Sales verbal skill generation method, device and system and medium

Abstract

The invention discloses a sales verbal skill generation method, device and system and a medium. The method comprises the following steps: constructing a dialogue model; according to a sales verbal skill generation task, finely adjusting the dialogue model; generating a prompt template according to a sales verbal skill demand; the prompt template is input into the dialogue model after fine tuning for training; and utilizing the trained dialogue model to generate a corresponding sales verbal skill according to the sales scene. According to the technical scheme of the invention, the large language model is trained to generate the personalized sales verbal skill with high pertinence according to different sales scenes and customer demands, the smoothness and naturalness of the sales verbal skill can be improved, and the sales verbal skill can be more vivid according to different sales scenes and customer demands through fine adjustment. Sales talking skills more conforming to the actual situation are generated, so that the sales effect is improved.

Description

Technical Field

[0001] The present invention relates to the field of language processing technology, and in particular to a sales speech generation method, device, system and medium. Background Art

[0002] Traditional sales script generation methods usually rely on the experience and intuition of salespeople. This approach relies on human subjectivity to a certain extent, but there are some problems. First, there are differences between individual salespeople, including differences in experience, eloquence, and adaptability. These differences may lead to inconsistencies in sales scripts, thereby affecting customer experience and sales results. Second, training excellent salespeople requires high costs and the training process takes a long time. Third, excellent sales talents are scarce and difficult to replicate. These factors limit the efficiency and quality of sales script generation.

[0003] With the development of artificial intelligence technology, more and more companies are trying to use natural language processing technology to generate sales scripts. Natural language processing technology can analyze a large amount of sales data, learn excellent sales script patterns, and then generate corresponding sales scripts according to different scenarios and customer needs. This method can improve sales efficiency, reduce labor costs, and achieve rapid replication and promotion of excellent sales scripts.

[0004] However, existing natural language processing methods often lack specificity and personalization when generating sales pitches. Many models can only generate general sales pitches, but cannot be customized according to the specific circumstances and needs of customers. This results in poor results in the generated pitches, which are difficult to really impress customers.

[0005] Therefore, a new sales script generation method is urgently needed to improve sales effectiveness. This method should be able to combine artificial intelligence technology and sales staff's expertise, understand customer needs and preferences through deep learning and big data analysis, and then generate targeted and personalized sales scripts. At the same time, this method should also have the ability to continuously learn and optimize to adapt to the ever-changing market environment and customer needs. Summary of the invention

[0006] In view of the above-mentioned deficiencies in the prior art, the purpose of the present invention is to provide a sales script generation method, device, system and medium that can be applied to language processing or other related fields. Its main purpose is to be able to generate highly targeted and personalized sales scripts according to different sales scenarios and customer needs.

[0007] The technical solution of the present invention is as follows:

[0008] A first aspect of the present invention provides a method for generating sales talk, comprising:

[0009] Build a dialogue model;

[0010] Generating tasks based on sales scripts and fine-tuning the dialogue model;

[0011] Generate prompt templates based on sales script requirements;

[0012] Inputting the prompt template into the fine-tuned dialogue model for training;

[0013] Use the trained dialogue model to generate corresponding sales scripts based on sales scenarios.

[0014] In one embodiment, the step of building a dialogue model includes:

[0015] Get the resources of the model to be trained and set up the running environment of the dialogue model;

[0016] Use the GLM hybrid objective function to pre-train 1.4T Chinese and English identifiers to obtain a pre-trained model;

[0017] The pre-trained model is trained to align human preferences to obtain the dialogue model.

[0018] In one embodiment, the step of generating tasks according to sales talk and fine-tuning the dialogue model includes:

[0019] Inject learnable prompts tokens into each layer of the Transformer architecture, wherein the prompts tokens include tokens set according to the sales script generation task and / or continuous vectors obtained through training;

[0020] Embed the Prompts tokens into the vector space of the dialogue model, and input them together with the original input sequence into each Transformer layer of the dialogue model for calculation, so as to enhance the dialogue model's understanding of the context and semantic information of the input sequence;

[0021] The prompts tokens are used to strengthen the prompt information of the sales script generation task at different depths of the dialogue model, so as to adjust the dialogue model for the prompts of each layer of the Transformer architecture.

[0022] In one embodiment, embedding the Prompts tokens into the vector space of the dialog model and inputting them together with the original input sequence into each Transformer layer of the dialog model for calculation, thereby enhancing the dialog model's understanding of the context and semantic information of the input sequence, includes:

[0023] Embed the Prompts tokens into the vector space of the dialogue model, and combine the Prompts tokens with the original user input sequence to obtain a new user input sequence;

[0024] Inputting the new user input sequence into the Transformer layer of the dialogue model for calculation;

[0025] The Transformer layer extracts key features and context information according to the new user input sequence.

[0026] In one embodiment, generating a prompt template according to sales speech requirements includes:

[0027] Formulate the task of generating sales scripts;

[0028] Creating a specified template according to the generation task;

[0029] Inserting examples or guiding statements into the specified template and using placeholders to represent dynamically changing parts, thereby obtaining a first-generation prompt template;

[0030] The initial prompt template is tested and iteratively optimized to obtain the prompt template.

[0031] In one embodiment, the testing and iterative optimization of the initial prompt template to obtain the prompt template includes:

[0032] Inputting the initial prompt template into the fine-tuned dialogue model;

[0033] Determine whether the output of the fine-tuned dialogue model meets expectations. If not, adjust the initial prompt template to obtain the prompt template.

[0034] In one embodiment, the step of inputting the prompt template into the fine-tuned dialogue model for training includes:

[0035] Inputting the prompt template into the fine-tuned dialogue model, and setting different input modes for the fine-tuned dialogue model;

[0036] The fine-tuned dialogue model outputs expected outputs corresponding to the input patterns according to different input models.

[0037] A second aspect of the present invention provides a sales speech generating device, comprising:

[0038] Get the module to build the dialogue model;

[0039] A first processing module is used to generate tasks according to sales talk and fine-tune the dialogue model;

[0040] The second processing module is used to generate a prompt template according to the sales speech requirements, and input the prompt template into the fine-tuned dialogue model for training;

[0041] The third processing module is used to use the trained dialogue model to generate corresponding sales scripts according to the input sales scenario.

[0042] A third aspect of the present invention provides a sales script generation system, the system comprising at least one processor; and

[0043] a memory communicatively connected to the at least one processor; wherein,

[0044] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the above-mentioned sales script generation method.

[0045] A fourth aspect of the present invention provides a non-volatile computer-readable storage medium, which stores computer-executable instructions. When the computer-executable instructions are executed by one or more processors, the one or more processors can execute the above-mentioned sales script generation method.

[0046] Beneficial effects: The present invention discloses a sales script generation method, device, system and medium. Compared with the prior art, the embodiments of the present invention construct a dialogue model; fine-tune the dialogue model according to the sales script generation task; generate a prompt template according to the sales script demand; input the prompt template into the fine-tuned dialogue model for training; and use the trained dialogue model to generate corresponding sales scripts according to the sales scenario. According to the technical solution of the present invention, by training a large language model, it can generate highly targeted and personalized sales scripts according to different sales scenarios and customer needs, and at the same time, it can also improve the fluency and naturalness of the sales script. Through fine-tuning, it can generate sales scripts that are more in line with the actual situation according to different sales scenarios and customer needs, thereby improving sales results. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The present invention will be further described below with reference to the accompanying drawings and embodiments, in which:

[0048] Figure 1 A flow chart of a method for generating sales speech provided by an embodiment of the present invention;

[0049] Figure 2A flowchart of step S100 in the sales speech generation method provided in an embodiment of the present invention;

[0050] Figure 3 A flowchart of step S200 in the sales speech generation method provided in an embodiment of the present invention;

[0051] Figure 4 A flowchart of step S220 in the sales speech generation method provided in an embodiment of the present invention;

[0052] Figure 5 A flowchart of step S300 in the sales speech generation method provided in an embodiment of the present invention;

[0053] Figure 6 A flowchart of step S340 in the sales speech generation method provided in an embodiment of the present invention;

[0054] Figure 7 A flowchart of step S400 in the sales speech generation method provided in an embodiment of the present invention;

[0055] Figure 8 A schematic diagram of the functional modules of a sales speech generating device provided by an embodiment of the present invention;

[0056] Fig. 9 A schematic diagram of the hardware structure of a sales script generation system provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0057] In order to make the purpose, technical solution and effect of the present invention clearer and more specific, the present invention is further described in detail below. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. The embodiments of the present invention are described below in conjunction with the accompanying drawings.

[0058] Sales script generation is a systematic approach designed to help salespeople or teams develop and optimize scripts and conversation guides for communicating with customers. This process is usually based on in-depth research of the target market and a clear articulation of the value of the product or service. By analyzing customer needs, pain points, and purchasing behaviors, salespeople can create personalized sales scripts to more effectively promote products and convince customers to buy. Sales script generation may include developing opening remarks, questioning strategies, solution presentations, handling objections, and closing deals. Leveraging data and feedback is key to optimizing sales scripts, which requires constant testing and adjustment of scripts to adapt to changing market conditions. Traditional sales script generation methods often rely too much on the personal experience and intuition of salespeople, which not only leads to inconsistent communication methods within the sales team, but also brings high training costs and difficulty in replication. Differences in skills and experience between different salespeople may lead to fluctuations in customer experience, which in turn affects overall sales results and brand image. In the fast-paced modern business environment, this approach has failed to meet the market's requirements for efficiency and personalization.

[0059] With the advancement of artificial intelligence (AI) technology, natural language processing (NLP) has begun to be applied to the generation of sales scripts, with the goal of improving sales efficiency and reducing costs through automation and intelligence. AI can analyze large amounts of sales data, learn successful sales models, and automatically generate scripts. This method is expected to solve the problems caused by relying on manual experience and achieve rapid replication and promotion of sales scripts.

[0060] However, current natural language processing methods often lack sufficient targeting and personalization when generating sales scripts. Since each customer's situation is different, a universal script template may not touch on the customer's specific needs and concerns, and the effect may not be satisfactory. To improve the effectiveness of the script, it is necessary to combine these technologies more deeply with customer data and market analysis to create more accurate and personalized sales scripts.

[0061] In order to solve the above problems, the present invention proposes a sales speech generation method, please refer to Figure 1 , Figure 1A flowchart of an embodiment of the sales script generation method provided by the present invention. The sales script generation method provided in this embodiment is applied to a system consisting of a terminal device, a network and a server, wherein the network is a medium for providing a communication link between the terminal device and the server, which may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.; the operating system on the terminal device may include a handheld device operating system (iPhone operating system, iOS system), Android system or other operating systems, and the terminal device is connected to the server through the network to achieve interaction, thereby performing operations such as receiving or sending data, and specifically may be various electronic devices with a display screen and supporting web browsing, including but not limited to smart phones, tablet computers, portable computers and desktop servers, etc. Figure 1 As shown, the method specifically comprises the following steps:

[0062] S100, building a dialogue model;

[0063] S200, generating tasks according to sales talk and fine-tuning the dialogue model;

[0064] S300, generate prompt templates according to sales talk requirements;

[0065] S400, inputting the prompt template into the fine-tuned dialogue model for training;

[0066] S500: Generate corresponding sales scripts according to sales scenarios using the trained dialogue model.

[0067] In this embodiment, an effective dialogue model is first constructed, such as the ChatGLM2-6b large model. This process involves selecting appropriate machine learning frameworks and algorithms that must be able to understand and generate natural language. Typically, this includes using architectures such as recurrent neural networks or more advanced transformer models. In the early stages of training, the model needs to learn on a wide range of text data to master the basic structure and rules of the language.

[0068] Among them, ChatGLM2-6b is a version of the GLM (General Language Model) model open sourced by Tsinghua University, focusing on the research and application of generative language models. This model has mastered a wealth of language rules and knowledge information through training with a large amount of text data. It can perform a variety of tasks, such as text generation, question-answering systems, machine translation, etc., and demonstrates strong natural language processing capabilities. As a variant of a larger version, ChatGLM2-6b has larger model parameters and stronger data processing capabilities, thus providing a more accurate and smooth user experience.

[0069] Secondly, the dialogue model is fine-tuned according to the sales script generation task. It is understandable that once the basic model is trained, it needs to be further adjusted to adapt to the specific sales environment. Fine-tuning the dialogue model is done by using a specific dataset in the sales field to make the model more familiar with sales terms and common sales scenarios. This step is conducive to connecting the basic model training with sales practice. After fine-tuning the dialogue model, it can ensure that the dialogue model can understand and respond to specific needs in sales.

[0070] After fine-tuning the model, a prompt template can be generated according to the sales script requirements. The prompt template provides a specific input example for the fine-tuned dialogue model, guiding it on how to generate sales scripts that meet actual needs. Each prompt contains necessary information fields, such as product features, target customers, and expected goals, which are the basis for generating personalized and effective scripts. This step ensures the consistency and relevance of the model output with the sales strategy.

[0071] After the prompt template is generated, it is fed into the fine-tuned dialogue model for training. At this stage, the fine-tuned dialogue model will learn how to generate accurate and useful sales pitches based on the provided template. During the training process, the fine-tuned dialogue model continuously optimizes the sentences it generates to ensure that they are both fluent and persuasive. This step is a continuation of the fine-tuning of the dialogue model, further educating and adjusting the dialogue model through the actual application of prompts.

[0072] Finally, the trained dialogue model is used to generate corresponding sales scripts according to the input sales scenario. This step transfers the application of the dialogue model from theory to practice, and formally puts it into real sales activities. The generated scripts need to reflect customer needs and address customer concerns while promoting the sales process. The success of the entire process depends largely on the logical connection and tight integration of each step, ensuring that each link can seamlessly support the next step, thereby improving the quality and effectiveness of the sales script as a whole. In this way, the potential of artificial intelligence can be maximized to achieve automation and personalization of sales scripts.

[0073] Technical effect: By training a large language model, it can generate highly targeted and personalized sales scripts according to different sales scenarios and customer needs, and at the same time improve the fluency and naturalness of sales scripts. Through fine-tuning, it can generate more realistic sales scripts according to different sales scenarios and customer needs, thereby improving sales results. The sales scripts generated by the large language model can not only improve the work efficiency of sales staff, but also significantly improve the customer's purchasing experience. The model can generate answers in real time, helping sales staff to respond quickly in a fast-paced sales environment while maintaining the personalization and contextual relevance of the scripts. In addition, because the model's generation capability is based on extensive and diverse training data, it also shows good adaptability and flexibility when dealing with unknown or rare sales scenarios.

[0074] In one embodiment, Figure 2 As shown, step 100 of constructing a dialogue model includes:

[0075] S110, obtaining resources of the model to be trained and setting the operating environment of the dialogue model;

[0076] S120, pre-training 1.4T Chinese and English identifiers using the GLM hybrid objective function to obtain a pre-trained model;

[0077] S130: Perform human preference alignment training on the pre-trained model to obtain the dialogue model.

[0078] In this embodiment, it is first necessary to obtain the model resources for training the model to be trained and set up a suitable operating environment. This involves preparing high-performance computing devices, such as GPUs or TPUs, and ensuring that there is enough storage space to accommodate a large amount of training data and model parameters. Model resources include model parameter files, dependency libraries, etc. Ensure that the model can be correctly loaded and run in a local or server environment, and prepare for subsequent fine-tuning and application. The configuration of the operating environment includes installing the necessary software libraries and dependencies, and ensuring a smooth data input and output process during training.

[0079] Next, the model to be trained is pre-trained using the GLM (General Language Model) hybrid objective function to obtain the pre-trained model. This step usually involves large-scale data sets, such as 1.4T Chinese and English identifiers, which contain rich language information and can help the model to be trained learn the basic laws and structures of the language. The pre-training process is completed through unsupervised learning, which means that the model to be trained needs to self-learn how to extract useful features from the text without the need for specific label guidance.

[0080] Among them, the GLM hybrid objective function is a technique for training large language models (LLMs), which allows the model to be pre-trained on a large amount of unlabeled text data to improve the understanding and generation of natural language. This pre-training method helps the model learn the common patterns and structures of language, so that it can perform well in various natural language processing tasks.

[0081] In the process of building a dialogue model, pre-training with the GLM hybrid objective function is a key step. For example, the ChatGLM2-6B model is based on the GLM hybrid objective function, pre-trained with 1.4T Chinese and English identifiers, and trained with human preference alignment, which significantly improves performance.

[0082] In addition, the GLM hybrid objective function is not only suitable for the pre-training stage of large language models, but can also be used in combination with other training strategies such as supervised fine-tuning, feedback self-service, human feedback reinforcement learning and other technologies to further enhance the model's dialogue generation capabilities.

[0083] Once the pre-trained model is built, the next step is to train the pre-trained model for human preference alignment to obtain the final dialogue model. This training process ensures that the output generated by the dialogue model is not only grammatically correct, but also in line with human preferences and habits. This is usually achieved through reinforcement learning, in which the dialogue model adjusts its output strategy based on human feedback to generate more natural responses that are more in line with human communication habits.

[0084] The entire training process is an iterative process, with each step building on the previous step to gradually improve the model's performance. This phased training method enables the dialogue model to better adapt to complex language environments and changing communication needs.

[0085] In one embodiment, Figure 3 As shown, the step of generating tasks according to sales talk and fine-tuning the dialogue model in step S200 includes:

[0086] S210, injecting learnable Prompts tokens into each layer of the Transformer architecture, wherein the Prompts tokens include tokens set according to the sales speech generation task and / or continuous vectors obtained through training;

[0087] S220, embedding the Prompts tokens into the vector space of the dialogue model, and inputting them together with the original input sequence into each Transformer layer of the dialogue model for calculation, so as to enhance the dialogue model's understanding of the context and semantic information of the input sequence;

[0088] S230. Utilize the prompts tokens to strengthen the prompt information of the sales script generation task at different depths of the dialogue model, so as to adjust the dialogue model for the prompts of each layer of the Transformer architecture.

[0089] In this embodiment, learnable prompts tokens are injected into each layer of the Transformer architecture to enhance the dialogue model's understanding of the sales script generation task. The prompts tokens include not only tokens customized for sales scripts, but may also include continuous vectors obtained through training, which work together in the model's internal calculations to improve the model's ability to generate sales scripts.

[0090] Among them, the Transformer architecture is a deep learning model that is mainly used to process serialized data, and Prompt tokens are a piece of text used to guide a language model to generate a specific output. In natural language processing (NLP), Prompt (prompt) in the field of artificial intelligence and natural language processing (NLP) is a special text input into the model to guide or instruct the model to perform a specific task or generate a specific type of output. Simply put, Prompt is like a task instruction to AI, telling it what kind of results we need. Prompt tokens can be part of the input to help the model understand the nature of the task and affect the text it generates. For example, in a text generation task, Prompt tokens can be a question or a guiding text, and the model generates corresponding answers or continuations based on these prompts.

[0091] In order to make more effective use of these prompts tokens, they are embedded into the vector space of the dialogue model and fed into each Transformer layer together with the original input sequence. This allows the dialogue model to take into account this additional prompt information when processing the input, thereby better understanding the context and semantic content. This integration makes prompts tokens part of the model's decision-making process, rather than just external guidance.

[0092] In addition, at the bottom layer of the dialogue model, the Prompts tokens are introduced to provide initial sales-related guidance. These tokens serve as additional input signals to guide the dialogue model to focus on the key elements of the sales task. As the level of the dialogue model deepens, the influence of these Prompts tokens gradually increases. This means that at the middle and top layers of the dialogue model, Prompts tokens have a greater impact on the decision-making process of the dialogue model, making the generated words closer to the core goals of the sales strategy.

[0093] The advantage of this approach is that it allows the dialogue model to gradually learn and integrate sales strategies at different levels of abstraction. At the bottom level, the dialogue model learns basic grammatical and semantic information; at the middle level, it starts to consider more complex sales concepts and strategies; and at the top level, the dialogue model is able to fuse this information to produce highly customized and sales-strategic dialogue.

[0094] Ultimately, this approach of reinforcing sales task information at different depths ensures that at the final stage of generation, the conversation model can produce more accurate and effective sales pitches. This is of great significance for improving sales conversion rates and customer satisfaction, and demonstrates the potential of deeply customized models to achieve more refined and goal-driven conversation generation.

[0095] The advantage of this approach is that it allows the dialogue model to maintain language modeling capabilities while enhancing the generation of sales pitches in a targeted manner. In this way, the dialogue model can adapt to specific sales scenarios and needs while learning the general laws of language. This improves the relevance of the dialogue model output and may increase sales conversion rates.

[0096] In one embodiment, Figure 4 As shown, in step S220, the prompts tokens are embedded in the vector space of the dialogue model, and are input together with the original input sequence into each Transformer layer of the dialogue model for calculation, so as to enhance the dialogue model's understanding of the context and semantic information of the input sequence, including:

[0097] S221, embedding the Prompts tokens into the vector space of the dialogue model, and combining the Prompts tokens with the original user input sequence to obtain a new user input sequence;

[0098] S222, inputting the new user input sequence into the Transformer layer of the dialogue model for calculation;

[0099] S223: The Transformer layer extracts key features and context information according to the new user input sequence.

[0100] In this embodiment, the prompts tokens are embedded into the vector space of the dialogue model, which can require the prompts tokens to be converted into a vector form that the dialogue model can understand and process. This process usually involves finding the embedding vectors corresponding to each token, which have fixed positions and meanings in the predefined vector space. After the embedding is completed, these vectors can be combined with the original user input sequence to form a new input sequence, which is rich in additional information.

[0101] The combined input sequence is then fed into the Transformer layer of the dialogue model for calculation. The Transformer layer is the core of the dialogue model and can process the input sequence through the self-attention mechanism to extract key features and context information. The self-attention mechanism enables the dialogue model to consider the information of other tokens related to a token when processing it, so that it can better understand the overall structure and meaning of the language.

[0102] In the Transformer layer, each token is assigned an importance weight, which determines the contribution of each part in generating the final speech. For example, if the user repeatedly mentions a specific question or requirement in the speech, the dialogue model will increase the influence of this part of the content based on the self-attention mechanism, so as to highlight this point in the generated answer.

[0103] This processing method ensures that the dialogue model can fully utilize the guidance information contained in the prompts tokens without losing the important context of the user's original input. In this way, the model can generate personalized sales scripts that are consistent with sales strategies and meet the actual needs of users, thereby improving user experience and sales efficiency.

[0104] In one embodiment, Figure 5 As shown, step S300 generates a prompt template according to sales speech requirements, including:

[0105] S310, formulate the task of generating sales scripts;

[0106] S320, creating a specified template according to the generation task;

[0107] S330, inserting examples or guiding statements into the specified template, and using placeholders to represent dynamically changing parts, to obtain a primary prompt template;

[0108] S340, testing and iteratively optimizing the initial prompt template to obtain the prompt template.

[0109] In this embodiment, the generation task of formulating sales scripts is the first step in creating an effective sales dialogue model. This step involves clearly defining the goals that the model needs to achieve, such as improving customer engagement, increasing product sales, or optimizing customer satisfaction. The formulation of the generation task directly affects the structure and content of the subsequent templates, so it needs to be precisely set according to specific sales goals and customer needs.

[0110] Next, create the specified templates based on the generated tasks. These templates serve as the basic structure for generating sales pitches. They need to contain the necessary information framework so that users can be guided to the expected purchase path in subsequent conversations. The design of the templates should be concise and clear, while being flexible enough to meet the needs of different sales scenarios.

[0111] Inserting examples or guiding statements into the specified template and using the placeholders to represent the dynamically changing parts will result in the first generation of prompt templates. These example statements not only provide a reference for the style and tone of the speech, but also demonstrate how to effectively showcase product features and handle customer concerns. The placeholders are used to replace the corresponding dynamic content, such as customer names, product specifications, or the latest discount information, based on real-time data when actually generating the speech.

[0112] The initial prompt template is tested and iteratively optimized to obtain the final prompt template. This process usually involves multiple rounds of experiments and evaluations to ensure that the template can effectively generate sales scripts that meet expectations in actual applications. Optimization may include adjusting the expression of sentences, improving the location and function of placeholders, or adding new sample sentences to improve the persuasiveness of the scripts.

[0113] The entire process is a systematic process, from task setting to template creation and optimization, each step is designed to improve the effectiveness of sales talk and ensure that it can work in different sales situations. This approach ensures that the sales team can provide consistent and efficient services when interacting with customers, thereby improving overall sales performance.

[0114] In one embodiment, Figure 6 As shown, step S340 tests and iteratively optimizes the initial prompt template to obtain the prompt template, including:

[0115] S341, inputting the initial prompt template into the fine-tuned dialogue model;

[0116] S342: Determine whether the output of the fine-tuned dialogue model meets expectations. If not, adjust the initial prompt template to obtain the prompt template.

[0117] In this embodiment, inputting the initial prompt template into the fine-tuned dialogue model is a key implementation step, which involves applying the template to an actual sales dialogue scenario. In this process, the initial prompt template is provided as input to the dialogue model, and the dialogue model generates corresponding sales scripts based on the initial prompt template. These corresponding sales scripts are a direct reflection of the dialogue model's understanding and application of the initial prompt template, and therefore, the corresponding sales scripts can be used to evaluate the effectiveness of the template.

[0118] Judging whether the output of the fine-tuned dialogue model (generating the corresponding sales script according to the initial prompt template) meets expectations is a key step in ensuring the quality and effectiveness of sales scripts. This judgment process usually requires human participation, through professionals or sales experts to evaluate the authenticity, persuasiveness and applicability of the dialogue. If the output of the model does not meet the established sales goals, or fails to effectively guide customers towards purchase, then the initial prompt template needs to be adjusted.

[0119] Adjustments include rewriting sample sentences to make them more attractive and persuasive; adjusting the position of placeholders to ensure that key information is appropriately highlighted in the sales pitch; or adding new guiding sentences to better guide the sales process. The purpose of these adjustments is to improve the effectiveness of the template so that it can work better with the fine-tuned dialogue model to generate sales pitches that are more in line with actual needs.

[0120] It is understandable that the way to adjust the Prompt template is not limited to rewriting the sample sentences and adjusting the placeholder positions, but also involves a series of more in-depth optimization strategies. These strategies are designed to ensure that the generated sales script is not only attractive and persuasive, but also accurately reflects the core elements of the sales strategy and effectively guides potential customers towards a purchase decision.

[0121] Rewriting sample sentences is a common optimization method that can deeply understand the needs and preferences of target customers and the unique selling points of products or services. Through carefully designed sentences, customers' interests can be stimulated, their doubts can be eliminated, and their desire to buy can be stimulated. These sentences need to be highly targeted and emotionally resonant in order to have the greatest impact in the conversation.

[0122] Adjusting the position of placeholders is also a key step, which determines the timing and manner of presenting key information in the speech. By carefully planning the position of placeholders, you can ensure that key information is highlighted at the appropriate moment in the conversation, thereby enhancing the persuasiveness of the speech. For example, placing key benefit points at the beginning or end of the conversation can better attract the customer's attention and leave a deep impression.

[0123] Adding new instructional phrases is another way to improve the effectiveness of your sales pitch. These phrases can provide clear action guidelines to help salespeople guide customers down the sales funnel. For example, an effective instructional phrase might prompt salespeople to ask customers about their needs or introduce a product’s unique features at the right time.

[0124] In addition to the above methods, you can also consider using more diverse language styles, adding storytelling elements, or including customer testimonials and case studies. These elements can make sales pitches more vivid and persuasive, while providing more evidence to support sales points.

[0125] In short, the purpose of these adjustments is to improve the effectiveness of the Prompt template so that it can work with the fine-tuned dialogue model to generate sales scripts that are more in line with actual needs. Through continuous testing and iteration, the effectiveness of sales scripts in actual applications can be ensured, thereby helping sales staff improve conversion rates and ultimately achieve sales goals.

[0126] After adjusting the initial prompt template, once the prompt template is obtained, it can be input into the dialogue model again for testing to ensure that all adjustments have been implemented correctly and the output of the dialogue model now meets expectations. This process may require multiple iterations until the best template design is found. Through this meticulous adjustment and optimization process, it can be ensured that the sales script is both in line with the brand's voice and can effectively stimulate the potential customer's willingness to buy, thereby improving the efficiency and success rate of the entire sales process.

[0127] In one embodiment, Figure 7 As shown, the step of inputting the prompt template into the fine-tuned dialogue model for training in step S400 includes:

[0128] S410, inputting the prompt template into the fine-tuned dialogue model, and setting different input modes for the fine-tuned dialogue model;

[0129] S420: The fine-tuned dialogue model outputs an expected output corresponding to the input mode according to different input models.

[0130] In this embodiment, the Prompt template is input into the fine-tuned model, and different input modes are set, such as question mode, product recommendation mode, or objection handling mode. In each mode, the style and content of the answers expected to be generated by the model will be different. In the fine-tuned dialogue model, we can set a variety of different input modes so that the model can adapt to different types of dialogue scenarios. These input modes can include question-answer pairs, dialogue fragments, multi-round dialogues, etc. By setting different input modes, we can let the model learn to generate appropriate responses in different situations.

[0131] For example, for the input mode of question-answer pairs, we can provide a question and multiple candidate answers, and then let the dialogue model select the most appropriate answer. In this case, the expected output of the dialogue model should be an optimal answer.

[0132] For the input mode of dialogue fragments, we can provide a fragment containing multiple dialogue turns and let the dialogue model generate the response for the next turn. In this case, the expected output of the dialogue model should be a coherent and reasonable response.

[0133] For the input mode of multi-round dialogue, we can provide a complete dialogue record containing multiple dialogue rounds, and then let the dialogue model generate the response for the next round. In this case, the expected output of the dialogue model should be a response that can continue the dialogue and maintain coherence.

[0134] In this way, we can train the fine-tuned dialogue model to produce expected outputs under different input modes, thereby improving the practicality and adaptability of the model.

[0135] Another embodiment of the present invention provides a sales speech generating device, such as Figure 8 As shown, the device 1 comprises:

[0136] A construction module 11 is used to construct a dialogue model;

[0137] A fine-tuning module 12, used to generate tasks according to sales talk and fine-tune the dialogue model;

[0138] Processing module 13, used to generate prompt template according to sales speech requirements;

[0139] A training module 14, used for inputting the prompt template into the fine-tuned dialogue model for training;

[0140] The generation module 15 is used to generate corresponding sales scripts according to the sales scenario using the trained dialogue model.

[0141] The module referred to in the present invention refers to a series of computer program instruction segments that can perform specific functions, which is more suitable for describing the execution process of the sales script generation method than a program. For the specific implementation method of each module, please refer to the corresponding method embodiment above, which will not be repeated here.

[0142] In one embodiment, the device 1 further comprises:

[0143] An input module, used to inject learnable prompts tokens into each layer of the Transformer architecture, wherein the prompts tokens include tokens set according to the sales script generation task and / or continuous vectors obtained through training;

[0144] A calculation module, used to embed the Prompts tokens into the vector space of the dialogue model, and input them together with the original input sequence into each Transformer layer of the dialogue model for calculation, so as to enhance the dialogue model's understanding of the context and semantic information of the input sequence;

[0145] The adjustment module is used to use the prompts tokens to strengthen the prompt information of the sales speech generation task at different depths of the dialogue model to adjust the prompts of each layer of the Transformer architecture.

[0146] In one embodiment, the device 1 further comprises:

[0147] The formulation module is used to formulate the generation task of sales scripts;

[0148] A creation module, used to create a specified template according to the generation task;

[0149] An insertion module is used to insert examples or guiding statements into the specified template, and use placeholders to represent dynamically changing parts, thereby obtaining an initial prompt template;

[0150] The verification module is used to test and iteratively optimize the initial prompt template to obtain the prompt template.

[0151] In one embodiment, the building module 11 includes:

[0152] An acquisition unit is used to acquire resources of the model to be trained and set the operating environment of the dialogue model;

[0153] A pre-training unit is used to pre-train 1.4T Chinese and English identifiers using the GLM hybrid objective function to obtain a pre-trained model;

[0154] The alignment training unit is used to perform human preference alignment training on the pre-trained model to obtain the dialogue model.

[0155] In one embodiment, the computing module comprises:

[0156] A merging unit, used to embed the Prompts tokens into the vector space of the dialogue model, and combine the Prompts tokens with the original user input sequence to obtain a new user input sequence;

[0157] A computing unit, used for inputting the new user input sequence into the Transformer layer of the dialogue model for computing;

[0158] The extraction unit is used for the Transformer layer to extract key features and context information according to the new user input sequence.

[0159] In one embodiment, the verification module includes:

[0160] An input unit, used for inputting the initial prompt template into the fine-tuned dialogue model;

[0161] The comparison unit is used to determine whether the output of the fine-tuned dialogue model meets expectations. If it does not meet expectations, the initial prompt template is adjusted to obtain the prompt template.

[0162] In one embodiment, the training module 14 includes:

[0163] A setting unit, used for inputting the prompt template into the fine-tuned dialogue model, and setting different input modes for the fine-tuned dialogue model;

[0164] The output unit is used for the fine-tuned dialogue model to output the expected output corresponding to the input mode according to different input models.

[0165] Another embodiment of the present invention provides a sales speech generation system, such as Fig. 9 As shown, the system 10 includes:

[0166] One or more processors 110 and memory 120, Fig. 9A processor 110 is used as an example for description. The processor 110 and the memory 120 may be connected via a bus or other means. Fig. 9 The example of connecting through bus is taken in the following.

[0167] The processor 110 is used to complete various control logics of the system 10, and it can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), a single-chip microcomputer, an ARM (Acorn RISC Machine) or other programmable logic device, a discrete gate or transistor logic, a discrete hardware component, or any combination of these components. In addition, the processor 110 can also be any traditional processor, microprocessor or state machine. The processor 110 can also be implemented as a combination of computing devices, for example, a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors combined with a DSP and / or any other such configuration.

[0168] The memory 120 is a non-volatile computer-readable storage medium that can be used to store non-volatile software programs, non-volatile computer executable programs and modules, such as program instructions corresponding to the sales speech generation method in the embodiment of the present invention. The processor 110 executes various functional applications and data processing of the system 10 by running the non-volatile software programs, instructions and units stored in the memory 120, that is, implements the sales speech generation method in the above method embodiment.

[0169] The memory 120 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and an application required by at least one function; the data storage area may store data created according to the use of the system 10, etc. In addition, the memory 120 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 120 may optionally include a memory remotely arranged relative to the processor 110, and these remote memories may be connected to the system 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0170] One or more units are stored in the memory 120. When executed by one or more processors 110, the sales speech generation method in any of the above method embodiments is executed, for example, the following steps are executed:

[0171] Build a dialogue model;

[0172] Generating tasks based on sales scripts and fine-tuning the dialogue model;

[0173] Generate prompt templates based on sales script requirements;

[0174] Inputting the prompt template into the fine-tuned dialogue model for training;

[0175] Use the trained dialogue model to generate corresponding sales scripts based on the input sales scenarios.

[0176] An embodiment of the present invention provides a non-volatile computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions. When the computer-executable instructions are executed by one or more processors, the sales script generation method in any of the above method embodiments is executed, for example, the following steps are executed:

[0177] Build a dialogue model;

[0178] Generating tasks based on sales scripts and fine-tuning the dialogue model;

[0179] Generate prompt templates based on sales script requirements;

[0180] Inputting the prompt template into the fine-tuned dialogue model for training;

[0181] Use the trained dialogue model to generate corresponding sales scripts based on the input sales scenarios.

[0182] As an example, non-volatile storage media can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) as external cache memory. By way of illustration and not limitation, RAM can be obtained in many forms such as synchronous RAM (SRAM), dynamic RAM, (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), Synchlink DRAM (SLDRAM), and direct Rambus RAM (DRRAM). The disclosed memory components or memories of the operating environment described herein are intended to include one or more of these and / or any other suitable types of memory.

[0183] In summary, the present invention discloses a sales speech generation method, device, system and medium, the method includes: constructing a dialogue model; fine-tuning the dialogue model according to the sales speech generation task; generating a prompt template according to the sales speech demand; inputting the prompt template into the fine-tuned dialogue model for training; using the trained dialogue model to generate corresponding sales speech according to the sales scenario. According to the technical solution of the present invention, by training a large language model, it can generate highly targeted and personalized sales speech according to different sales scenarios and customer needs, and at the same time, it can also improve the fluency and naturalness of the sales speech. Through fine-tuning, it can generate sales speech that is more in line with the actual situation according to different sales scenarios and customer needs, thereby improving sales results.

[0184] Of course, those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing related hardware (such as a processor, a controller, etc.) through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium, and the computer program can include the processes of the above-mentioned method embodiments when executed. The storage medium can be a memory, a disk, a floppy disk, a flash memory, an optical storage device, etc.

[0185] It should be understood that the application of the present invention is not limited to the above examples. For ordinary technicians in this field, improvements or changes can be made based on the above description. All these improvements and changes should fall within the scope of protection of the claims attached to the present invention.

Claims

1. A sales script generation method, characterized in that: include: Build a dialogue model; Generating tasks based on sales scripts and fine-tuning the dialogue model; Generate prompt templates based on sales script requirements; Inputting the prompt template into the fine-tuned dialogue model for training; Use the trained dialogue model to generate corresponding sales scripts based on sales scenarios.

2. The sales script generation method according to claim 1, characterized in that: The step of building the dialogue model includes: Get the resources of the model to be trained and set up the running environment of the dialogue model; Use the GLM hybrid objective function to pre-train 1.4T Chinese and English identifiers to obtain a pre-trained model; The pre-trained model is trained to align human preferences to obtain the dialogue model.

3. The sales script generation method according to claim 1, characterized in that: The step of generating tasks according to sales talk and fine-tuning the dialogue model includes: Inject learnable prompt tokens into each layer of the Transformer architecture, where the prompt tokens include tokens set according to the sales script generation task and / or continuous vectors obtained through training; Embed the Prompts tokens into the vector space of the dialogue model, and input them together with the original input sequence into each Transformer layer of the dialogue model for calculation, so as to enhance the dialogue model's understanding of the context and semantic information of the input sequence; The prompts tokens are used to strengthen the prompt information of the sales script generation task at different depths of the dialogue model to adjust the prompts of each layer of the Transformer architecture.

4. The method for generating sales scripts according to claim 3, characterized in that: The step of embedding the Prompts tokens into the vector space of the dialog model and inputting the Prompts tokens together with the original input sequence into each Transformer layer of the dialog model for calculation, thereby enhancing the dialog model's understanding of the context and semantic information of the input sequence, includes: Embed the Prompts tokens into the vector space of the dialogue model, and combine the Prompts tokens with the original user input sequence to obtain a new user input sequence; Inputting the new user input sequence into the Transformer layer of the dialogue model for calculation; The Transformer layer extracts key features and context information according to the new user input sequence.

5. The sales script generation method according to claim 1, characterized in that: The prompt template is generated according to the sales speech requirements, including: Formulate the task of generating sales scripts; Creating a specified template according to the generation task; Inserting examples or guiding statements into the specified template and using placeholders to represent dynamically changing parts, thereby obtaining a first-generation prompt template; The initial prompt template is tested and iteratively optimized to obtain the prompt template.

6. The method for generating sales scripts according to claim 5, characterized in that: The step of testing and iteratively optimizing the initial prompt template to obtain the prompt template includes: Inputting the initial prompt template into the fine-tuned dialogue model; Determine whether the output of the fine-tuned dialogue model meets expectations. If not, adjust the initial prompt template to obtain the prompt template.

7. The sales script generation method according to claim 1, characterized in that: The step of inputting the prompt template into the fine-tuned dialogue model for training includes: Inputting the prompt template into the fine-tuned dialogue model, and setting different input modes for the fine-tuned dialogue model; The fine-tuned dialogue model outputs expected outputs corresponding to the input patterns according to different input models.

8. A sales speech generating device, characterized in that: include: Get the module to build the dialogue model; A fine-tuning module, used to generate tasks according to sales talk and fine-tune the dialogue model; The processing module is used to generate prompt templates according to sales script requirements; A training module, used for inputting the prompt template into the fine-tuned dialogue model for training; The generation module is used to use the trained dialogue model to generate corresponding sales scripts according to the input sales scenario.

9. A sales script generation system, characterized in that: The system includes at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the sales script generation method described in any one of claims 1-7.

10. A non-volatile computer-readable storage medium, characterized in that: The non-volatile computer-readable storage medium stores computer-executable instructions. When the computer-executable instructions are executed by one or more processors, the one or more processors can execute the sales script generation method described in any one of claims 1-7.

Images