Product innovation tool recommendation method and system based on pre-training model

Abstract

The invention provides a product innovation tool recommendation method and system based on a pre-training model. The method comprises the steps of obtaining a design requirement standard text for a target product; obtaining a preset innovation tool list, and filling the preset innovation tool list and the design requirement standard text into a preset prompt word template to obtain a prompt project text; reasoning the prompt project text through the pre-training model to obtain a recommendation tool and a recommendation reason corresponding to the design requirement standard text, the recommendation tool being at least one innovation tool in a preset innovation tool list. Potential rules of tool recommendation have been learned in large-scale data training through the pre-training model, and relevance between requirements and tools can be identified more efficiently in combination with normalized processing of the prompt project text on input information, so that recommendation tools and recommendation reasons output by reasoning of the pre-training model are more targeted and logical, and the recommendation efficiency is improved. Therefore, the recommendation accuracy is remarkably improved.

Description

Technical Field

[0001] This application relates to the technical fields of artificial intelligence, large language model applications, and industrial design and engineering. Specifically, it relates to a product innovation tool recommendation method and system based on a pre-trained model. Background Technology

[0002] Currently, innovation tools are needed for product innovation design analysis. However, traditional innovation tool recommendation methods largely rely on similarity matching of historical data. In practice, this reliance on historical data similarity matching is ill-suited to the dynamic design needs of product innovation. For example, if user needs do not perfectly match the keywords in historical case studies, the recommendation tool becomes ineffective. Therefore, the accuracy of current product innovation tool recommendations is relatively low. Summary of the Invention

[0003] The purpose of this application is to provide a product innovation tool recommendation method and system based on a pre-trained model, in order to improve the problem of low accuracy in recommending product innovation tools.

[0004] This application provides a product analysis tool innovation tool recommendation method based on a pre-trained model, comprising: obtaining standard text of design requirements for a target product; obtaining a preset list of innovation tools, and filling the preset list of innovation tools and the standard text of design requirements into a preset prompt word template to obtain prompt engineering text; and using a pre-trained model to reason about the prompt engineering text to obtain recommended tools and reasons for recommendation corresponding to the standard text of design requirements, wherein the recommended tool is at least one innovation tool in the preset list of innovation tools. In the implementation of the above scheme, by combining the standard text of design requirements with the semantic reasoning ability of the pre-trained model, the standard text of design requirements is standardized into the input format of the preset prompt word template, reducing the interference of natural language ambiguity on model understanding. Moreover, the pre-trained model has learned the potential rules of tool recommendation in large-scale data training. Combined with the standardized processing of input information by the prompt engineering text, it can more efficiently identify the correlation between requirements and tools. Therefore, the recommended tools and reasons for recommendation output by the pre-trained model are more targeted and logical, thereby significantly improving the recommendation accuracy.

[0005] Optionally, in this embodiment, filling the preset list of innovative tools and the standard text of design requirements into a preset prompt word template includes: for each innovative tool in the preset list of innovative tools, obtaining the explanatory definition text of the innovative tool from a tool database, the tool database storing the preset list of innovative tools and the explanatory definition of each innovative tool; filtering candidate innovative tools from the preset list of innovative tools whose similarity between the explanatory definition text and the standard text of design requirements is greater than a preset threshold; and filling the candidate innovative tools and the standard text of design requirements into the preset prompt word template. In the implementation of the above solution, by extracting the explanatory definition text of each innovative tool from the tool database and filtering based on its similarity with the standard text of design requirements, tools unrelated to the requirements can be accurately filtered, avoiding misrecommendation caused by vague tool function descriptions or keyword mismatches, thereby significantly improving the accuracy of the recommendation results.

[0006] Optionally, in this embodiment, after obtaining the recommended tools and reasons corresponding to the standard design requirement text, the method further includes: filtering the recommended tools and reasons corresponding to the standard design requirement text using regular expressions to obtain structured data; and sending the structured data to the terminal device so that the terminal device can parse the innovative tool type and recommendation reason from the structured data. In the implementation of the above scheme, filtering the recommended tools and reasons using regular expressions can convert unstructured natural language text into a standardized data format (such as field-based tags or JSON structure), facilitating the subsequent automated processing, storage, and retrieval of the recommendation results by the system, and improving the operability and consistency of the data.

[0007] Optionally, in this embodiment, after sending structured data to the terminal device, the method further includes: receiving a problem description text and the innovative tool type of the recommended tool sent by the terminal device; searching for a technique analysis prompt word template corresponding to the innovative tool type from a prompt template library, wherein the prompt template library stores the correspondence between innovative tool types and technique analysis prompt word templates; filling the problem description text into the technique analysis prompt word template to obtain technique analysis prompt engineering text; and inferring the technique analysis prompt engineering text through a pre-trained model to obtain the technique analysis result of the target product. In the implementation of the above solution, by matching the technique analysis prompt word template corresponding to the innovative tool type from the prompt template library, the analysis logic and methodology of different tools can be dynamically adapted, enabling the pre-trained model to focus on the applicable scenarios and analysis points of specific tools during the inference process, thereby generating technique analysis results that are more in line with the characteristics of the tools and improving the relevance and accuracy of the technique analysis results of the target product.

[0008] Optionally, in this embodiment, after obtaining the technical analysis results of the target product, the method further includes: determining whether valid indication information of the technical analysis results of the target product has been received; if so, adding the design requirement standard text and the corresponding recommendation tools and reasons to the design requirement standard text to the fine-tuning training dataset. In the implementation of the above scheme, by incorporating the design requirement text, recommendation tools, and reasons corresponding to the valid indication information into the fine-tuning training dataset, the knowledge base of the pre-trained model can be dynamically updated, making it more closely aligned with the needs and tool matching rules of actual application scenarios in subsequent inference. This closed-loop self-feedback fine-tuning mechanism enables the model to gradually learn user preferences and design logic, improving the accuracy and consistency of recommendation and analysis results.

[0009] Optionally, in this embodiment, after adding the standard design requirement text and the corresponding recommendation tools and reasons to the fine-tuning training dataset, the method further includes: determining whether the total number of valid indication information from the technique analysis results exceeds a preset threshold; if so, using the standard design requirement text in the fine-tuning training dataset as training data and the recommendation tools and reasons in the fine-tuning training dataset as training labels, the pre-trained model is fine-tuned. In the implementation of the above scheme, by determining whether the total number of valid indication information exceeds a preset threshold and initiating fine-tuning training when the condition is met, the parameters of the pre-trained model can be dynamically adjusted based on user feedback data, gradually adapting it to the semantic features and tool matching rules of a specific domain (such as product design). This mechanism allows the model to continuously absorb valid recommendation logic in new scenarios during iteration, improving the accuracy and generalization ability of subsequent reasoning.

[0010] Optionally, in this embodiment, before fine-tuning the pre-trained model, the method further includes: obtaining a target language model and inserting a preset matrix into the target language model to obtain a pre-trained model; fine-tuning the pre-trained model includes: updating the preset matrix in the pre-trained model while freezing the model parameters of the target language model in the pre-trained model. In the implementation of the above scheme, by inserting a preset matrix into the target language model and updating only that matrix, rather than adjusting the entire model parameters, the number of parameters that need to be optimized in the fine-tuning stage can be significantly reduced. Since the original parameters of the target language model are frozen, large-scale gradient calculations and parameter updates are avoided, thereby significantly reducing computational overhead and memory usage, making the fine-tuning process more efficient and suitable for resource-constrained scenarios.

[0011] This application embodiment also provides a product innovation tool recommendation system based on a pre-trained model, including: a terminal device and a server; the terminal device is used to send standard text of design requirements for a target product to the server; the server is used to receive the standard text of design requirements for the target product sent by the terminal device; obtain a preset innovation tool list, and fill the preset innovation tool list and the standard text of design requirements into a preset prompt word template to obtain prompt engineering text; and use a pre-trained model to reason about the prompt engineering text to obtain the recommended tool and recommendation reason corresponding to the standard text of design requirements, wherein the recommended tool is at least one innovation tool in the preset innovation tool list.

[0012] This application embodiment also provides a product innovation tool recommendation device based on a pre-trained model, including: a standard text acquisition module, used to acquire standard text of design requirements for a target product; a prompt text acquisition module, used to acquire a preset innovation tool list and fill the preset innovation tool list and the standard text of design requirements into a preset prompt word template to obtain prompt engineering text; and a recommendation information acquisition module, used to reason about the prompt engineering text through a pre-trained model to obtain the recommended tool and recommendation reason corresponding to the standard text of design requirements, wherein the recommended tool is at least one innovation tool in the preset innovation tool list.

[0013] Optionally, in this embodiment, the prompt text acquisition module includes: a tool definition acquisition submodule, used to acquire the explanatory definition text of each innovative tool in the preset innovative tool list from a tool database, wherein the tool database stores the preset innovative tool list and the explanatory definition of each innovative tool; an innovative tool filtering submodule, used to filter candidate innovative tools from the preset innovative tool list whose similarity between the explanatory definition text and the design requirement standard text is greater than a preset threshold; and a prompt template filling submodule, used to fill the candidate innovative tools and the design requirement standard text into a preset prompt word template.

[0014] Optionally, in this embodiment of the application, the product innovation tool recommendation device based on the pre-trained model further includes: a structured data acquisition module, used to filter the recommended tools and recommendation reasons corresponding to the standard text of design requirements using regular expressions to obtain structured data; and a structured data sending module, used to send the structured data to the terminal device so that the terminal device can parse the innovation tool type and recommendation reason of the recommended tool from the structured data.

[0015] Optionally, in this embodiment, the product innovation tool recommendation device based on a pre-trained model further includes: a tool type receiving module, used to receive a problem description text and the innovative tool type of the recommended tool sent by a terminal device; a prompt template search module, used to search for a technique analysis prompt word template corresponding to the innovative tool type from a prompt template library, wherein the prompt template library stores the correspondence between innovative tool types and technique analysis prompt word templates; a prompt text obtaining module, used to fill the problem description text into the technique analysis prompt word template to obtain technique analysis prompt engineering text; and an analysis result obtaining module, used to infer the technique analysis prompt engineering text through a pre-trained model to obtain the technique analysis result of the target product.

[0016] Optionally, in this embodiment of the application, the product innovation tool recommendation device based on the pre-trained model further includes: an effective indication judgment module, used to determine whether effective indication information of the technical analysis result of the target product has been received; and a training data addition module, used to add the design requirement standard text and the recommendation tool and recommendation reason corresponding to the design requirement standard text to the fine-tuning training dataset if effective indication information of the technical analysis result of the target product has been received.

[0017] Optionally, in this embodiment of the application, the product innovation tool recommendation device based on the pre-trained model further includes: an indication quantity judgment module, used to judge whether the total number of valid indication information of the technique analysis results is greater than a preset quantity threshold; and a model fine-tuning training module, used to fine-tune the pre-trained model by using the design requirement standard text in the fine-tuning training dataset as training data and the recommended tools and recommendation reasons in the fine-tuning training dataset as training labels if the total number of valid indication information of the technique analysis results is greater than the preset quantity threshold.

[0018] Optionally, in this embodiment of the application, the product innovation tool recommendation device based on the pre-trained model further includes: a preset matrix insertion module, used to obtain a target language model and insert a preset matrix into the target language model to obtain a pre-trained model; and a preset matrix update module, used to update the preset matrix in the pre-trained model while freezing the model parameters of the target language model in the pre-trained model.

[0019] This application also provides an electronic device, including a processor and a memory, wherein the memory stores machine-readable instructions executable by the processor, and the machine-readable instructions are executed by the processor to perform the methods described above.

[0020] This application also provides a computer-readable storage medium storing a computer program, which is executed by a processor to perform the methods described above.

[0021] This application also provides a computer program product, including: a computer program or computer instructions, which are executed by a processor to perform the method described above. Attached Figure Description

[0022] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 The diagram shown is a flowchart illustrating the product analysis tool and innovation tool recommendation method based on a pre-trained model provided in an embodiment of this application. Figure 2 The illustration shown is a schematic diagram of a product innovation tool recommendation system based on a pre-trained model provided in an embodiment of this application; Figure 3 The diagram shown is a framework diagram of the software architecture system provided in the embodiments of this application; Figure 4 The diagram shown is a structural schematic of a product innovation tool recommendation device based on a pre-trained model provided in an embodiment of this application. Figure 5 The diagram shows a structural schematic of an electronic device provided in an embodiment of this application. Detailed Implementation

[0024] 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. It should be understood that the accompanying drawings in the embodiments of this application are for illustrative and descriptive purposes only and are not intended to limit the protection scope of the embodiments of this application. Furthermore, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in the embodiments of this application illustrate operations implemented according to some embodiments of this application. It should be understood that the operations in the flowcharts may not be implemented in sequence, and steps without logical contextual relationships may be reversed or implemented simultaneously. In addition, those skilled in the art, guided by the content of the embodiments of this application, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts.

[0025] Furthermore, the described embodiments are merely a part of the embodiments of this application, and not all of them. The components of the embodiments of this application described and illustrated herein can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed embodiments of this application, but merely to illustrate selected embodiments of this application.

[0026] It is understood that the terms "first" and "second" in the embodiments of this application are used to distinguish similar objects. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, and that "first" and "second" do not necessarily imply that they are different. In the description of the embodiments of this application, the term "and / or" is merely a description of the association relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship. The term "multiple" refers to two or more (including two), and similarly, "multiple groups" refers to two or more groups (including two groups).

[0027] It should be noted that the product analysis tool innovation tool recommendation method based on pre-trained models provided in this application can be executed by electronic devices. Here, electronic devices refer to device terminals or servers with the function of executing computer programs. Device terminals include, for example, smartphones, personal computers, tablets, personal digital assistants, or mobile internet devices. Servers refer to devices that provide computing services through a network. Servers include, for example, x86 servers and non-x86 servers. Non-x86 servers include, for example, mainframes, minicomputers, and UNIX servers.

[0028] In related technologies, product innovation design analysis largely relies on historical data similarity matching to recommend innovation tools. The core logic of this method is to find the closest match from an existing case library based on the similarity of keywords or numerical features. However, this reliance on historical data makes it difficult to adapt to the dynamic design needs of product innovation. Traditional similarity matching only focuses on the superficial similarity of keywords or values, failing to understand the deeper logic of user needs. For example, if a user's need is "a material that can withstand high temperatures and is lightweight," but historical cases only record "high-temperature materials," the system may ignore the crucial constraint of "lightweight" because the user's need doesn't perfectly match the keywords in the historical data. This could lead to the recommendation of an unsuitable innovation tool, rendering it ineffective for the user. Therefore, the accuracy of current product innovation tool recommendations is relatively low.

[0029] For the above issues, please refer to Figure 1 The illustrated flowchart represents a product analysis tool innovation tool recommendation method based on a pre-trained model, provided in this application embodiment. The core idea of ​​this method is to combine standard design requirement text with the semantic reasoning capabilities of a pre-trained model to more efficiently identify the correlation between requirements and tools. This makes the recommended innovation tools and their reasons for recommendation more targeted and logical, thereby significantly improving recommendation accuracy. The implementation methods of the aforementioned product analysis tool innovation tool recommendation method based on a pre-trained model may include: Step S110: Obtain the standard text of design requirements for the target product.

[0030] Target products refer to products that require recommendations of innovative tools for the target user's terminal devices, such as electromechanical products, network products, or communication equipment. For ease of understanding and explanation, electromechanical products will be used as an example below.

[0031] Design requirement standard text refers to standardized description text generated by users based on their design requirements for a target product, using a pre-defined structured template. This structured template could be something like, "How can we achieve the [functional description] of [product or system name] to reach the [expected design goal]?" Step S120: Obtain a list of preset innovation tools and fill the list of preset innovation tools and the standard text of design requirements into a preset prompt word template to obtain the prompt engineering text.

[0032] The innovation tool list refers to a collection of tools used to analyze target products based on standard design requirements texts. Examples of the aforementioned preset innovation tool list include: fault analysis tools, fishbone analysis tools, cause-and-effect analysis tools, 5 Whys innovation tools, requirement transformation innovation tools, component analysis tools, function-oriented search innovation tools, FS analysis tools, target mining innovation tools, product inspection innovation tools, attribute association innovation tools, technology forecasting innovation tools, appearance design innovation tools, and emotional design innovation tools.

[0033] A preset prompt template is a structured input format used to combine design requirements with a preset tool list, guiding a pre-trained model to generate recommended results. For example, the prompt template above could be in the form of: "You are an expert in recommending innovative tools in the field of electromechanical innovative design. Based on requirements, you recommend innovative tools. The design requirements are: {requirement text}. Available tools: {tool list}. Please recommend the most suitable tool and explain your reasoning."

[0034] The prompt engineering text is the complete input text generated by filling a preset prompt word template with the standard design requirement text and a preset tool list, and is used to drive model inference. For example, the filled-in prompt engineering text could be something like: "You are an expert in pushing innovative tools in the field of electromechanical innovative design. You are pushing innovative tools based on requirements. The design requirement is: How to solve the problem of difficult disassembly of connectors in high-temperature environments to achieve rapid disassembly? Available tools: Fault innovation tool, Fishbone innovation tool, Causal innovation tool, 5 Whys innovation tool, Requirement transformation innovation tool, Component innovation tool, Function-oriented search innovation tool, FS innovation tool, Target mining innovation tool, Product inspection innovation tool, Attribute association innovation tool, Technology prediction innovation tool, Appearance design innovation tool, Emotional design innovation tool, etc. Please recommend the most suitable tool and explain your reasons."

[0035] Step S130: Use a pre-trained model to reason about the prompt engineering text to obtain the recommended tools and reasons for the recommendations corresponding to the standard design requirements text. The recommended tools are at least one innovative tool from the preset list of innovative tools.

[0036] Pre-trained models are language models trained on large-scale data. They possess the ability to understand natural language and generate structured content, and are used to infer from prompt engineering text and output analysis results. The aforementioned pre-trained models can employ Deepseek series models (such as Deepseek R1-7B), Qwen series models, Llama series models (such as Llama 3-8B), etc. These models can all process complex text and generate analysis results that conform to template logic.

[0037] The recommended tool is at least one innovative tool that matches the design requirements and is recommended by the pre-trained model from a pre-defined tool list. The reasons for the recommendation are textual explanations generated by the pre-trained model that illustrate the logic behind the matching between the recommended tool and the design requirements.

[0038] Understandably, the structured format of the prompt template above can also include: role definition, task definition, and example input / output. For example, a structured prompt template like this: "Role Definition: You are an expert in pushing innovative tools in the field of electromechanical innovative design. You push innovative tools based on needs. Task Definition: You need to deeply understand the tool definition, carefully analyze the problem, and recommend the most suitable innovative tool based on the problem. Please strictly follow the following output format: Analyze the problem: {question}; Optional innovative tools: {", ".join(TOOLS_LIST)}; Select the most suitable tool based on the following tool definition: {context}; Push tool: <tool name, must be from the above list>; Push reason: <explain applicability based on the tool definition>; Please ensure the tool name strictly comes from the above list; Example input / output: "Input requirement: The sintering furnace has a problem of uneven temperature during use. How to solve this problem to achieve the design goal of clearly identifying the cause of the problem and solving the temperature distribution problem? Push tool: <causal analysis>; Push reason: <Causal analysis is a systematic tool that can help identify the root cause of the problem and its impact path.>" For the problem of uneven temperature in sintering furnaces, causal analysis can help us explore the various potential factors causing temperature fluctuations, including the interactions between operating parameters, equipment status, and environmental conditions. By constructing a causal model, we can clearly see which factors directly affect the temperature distribution, and then develop targeted solutions.

[0039] In the implementation of the above solution, by combining the standard text of design requirements with the semantic reasoning ability of the pre-trained model, the standard text of design requirements is standardized into an input format of preset prompt word templates, which reduces the interference of natural language ambiguity on the model's understanding. Moreover, the pre-trained model has learned the potential rules of tool recommendation in large-scale data training. Combined with the standardized processing of input information by prompt engineering text, it can more efficiently identify the correlation between requirements and tools. Therefore, the recommended tools and reasons for recommendation output by the pre-trained model are more targeted and logical, thereby significantly improving the recommendation accuracy.

[0040] The following explains some of the innovation tools in the list of innovation tools: Fault analysis tools analyze the probability of failure by establishing top and intermediate events in a fault tree, identify the key nodes that cause failure, understand the causes of system failure, and discuss the relationships and probabilities between the causes of system failure. From this, the best way to solve problems, reduce risks, and gain innovation opportunities can be found.

[0041] Fishbone analysis is a visual analysis tool that analyzes problems from six aspects: people, machines, materials, methods, environment, and measurement, thereby finding directions for solving problems. Fishbone diagram analysis is hierarchical and well-organized, which helps to focus on the causes of problems rather than the symptoms.

[0042] Causal analysis tools analyze the original problems faced by a product using a causal chain model, continuously analyzing the deeper causes of each cause to obtain the root cause of the original problem and gain innovative opportunities to solve it.

[0043] The 5 Whys innovation tool starts with an existing problem and uses five "why" questions to continuously investigate the cause of the problem, directly and efficiently identifying the root cause of the existing problem. By solving the root cause, the existing problem is solved at its source.

[0044] Demand transformation innovation tools utilize QFD analysis tools to associate product design requirements with actionable engineering feature parameters through an intuitive matrix framework, identify various conflicts, and clarify the key conflicts that need to be improved.

[0045] Component analysis tools decompose a system based on the functional analysis of its components, establish a component analysis model of the system, and obtain functional relationships such as standard, insufficient, excessive, and harmful relationships between components. This helps designers to understand the interactions between components in detail, discover the root causes of problems in the system, and clarify the direction for system improvement.

[0046] Function-oriented search innovation tools are used to solve the problem of how to implement a single function, providing designers with specific knowledge on function implementation through function-oriented search.

[0047] FS analysis tools decompose complex functions into several sub-functions through functional decomposition, build a function tree that reflects the relationships between functions, and then provide specific knowledge to support the implementation of sub-functions through function-oriented search.

[0048] Goal mining innovation tools are designed to address the problem of designers having a vague goal but not knowing how to achieve it. Through a systematic framework of goal mining, a top-down, step-by-step derivation approach is used to clarify the means of achieving the goal. First, the desired effect and the "goal" are clearly defined. Second, the basic approach and "direction" for achieving the "goal" are further identified. Next, the process is refined to identify the corresponding methods and "concepts" supporting the "direction," ultimately forming the target function. Throughout this process, through continuous higher-level and lateral thinking, the target function for solving the problem is obtained.

[0049] The product inspection and innovation tool examines the structure of existing products from nine different perspectives and through 75 specific questions, including usage, analogy, addition, reduction, and change, overcoming designers' reluctance to ask questions. Guided by questions from different angles, it explores directions for technological expansion of existing products and improves the economic efficiency of existing technologies.

[0050] Attribute-related innovation tools start from the perspective of internal and external factors of a problem. Based on existing products, they use the method of associating the internal attributes of products with the attributes of the external environment to continuously explore the value of existing products and improve the economic benefits of existing product technologies.

[0051] Technology forecasting and innovation tools, based on the eight laws of evolution, predict and recommend trends in the improvement and development of product technology systems as they achieve their functions. This helps designers determine which stage of the technological system evolution process their current R&D product is in, and then, based on the content of technology evolution theory, better predict the future development direction of the product.

[0052] Product design innovation tools assist designers in obtaining image materials to inspire their designs by using product association word graph retrieval. This overcomes designers' limitations in associative thinking, supports divergent thinking, and generates creative ideas. The goal is to make products novel and attractive through product design.

[0053] Emotional design innovation tools utilize a lexical search method to assist designers in acquiring imagery that sparks inspiration during the conceptual design phase of product emotional design. This overcomes designers' limitations in associative thinking, supports divergent thinking, and fosters creative ideas. Simultaneously, the search, through semantic expansion, bridges the "semantic gap" designers face when using emotional vocabulary to find inspiration, providing more comprehensive and accurate inspiration sources. This also helps in meeting consumers' spiritual and cultural needs when designing product appearance.

[0054] As an optional implementation of step S110 above, for example, the user can input the standard text of the target product's design requirements through the human-computer interaction interface or application programming interface (API) of the terminal device. After receiving the standard text of the target product's design requirements through the human-computer interaction interface or API, the terminal device can send the standard text of the target product's design requirements to an electronic device (such as a server). The electronic device (such as a server) can directly receive the standard text of the target product's design requirements from the terminal device, or it can obtain the standard text of the target product's design requirements from the terminal device through the API interface.

[0055] Understandably, when users input the standard design requirements text for the target product, they can also set a preset standardized requirement expression format on the human-computer interaction interface. This constrains the user's input requirement expression format, improves the ambiguity when users express requirements using natural language, and ensures that the pre-trained model can guarantee the requirement-specificity and design efficiency of the product innovation process during inference. Suppose the preset standardized requirement expression format is "[Product or System Name] has generated a [Problem Description] problem in [Problem Scenario]. How to solve this problem to achieve the [Expected Design Goal]?", the initial requirement text prepared by the user before inputting the standard design requirements text for the target product might be: "The quick-detachable connector is a structure used to connect the upper tube seat and the guide tube in a nuclear reactor fuel assembly. When the fuel rods reach the expected operating time, the connector needs to be quickly disassembled at a depth of 4 meters underwater to replace the fuel assembly. However, due to the small operating range of the overall device and the tight operating time, traditional connector structures are difficult to meet the requirements." However, the text that users input on the human-computer interaction interface according to the preset standardized requirement expression format might be: "Select the format of existing product fault and problem solution for standardized expression: [Connector] has a problem of [difficult to quickly disassemble and assemble] during [use]. How to solve this problem to achieve the design goal of [high-strength and reliable connection, quick disassembly, easy operation in restricted environments, and minimizing the generation of waste parts]?"

[0056] In the process of constraining the user's input requirement expression format as described above, specific preset expression formats can be set according to the specific requirement category. For example, if the user wants to input a requirement category of product function implementation, then the preset expression format for the standard text of the design requirement for this product function implementation category is "How to implement the [Function Description] of [Product or System Name] to achieve the [Expected Design Goal]?". Similarly, if the user wants to input a requirement category of product fault problem solving, then the preset expression format for the standard text of the design requirement for this product fault problem solving category is "[Product or System Name] has generated the [Problem Description] problem in the [Problem Scenario]. How to solve this problem to achieve the [Expected Design Goal]?". Similarly, if the user wants to input a requirement category of product optimization and improvement, then the preset expression format for the standard text of the design requirement for this product optimization and improvement category is "How to expand or improve [Product or System Name] to achieve the [Expected Design Goal]?". Similarly, if the user wants to input a requirement category of product appearance design, then the default expression format for the standard text of the design requirement for that product appearance design category is "How to design the appearance description of [product or system name] to achieve the [expected design goal]?".

[0057] As an optional implementation of step S120 above, the above-described implementation of filling the preset innovation tool list and design requirement standard text into the preset prompt template may include: Step S121: For each innovative tool in the preset list of innovative tools, retrieve the explanation definition text of the innovative tool from the tool database. The tool database stores the preset list of innovative tools and the explanation definition of each innovative tool.

[0058] The implementation of step S121 above is as follows: It is understood that since there are many kinds of innovative tools in the preset innovative tool list stored in the tool database, the total number of characters in these preset innovative tool lists and the explanation definition of each innovative tool may exceed the context limit of the pre-trained model. Therefore, the explanation definition text of the innovative tool can be obtained from the tool database first, and the innovative tools in the tool database can be filtered using the similarity between the explanation definition text of each innovative tool and the standard text of the design requirements. A filtered list of innovative tools can then be obtained, so that the pre-trained model can directly select the innovative tools to be recommended from the filtered list. This effectively avoids the situation where the total number of characters in the preset innovative tool list and the explanation definition of each innovative tool may exceed the context limit of the pre-trained model, and allows the pre-trained model to process fewer characters, thereby improving the recommendation efficiency of innovative tools.

[0059] Step S122: Select candidate innovation tools from the preset innovation tool list whose similarity between the explanation definition text and the design requirement standard text is greater than a preset threshold.

[0060] The implementation of step S122 above is as follows: First, the similarity between the explanatory definition text and the standard design requirement text is calculated by calling the code cosine_similarity(question_vec, tool_vectors). Then, candidate innovation tools with a similarity greater than a preset threshold between the explanatory definition text and the standard design requirement text are selected from the preset list of innovation tools. Here, question_vec represents the text vector of the standard design requirement text, tool_vectors represents the text vector of the explanatory definition text of the innovation tool, and cosine_similarity represents the cosine similarity between the text vector of the standard design requirement text and the text vector of the explanatory definition text of the innovation tool. The text vector of the standard design requirement text can be obtained by calculating it using a Term Frequency-Inverse Document Frequency (TF-IDF) vectorizer.

[0061] Step S123: Fill the candidate innovation tools and design requirement standard text into the preset prompt word template to obtain the prompt engineering text.

[0062] For example, the implementation of step S123 above is as follows: Suppose the standard design requirement text is "[Connector] has a problem of [difficulty in quick assembly and disassembly] during [use]. How to solve this problem to achieve the design goal of [high-strength and reliable connection, quick disassembly, easy operation in confined environments, and minimizing waste]?", and the candidate innovation tools are "fault innovation tool, fishbone innovation tool, causal innovation tool, 5 Whys innovation tool, requirement transformation innovation tool, component innovation tool", and the prompt template is "You are an expert in pushing innovation tools in the field of electromechanical innovation design. Push innovation tools according to requirements. Among them, the design requirement is: {requirement text}. Available tools: {tool list}. Please recommend the most suitable tool." "Please recommend the most suitable tool and explain your reasoning." After filling the preset prompt template with candidate innovation tools and standard design requirement text, the resulting prompt engineering text could be: "You are an expert in pushing innovation tools in the field of electromechanical innovation design. Based on the requirements, you are pushing innovation tools. The design requirement is: 'The connector has encountered a problem of difficulty in quick assembly and disassembly during use. How can this problem be solved to achieve the design goals of high-strength and reliable connection, quick disassembly, easy operation in confined environments, and minimizing waste?' Available tools: Fault Innovation Tool, Fishbone Innovation Tool, Cause-and-Effect Innovation Tool, 5 Whys Innovation Tool, Requirement Transformation Innovation Tool, Component Innovation Tool. Please recommend the most suitable tool and explain your reasoning."

[0063] Understandably, since the role defined for the pre-trained model in the aforementioned preset prompt word template is that of an innovation tool recommendation expert in the field of electromechanical innovation design, whose task is to recommend the most suitable innovation tools to users based on the innovative design requirements input by the designers. In the implementation of the above scheme, before filling the prompt word template, candidate innovation tools are filtered out using a similarity threshold. This reduces the input of redundant tool information and avoids interference from irrelevant tool features or descriptions during the model's reasoning process, thereby improving the model's ability to focus on the logical connection between requirements and tools and enhancing the stability of the recommendation results. Furthermore, by combining the filtered candidate tools and their explanatory definitions with the design requirement text before inputting them into the prompt word template, the model's reasoning process is based on a more complete semantic context (including requirement features, tool functions, and the degree of matching between the two), thereby generating more logical and explanatory recommendation reasons, increasing user trust in the recommendation results and the adoption rate.

[0064] As an optional implementation of step S130 above, the above implementation of reasoning about the prompt engineering text through a pre-trained model may include: obtaining the model calling program of the pre-trained model, inputting the prompt engineering text into the pre-trained model through the model calling program, so that the pre-trained model can reason about the prompt engineering text, thereby obtaining the recommended tools and reasons for the recommendation corresponding to the design requirement standard text output by the pre-trained model.

[0065] As an optional implementation of the above-mentioned product analysis tool innovation tool recommendation method based on pre-trained models, after obtaining the recommended tools and reasons for recommendation corresponding to the standard text of design requirements, it may further include: Step S141: The electronic device uses regular expressions to filter the recommended tools and reasons corresponding to the standard text of design requirements to obtain structured data.

[0066] For example, the implementation of step S141 above can be as follows: It is understood that the structured data may include an "Analysis Problem" field, and the electronic device can use the regular expression "r"Analysis Problem: (.*?)\n"" to filter out the "Analysis Problem" string from the design requirement standard text. Similarly, the structured data may include a "Recommendation Tool" field, and the electronic device can use the regular expression "r"Recommendation Tool: ([^\n]+)"" to filter out the "Recommendation Tool" string from the design requirement standard text. Similarly, the structured data may include a "Recommendation Reason" field, and the electronic device can use the regular expression "r"Recommendation Reason: ([^\n]+)"" to filter out the "Recommendation Reason" string from the design requirement standard text. After the recommendation tool is verified, the analysis problem string, recommendation tool string, and recommendation reason string are encapsulated into structured data. The specific process for verifying the aforementioned recommendation tool may include: determining whether the recommendation tool is an innovative tool in a preset list of innovative tools; if the recommendation tool is an innovative tool in the preset list, the verification result of the recommendation tool is determined to be verified as passed; otherwise, the verification result of the recommendation tool is determined to be verified as failed. It is understood that the structured data directly contains clear identifiers of the type of recommendation tool and the reasons for the recommendation, enabling terminal devices to quickly extract key information without relying on complex natural language understanding logic, reducing computational resource consumption during the parsing process, and accelerating the response speed of the user interface.

[0067] Step S142: The electronic device sends structured data to the terminal device so that the terminal device can parse the innovative tool type and recommendation reason from the structured data.

[0068] For example, in implementing step S142 above, after obtaining structured data, the electronic device can also send structured data to the terminal device. This structured data can be in a preset string format, JSON format, XML format, or YAML format, etc. If the structured data is in a preset string format, the terminal device, after receiving the structured data, can use the regular expression "r" to analyze the problem: (.*?)\n"" to filter out the string "analyze the problem" from the structured data. It can also use the regular expression "r" to recommend tools: ([^\n]+)"" to filter out the string "recommend tools" from the structured data. Furthermore, it can use the regular expression "r" to recommend reasons: ([^\n]+)"" to filter out the string "recommendation reasons" from the structured data. In the implementation of the above scheme, by using a unified structured data format, it is possible to ensure that different terminal devices (such as PCs, mobile devices, or embedded systems) parse the recommendation results in a consistent manner, avoiding information loss or misreading due to differences in terminal processing capabilities, and improving the overall system stability and scalability.

[0069] As an optional implementation of the above-mentioned product analysis tool innovation tool recommendation method based on pre-trained models, after sending structured data to the terminal device, it may further include: Step S151: The electronic device receives the problem description text and the innovative tool type of the recommended tool sent by the terminal device.

[0070] An example implementation of step S151 above is as follows: A user on a terminal device can input a problem description text for the target product (e.g., "The motor experiences overheating and shutdown when operating in a high-temperature environment"), and select the corresponding innovation tool type on the terminal device (e.g., "Causal Innovation Tool"). Then, the user submits the problem description text and innovation tool type to an electronic device (e.g., a server). The electronic device (e.g., a server) can receive the problem description text and innovation tool type from the terminal device, or obtain them from the terminal device via an API interface. In the implementation of the above solution, by having the terminal device provide the problem description and tool type, and the system dynamically generates corresponding technical analysis prompts and engineering text, a closed-loop optimization of human-computer interaction can be achieved. This allows users to obtain accurate analysis results based on the latest needs without manually adjusting the analysis process, reducing repetitive operations and improving overall collaboration efficiency.

[0071] Step S152: Search for the technique analysis prompt word template corresponding to the innovation tool type in the prompt template library. The prompt template library stores the correspondence between innovation tool types and technique analysis prompt word templates.

[0072] For example, the implementation of step S152 described above can be as follows: The prompt template library can pre-store the correspondence between various innovation tool types and technique analysis prompt templates. If the user selects a "causal innovation tool" as the technique innovation tool type, the corresponding technique analysis prompt template can be found and retrieved from the prompt template library. The technique analysis prompt template corresponding to the causal innovation tool can be represented as: "You are a causal analysis expert in the electromechanical field. Please provide your answer strictly according to the following format: Question: {question}; 1. Possible causes: 2. In-depth analysis: 3. Suggested solutions: Please add the corresponding content after each section, without omitting any headings. Analysis:". The correspondence between innovation tool types and technique analysis prompt templates stored in the prompt template library provides a flexible expansion interface for the system. When a new innovation tool is added, only the corresponding template needs to be added to support its technique analysis capabilities, avoiding system reconstruction costs caused by tool iteration and enhancing the adaptability and sustainability of the solution.

[0073] Step S153: Fill the problem description text into the technique analysis prompt template to obtain the technique analysis prompt engineering text.

[0074] For example, the implementation of step S153 above can be described as follows: Assume the problem description text received by the electronic device (e.g., a server) is "The motor experiences overheating and shutdown when operating in a high-temperature environment," and the innovation tool type is "Causal Innovation Tool." Then, the structured prompt template corresponding to the causal innovation tool can be represented as: "You are a causal analysis expert in the electromechanical field. Please provide your answer strictly according to the following format: Problem: The motor experiences overheating and shutdown when operating in a high-temperature environment; 1. Possible causes: 2. In-depth analysis: 3. Suggested solutions: Please add the corresponding content after each section, without omitting any headings. Analysis:." It is understandable that by filling the problem description text into the technique analysis prompt template, the pre-trained model's inference can decompose user needs into structured analysis steps (such as parameter settings, constraints, verification methods, etc.), making the analysis process logically coherent and traceable, reducing the difficulty of solving complex design problems, and improving design efficiency.

[0075] Step S154: Use a pre-trained model to reason about the technical analysis prompts in the engineering text to obtain the technical analysis results of the target product.

[0076] For example, the implementation of step S154 above involves obtaining the model calling program of the pre-trained model, inputting the technique analysis prompt engineering text into the pre-trained model through the model calling program, so that the pre-trained model can perform inference analysis on the technique analysis prompt engineering text and output the technique analysis results of the target product.

[0077] As an optional implementation of the aforementioned product analysis tool innovation tool recommendation method based on a pre-trained model, after obtaining the technical analysis results of the target product, a self-feedback fine-tuning process can be used to obtain a fine-tuning training dataset to train the aforementioned pre-trained model. The implementation of this self-feedback fine-tuning process may include: Step S161: Determine whether valid indication information of the technical analysis results of the target product has been received.

[0078] An example implementation of step S161 above is as follows: A user can send a valid indication message of the technical analysis result of the target product to an electronic device (such as a server) through a terminal device. This valid indication message can be 1 or 0, where 1 indicates that the technical analysis result of the target product is valid, and 0 indicates that the technical analysis result of the target product is invalid. Then, the electronic device (such as a server) can use an executable program to determine whether it has received the valid indication message of the technical analysis result of the target product. If the technical analysis result of the target product is 1, it means that the technical analysis result of the target product is valid; similarly, if the technical analysis result of the target product is 0, it means that the technical analysis result of the target product is invalid.

[0079] Step S162: If a valid indication of the technical analysis result of the target product is received, the design requirement standard text and the corresponding recommended tools and reasons are added to the fine-tuning training dataset.

[0080] For example, in the implementation of step S162 above: if the received technical analysis result of the target product is 1, then the standard design requirement text, along with the corresponding recommended tools and reasons, can be added to the fine-tuning training dataset. It is understood that the fine-tuning training dataset can include not only the standard design requirement text and the corresponding recommended tools and reasons added after receiving valid user instructions, but also sample data mined from existing product innovation design cases. This sample data can include: initial innovation design requirement texts mined from existing product innovation design cases organized into requirement input texts according to the aforementioned preset standardized requirement expression format, and these requirement input texts determined as the standard design requirement texts. Furthermore, considering the integrated use of multiple innovation tools in some innovation design cases, the requirements for each innovation tool's usage stage should be organized into standardized requirement expression texts to ensure a reasonable one-to-one correspondence between the standard design requirement texts and the innovation tools. Then, we analyze the innovative tools used in the above design requirement standard text from existing product innovation design cases, as well as the reasons and justifications for using these tools. This ensures a strong logical relationship between the design requirement standard text and the innovative tools, forming a structured "push tool + push reason" model output format text corresponding to the design requirement standard text.

[0081] Understandably, incorporating effective feedback data into the fine-tuning training dataset allows the model to learn user preference patterns regarding recommended tools and reasons (such as high-frequency adoption of certain tools or acceptance of specific recommendation logic) during the fine-tuning process. This results in recommendations that better match users' actual needs, reducing recommendation bias caused by domain differences in traditional general-purpose models. Furthermore, by accumulating effective feedback and periodically fine-tuning the model, its dependence on static historical data can be reduced, enabling it to quickly adapt to new demands, tools, or scenarios emerging in the product design domain. This avoids recommendation failures caused by lag in demand changes, enhancing the system's real-time performance and flexibility.

[0082] In the implementation of the above scheme, by incorporating the design requirement text, recommendation tools and reasons corresponding to the effective instruction information into the fine-tuning training dataset, the knowledge base of the pre-trained model can be dynamically updated, making it more in line with the needs and tool matching rules of the actual application scenario in subsequent reasoning. This closed-loop self-feedback fine-tuning mechanism enables the model to gradually learn user preferences and design logic, thereby improving the accuracy and consistency of recommendation and analysis results.

[0083] As an optional implementation of the aforementioned product analysis tool and innovation tool recommendation method based on a pre-trained model, after adding the design requirement standard text and the corresponding recommendation tools and reasons to the fine-tuning training dataset, it further includes: Step S171: Determine whether the total number of valid indicator information in the technique analysis results is greater than a preset threshold.

[0084] For example, in the implementation of step S171 above, an electronic device (such as a server) can determine whether the total number of valid indication information of the technique analysis results is greater than a preset number threshold. The preset number threshold can be set according to specific circumstances, such as 100 or 120.

[0085] Step S172: If the total number of valid indication information from the technique analysis results is greater than the preset threshold, then the pre-trained model is fine-tuned by using the standard text of design requirements in the fine-tuning training dataset as training data and the recommended tools and reasons for recommendation in the fine-tuning training dataset as training labels.

[0086] For example, in implementing step S172 above: if the total number of valid indicator information from the technique analysis results exceeds a preset threshold, the design requirement standard text from the fine-tuning training dataset is input into the pre-trained model so that the pre-trained model outputs prediction tools and prediction rationale. Then, the total loss value is calculated based on the prediction tools, prediction rationale, recommendation tools, and recommendation rationale from the fine-tuning training dataset, and the preset matrix in the pre-trained model is updated based on the total loss value. Optionally, during the fine-tuning training of the pre-trained model described above, the fine-tuning training dataset can also be divided into a training set, a validation set, and a test set according to a preset data ratio (e.g., 8:1:1). The training set is used to fine-tune the pre-trained model, the validation set is used to verify the training effect of the pre-trained model, and the test set is used to test the generalization performance of the pre-trained model. It is understood that triggering fine-tuning training only when the number of valid feedbacks reaches a preset threshold can avoid low-quality or redundant data interfering with model learning, ensure that the training dataset contains high-value samples, reduce unnecessary computational overhead, and improve the stability and convergence efficiency of model updates.

[0087] As an optional implementation of step S172 above, before fine-tuning the pre-trained model, the following is also included: Step S172a: Obtain the target language model and insert a preset matrix into the target language model to obtain a pre-trained model.

[0088] An example implementation of step S172a above is as follows: It is understood that Low-Rank Adaptation (LoRA) technology can be used to fine-tune the pre-trained model. Specifically, a target language model, such as Deepseek R1-7B, can be obtained first. Then, a preset matrix (e.g., a pre-set low-rank matrix) can be inserted into the preset neural network layers of the target language model (e.g., Deepseek R1-7B) to obtain the pre-trained model. This Low-Rank Adaptation (LoRA) technology enables the model to quickly complete the training process of the dataset in the field of electromechanical innovation design when performing tasks in this field, while avoiding the need to retrain all parameters of the target language model (e.g., Deepseek R1-7B), thus effectively saving computational and memory resources. The introduction of the aforementioned preset matrix provides the model with additional trainable parameters, enabling it to quickly absorb knowledge from specific domains or tasks while maintaining the generality of the original language model. Furthermore, by fine-tuning the preset matrix, the model can learn domain-related feature representations without retraining the entire model, thereby achieving flexible adaptation to new scenarios with limited data and improving generalization performance.

[0089] As an optional implementation of step S172 above, the above implementation of fine-tuning the pre-trained model may include: Step S172b: With the model parameters of the target language model in the pre-trained model frozen, update the preset matrix in the pre-trained model.

[0090] Understandably, even with the model parameters of the target language model frozen in the pre-trained model, the hyperparameters of the pre-trained model can still be set. Specific steps for setting the hyperparameters of the pre-trained model include: setting the sequence length to 512, the learning rate to le-5, the computational precision to bf16, the number of training epochs to 5, the batch size to 2, the gradient accumulation to 4, the LoRA rank to 16, and the scaling factor to 32, etc.

[0091] An example implementation of step S172b above is as follows: With the model parameters of the target language model (e.g., Deepseek R1-7B) in the pre-trained model frozen, the design requirement standard text in the fine-tuning training dataset is used as training data, and the recommendation tools and reasons in the fine-tuning training dataset are used as training labels. The preset matrix (e.g., a low-rank matrix) in the pre-trained model is then updated. This allows the pre-trained model to retain the general language understanding capabilities of Deepseek R1-7B while also improving its understanding of the electromechanical innovation design domain of the target product (e.g., electromechanical products). In the implementation of the above scheme, freezing the original parameters of the target language model avoids disturbances to the pre-training weights during fine-tuning, reducing model crashes or performance fluctuations caused by parameter updates. The low-rank characteristic of the preset matrix (designed as a low-rank matrix) limits model complexity, suppresses overfitting to the training data, and enables the model to maintain high robustness and convergence stability even with limited data.

[0092] Please see Figure 2 The illustration shows a schematic diagram of a product innovation tool recommendation system based on a pre-trained model provided in an embodiment of this application; this embodiment of the application also provides a product innovation tool recommendation system 200 based on a pre-trained model, including: a terminal device 210 and a server 220; Terminal device 210 is used to send standard text of design requirements for the target product to the server; Server 220 is used to receive standard text of design requirements for the target product sent by terminal device; obtain a preset list of innovative tools, and fill the preset list of innovative tools and the standard text of design requirements into a preset prompt word template to obtain prompt engineering text; and use a pre-trained model to reason about the prompt engineering text to obtain recommended tools and reasons for recommendation corresponding to the standard text of design requirements, wherein the recommended tools are at least one innovative tool in the preset list of innovative tools.

[0093] Please see Figure 3 The diagram shows a framework of the software architecture system provided in this application embodiment. It can be understood that the above-mentioned product innovation tool recommendation method based on pre-trained models can be applied to the development of a software architecture system for product innovation tool recommendation. This software architecture system may include: an interaction layer, a functional layer, a model layer, and a data layer. These four layers can ensure that, in addition to the basic function of tool push, the system also meets the functions of user management of personal login information, management system tool information, and interface with computer-aided innovation platforms, thereby enhancing the simplicity, practicality, and scalability of the software architecture system.

[0094] The interaction layer is used to realize the interaction of the personal center, tool push and tool information, and provide designers with operations such as personal information management, innovative tool push and tool information management.

[0095] The functional layer is used to implement the innovative tool push function based on strategy selection. Here, strategy selection refers to choosing from the preset prompt word templates mentioned above. For example, if the software architecture system defaults to the first preset prompt word template, the user can choose the second preset prompt word template, at which point the software architecture system can set the second preset prompt word template as its default prompt word template. The functional layer can include: viewing, modifying, and deleting personal information, as well as self-feedback fine-tuning functions. It can also include: analyzing input needs and viewing push results. Furthermore, it can include: viewing, modifying, adding, and deleting innovative tools.

[0096] The model layer, which manages various pre-trained models, is the core layer of the entire push system. This model layer can include tool push models, model output filtering, and model self-feedback fine-tuning.

[0097] The data layer provides data source support for the push system by building a knowledge base for database tools, fine-tuning models, and storing and accessing user information.

[0098] Please see Figure 4 The diagram shown is a structural schematic of a product innovation tool recommendation device based on a pre-trained model provided in an embodiment of this application; this application also provides a product innovation tool recommendation device 300 based on a pre-trained model, including: The standard text acquisition module 310 is used to acquire the standard text of design requirements for the target product.

[0099] The prompt text acquisition module 320 is used to obtain a preset list of innovative tools and fill the preset prompt word template with the preset list of innovative tools and the standard text of design requirements to obtain the prompt engineering text.

[0100] The recommendation information acquisition module 330 is used to reason about the prompt engineering text through a pre-trained model to obtain the recommended tools and reasons for the recommendation corresponding to the standard text of design requirements. The recommended tools are at least one innovative tool in the preset list of innovative tools.

[0101] As an optional implementation of the above-mentioned device, the prompt text acquisition module includes: The tool definition acquisition submodule is used to retrieve the explanatory definition text of each innovative tool in the preset list of innovative tools from the tool database. The tool database stores the preset list of innovative tools and the explanatory definition of each innovative tool.

[0102] The innovation tool filtering submodule is used to filter candidate innovation tools from the preset innovation tool list that have a similarity greater than a preset threshold between the explanation definition text and the design requirement standard text.

[0103] The prompt template filler submodule is used to fill the preset prompt word templates with candidate innovation tools and standard text of design requirements.

[0104] As an optional implementation of the above-mentioned device, the product innovation tool recommendation device based on the pre-trained model further includes: The structured data acquisition module uses regular expressions to filter the recommended tools and reasons corresponding to the standard text of design requirements, thereby obtaining structured data.

[0105] The structured data sending module is used to send structured data to the terminal device so that the terminal device can parse the innovative tool type and recommendation reason from the structured data.

[0106] As an optional implementation of the above-mentioned device, the product innovation tool recommendation device based on the pre-trained model further includes: The tool type receiving module is used to receive problem description text and innovative tool types recommended by terminal devices.

[0107] The prompt template search module is used to search for the technique analysis prompt word templates corresponding to the innovation tool types in the prompt template library. The prompt template library stores the correspondence between innovation tool types and technique analysis prompt word templates.

[0108] The prompt text acquisition module is used to fill the problem description text into the technique analysis prompt word template to obtain the technique analysis prompt engineering text.

[0109] The analysis results acquisition module is used to infer the technical analysis prompts from the engineering text using a pre-trained model to obtain the technical analysis results of the target product.

[0110] As an optional implementation of the above-mentioned device, the product innovation tool recommendation device based on the pre-trained model further includes: The valid indication judgment module is used to determine whether valid indication information of the technical analysis results of the target product has been received.

[0111] The training data addition module is used to add the design requirement standard text, the corresponding recommendation tool, and the recommendation reason to the fine-tuning training dataset if valid indication information of the technical analysis results of the target product is received.

[0112] As an optional implementation of the above-mentioned device, the product innovation tool recommendation device based on the pre-trained model further includes: The indicator quantity judgment module is used to determine whether the total number of valid indicator information results of the technique analysis is greater than a preset quantity threshold.

[0113] The model fine-tuning training module is used to fine-tune the pre-trained model if the total number of valid indication information from the technique analysis results exceeds a preset threshold. This involves using the standard text of design requirements in the fine-tuning training dataset as training data and the recommended tools and reasons for recommendation in the fine-tuning training dataset as training labels.

[0114] As an optional implementation of the above-mentioned device, the product innovation tool recommendation device based on the pre-trained model further includes: The preset matrix insertion module is used to obtain the target language model and insert a preset matrix into the target language model to obtain a pre-trained model.

[0115] The preset matrix update module is used to update the preset matrix in the pre-trained model while freezing the model parameters of the target language model in the pre-trained model.

[0116] It should be understood that this device corresponds to the above-described embodiment of the product analysis tool innovation tool recommendation method based on pre-trained models, and is capable of performing the various steps involved in the above-described method embodiment. The specific functions of this device can be found in the description above, and detailed descriptions are appropriately omitted here. The device includes at least one software functional module that can be stored in memory or embedded in the device's operating system (OS) in the form of software or firmware.

[0117] Please see Figure 5 The diagram shows a structural schematic of an electronic device provided in an embodiment of this application. An electronic device 400 provided in this application includes a processor 410 and a memory 420. The memory 420 stores machine-readable instructions executable by the processor 410. When the machine-readable instructions are executed by the processor 410, the method described above is performed.

[0118] This application embodiment also provides a computer-readable storage medium 430, on which a computer program is stored. This computer program is executed by a processor 410 to perform the methods described above. The computer-readable storage medium 430 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0119] This application also provides a computer program product, including: a computer program or computer instructions, which are executed by a processor to perform the method described above.

[0120] 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 apparatus embodiments, since they are basically similar to method embodiments, the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0121] It should be understood that the disclosed apparatus and methods can also be implemented in other ways, as provided in the embodiments of this application. The apparatus embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this application. In this regard, each block in the flowchart or block diagram may represent a module, segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the accompanying drawings. For example, two consecutive blocks may actually be executed substantially in parallel, or they may sometimes be executed in reverse order, depending primarily on the functions involved.

[0122] Furthermore, the functional modules of each embodiment in this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part. In addition, in the description of this specification, the reference to terms such as "one embodiment," "some embodiments," "example," "specific example," "some examples," etc., means that the specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in a suitable manner in any one or more embodiments or examples. Furthermore, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0123] The above description is only an optional implementation of the embodiments of this application, but the protection scope of the embodiments of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the embodiments of this application should be covered within the protection scope of the embodiments of this application.

Claims

1. A product innovation tool recommendation method based on a pre-trained model, characterized in that, include: Obtain the standard text of design requirements for the target product; Obtain a list of preset innovation tools, and fill the list of preset innovation tools and the standard text of the design requirements into a preset prompt word template to obtain the prompt engineering text; By reasoning through the prompt engineering text using a pre-trained model, recommended tools and reasons for recommendation are obtained corresponding to the standard design requirement text. The recommended tools are at least one innovative tool from the preset list of innovative tools.

2. The method according to claim 1, characterized in that, The step of filling the preset list of innovative tools and the standard text of design requirements into the preset prompt template includes: For each innovative tool in the preset list of innovative tools, the explanatory definition text of that innovative tool is retrieved from the tool database, which stores the preset list of innovative tools and the explanatory definition of each innovative tool. From the preset list of innovation tools, candidate innovation tools whose similarity between the explanatory definition text and the standard text of the design requirements is greater than a preset threshold are selected. Fill the preset prompt word template with the candidate innovation tools and the standard text of the design requirements.

3. The method according to claim 1, characterized in that, After obtaining the recommended tools and reasons corresponding to the standard text of the design requirements, the method further includes: Regular expressions are used to filter the recommended tools and reasons corresponding to the standard text of the design requirements to obtain structured data; The structured data is sent to the terminal device so that the terminal device can parse the innovative tool type and the reason for recommendation from the structured data.

4. The method according to claim 3, characterized in that, After sending the structured data to the terminal device, the method further includes: Receive the problem description text and the innovative tool type of the recommended tool sent by the terminal device; The technique analysis prompt template corresponding to the innovation tool type is searched in the prompt template library, which stores the correspondence between innovation tool types and technique analysis prompt templates; Fill the problem description text into the technique analysis prompt template to obtain the technique analysis prompt engineering text; The pre-trained model is used to infer the technical analysis prompts from the engineering text to obtain the technical analysis results of the target product.

5. The method according to claim 4, characterized in that, After obtaining the technical analysis results of the target product, the method further includes: Determine whether a valid indication message regarding the technical analysis results of the target product has been received; If so, the design requirement standard text, along with the corresponding recommendation tool and recommendation reason, will be added to the fine-tuning training dataset.

6. The method according to claim 5, characterized in that, After adding the design requirement standard text and the corresponding recommendation tool and reasons to the design requirement standard text to the fine-tuning training dataset, the method further includes: Determine whether the total number of valid indications from the technique analysis results is greater than a preset threshold. If so, the pre-trained model is fine-tuned using the standard text of design requirements in the fine-tuning training dataset as training data and the recommendation tools and reasons for recommendation in the fine-tuning training dataset as training labels.

7. The method according to claim 6, characterized in that, Before fine-tuning the pre-trained model, the method further includes: Obtain the target language model, and insert a preset matrix into the target language model to obtain the pre-trained model; The fine-tuning of the pre-trained model includes: While freezing the model parameters of the target language model in the pre-trained model, update the preset matrix in the pre-trained model.

8. A product innovation tool recommendation system based on a pre-trained model, characterized in that, include: Terminal devices and servers; The terminal device is used to send standard text of design requirements for the target product to the server; The server is used to receive standard text of design requirements for the target product sent by the terminal device. Obtain a list of preset innovation tools, and fill the list of preset innovation tools and the standard text of the design requirements into a preset prompt word template to obtain the prompt engineering text; By reasoning through the prompt engineering text using a pre-trained model, recommended tools and reasons for recommendation are obtained corresponding to the standard design requirement text. The recommended tools are at least one innovative tool from the preset list of innovative tools.

9. An electronic device, characterized in that, include: A processor and a memory, the memory storing machine-readable instructions executable by the processor, the machine-readable instructions being executed by the processor to perform the method of any one of claims 1 to 7.

10. A computer program product, characterized in that, include: A computer program or computer instructions that, when executed by a processor, perform the method according to any one of claims 1 to 7.

Images