An intelligent translation quotation generation method based on a large model

Abstract

The application provides a large model-based intelligent translation quotation generation method, which comprises manuscript uploading and preprocessing, manuscript content multi-dimensional analysis, manuscript basic attribute acquisition, translator resource saturation calculation, market reference data acquisition, comprehensive prompt word construction, large model calling, comprehensive output quotation report, reception of project manager feedback and adjustment, and feedback data-based system optimization; pricing is based on transparency: each single quotation is accompanied by detailed basis, which is convenient for PM to explain to customers and convenient for internal audit; resource allocation optimization: translator resource visualization and prediction function, which avoids the situation that the priced project cannot be executed due to insufficient translators.

Description

Technical Field

[0001] This invention relates to the translation service industry, and more particularly to an intelligent translation quote generation method based on a large model. Background Technology

[0002] Pricing in the translation service industry has long relied on manual judgment. After receiving a client's manuscript, the project manager (PM) needs to analyze multiple dimensions, including the technical difficulty, terminology density, typesetting requirements, translator availability, and market conditions, before providing a quote based on experience. This process typically takes 2-3 hours, and there are significant differences in quotes between different PMs, making it difficult to explain the pricing rationale to the client.

[0003] Traditional automated quotation solutions have significant limitations: rule engines struggle to handle complex and ever-changing manuscript situations, resulting in high maintenance costs; and ordinary machine learning models lack the ability to deeply understand specialized content. The emergence of large language models offers a new technical approach to solving this problem.

[0004] In the traditional translation industry, pricing relies heavily on the personal experience of project managers, leading to the following problems: Low pricing efficiency: Project managers (PMs) need to manually analyze the text, search for translators, and calculate prices, a process that is time-consuming, typically taking 2-3 hours, making it difficult to meet clients' demands for rapid response. Inconsistent pricing standards: Differences in experience and judgment standards among PMs result in inconsistent pricing for similar documents, impacting client experience and the company's pricing strategy. Difficulty in comprehensively considering multiple factors: Translation pricing requires comprehensive consideration of multiple dimensions, including text content, technical difficulty, format, word count, translation direction, typesetting requirements, and translator resource saturation, making it difficult to accurately quantify the weight of each factor manually. Inaccurate translator resource prediction: Traditional methods struggle to accurately predict future translator resource saturation, potentially leading to inadequate consideration of supply and demand in pricing, affecting project delivery and cost control. Lack of transparency in pricing basis: Traditional pricing methods struggle to clearly explain the pricing basis to clients, impacting client trust. High reliance on PM experience: Novice PMs require extensive training to independently provide pricing, increasing corporate training costs.

[0005] Some translation companies attempt to use rule engines or simple machine learning models for pricing, but these methods have the following drawbacks: Rule engines: struggle to handle complex and ever-changing text features, require manual maintenance of numerous rules, lack flexibility, and have high maintenance costs. Traditional machine learning models: require extensive training with labeled data and struggle to understand the semantic content and technical complexity of the text. Lack of intelligent analysis capabilities: unable to deeply understand texts with dense technical terminology and complex sentence structures. Summary of the Invention

[0006] The purpose of this invention is to address the shortcomings of the prior art by providing an intelligent translation quote generation method based on a large model. This method aims to utilize the understanding and reasoning capabilities of the large model to automate the generation of translation quotes, thereby significantly improving efficiency while ensuring quote accuracy.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: This invention provides a method for generating intelligent translation quotes based on a large model, comprising the following steps: S1. Manuscript upload and preprocessing; S2. Multi-dimensional analysis of manuscript content; S3. Obtain the basic attributes of the manuscript; S4. Calculate translator resource saturation; S5. Market reference data acquisition; S6. Construction of comprehensive prompt words; S7, Large Model Call; S8. Comprehensive output quotation report; S9. Receive feedback and adjustments from the project manager; S10, Optimize the system based on feedback data.

[0008] Furthermore, S1 specifically refers to: Clients upload translated documents via a web interface. The intelligent translation pricing system receives the files and records the file name, file type, file size, upload time, client ID, client level, past cooperation history, and client budget range. Based on the file type, it selects an appropriate parser to extract the text content. For complex formats such as PDF and Word, it uses a specialized parsing tool to preserve formatting information and stores the parsed text content in the document database.

[0009] Furthermore, S2 includes: S201, Domain Classification Analysis: Construct domain-specific category suggestions, requiring a large language model to identify the domain to which the manuscript belongs; S202, Professional Difficulty Assessment: Construct professional difficulty assessment prompts and require a large language model to assess the professional difficulty of the manuscript; S203, Format Recognition: Construct format recognition prompts, requiring a large language model to identify special format elements in the manuscript; S204. Content Quality Assessment: Construct content quality assessment prompts and require a large language model to evaluate the content quality of the manuscript.

[0010] Furthermore, S3 specifically refers to: Manuscript size: a quantifiable metric; File type: Identifies file format; Translation focus: Identifying source and target language pairs; Layout requirements: Determine whether layout services are needed and the complexity of the layout; Delivery time requirements: Identify the customer's delivery time requirements.

[0011] Furthermore, S4 includes: S401, Real-time query of translator resources: Query the translator database to obtain the number of currently available translators, categorize the number of available translators by field, and categorize the number of available translators by ability level; query the number of currently ongoing projects and their estimated completion time, and calculate the current translator workload. Current translator load factor = Number of current projects / Number of available translators; S402. Comprehensive assessment of translator saturation: A comprehensive saturation index is calculated based on the current translator workload, ranging from 0 to 100. A higher value indicates a higher saturation. The price is adjusted according to the saturation index: the price is increased when the saturation is high.

[0012] Furthermore, S5 specifically includes: Retrieve historical order reference data from the order system: Historical pricing data: Historical pricing and transaction prices for similar articles; Translator Profiles: Descriptions of translators' abilities, past performance, and rates at different levels; Translation unit price: Baseline translation unit prices for different fields and difficulty levels; Typesetting unit price: The unit price for typesetting services with different typeset complexities; Parameter data is obtained by using a large model and manually inputting market data. Market conditions: industry evaluation prices, competitor prices, and market price fluctuation trends.

[0013] Furthermore, S6 specifically includes: Construct comprehensive prompt words, input all analysis results and data into a large language model, including: Manuscript information: Title, field, size, translation focus; Basic attributes of a manuscript: subject category, level of specialization, format recognition, and content quality; Translator resource saturation: current saturation, future forecast, and comprehensive saturation index; Market reference data: historical quotes, translator unit price, benchmark unit price, market conditions; Pricing rules and constraints: pricing strategies, difficulty coefficients, saturation coefficients, prices, and difficulty limits for different fields; Output format requirements: quotation amount, pricing basis, translator level recommendation, and selected option.

[0014] Furthermore, S7 includes: The intelligent translation pricing system selects a suitable large language model for use, employing a hybrid large model approach: Prioritize using open-source models to process routine articles. For high-difficulty, high-value articles, use commercial APIs to improve accuracy. Set up a model switching strategy: select the model based on article difficulty, client level, and historical conversion rate. Inputting comprehensive prompts into a large language model generates a pricing scheme, including a recommended price, explanation of the pricing basis, suggested translator level, available pricing options, and risk warnings.

[0015] Furthermore, S8 specifically includes: The output of the large language model is formatted into a structured quotation report, which includes: basic manuscript information, content analysis summary, translator resource status, recommended quotation amount, detailed explanation of pricing basis, translator level suggestions and reasons, comparison of 3 selectable quotation options, risk warnings and suggestions, and the quotation report is displayed on a web interface, supporting interactive viewing; Specifically, S9 is: PMs can accept recommended quotes and send them directly to clients; PMs can adjust quote amounts, and the system records the reasons for the adjustments; PMs make personal judgments regarding client budget constraints, long-term cooperation discounts, and handling of special circumstances; PMs can select different quote options and modify translator level recommendations, and all adjustments are recorded in the historical quote database.

[0016] Furthermore, S10 specifically includes: Record the discrepancy between the final transaction price and the quoted price, analyze the reasons for the adjustment, and identify the shortcomings of the pricing strategy; Regularly evaluate the accuracy of quotes: Accuracy = Number of orders with transaction prices within ±10% of the recommended quote / Total number of orders; Conversion rate = Number of completed orders / Total number of quoted orders; Fine-tune the large language model using historical transaction data; Build a training dataset: manuscript features + transaction price; update the model using efficient fine-tuning methods; update the model version regularly; adjust pricing rules and coefficients according to market changes to continuously optimize the performance and accuracy of the intelligent translation pricing system.

[0017] The beneficial effects of this invention are: transparent pricing basis: each quotation comes with detailed basis, which makes it easy for PM to explain to customers and also facilitates internal review; Resource allocation optimization: Translator resource visualization and prediction functions to avoid projects that have already been quoted being unable to be executed due to insufficient translators; Continuous optimization: The transaction data feedback mechanism enables the system to learn market patterns and continuously improve the accuracy of its quotes; Cost controllable: The hybrid model strategy ensures that simple manuscripts do not consume excessive resources, while complex manuscripts receive sufficient intelligent support; Multiple options are available: three pricing options—conservative, standard, and aggressive—to meet the needs of different customers. Reduce training costs: The system provides professional pricing suggestions, allowing even novice PMs to get started quickly; Capability to handle complex content: The large model has a good understanding of manuscripts with dense technical terms and complex sentence structures, which makes up for the shortcomings of the rule engine. Attached Figure Description

[0018] Figure 1 A flowchart of an intelligent translation quotation generation method based on a large model; Figure 2 A flowchart for manuscript analysis; Figure 3 Flowchart for calculating translator saturation and workload; Figure 4 Flowchart for calling large models; Figure 5 Here is a flowchart of the feedback mechanism; Figure 6 This is an intelligent translation quotation generation system. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0020] Please see Figures 1 to 5 A method for generating intelligent translation quotes based on a large model includes the following steps: S1. Manuscript upload and preprocessing; S2. Multi-dimensional analysis of manuscript content; S3. Obtain the basic attributes of the manuscript; S4. Calculate translator resource saturation; S5. Market reference data acquisition; S6. Construction of comprehensive prompt words; S7, Large Model Call; S8. Comprehensive output quotation report; S9. Receive feedback and adjustments from the project manager; S10, Optimize the system based on feedback data.

[0021] Furthermore, S1 specifically refers to: Clients upload translated documents via a web interface. The intelligent translation pricing system receives the files and records the file name, file type, file size, upload time, client ID, client level, historical cooperation information, and client budget range. Based on the file type, it selects an appropriate parser to extract the text content. For complex formats such as PDF and Word, it uses a specialized parsing tool to preserve formatting information and stores the parsed text content in the document database.

[0022] Furthermore, S2 includes: S201, Domain Classification Analysis: Construct domain-specific category suggestions, requiring a large language model to identify the domain to which the manuscript belongs; In a specific embodiment, the prompt word example is as follows: Please analyze the following text and determine its professional field. Optional fields include: medicine, law, technology, finance, literature, business, and others. Please only return the field name; no explanation is required.

[0023] Text content: "[Manuscript text]". The large language model returns domain classification results, such as "medicine", "technology", etc. The system records the domain classification results and confidence scores.

[0024] S202, Professional Difficulty Assessment: Construct professional difficulty assessment prompts and require a large language model to assess the professional difficulty of the manuscript; In a specific embodiment, the prompt word example is as follows: Please assess the level of technical difficulty of the following text.

[0025] The evaluation dimensions include: Terminology density: The proportion of technical terms in a text; Sentence complexity: The length and structural complexity of a sentence; Technical Depth: The technical depth and professionalism of the content. Please rate it on a scale of 1 to 5, with 1 being the easiest and 5 being the most difficult. Return format: Terminology Density: X points, Sentence Complexity: Y points, Technical Depth: Z points, Overall Difficulty: W points.

[0026] Text content: "[Manuscript text]". The large language model returns scores for each dimension and a comprehensive difficulty score. The system records the difficulty assessment results.

[0027] S203, Format Recognition: Construct format recognition prompts, requiring a large language model to identify special format elements in the manuscript; In a specific embodiment, the prompt word example is as follows: Please analyze the following text and identify any special formatting elements it contains. Check for: tables, charts, formulas, images, special formatting marks, hyperlinks, etc. For each formatting element, please count its number and indicate its location. Text content: [Manuscript text]

[0028] The large language model returns a list of format elements and statistical information. The system combines the results from the file parser to comprehensively determine the format complexity.

[0029] S204. Content Quality Assessment: Construct content quality assessment prompts and require a large language model to evaluate the content quality of the manuscript.

[0030] In a specific embodiment, the prompt word example is as follows: Please evaluate the content quality of the following text.

[0031] The evaluation dimensions include: Duplicate content: Does the text contain a large amount of duplicate content? Terminology consistency: Whether the use of technical terms is consistent; Original text quality: Whether the original text is clear and accurate; Translatability: Is the text suitable for translation? Are there any ambiguities? Please provide the evaluation results and suggestions for improvement. Text content: [Manuscript text].

[0032] The large language model returns content quality assessment results and improvement suggestions. The system records the quality assessment results for use in adjusting pricing.

[0033] Furthermore, S3 specifically refers to: Manuscript size: quantifiable metrics; for example: word count, character count, paragraph count, and page count.

[0034] File type: Identifies file format; Translation focus: Identifying source and target language pairs; Layout requirements: Determine whether layout services are needed and the complexity of the layout; Delivery time requirements: Identify the customer's delivery time requirements.

[0035] Furthermore, S4 includes: S401, Real-time query of translator resources: Query the translator database to obtain the number of currently available translators, categorize the number of available translators by field, and categorize the number of available translators by ability level; query the number of currently ongoing projects and their estimated completion time, and calculate the current translator workload. Current translator load factor = Number of current projects / Number of available translators; S402. Comprehensive assessment of translator saturation: A comprehensive saturation index is calculated based on the current translator workload, ranging from 0 to 100. A higher value indicates a higher saturation. The price is adjusted according to the saturation index: the price is increased when the saturation is high.

[0036] Furthermore, S5 specifically includes: Retrieve historical order reference data from the order system: Historical pricing data: Historical pricing and transaction prices for similar articles; Translator Profiles: Descriptions of translators' abilities, past performance, and rates at different levels; Translation unit price: Baseline translation unit prices for different fields and difficulty levels; Typesetting unit price: The unit price for typesetting services with different typeset complexities; Parameter data is obtained by using a large model and manually inputting market data. Market conditions: industry evaluation prices, competitor prices, and market price fluctuation trends.

[0037] Furthermore, S6 specifically includes: Construct comprehensive prompt words, input all analysis results and data into a large language model, including: Manuscript information: Title, field, size, translation focus; Basic attributes of a manuscript: subject category, level of specialization, format recognition, and content quality; Translator resource saturation: current saturation, future forecast, and comprehensive saturation index; Market reference data: historical quotes, translator unit price, benchmark unit price, market conditions; Pricing rules and constraints: pricing strategies, difficulty coefficients, saturation coefficients, prices, and difficulty limits for different fields; Output format requirements: quotation amount, pricing basis, translator level recommendation, and selected option.

[0038] Furthermore, S7 includes: The intelligent translation pricing system selects a suitable large language model for use, employing a hybrid large model approach: Prioritize using open-source models to process routine articles. For high-difficulty, high-value articles, use commercial APIs to improve accuracy. Set up a model switching strategy: select the model based on article difficulty, client level, and historical conversion rate. Specifically, the open-source models are LLaMA, ChatGLM, and DeepSeek.

[0039] Inputting comprehensive prompts into a large language model generates a pricing scheme, including a recommended price, explanation of the pricing basis, suggested translator level, available pricing options, and risk warnings.

[0040] Furthermore, S8 specifically includes: The output of the large language model is formatted into a structured quotation report, which includes: basic manuscript information, content analysis summary, translator resource status, recommended quotation amount, detailed explanation of pricing basis, translator level suggestions and reasons, comparison of 3 selectable quotation options, risk warnings and suggestions, and the quotation report is displayed on a web interface, supporting interactive viewing; Specifically, S9 is: PMs can accept recommended quotes and send them directly to clients; PMs can adjust quote amounts, and the system records the reasons for the adjustments; PMs make personal judgments regarding client budget constraints, long-term cooperation discounts, and handling of special circumstances; PMs can select different quote options and modify translator level recommendations, and all adjustments are recorded in the historical quote database.

[0041] Furthermore, S10 specifically includes: Record the discrepancy between the final transaction price and the quoted price, analyze the reasons for the adjustment, and identify the shortcomings of the pricing strategy; Regularly evaluate the accuracy of quotes: Accuracy = Number of orders with transaction prices within ±10% of the recommended quote / Total number of orders; Conversion rate = Number of completed orders / Total number of quoted orders; Fine-tune the large language model using historical transaction data; Build a training dataset: manuscript features + transaction price; update the model using efficient fine-tuning methods; update the model version regularly; adjust pricing rules and coefficients according to market changes to continuously optimize the performance and accuracy of the intelligent translation pricing system.

[0042] like Figure 6 As shown, an intelligent translation quotation generation method based on a large model is implemented using an intelligent translation quotation generation system, which includes nine modules: manuscript receiving, content analysis, attribute acquisition, translator resource calculation, market reference data acquisition, comprehensive prompt word construction, large model invocation, quotation generation, and feedback optimization.

[0043] This invention uses large language model technology to comprehensively analyze multiple factors such as manuscript content, professional difficulty, format characteristics, and translator resource saturation, automatically generating accurate and reasonable translation service quotes, and providing pricing basis and translator level suggestions.

[0044] This invention mainly solves four major problems: Efficiency: Reduces quotation time from hours to minutes, significantly improving work efficiency.

[0045] Standardization: Based on a unified multi-dimensional analysis model and pricing strategy, ensure consistency in pricing for similar manuscripts.

[0046] Comprehensive: The system takes into account multiple factors to generate a more accurate and reasonable quote.

[0047] Intelligence: Accurately assesses translator resource saturation through a combination of real-time querying and predictive analysis.

[0048] Core innovations: Multi-dimensional comprehensive analysis: Unlike traditional pricing methods that are based solely on word count, this system comprehensively considers multiple dimensions such as manuscript content analysis, field classification, format complexity, translator resource availability, and market conditions, and provides a price quote based on a comprehensive judgment by a large model; Hybrid model invocation strategy: Automatically select the model type based on the difficulty and value of the manuscript—use open source models for simple manuscripts to control costs, and call commercial APIs for complex manuscripts to ensure accuracy, thus achieving a balance between cost and effectiveness; Translator resource prediction mechanism: Not only can the current status of translators be queried in real time, but also the translator saturation in the future can be predicted based on historical order data and industry trends, and the supply and demand relationship can be taken into account in pricing. Feedback and learning capabilities: The system records data for each quote, transaction, and price adjustment, regularly analyzes the transaction rate and quote deviation, and continuously optimizes pricing strategies, making the system more and more accurate with the accumulation of usage time.

[0049] The embodiments described above are merely illustrative of implementation methods of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be defined by the appended claims.

Claims

1. A method for generating intelligent translation quotes based on a large model, characterized in that, Includes the following steps: S1. Manuscript upload and preprocessing; S2. Multi-dimensional analysis of manuscript content; S3. Obtain the basic attributes of the manuscript; S4. Calculate translator resource saturation; S5. Market reference data acquisition; S6. Construction of comprehensive prompt words; S7, Large Model Call; S8. Comprehensive output quotation report; S9. Receive feedback and adjustments from the project manager; S10, Optimize the system based on feedback data.

2. The intelligent translation quotation generation method based on a large model according to claim 1, characterized in that, Specifically, S1 is: Clients upload translated documents via a web interface. The intelligent translation pricing system receives the files and records the file name, file type, file size, upload time, client ID, client level, historical cooperation information, and client budget range. Based on the file type, it selects an appropriate parser to extract the text content. For complex formats such as PDF and Word, it uses a specialized parsing tool to preserve formatting information and stores the parsed text content in the document database.

3. The intelligent translation quotation generation method based on a large model according to claim 2, characterized in that, S2 includes: S201, Domain Classification Analysis: Construct domain-specific category suggestions, requiring a large language model to identify the domain to which the manuscript belongs; S202, Professional Difficulty Assessment: Construct professional difficulty assessment prompts and require a large language model to assess the professional difficulty of the manuscript; S203, Format Recognition: Construct format recognition prompts, requiring a large language model to identify special format elements in the manuscript; S204. Content Quality Assessment: Construct content quality assessment prompts and require a large language model to evaluate the content quality of the manuscript.

4. The intelligent translation quotation generation method based on a large model according to claim 3, characterized in that, Specifically, S3 is: Manuscript size: a quantifiable metric; File type: Identifies file format; Translation focus: Identifying source and target language pairs; Layout requirements: Determine whether layout services are needed and the complexity of the layout; Delivery time requirements: Identify the customer's delivery time requirements.

5. The intelligent translation quotation generation method based on a large model according to claim 4, characterized in that, S4 includes: S401, Real-time query of translator resources: Query the translator database to obtain the number of currently available translators, categorize the number of available translators by field, and categorize the number of available translators by ability level; query the number of currently ongoing projects and their estimated completion time, and calculate the current translator workload. Current translator load factor = Number of current projects / Number of available translators; S402. Comprehensive assessment of translator saturation: A comprehensive saturation index is calculated based on the current translator workload, ranging from 0 to 100. A higher value indicates a higher saturation. The price is adjusted according to the saturation index: the price is increased when the saturation is high.

6. The intelligent translation quotation generation method based on a large model according to claim 5, characterized in that, Specifically, S5 is: Retrieve historical order reference data from the order system: Historical pricing data: Historical pricing and transaction prices for similar articles; Translator Profiles: Descriptions of translators' abilities, past performance, and rates at different levels; Translation unit price: Baseline translation unit prices for different fields and difficulty levels; Typesetting unit price: The unit price for typesetting services with different typeset complexities; Parameter data is obtained by using a large model and manually inputting market data. Market conditions: industry evaluation prices, competitor prices, and market price fluctuation trends.

7. The intelligent translation quotation generation method based on a large model according to claim 6, characterized in that, Specifically, S6 is: Construct comprehensive prompt words, input all analysis results and data into a large language model, including: Manuscript information: Title, field, size, translation focus; Basic attributes of a manuscript: subject category, level of specialization, format recognition, and content quality; Translator resource saturation: current saturation, future forecast, and comprehensive saturation index; Market reference data: historical quotes, translator unit price, benchmark unit price, market conditions; Pricing rules and constraints: pricing strategies, difficulty coefficients, saturation coefficients, prices, and difficulty limits for different fields; Output format requirements: quotation amount, pricing basis, translator level recommendation, and selected option.

8. The intelligent translation quotation generation method based on a large model according to claim 7, characterized in that, S7 includes: The intelligent translation pricing system selects a suitable large language model for use, employing a hybrid large model approach: Prioritize using open-source models to process routine articles. For high-difficulty, high-value articles, use commercial APIs to improve accuracy. Set up a model switching strategy: select the model based on article difficulty, client level, and historical conversion rate. Inputting comprehensive prompts into a large language model generates a pricing scheme, including a recommended price, explanation of the pricing basis, suggested translator level, available pricing options, and risk warnings.

9. The intelligent translation quotation generation method based on a large model according to claim 8, characterized in that, Specifically, S8 is: The output of the large language model is formatted into a structured quotation report, which includes: basic manuscript information, content analysis summary, translator resource status, recommended quotation amount, detailed explanation of pricing basis, translator level suggestions and reasons, comparison of 3 selectable quotation options, risk warnings and suggestions, and the quotation report is displayed on a web interface, supporting interactive viewing; Specifically, S9 is: PMs can accept recommended quotes and send them directly to clients; PMs can adjust quote amounts, and the system records the reasons for the adjustments; PMs make personal judgments regarding client budget constraints, long-term cooperation discounts, and handling of special circumstances; PMs can select different quote options and modify translator level recommendations, and all adjustments are recorded in the historical quote database.

10. The intelligent translation quotation generation method based on a large model according to claim 9, characterized in that, Specifically, S10 is: Record the discrepancy between the final transaction price and the quoted price, analyze the reasons for the adjustment, and identify the shortcomings of the pricing strategy; Regularly evaluate the accuracy of quotes: Accuracy = Number of orders with transaction prices within ±10% of the recommended quote / Total number of orders; Conversion rate = Number of completed orders / Total number of quoted orders; Fine-tune the large language model using historical transaction data; Build a training dataset: manuscript features + transaction price; update the model using efficient fine-tuning methods; update the model version regularly; adjust pricing rules and coefficients according to market changes to continuously optimize the performance and accuracy of the intelligent translation pricing system.

Images