Economic analysis report generation method and device based on large model context protocol

Abstract

The invention discloses an economic analysis report generation method and device based on a large model context agreement, and the method comprises the steps: introducing the large model context agreement, carrying out the standardized definition of the large model context agreement through a special protocol description language, and enabling the protocol to structurally package all elements (a data source, a knowledge base, constraints and steps) needed by the generation of a report; and then, automatically converting the precisely defined protocol into a Prompt template which can be understood by the large model and is pre-input with verified data through a protocol analyzer, so as to replace a complex and unstable manual prompt word project, thereby realizing effective standardization, precise control and automation of the process of generating the economic analysis report by the large model. And the quality, consistency and efficiency of the final report are ensured.

Description

Technical Field

[0001] The present application relates to the field of intelligent generation, and more specifically, to a method and device for generating an economic analysis report based on a large model context protocol. Background Art

[0002] Economic analysis reports play a vital role in modern business decision-making, policy-making, and market research. They provide business managers, government agencies, and investors with key insights into macroeconomic trends, industry dynamics, and specific market conditions, and serve as an important basis for strategic planning, risk assessment, and resource allocation. However, traditionally, writing a high-quality economic analysis report typically requires a significant investment of human resources and time, involving complex data collection, cleaning, processing, analysis, and the application of professional knowledge. This is not only inefficient and costly, but also difficult to ensure the timeliness of the report and the consistency of the analytical caliber. Especially when dealing with massive, multi-source, heterogeneous data, the difficulty and risk of error in manual processing increase significantly. Therefore, seeking automated and intelligent economic analysis report generation solutions to improve efficiency, reduce costs, and ensure report quality has become an urgent need in the industry.

[0003] The rapid development of artificial intelligence (AI), particularly large language models (LLMs), has opened up new possibilities for automated economic analysis report generation thanks to their powerful natural language understanding and generation capabilities. Some attempts have involved directly leveraging general-purpose large models for report writing, with users describing their needs in natural language and expecting the model to directly output the report. However, this direct application approach often faces numerous challenges. First, for rigorously structured and logical economic analysis reports, simple, open-ended natural language instructions struggle to accurately and comprehensively convey the complex elements required, such as specific data sources, analytical dimensions, required constraints, and detailed generation steps. Second, large models have limited capabilities for processing and integrating multi-source, real-time data, making it difficult to ensure the accuracy and timeliness of input data. Furthermore, their generation process lacks sufficient transparency and control, resulting in outputs that may not meet expectations or even contain erroneous information or "illusions." Furthermore, relying directly on manually written, complex, and sophisticated prompts to guide large models is inherently time-consuming and highly skilled, making it difficult to standardize and scale, and ensuring the stability and consistency of generated reports.

[0004] Therefore, an optimized intelligent generation solution for economic analysis reports is expected. Summary of the Invention

[0005] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides an economic analysis report generation method based on a large model context protocol, which introduces a large model context protocol and standardizes it through a special protocol description language. The protocol structurally encapsulates all the elements required for report generation (data source, knowledge base, constraints, steps); then, the protocol parser automatically converts this precisely defined protocol into a prompt template that can be understood by the large model and has pre-verified data, thereby replacing the complex and unstable manual prompt word engineering, thereby achieving effective standardization, precise control and automation of the process of generating economic analysis reports for large models, ensuring the quality, consistency and efficiency of the final report.

[0006] According to one aspect of the present application, a method for generating an economic analysis report based on a large model context protocol is provided, which includes: Based on the reporting requirements, a protocol description language is used to define the big model context protocol, which includes data source, knowledge base, constraints and generation steps. Based on the data source defined in the big model context protocol, connect to the specified database or API through the multi-source protocol adapter to capture the required data, and perform data quality verification on the required data based on the rules in the big model context protocol to obtain verified data; The protocol parser converts the large model context protocol into a Prompt template that can be parsed by the large model, and inserts the verified data into a predetermined position of the Prompt template to obtain a Prompt template with input data; wherein the protocol parser converts the large model context protocol into a Prompt template that can be parsed by the large model; The Prompt template with input data is input into the economic analysis report generation engine based on the large model to obtain the economic analysis report.

[0007] According to another aspect of the present application, there is provided an economic analysis report generation device based on a large model context protocol, comprising: A large model context protocol definition module is used to define a large model context protocol using a protocol description language based on report requirements. The large model context protocol includes data sources, knowledge bases, constraints, and generation steps. The data quality verification module is used to connect to the specified database or API through the multi-source protocol adapter based on the data source defined in the large model context protocol to capture the required data, and perform data quality verification on the required data based on the rules in the large model context protocol to obtain verified data; A prompt template generation module is used for the protocol parser to convert the large model context protocol into a prompt template that can be parsed by the large model, and insert the verified data into the predetermined position of the prompt template to obtain a prompt template with input data; The economic analysis report generation module is used to input the prompt template with input data into the economic analysis report generation engine based on the large model to obtain the economic analysis report.

[0008] Compared with the existing technology, the present application provides a method for generating an economic analysis report based on a large model context protocol. The method introduces a large model context protocol and standardizes it through a special protocol description language. The protocol structuredly encapsulates all the elements required for report generation (data source, knowledge base, constraints, steps); then, the protocol parser automatically converts this precisely defined protocol into a prompt template that can be understood by the large model and has pre-verified data, thereby replacing the complex and unstable manual prompt word engineering, thereby achieving effective standardization, precise control and automation of the process of generating economic analysis reports for large models, ensuring the quality, consistency and efficiency of the final report. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] The above and other purposes, features, and advantages of the present application will become more apparent through a more detailed description of the embodiments of the present application in conjunction with the accompanying drawings. The accompanying drawings are intended to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0010] Figure 1 A flowchart of a method for generating an economic analysis report based on a large model context protocol according to an embodiment of the present application; Figure 2 A data flow diagram of a method for generating an economic analysis report based on a large model context protocol according to an embodiment of the present application; Figure 3 This is a flowchart of sub-step S3 of the method for generating an economic analysis report based on a large model context protocol according to an embodiment of the present application; Figure 4 This is a flowchart of sub-step S32 of the method for generating an economic analysis report based on a large model context protocol according to an embodiment of the present application; Figure 5 This is a block diagram of an economic analysis report generation device based on a large model context protocol according to an embodiment of the present application. DETAILED DESCRIPTION

[0011] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described herein.

[0012] As used in this application and the claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" are not intended to refer to the singular but may include the plural. Generally speaking, the terms "comprises" and "include" only indicate the inclusion of the steps and elements specifically identified, and these steps and elements do not constitute an exclusive list. A method or apparatus may also include other steps or elements.

[0013] Although the present application makes various references to certain modules in the system according to embodiments of the present application, any number of different modules can be used and run on the user terminal and / or server. The modules are illustrative only, and different aspects of the system and method can use different modules.

[0014] Flowcharts are used in this application to illustrate the operations performed by the systems according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, the various steps may be processed in reverse order or simultaneously, as needed. Furthermore, other operations may be added to these processes, or one or more operations may be removed from these processes.

[0015] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described herein.

[0016] In the technical solution of the present application, a method for generating an economic analysis report based on a large model context protocol is proposed. Figure 1 A flowchart of a method for generating an economic analysis report based on a large model context protocol according to an embodiment of the present application. Figure 2 This is a data flow diagram of the method for generating an economic analysis report based on a large model context protocol according to an embodiment of the present application. Figure 1 and Figure 2As shown, the economic analysis report generation method based on the big model context protocol according to the embodiment of the present application includes the following steps: S1, based on the report requirements, the big model context protocol is defined using a protocol description language, and the big model context protocol includes a data source, a knowledge base, constraints and generation steps; S2, based on the data source defined in the big model context protocol, connecting to the specified database or API through a multi-source protocol adapter to capture the required data, and performing data quality verification on the required data based on the rules in the big model context protocol to obtain verified data; S3, the protocol parser converts the big model context protocol into a Prompt template that can be parsed by the big model, and inserts the verified data into a predetermined position of the Prompt template to obtain a Prompt template with input data; wherein, the protocol parser converts the big model context protocol into a Prompt template that can be parsed by the big model; S4, inputting the Prompt template with input data into the economic analysis report generation engine based on the big model to obtain an economic analysis report.

[0017] Specifically, S1 uses a protocol description language to define the big model context protocol based on reporting requirements. The big model context protocol includes data sources, knowledge bases, constraints, and generation steps. This includes the report's topic (e.g., macroeconomic trends, industry segmentation analysis, corporate financial health, etc.), as well as the intended readership (e.g., management, investors, regulators), data granularity and timeliness requirements (e.g., quarterly data vs. real-time data), desired presentation format (text analysis, chart visualization, or structured summary), compliance standards (e.g., accounting standards or information disclosure regulations), and specific requirements for conclusion reliability. It is worth mentioning that the required data sources are clearly specified in the large model context protocol, such as specifying access to a country's statistical bureau database or a specific financial API; at the same time, relevant knowledge bases are included so that the large model can retrieve authoritative background materials or historical cases as support when generating content; in addition, the constraints must be listed in detail, including but not limited to data timeliness (ensuring that all referenced information is the latest or within a specified period), model coverage (limiting analysis dimensions and depth, and preventing deviations from the topic), chart requirements (such as the need to include a year-on-year growth trend line chart or a specific type of bar chart), and compliance clauses (ensuring that the output content complies with laws, regulations, and industry standards). In the technical solution of this application, by defining the large model context protocol using a protocol description language based on reporting requirements, the system can minimize manual intervention and subjective errors, improve the system's adaptability to diverse needs in complex scenarios, and at the same time strengthen the consistency, professionalism, and credibility of the output results through a constraint mechanism, thereby achieving end-to-end automatic, high-quality economic analysis report generation.

[0018] In one example, for example, by clearly defining data timeliness constraints, it is possible to effectively prevent expired data from being mixed into the analysis process, thereby improving the novelty and reference value of the conclusions; by setting coverage and compliance, it is possible to avoid irrelevant redundant content and illegal statements, so that the final generated report not only focuses on key issues but also complies with legal and industry regulatory requirements, significantly improving the response efficiency and output quality of the economic analysis report generation system driven by large models to different business scenarios and dynamically changing needs.

[0019] Specifically, S2, based on the data source defined in the big model context protocol, connects to the designated database or API via a multi-source protocol adapter to capture the required data. The data quality of the required data is then verified based on the rules in the big model context protocol to produce verified data. A multi-source protocol adapter is a middleware component that can interface with data interfaces of different types and formats (such as SQL databases, RESTful APIs, CSV files, and even web crawlers). It flexibly parses various interface protocol specifications and dynamically loads corresponding drivers or parsing templates based on the data requirements preset in the big model context protocol, enabling integrated access to heterogeneous data sources. For example, in practical applications, if it is necessary to simultaneously obtain macroeconomic indicators (such as GDP growth rate), industry production, sales, and inventory data, and financial market trends, a multi-source protocol adapter can automatically identify the data interface corresponding to each data type and unify the data capture logic for each data type, efficiently integrating information previously dispersed across different platforms or systems into a single processing flow. This not only significantly improves data collection efficiency but also lays a solid foundation for subsequent standardized processing and in-depth analysis. However, simply achieving multi-channel, efficient data capture is far from enough. Because raw economic data often contains issues such as missing fields, abnormal values, and even delayed collection times, if it is directly input into the big model without screening, it will inevitably introduce noise information, affecting the accuracy and authority of the report generation results. Therefore, in the technical solution of this application, the required data is verified for data quality based on the rules in the big model context protocol to obtain verified data.

[0020] In one example, the required fields and their reasonable intervals are first clarified according to the agreement, and missing items or data exceeding the threshold range are eliminated or corrected; secondly, for information from multiple channels but pointing to the same indicator, authoritative sources are given priority through cross-comparison and consistency verification, and algorithms such as weighted average can be used to eliminate abnormal fluctuations; in addition, the collection time and source of each piece of data can be recorded through a metadata tagging mechanism to achieve traceability and transparent management in the subsequent generation process. In this way, the system can ensure that the data entering the Prompt template generation link is strictly screened, highly reliable, and structurally standardized. This not only significantly improves the professional level and consistency of the content of the economic analysis report, but also effectively reduces the risk of "illusion" of large models caused by unstable underlying inputs, and fundamentally guarantees the actual decision-making reference value of automated output results for various user groups such as corporate managers, investors and even regulators.

[0021] In particular, in S3, the protocol parser converts the large model context protocol into a prompt template that can be parsed by the large model, and inserts the verified data into a predetermined position in the prompt template to obtain a prompt template with the input data. The protocol parser converts the large model context protocol into a prompt template that can be parsed by the large model. It should be understood that economic analysis reports inherently place high demands on structural rigor, logical integrity, and data accuracy. Traditional approaches that rely on manually written prompts or direct natural language input often struggle to accurately express complex, multi-dimensional, multi-constrained, and step-by-step task instructions, leading to problems such as disorganized structure, omission of key elements, and semantic deviation in the generated content. Furthermore, relying solely on the large model's inherent understanding capabilities, in a multi-source, heterogeneous, and dynamically changing data environment, it is difficult to ensure the consistency of content with facts and the reliability of analytical conclusions. Therefore, in the technical solution of this application, a protocol parser is used to convert the large model context protocol into a prompt template that can be parsed by the large model. The protocol parser can automatically identify and decompose the various tasks and constraints in the protocol, converting them into a prompt template that the large model can accurately understand and execute. This process not only involves breaking down the overall task flow but also breaking down each analysis step into specific, controllable prompt statements. Embedded coding technology is used to optimize the robustness of semantic expression, ensuring that each instruction can be processed stably and efficiently by the large model. Crucially, after the prompt template framework is constructed, the protocol parser accurately inserts verified data into the preset locations within the prompt template based on data capture and quality verification results. This mechanism significantly enhances the data-driven nature of the generated content. On the one hand, it ensures that the core conclusions and chart analysis in the report are based on reliable data, effectively preventing the "illusions" of the large model caused by uncertain or erroneous inputs. On the other hand, it also enables traceability and transparent management of information throughout the entire process, from data collection to text generation, making the final output more trustworthy and auditable. This approach not only significantly improves the adaptability of the automatic economic analysis report generation system to the diverse needs of complex business scenarios, but also effectively addresses the problems of human subjectivity, instruction ambiguity, and inefficiency inherent in traditional methods.

[0022] In a specific example of this application, Figure 3As shown, the S3 includes: S31, decomposing the task in the large model context protocol into multiple steps, and generating a Prompt statement for each step to obtain a sequence distribution of Prompt statements as an initial Prompt template; S32, performing feature-level semantic expression robustness optimization on each Prompt statement in the sequence distribution of Prompt statements to obtain a semantic sequence distribution of optimized Prompt statements; S33, obtaining a Prompt template that can be parsed by the large model based on the semantic sequence distribution of the optimized Prompt statements.

[0023] Specifically, the S31 decomposes the tasks in the large model context protocol into multiple steps, and generates a Prompt statement for each step to obtain a sequence distribution of Prompt statements as the initial Prompt template. It should be understood that the economic analysis report itself is a highly structured, logically rigorous and content-rich professional document, which usually contains multiple organic components such as data summary, trend analysis, chart display, risk warning, conclusion and suggestions. Directly relying on a single instruction or extensive natural language description makes it difficult for the large model to accurately understand and orderly execute complex business needs, which can easily lead to a chaotic report content structure, missing key points or logical jumps. Therefore, in an embodiment of the present application, first, the tasks in the large model context protocol are decomposed into multiple steps, and a Prompt statement with clear semantics and specific goals is generated for each step, so that each instruction can accurately guide the large model to complete the corresponding subtask. This step-by-step design not only improves the information density and execution controllability of Prompt expression, but also effectively reduces the risk of misjudgment caused by a single instruction that is too long or ambiguous. In this way, the system can automatically construct initial prompt templates adapted to the needs of different scenarios based on the protocol definition, laying a solid foundation for subsequent robustness optimization and high-quality report generation. At the same time, this standardized, multi-step, multi-instruction collaboratively driven large-scale model calling method significantly improves the automation system's adaptability to the diverse needs of complex business scenarios, achieving consistency, standardization, and efficiency throughout the entire process of economic analysis reports, from data collection to intelligent writing. This effectively solves the pain points of information fragmentation, inefficiency, and difficult quality control in traditional manual operations, providing more reliable and insightful data support for business managers and decision makers.

[0024] Specifically, the S32 performs feature-level semantic expression robustness optimization on each Prompt statement in the sequence distribution of Prompt statements to obtain an optimized semantic sequence distribution of Prompt statements. In a specific example of the present application, Figure 4As shown, the S32 includes: S321, performing contextual semantic encoding on each Prompt statement in the sequence distribution of the Prompt statement to obtain a sequence distribution of the context Prompt statement semantic embedding coding vector; S322, performing local semantic significant optimization on each context Prompt statement semantic embedding coding vector in the sequence distribution of the context Prompt statement semantic embedding coding vector to obtain a sequence distribution of optimized context Prompt statement semantic embedding coding vectors as the semantic sequence distribution of the optimized Prompt statement.

[0025] More specifically, the S321 performs contextual semantic encoding on each Prompt statement in the sequence distribution of the Prompt statement to obtain a sequence distribution of the contextual Prompt statement semantic embedding coding vector. That is, in an embodiment of the present application, first, the Prompt statement embedding matrix is used to pass each Prompt statement in the sequence distribution of the Prompt statement through a word embedding encoder based on Word2Vec to obtain a sequence distribution of the Prompt statement semantic embedding coding vector. It should be understood that the generation of economic analysis reports involves a large number of structured, multi-level, and highly professional information instructions. Each Prompt statement not only carries specific data requests, analysis tasks or constraints, but also needs to form a close logical association with other steps. If these instructions are expressed only in traditional character strings or discrete symbols, it is difficult to capture the deep semantic connections and context dependencies contained therein, thereby affecting the accurate understanding and execution of the overall task process by the large model.

[0026] As a classic and efficient word vector learning algorithm, Word2Vec's core idea is to map each word into a low-dimensional continuous vector, so that words with similar context or co-occurrence relationship in the real language environment have a closer vector distance. Specifically, it uses large-scale text data in an unsupervised way to automatically learn the potential distributed representation between words. This representation can not only reflect the information of the word itself, but also effectively capture the complex and subtle semantic and syntactic relationship between it and other words. Therefore, in the technical solution of the present application, each Prompt statement in the sequence distribution of the Prompt statement is embedded in the matrix through a word embedding encoder based on Word2Vec to obtain the sequence distribution of the Prompt statement semantic embedding coding vector. The obtained sequence distribution of the Prompt statement semantic embedding coding vector not only significantly improves the model's perception of the intrinsic connection between different task links and the information flow path, but also provides a solid data foundation for operations such as local saliency optimization and robustness enhancement. Embedding the Prompt sequence through Word2Vec not only ensures the comparability and integration of instructions in each step, but also greatly enhances the protocol parser's ability to process fine-grained information structure and perform deep logical modeling in scenarios with diverse economic analysis tasks and high professional requirements.

[0027] Furthermore, the sequence distribution of the semantic embedding encoding vectors of the Prompt statements is encoded based on the task step context semantic association of the bidirectional LSTM model to obtain the sequence distribution of the contextual Prompt statement semantic embedding encoding vectors. It should be understood that in the automatic generation scenario of economic analysis reports based on large-scale model context protocols, economic analysis tasks usually involve multiple steps, strong logical dependencies, and complex information flows. Each Prompt statement not only carries independent subtask instructions, but also has close semantic associations and data dependencies with the previous and next steps. If the Prompt statements are processed one by one in isolation, it is difficult to capture the global context information, and it is impossible to reflect the logical progression and mutual constraints between the various links, which will directly affect the coherence, professionalism, and reasoning depth of the final report content.

[0028] Bidirectional LSTM is an improved recurrent neural network structure. Its core idea is to simultaneously introduce information flows in both the forward (from front to back) and reverse (from back to front) directions. When processing sequence data, it can not only capture the dependency between the current element and the previous history, but also synchronously model the possible connection between the current element and the future link. The LSTM unit effectively solves the problem of gradient vanishing or exploding in traditional RNN in long-distance dependency modeling through a gating mechanism, enabling the model to memorize and dynamically adjust important information for a long time. Therefore, in the technical solution of the present application, the sequence distribution of the Prompt statement semantic embedding encoding vector is encoded based on the task step context semantic association of the bidirectional LSTM model to obtain the sequence distribution of the context Prompt statement semantic embedding encoding vector. By encoding the task step context semantic association of the bidirectional LSTM model based on the task step context semantic association of the bidirectional LSTM model for the sequence distribution of the Prompt statement semantic embedding encoding vector, the system can fully perceive the implicit data flow, causal reasoning and constraint transfer between each subtask in the entire economic analysis process. For example, the data output during the "macro data collection" phase directly impacts subsequent steps like "trend analysis" and "chart generation," while "compliance review" requires a comprehensive assessment based on all preceding content. Through Bi-LSTM, each prompt instruction carries the information required for the complete process, ensuring that the generated large model input template maintains both fine-grained differentiation and overall consistency, ensuring a closed-loop business logic. This mechanism provides a solid foundation for subsequent deep feature processing, such as robustness optimization and significance reshaping. This ensures that the resulting economic analysis report is more aligned with actual business needs, more professional, credible, and highly interpretable.

[0029] More specifically, S322 performs local semantic optimization on each contextual prompt statement semantic embedding vector in the sequence distribution of contextual prompt statement semantic embedding vectors to obtain a sequence distribution of optimized contextual prompt statement semantic embedding vectors as the optimized semantic sequence distribution of prompt statements. It should be understood that while the contextual prompt statement semantic embedding vectors generated by the protocol parser capture the correlation between task steps through a bidirectional LSTM, they still face challenges in practical applications such as semantic noise interference and inaccurate feature representation. Specifically, the contextual prompt statement semantic embedding vectors carry redundant semantic components (such as residual features of non-critical constraints) and noise interference (such as semantic shifts caused by dynamic adjustments to protocol parameters) during cross-step propagation. This can cause the generation engine to lose focus when parsing prompts, leading to structural redundancy and weakened logic in the report content. For example, when processing multi-source heterogeneous economic data, the semantic encoding vectors of different protocol steps may contain outdated statistical caliber features or conflicting model coverage parameters. Without local semantic purification, the causal reasoning accuracy of subsequent generation steps will be directly affected. Therefore, in the technical solution of the present application, local semantic significant optimization is performed on each context Prompt statement semantic embedding coding vector in the sequence distribution of the context Prompt statement semantic embedding coding vector to obtain the sequence distribution of the optimized context Prompt statement semantic embedding coding vector as the semantic sequence distribution of the optimized Prompt statement.

[0030] Specifically, a sliding convolution window is first used to capture key local patterns within the semantic embedding encoding vector of the contextual prompt statement (such as the coupling relationship between specific constraints). Semantic compression is then used to remove redundant dimensions (such as eliminating the interference of outdated data timeliness tags). Finally, a gain operator is used to dynamically reshape the weight distribution of semantic units. The semantic embedding encoding vector of the contextual prompt statement, after local semantic optimization, is able to withstand the cascading propagation of multi-source data noise and accurately map protocol parameter adjustments (such as shortening the data timeliness window) to adaptive scaling of semantic granularity. Specifically, when fields are missing from external data sources, the optimized semantic embedding encoding vector of the contextual prompt statement automatically identifies the core semantic units of key economic indicators through feature effective component statistics, and uses a phase reshaping gain operator to enhance their expression strength, allowing the generation engine to maintain the integrity of the core argument even in missing data scenarios. At the same time, this mechanism ensures that multi-step prompt statements form a topologically coherent enhanced vector field in the semantic space by maintaining the directional symmetry of the characteristic phase, thereby avoiding the decoupling problem of formatting instructions and content generation commonly seen in traditional methods (such as the misalignment of chapter titles and data analysis paragraphs), and ultimately outputting high-quality economic analysis reports that are both professional and insightful.

[0031] More specifically, first, the semantic embedding coding vectors of each context prompt statement are reconstructed into semantic units based on one-dimensional convolutional coding to obtain the sequence distribution of the semantic local unit coding vectors of the context prompt statement. It should be understood that when the protocol parser breaks down the macroeconomic analysis task into multi-step prompt statements, it is difficult for traditional global semantic coding to effectively capture the key constraint boundaries and logical association patterns within the instructions. For example, the implicit coupling relationship between data screening rules and compliance verification requirements is easily weakened. Therefore, in the technical solution of the present application, the semantic embedding coding vectors of each context prompt statement are reconstructed into semantic units based on one-dimensional convolutional coding to obtain the sequence distribution of the semantic local unit coding vectors of the context prompt statement.

[0032] That is, through the sliding window mechanism of the one-dimensional convolution kernel, atomic semantic units with economic analysis value are decoupled from the high-dimensional semantic embedding vector. In this process, the convolution kernel dynamically perceives the phase changes of adjacent semantic dimensions during the sliding process, and reconstructs the originally diffuse semantic features into local units with clear functional orientations, such as data verification units and analysis model call units. Through the collaborative operation of multi-scale convolution kernel groups, the integrity of the macro-instruction framework is preserved, and the inherent topological structure of the micro-constraint conditions is refined. This reconstruction mechanism enables the semantic unit to adaptively distinguish the weight distribution of core economic indicators and auxiliary parameters, thereby providing a clearly structured intermediate feature representation for subsequent semantic compression and gain reshaping, ensuring that the generation engine accurately parses the multi-dimensional constraint logic implicit in the protocol, and ultimately outputs an economic analysis report that meets professional standards and has decision-making value.

[0033] In a specific example of the present application, the semantic unit reconstruction of each context prompt sentence semantic embedding coding vector based on one-dimensional convolutional coding is performed using the following semantic reconstruction formula to obtain a sequence distribution of the context prompt sentence semantic local unit coding vector; wherein, the semantic reconstruction formula is: ;in, is the contextual Prompt sentence semantic embedding encoding vector, It is a one-dimensional convolutional coding process. is the semantic unit reconstruction step, are the first, second, and third in the sequence distribution of the semantic local unit encoding vector of the context prompt sentence. and The semantic local unit encoding vector of the context prompt sentence.

[0034] Next, semantic compression is performed on each context prompt statement semantic local unit encoding vector in the sequence distribution of the context prompt statement semantic local unit encoding vector to obtain a sequence distribution of the context prompt statement semantic compressed local unit encoding vector. It should be understood that after the prompt statement sequence converted by the protocol in the economic analysis scenario is reconstructed by one-dimensional convolution, the local semantic unit still has information redundancy and noise interference problems. Specifically, when the protocol parser decomposes a multi-step task into discrete semantic units (such as instruction modules such as data extraction, model call, and compliance verification), the local encoding vector generated by the convolution reconstruction may carry repeated constraint descriptions or unnecessary auxiliary parameters, such as regional economic indicator screening rules that appear repeatedly in multiple steps, which will weaken the expression strength of the core analysis logic. Therefore, in the technical solution of the present application, semantic compression is performed on each context prompt statement semantic local unit encoding vector in the sequence distribution of the context prompt statement semantic local unit encoding vector to obtain a sequence distribution of the context prompt statement semantic compressed local unit encoding vector.

[0035] Specifically, through selective feature distillation, redundant components with low relevance to the current analysis task (such as outdated data format descriptions) are stripped from local semantic units, while the semantic density of key constraints (such as real-time data timeliness thresholds) is strengthened. This compression mechanism dynamically evaluates the information contribution of each dimension within a semantic unit, suppressing the weight distribution of non-essential features. For example, this reduces the activation intensity of redundant fields and improves the semantic purity of core economic indicator analysis instructions. The sequence distribution of the compressed context prompt statement semantic local unit encoding vectors can accurately carry the key generation elements defined in the protocol, allowing the subsequent semantic reconstruction phase to focus on the construction of the core analysis path.

[0036] In a specific example of the present application, semantic compression is performed on each context prompt sentence semantic local unit encoding vector in the sequence distribution of context prompt sentence semantic local unit encoding vectors using the following semantic compression formula to obtain a sequence distribution of context prompt sentence semantic compressed local unit encoding vectors; wherein the semantic compression formula is: ;in, is the one-norm of the vector, The first in the sequence distribution of the local unit encoding vector for the context prompt sentence semantic compression The context prompt sentence semantically compresses the local unit encoding vector.

[0037] Then, the statistics of the effective components of the context prompt statement semantics of each context prompt statement semantic compression local unit encoding vector in the sequence distribution of the context prompt statement semantic compression local unit encoding vector are calculated. It should be understood that when the protocol parser performs semantic reconstruction and compression on a multi-step generation task (such as a compound instruction involving cross-regional economic indicator comparison and compliance verification), the compressed context prompt statement semantics local unit encoding vector may carry an asymmetric semantic weight distribution. For example, there is a significant difference between the semantic density of the core economic indicator analysis instruction and the auxiliary format description, but the traditional method lacks a quantitative evaluation mechanism. Therefore, in the technical solution of the present application, the statistics of the effective components of the context prompt statement semantics of each context prompt statement semantic compression local unit encoding vector in the sequence distribution of the context prompt statement semantic compression local unit encoding vector are calculated. The information contribution of the compressed semantic unit is objectively measured through statistical modeling, and the effective components that are strongly associated with the core analysis task (such as key economic indicator calculation rules, compliance constraint thresholds) are identified, while locating redundant or low-value feature dimensions (such as repetitive data format descriptions). This statistical mechanism dynamically evaluates the predictive power and logical support strength of each local unit for the final report generation goal by constructing quantitative indicators of semantic effectiveness. For example, it strengthens the statistical weight of compliance directives such as "model coverage must exclude non-public data" and weakens the contribution assessment of redundant fields. By accurately quantifying the effectiveness distribution of semantic units, it provides an interpretable regulatory basis for subsequent gain operator calculations. This enables the semantic reconstruction phase to adaptively enhance the expressive strength of the core analytical logic, effectively avoiding the deviation in analytical conclusions caused by imbalanced feature weights in traditional methods.

[0038] In a specific example of the present application, the statistical number of the context prompt statement semantic effective components of each context prompt statement semantic compression local unit encoding vector in the sequence distribution of the context prompt statement semantic compression local unit encoding vector is calculated using the following statistical formula; wherein, the statistical formula is: ;in, For the The context prompt sentence semantic compression local unit encoding vector The eigenvalues at the positions, Indicates the active ingredient count, Preset threshold for training, for The corresponding statistics of the semantic effective components of the context prompt sentence, The number of vectors in the sequence distribution of semantic local unit encoding vectors for the context prompt sentence.

[0039] Furthermore, based on the statistics of the effective components of the semantically compressed local unit coding vectors of each context Prompt statement, the initial context Prompt statement semantic unit reshaping gain operator of each context Prompt statement semantically compressed local unit coding vector is calculated. Considering that the effectiveness distribution of local semantic units after the Prompt statement converted by the protocol in the economic analysis scenario has undergone semantic compression and statistical evaluation still lacks a dynamic control mechanism. Specifically, when the protocol parser identifies high effective component statistics of key economic indicator analysis instructions in a multi-step generation task (such as industry growth rate prediction model call instructions), the traditional static weight allocation method cannot adaptively strengthen the decision-making influence of the core semantic unit, resulting in the large model generation engine possibly weakening the strength of the key constraint conditions. Therefore, in the technical solution of the present application, based on the statistics of the effective components of the semantically compressed local unit coding vectors of each context Prompt statement semantically compressed local unit coding vector, the initial context Prompt statement semantic unit reshaping gain operator of each context Prompt statement semantically compressed local unit coding vector is calculated. That is, the quantitative evaluation results of semantically effective components are converted into operational dynamic enhancement coefficients, and by establishing a nonlinear mapping relationship between statistical values and feature weights, adaptive regulation of different semantic units is achieved. For example, for the high effective component statistics of compliance verification instructions, the gain operator calculation module will automatically generate an exponential enhancement coefficient, while attenuation suppression is applied to the statistical trough values described by redundant data formats. This dynamic calculation mechanism enables the semantic reconstruction process to intelligently adjust the phase intensity distribution of each semantic unit based on the core analysis objectives defined in the protocol. Through the refined regulation of the gain operator, it is ensured that key generation elements in the protocol (such as real-time data timeliness thresholds and model coverage restrictions) are given priority expression weights in the final Prompt template, thereby guiding the large model generation engine to accurately focus on high-value analysis paths, avoiding the focus ambiguity caused by the homogenization of feature weights in traditional methods, and significantly improving the logical rigor and professional decision-making reference value of the generated report.

[0040] More specifically, in an example of the present application, the initial context Prompt statement semantic unit reshaping gain operator of each context Prompt statement semantic compression local unit coding vector can be calculated by the following steps: first, based on the context Prompt statement semantic effective component statistics of each context Prompt statement semantic compression local unit coding vector, the semantic inhibition factor corresponding to each context Prompt statement semantic compression local unit coding vector is determined; then, based on the semantic inhibition factor corresponding to each context Prompt statement semantic compression local unit coding vector, the initial context Prompt statement semantic unit reshaping gain operator of each context Prompt statement semantic compression local unit coding vector is calculated.

[0041] In a specific example of the present application, the context prompt statement semantic unit reshaping gain operator of each context prompt statement semantic compression local unit encoding vector is calculated by the following formula; wherein, the formula is: ;in, for The corresponding polar angle, for The corresponding semantic inhibition factor, represents pi, represents the inverse tangent function, Reshape the gain operator for the initial context Prompt sentence semantic unit, for The corresponding initial context Prompt sentence semantic unit reshapes the gain operator.

[0042] In particular, it should be understood that multi-step constraints in the economic analysis protocol (such as the dynamic coupling of data timeliness and model coverage) will cause non-uniform distribution of semantic units in the vector space, resulting in directional deviations in the regulation of key economic characteristics by the gain operator. For example, in the cross-cycle economic forecasting task, the initial gain operator may weaken its ability to respond to emerging policy variables due to excessive concentration of historical data weights, resulting in inaccurate capture of the trend of industrial structure transformation in the generated report. Therefore, in order to establish a dynamic balance mechanism for the gain operator, in the preferred example of the present application, the semantic unit distribution missing correction is performed on the initial context prompt statement semantic unit reshaping gain operator to obtain the context prompt statement semantic unit reshaping gain operator.

[0043] Specifically, when protocol parameters fluctuate (e.g., temporary adjustments to compliance clauses), the correction module automatically identifies regions of semantic manifold distortion caused by parameter shifts through semantic unit dispersion detection and reconstructs the gain distribution based on directional symmetry constraints. For example, when dealing with sudden economic events (e.g., sharp exchange rate fluctuations), the modified initial context prompt statement semantic unit reshaping gain operator dynamically adjusts the semantic weights of the affected industry analysis modules while maintaining topological continuity of the upstream and downstream industry chain linkage features, avoiding the phenomenon of local features swallowing up global semantics, a common phenomenon in traditional methods. This correction process essentially constructs a flexible framework for expressing nonlinear relationships between economic variables, enabling the generation engine to adaptively modulate features when dealing with complex protocol constraints. The modified initial context prompt statement semantic unit reshaping gain operator re-anchors discrete semantic units onto a feature manifold with economic explanatory power through the flatness excitation mechanism in the gauge field. When faced with conflicting data from multiple sources (such as GDP calculations using different statistical calibers), the correction module suppresses the increase in feature entropy in areas where semantic dispersion is missing, ensuring that the expression strength of core economic indicators (such as the correlation between inflation and employment rates) in the Prompt template is not affected by noise. Ultimately, the contextual Prompt statement semantic units reshape the dynamic balance of gain operators, allowing the generated report to accurately reflect the economic logic network under the protocol constraints, while maintaining the rigor of the analytical conclusions and achieving semantic consistency across data sources and time dimensions.

[0044] In this example, the semantic unit reshaping gain operator of the initial context prompt statement is corrected for semantic unit dispersion loss using the following correction formula to obtain the context prompt statement semantic unit reshaping gain operator; wherein the correction formula is: ;in, is the semantic space flatness factor of the context prompt statement, is the factor representing the semantic flatness decomposition metric of the context prompt sentence, Reshape the gain operator for the contextual prompt sentence semantic unit.

[0045] Specifically, here, for the context prompt sentence semantic compression local unit encoding vector Statistics of semantically valid components of the corresponding context prompt sentence , in the calculation If each context prompt statement semantically compresses the local unit encoding vector Considered as a set of semantically compressed local unit encoding vectors of contextual prompt statements Based on the generator of the effective component, the generator is used as a subspace generating vector pointing to the set space, let , we would also expect the polar angle to be It satisfies the directional symmetry, so that the set space maintains the translation invariance of the effective components.

[0046] Therefore, if the set space is decomposed as a holomorphic structural metric, the holomorphic flatness of the space can be expressed as: ; Then, the effectiveness of local phase encoding is enhanced as a flatness decomposition metric representation based on the holomorphic structure to construct the single-mode coupling representation as a gauge field: ; That is, the gain operator is reshaped by the initial context Prompt sentence semantic unit The effectiveness enhancement representation is used to show that its individual pattern is the highest weight state of flatness excitation in the spatial holomorphic structure, so that the initial context Prompt sentence semantic unit can be updated to reshape the gain operator : .

[0047] Thus, the gain operator is reshaped while keeping the initial context Prompt statement semantic unit The set is based on the phase direction symmetry under the condition of flat spatial stability, thereby avoiding the loss of feature phase dispersion in the subsequent feature coupling enhancement process based on the reshaping of feature phase significance, and improving the expression effect of the enhanced feature vector.

[0048] Subsequently, based on a contextual prompt statement semantic unit reshaping gain operator that semantically compresses the local unit encoding vectors of each contextual prompt statement, semantic unit saliency reshaping is performed on the sequence distribution of the contextual prompt statement semantic unit encoding vectors to obtain an optimized contextual prompt statement semantic embedding encoding vector. It should be understood that multi-step protocol constraints (such as the dynamic superposition of data timeliness windows and compliance clauses) can cause the semantic unit weight distribution to deviate from the essential requirements of economic analysis. For example, when dealing with regional economic comparison tasks, traditional methods may overemphasize short-term volatility indicators (such as monthly PMI values) due to uncalibrated gain distribution, while weakening long-term core elements such as industrial structure transformation (such as changes in the proportion of high-value-added industries), resulting in a lack of strategic insight in the generated report. Therefore, to establish a decision-oriented expression mechanism for economic semantics, the technical solution of this application uses a contextual prompt statement semantic unit reshaping gain operator that semantically compresses the local unit encoding vectors of each contextual prompt statement, and semantic unit saliency reshaping is performed on the sequence distribution of the contextual prompt statement semantic unit encoding vectors to obtain an optimized contextual prompt statement semantic embedding encoding vector. That is, through the reshaping of semantic unit saliency, the system transforms the economic analysis priorities implicit in the protocol (such as policy sensitivity and data credibility) into topological constraints in the vector space, and reconstructs the semantic manifold using the dynamic modulation capability of the gain operator.

[0049] Specifically, when protocol parameters change dynamically (e.g., sudden trade policy adjustments triggering compliance constraint upgrades), the gain operator automatically strengthens the semantic response strength of the affected modules through feature phase reshaping. For example, when analyzing the transmission effects of monetary policy, this reshaping process enhances the correlation between interest rate changes and capital market reactions while suppressing historical inflation data noise that is irrelevant to the current analysis phase. This dynamic adjustment enables the generation engine to adapt to the complex interactions of protocol parameters, ensuring that the core economic logic forms a coherent semantic trajectory in the vector space rather than discrete feature fragments. The reshaped and optimized contextual prompt sentence semantic embedding encoding vector reshapes the gain operator through nonlinear regulation of the contextual prompt sentence semantic unit, reorganizing economic factors scattered across multiple steps (such as industry prosperity index and fiscal expenditure structure) into feature clusters with economic explanatory power. When faced with conflicting data from multiple sources (e.g., GDP accounting results from different statistical calibers), the system suppresses the semantic radiation range of low-confidence features to ensure that the generation of core conclusions (such as regional economic resilience assessment) is not interfered with by local data anomalies. Ultimately, the significance reshaping mechanism enables the generated reports to accurately capture the mutation patterns of microeconomic indicators while maintaining the stability of the macroeconomic framework, achieving an intelligent leap from data-driven to insight-driven.

[0050] In a specific example of the present application, the following saliency reshaping formula is used to perform semantic unit saliency reshaping on the sequence distribution of the contextual prompt sentence semantic local unit encoding vector to obtain an optimized contextual prompt sentence semantic embedding encoding vector; wherein, the saliency reshaping formula is: ;in, is the value of the natural exponential function with e as the base, Reshape the gain weight for the context prompt sentence semantic unit, To optimize the semantic embedding encoding vector of the context prompt sentence.

[0051] Specifically, the S33 is based on the semantic sequence distribution of the optimized Prompt statement to obtain a Prompt template that can be parsed by the large model. That is, in an embodiment of the present application, the sequence distribution of the optimized context Prompt statement semantic embedding coding vector is semantically decoded to obtain the sequence distribution of the Prompt optimization statement as a Prompt template that can be parsed by the large model. It should be understood that although the sequence distribution of the optimized context Prompt statement semantic embedding coding vector highly condenses the deep semantic features and global logical relationships of the task instructions, these high-dimensional vectors themselves are not directly natural language or structured instruction texts that can be understood and executed by the large model. Therefore, in the technical solution of the present application, the sequence distribution of the optimized context Prompt statement semantic embedding coding vector is semantically decoded to obtain the sequence distribution of the Prompt optimization statement as a Prompt template that can be parsed by the large model. That is, through the pre-trained decoder network, the semantic embedding coding vectors of each optimized context Prompt statement are inversely transformed, and are mapped back from the abstract high-dimensional continuous space to a text fragment with actual language expression ability.

[0052] During this process, the decoder gradually recovers the business commands corresponding to each step, ensuring that the content, sequence, and logic of each instruction are completely consistent with the original task decomposition and protocol requirements. Through high-quality semantic decoding, it can ensure that all data-driven information, constraints of each link, and professional terminology can be reflected in the final Prompt template in the best way, so that the subsequent reports generated by the large model have a high degree of consistency, accuracy, and controllability. In this way, an integrated closed loop from deep feature space to natural language expression to structured intelligent instruction set is achieved, so that the entire economic analysis report automatic generation system can stably and efficiently produce professional-level results that meet industry standards and user expectations, effectively promoting the application of intelligent economic analysis from theory to practical implementation.

[0053] Specifically, S4 involves inputting a Prompt template with input data into an economic analysis report generation engine based on a large model to generate an economic analysis report. It should be understood that economic analysis involves the coupling of multidimensional data and dynamic reasoning (such as nonlinear correlations between macroeconomic indicators), requiring the deep reasoning capabilities of a large model to transform structured instructions into strategically valuable insights. Therefore, to enable protocol-driven intelligent decision-making, the technical solution of this application involves inputting a Prompt template with input data into an economic analysis report generation engine based on a large model to generate an economic analysis report. In one specific example, the protocol-aware module first parses the step topology in the Prompt template and dynamically loads fine-tuned models for the corresponding domain (such as econometric forecasting models and risk assessment maps). Subsequently, at the data fusion layer, semantic alignment is performed between verified multi-source data (such as industry capacity data and policy texts) and prior rules in the knowledge base (such as economic cycle theory and compliance clauses). Finally, through the multi-round iterative generation mechanism of the large model, combined with dynamic allocation of attention weights, logical progression between report sections is ensured (such as the transition from current situation description to policy recommendations for trend forecasting). In this way, when faced with sudden economic events (such as sharp exchange rate fluctuations), the generation engine can dynamically adjust the parameter weights of the analysis model through the emergency response rules preset in the protocol (such as the risk threshold trigger mechanism), compare and deduce real-time market data with historical stress test results, and output report sections containing multiple scenario simulations. This capability enables the system to maintain the stability of the protocol framework while flexibly responding to uncertainty, avoiding the delayed conclusions caused by traditional static templates and significantly improving the decision-making support effectiveness of the economic analysis system.

[0054] In summary, according to the embodiment of the present application, the economic analysis report generation method based on the big model context protocol is explained, which introduces the big model context protocol and standardizes it through a special protocol description language. The protocol structuredly encapsulates all the elements required for report generation (data source, knowledge base, constraints, steps); then, the protocol parser automatically converts this precisely defined protocol into a prompt template that can be understood by the big model and has pre-verified data, thereby replacing the complex and unstable manual prompt word engineering, thereby achieving effective standardization, precise control and automation of the process of generating economic analysis reports for big models, ensuring the quality, consistency and efficiency of the final report.

[0055] Furthermore, a device for generating an economic analysis report based on a large model context protocol is also provided.

[0056] Figure 5 FIG. 1 is a block diagram of an economic analysis report generating device based on a large model context protocol according to an embodiment of the present application. Figure 5As shown, according to an embodiment of the present application, an economic analysis report generating device 300 based on a big model context protocol includes: a big model context protocol definition module 310, which is used to define a big model context protocol based on report requirements using a protocol description language, and the big model context protocol includes a data source, a knowledge base, constraints and generation steps; a data quality verification module 320, which is used to connect to a specified database or API through a multi-source protocol adapter based on the data source defined in the big model context protocol to capture required data, and perform data quality verification on the required data based on the rules in the big model context protocol to obtain verified data; a Prompt template generation module 330, which is used for a protocol parser to convert the big model context protocol into a Prompt template that can be parsed by the big model, and insert the verified data into a predetermined position of the Prompt template to obtain a Prompt template with input data; an economic analysis report generation module 340, which is used to input the Prompt template with input data into an economic analysis report generation engine based on the big model to obtain an economic analysis report.

[0057] As described above, the economic analysis report generation device 300 based on the large model context protocol according to the embodiment of the present application can be implemented in various wireless terminals, such as a server having an economic analysis report generation algorithm based on the large model context protocol. In one possible implementation, the economic analysis report generation device 300 based on the large model context protocol according to the embodiment of the present application can be integrated into the wireless terminal as a software module and / or hardware module. For example, the economic analysis report generation device 300 based on the large model context protocol can be a software module in the operating system of the wireless terminal, or can be an application developed for the wireless terminal; of course, the economic analysis report generation device 300 based on the large model context protocol can also be one of the many hardware modules of the wireless terminal.

[0058] Alternatively, in another example, the economic analysis report generating apparatus 300 based on the large model context protocol and the wireless terminal may also be separate devices, and the economic analysis report generating apparatus 300 based on the large model context protocol may be connected to the wireless terminal via a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.

[0059] While various embodiments of the present disclosure have been described above, the above descriptions are illustrative, non-exhaustive, and not intended to be limiting of the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is selected to best explain the principles of the embodiments, their practical applications, or improvements to existing technologies, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A method for generating an economic analysis report based on a large model context protocol, characterized in that: include: Based on the reporting requirements, a protocol description language is used to define the big model context protocol, which includes data source, knowledge base, constraints and generation steps. Based on the data source defined in the big model context protocol, connect to the specified database or API through the multi-source protocol adapter to capture the required data, and perform data quality verification on the required data based on the rules in the big model context protocol to obtain verified data; The protocol parser converts the large model context protocol into a Prompt template that can be parsed by the large model, and inserts the verified data into a predetermined position of the Prompt template to obtain a Prompt template with input data; wherein the protocol parser converts the large model context protocol into a Prompt template that can be parsed by the large model; The Prompt template with input data is input into the economic analysis report generation engine based on the large model to obtain the economic analysis report.

2. The method for generating an economic analysis report based on a large model context protocol according to claim 1, characterized in that: Constraints include data currency, model coverage, chart requirements, and regulatory compliance.

3. The method for generating an economic analysis report based on a large model context protocol according to claim 2, characterized in that: The protocol parser converts the large model context protocol into a Prompt template that can be parsed by the large model, including: Decompose the tasks in the large model context protocol into multiple steps, and generate a prompt statement for each step to obtain the sequence distribution of prompt statements as the initial prompt template; Perform feature-level semantic expression robustness optimization on each Prompt statement in the sequence distribution of Prompt statements to obtain the optimized semantic sequence distribution of Prompt statements; By optimizing the semantic sequence distribution of Prompt statements, a Prompt template that can be parsed by large models is obtained.

4. The method for generating an economic analysis report based on a large model context protocol according to claim 3, characterized in that: The semantic expression robustness of each prompt statement in the sequence distribution of prompt statements is optimized based on the feature level to obtain the optimized semantic sequence distribution of prompt statements, including: Performing contextual semantic encoding on each Prompt statement in the sequence distribution of Prompt statements to obtain a sequence distribution of contextual Prompt statement semantic embedding encoding vectors; Local semantic significant optimization is performed on each context prompt statement semantic embedding coding vector in the sequence distribution of the context prompt statement semantic embedding coding vector to obtain the sequence distribution of the optimized context prompt statement semantic embedding coding vector as the semantic sequence distribution of the optimized prompt statement.

5. The method for generating an economic analysis report based on a large model context protocol according to claim 4, characterized in that: Contextual semantic encoding is performed on each Prompt statement in the sequence distribution of Prompt statements to obtain a sequence distribution of contextual Prompt statement semantic embedding encoding vectors, including: Embed the Prompt sentence into the matrix and pass each Prompt sentence in the sequence distribution of the Prompt sentence through the Word2Vec-based word embedding encoder to obtain the sequence distribution of the Prompt sentence semantic embedding encoding vector; The sequence distribution of the semantic embedding encoding vector of the Prompt sentence is encoded with the task step context semantic association based on the bidirectional LSTM model to obtain the sequence distribution of the contextual Prompt sentence semantic embedding encoding vector.

6. The method for generating an economic analysis report based on a large model context protocol according to claim 5, characterized in that: Performing local semantic significant optimization on each context prompt sentence semantic embedding coding vector in the sequence distribution of the context prompt sentence semantic embedding coding vector to obtain the sequence distribution of the optimized context prompt sentence semantic embedding coding vector, including: Perform semantic unit reconstruction and semantic compression on the semantic embedding encoding vectors of each context prompt sentence to obtain the sequence distribution of the semantic compressed local unit encoding vectors of the context prompt sentence; Calculate the statistics of the effective semantic components of the context prompt statement of each context prompt statement semantic compression local unit encoding vector in the sequence distribution of the context prompt statement semantic compression local unit encoding vector; Calculate the context prompt sentence semantic unit reshaping gain operator of each context prompt sentence semantic compression local unit encoding vector based on the context prompt sentence semantic effective component statistics of each context prompt sentence semantic compression local unit encoding vector; Based on the context prompt sentence semantic unit reshaping gain operator of the semantically compressed local unit encoding vector of each context prompt sentence, the sequence distribution of the context prompt sentence semantic local unit encoding vector is reshaped by semantic unit saliency to obtain the optimized context prompt sentence semantic embedding encoding vector.

7. The method for generating an economic analysis report based on a large model context protocol according to claim 6, characterized in that: The semantic unit reconstruction and semantic compression processing are performed on the semantic embedding coding vectors of each context prompt sentence to obtain the sequence distribution of the semantic compression local unit coding vectors of the context prompt sentence, including: The semantic embedding coding vectors of each context prompt sentence are reconstructed into semantic units based on one-dimensional convolutional coding to obtain the sequence distribution of the semantic local unit coding vectors of the context prompt sentence; Semantic compression is performed on each context prompt statement semantic local unit encoding vector in the sequence distribution of the context prompt statement semantic local unit encoding vector to obtain a sequence distribution of the context prompt statement semantic compressed local unit encoding vector.

8. The method for generating an economic analysis report based on a large model context protocol according to claim 7, characterized in that: Based on the statistics of the effective semantic components of the context prompt sentence semantic compression local unit encoding vector of each context prompt sentence, the context prompt sentence semantic unit reshaping gain operator of each context prompt sentence semantic compression local unit encoding vector is calculated, including: Determining a semantic inhibition factor corresponding to each context prompt sentence semantic compression local unit encoding vector based on a statistical number of context prompt sentence semantic effective components of each context prompt sentence semantic compression local unit encoding vector; Based on the semantic suppression factor corresponding to each context prompt sentence semantic compression local unit encoding vector, the initial context prompt sentence semantic unit reshaping gain operator of each context prompt sentence semantic compression local unit encoding vector is calculated; The semantic unit reshaping gain operator of the initial context prompt statement is corrected by semantic unit scattering missing to obtain the context prompt statement semantic unit reshaping gain operator.

9. The method for generating an economic analysis report based on a large model context protocol according to claim 8, characterized in that: By optimizing the semantic sequence distribution of Prompt statements, we obtain Prompt templates that can be parsed by large models, including: The sequence distribution of the semantic embedding encoding vector of the optimized context Prompt statement is semantically decoded to obtain the sequence distribution of the Prompt optimization statement as a Prompt template that can be parsed by the large model.

10. An economic analysis report generation device based on a large model context protocol, characterized in that: include: A large model context protocol definition module is used to define a large model context protocol using a protocol description language based on report requirements. The large model context protocol includes data sources, knowledge bases, constraints, and generation steps. The data quality verification module is used to connect to the specified database or API through the multi-source protocol adapter based on the data source defined in the large model context protocol to capture the required data, and perform data quality verification on the required data based on the rules in the large model context protocol to obtain verified data; A prompt template generation module is used for the protocol parser to convert the large model context protocol into a prompt template that can be parsed by the large model, and insert the verified data into the predetermined position of the prompt template to obtain a prompt template with input data; The economic analysis report generation module is used to input the prompt template with input data into the economic analysis report generation engine based on the large model to obtain the economic analysis report.

Images