corresponding to dialogues that achieve positive business results are stored as success case triples. Based on the success case triples, the large language model is adjusted; and / or, the success case triples are transformed into basic strategies and updated to the strategy discourse library.

6. The method according to claim 5, characterized in that, The business outcome data includes transaction completion status, objection resolution effectiveness, user satisfaction, and agent adoption rate; the adjustment of the large language model based on the successful case triples includes: The large language model is adjusted by using the triples corresponding to successful business results, which include multi-dimensional input information, composite strategies, and natural language phrasing.

7. A script generation system, characterized in that, include: The acquisition unit is used to acquire multi-dimensional input information and generate a first prompt word based on the multi-dimensional input information; An integration unit is used to retrieve multiple basic strategies from a strategy script library based on the first prompt word, and integrate the multiple basic strategies to obtain a composite strategy. The strategy script library includes several basic strategies that represent the correspondence between strategy definitions and applicable customer characteristics. The first generation unit is used to generate a second prompt word based on the composite strategy and the multi-dimensional input information; The second generation unit is used to input the second prompt word into the large language model and generate natural language speech that conforms to the composite strategy.

8. A computer program product, characterized in that, The computer program product stores instructions that, when executed on a terminal device, cause the terminal device to perform the speech generation method as described in any one of claims 1-6.

9. An electronic device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the speech generation method as described in any one of claims 1-6.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed on a terminal device, cause the terminal device to perform the speech generation method as described in any one of claims 1-6.

Images