Method, device and equipment for analyzing city investment policy text and computer readable medium
By integrating and analyzing local government financing vehicle (LGFV) policy texts and market time-series data using a multimodal large model, the problem of subjective analysis bias by researchers has been solved, achieving more accurate and objective policy text analysis.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- 泰康保险集团股份有限公司
- Filing Date
- 2026-01-27
- Publication Date
- 2026-05-15
AI Technical Summary
In existing technologies, when analyzing local government financing vehicle (LGFV) policy texts, researchers' reliance on their own experience can easily introduce subjective biases, leading to analysis results that are not objective and accurate enough.
A pre-trained multimodal large model is used to fuse and analyze urban investment policy texts and market time series data. The first multimodal large model parses text semantics and image visual trend information, and the second large model outputs market trend analysis comments, text comments and prediction categories, reducing manual operation.
This improved the objectivity and accuracy of the analysis of local government financing vehicle (LGFV) policy texts, reduced subjective bias, and enhanced the reliability and credibility of the analysis results.
Smart Images
Figure CN122045255A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of computer technology, and in particular to a method, apparatus, device, and computer-readable medium for analyzing urban investment policy texts. Background Technology
[0002] Important policy documents issued by relevant departments are a major factor influencing the spread of local government financing vehicle (LGFV) bonds and are of reference value for the future trend of LGFV bond spreads. Therefore, it is necessary to analyze these policy documents when they are released.
[0003] In the process of realizing this invention, the inventors discovered that the prior art has at least the following problems: when researchers analyze urban investment policy texts, relying solely on their own experience can easily introduce subjective opinions, leading to significant analytical biases. Summary of the Invention
[0004] In view of this, embodiments of the present invention provide a method, apparatus, device, and computer-readable medium for analyzing urban investment policy texts, which can improve the objectivity and accuracy of the analysis.
[0005] To achieve the above objectives, according to one aspect of the present invention, a method for analyzing local government financing vehicle (LGFV) policy texts is provided, comprising: Obtain the local government financing vehicle (LGFV) policy text to be analyzed and the market time series data corresponding to the release time of the LGFV policy text, and convert the market time series data into the corresponding market trend chart to generate image data; The urban investment policy text and the image data are input into a pre-trained first multimodal large model. The first multimodal large model parses the semantics of the urban investment policy text and the visual trend information in the image data, establishes the association between the semantics and the visual trend information, and generates a fusion analysis result. The fusion analysis results and the preset historical labeled text are input into the pre-trained second model. The second model outputs market trend analysis and commentary on the market time series data, text commentary on the urban investment policy text, and prediction category of the urban investment policy text.
[0006] The process of acquiring the local government financing vehicle (LGFV) policy text to be analyzed and the market time series data corresponding to the release time of the LGFV policy text, and converting the market time series data into a corresponding market trend chart to generate image data, includes: Based on the release time of the aforementioned local government financing vehicle (LGFV) policy text, market time-series data within a preset time window up to the release time is extracted from the market database. The market time-series data includes one or more of the following: LGFV bond interest rates, interest-bearing debt size, net financing amount, and interest rate spread. For each type of market time series data, a corresponding time series trend chart is generated according to a preset time series, and the time series trend charts corresponding to the market time series data are merged to construct the image data.
[0007] The first multimodal large model parses the semantics of the text and the visual trend information in the image data, establishes the correlation between the semantics and the visual trend information, and generates a fusion analysis result, including: The urban investment policy text is input into the text encoder of the first multimodal large model to extract the semantics of the urban investment policy text to generate a text feature vector sequence, and the image data is input into the visual encoder of the first multimodal large model to extract visual trend information. Based on the text feature vector sequence, cross-attention calculation is performed on the visual trend information according to the attention mechanism to associate the semantics with the corresponding visual trend information, thereby generating the fusion analysis result.
[0008] The step of feeding the image data into the visual encoder of the first multimodal large model to extract visual trend information includes: The image data is input into the visual encoder of the first multimodal large model to identify the data curves, indicator change curves, and statistical feature curves in the market trend chart; The text of the image data is input into the text encoder of the first multimodal large model to generate a market trend analysis summary corresponding to the data curve, indicator change curve, and statistical feature curve, so as to construct the visual trend information based on the data curve, indicator change curve, statistical feature curve, and the corresponding market trend analysis summary.
[0009] The second major model outputs market trend analysis and commentary on the market time series data, textual commentary on the local government financing vehicle (LGFV) policy text, and prediction categories for the LGFV policy text, including: Using the Few-shot method, text category example pairs are established based on historical policy texts and their corresponding category labels, and trend category example pairs are established based on historical market trends and their category labels. Training data is then built using the definitions of the text category example pairs, the trend category example pairs, and the category labels. The second major model is trained based on the training data, and the category labels include loose, neutral, and tight.
[0010] The second major model outputs market trend analysis and commentary on the market time series data, textual commentary on the local government financing vehicle (LGFV) policy text, and prediction categories for the LGFV policy text, including: The second major model, based on the urban investment policy text and market time series data in the fusion analysis results, determines the corresponding historical annotation text. The historical annotation text includes the historical policy text, the historical market data description corresponding to the historical policy text, the category label of the historical policy text, the analysis and commentary of the historical policy text, and the analysis and commentary of the historical market data description. The second major model uses historical market data to describe analysis and commentary to determine market trend analysis and commentary for the market time series data, uses historical policy text analysis and commentary to determine textual commentary for the local government financing vehicle (LGFV) policy text, and uses the category labels of the historical policy text as the prediction category for the LGFV policy text.
[0011] Based on the urban investment policy texts and market time-series data in the fusion analysis results, the corresponding historical annotation texts are determined, including: Based on the urban investment policy texts, market time series data, and the correlation between the urban investment policy texts and the market time series data in the fusion analysis results, an urban investment policy text vector is established. By utilizing the similarity between the vector of the urban investment policy text and the vectors of multiple historical annotated texts, the historical annotated text corresponding to the urban investment policy text can be determined.
[0012] According to a second aspect of the present invention, an apparatus for analyzing local government financing vehicle (LGFV) policy texts is provided, comprising: The image module is used to acquire the local government financing vehicle (LGFV) policy text to be analyzed and the market time series data corresponding to the release time of the LGFV policy text, and convert the market time series data into the corresponding market trend chart to generate image data; The fusion module is used to input the urban investment policy text and the image data into a pre-trained first multimodal large model. The first multimodal large model parses the semantics of the urban investment policy text and the visual trend information in the image data, establishes the correlation between the semantics and the visual trend information, and generates fusion analysis results. The analysis module is used to input the fusion analysis results and preset historical labeled text into the pre-trained second large model. The second large model outputs market trend analysis and commentary on the market time series data, text commentary on the urban investment policy text, and prediction category of the urban investment policy text.
[0013] According to a third aspect of the present invention, an electronic device for analyzing local government financing vehicle (LGFV) policy texts is provided, comprising: One or more processors; Storage device for storing one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors perform the methods described above.
[0014] According to a fourth aspect of the present invention, a computer-readable medium is provided having a computer program stored thereon, which, when executed by a processor, implements the method as described above.
[0015] One embodiment of the above invention has the following advantages or beneficial effects: It acquires the local government financing vehicle (LGFV) policy text to be analyzed and the market time series data corresponding to the release time of the LGFV policy text, and converts the market time series data into a corresponding market trend chart to generate image data; it inputs the LGFV policy text and the image data into a pre-trained first multimodal large-scale model, which parses the semantics of the LGFV policy text and the visual trend information in the image data, establishes the correlation between the semantics and the visual trend information, and generates a fusion analysis result; it inputs the fusion analysis result and preset historical labeled text into a pre-trained second large-scale model, which outputs a market trend analysis commentary on the market time series data, a text commentary on the LGFV policy text, and a predicted category for the LGFV policy text. By using the first multimodal large-scale model to fuse the LGFV policy text and market time series data, and then outputting the corresponding commentary and category through the second large-scale model, the above process requires no manual operation, thus improving the objectivity and accuracy of the analysis.
[0016] The further effects of the aforementioned unconventional alternative methods will be explained below in conjunction with specific implementation methods. Attached Figure Description
[0017] The accompanying drawings are provided to better understand the invention and are not intended to unduly limit the scope of the invention. Wherein: Figure 1 This is a schematic diagram of the main process of the method for analyzing urban investment policy texts according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the process for generating image data according to an embodiment of the present invention; Figure 3 This is a flowchart illustrating the process of establishing the association between semantic and visual trend information and generating fusion analysis results according to an embodiment of the present invention. Figure 4 This is a schematic diagram of the process for training the second major model according to an embodiment of the present invention; Figure 5 This is a flowchart illustrating the output prediction content of the second major model according to an embodiment of the present invention; Figure 6 This is a schematic diagram of the original policy text classification model according to an embodiment of the present invention; Figure 7 This is a schematic diagram of the main structure of the apparatus for analyzing urban investment policy texts according to an embodiment of the present invention; Figure 8 This is an exemplary system architecture diagram in which embodiments of the present invention can be applied; Figure 9 This is a schematic diagram of the structure of a computer system suitable for implementing terminal devices or servers of the present invention. Detailed Implementation
[0018] The following description, in conjunction with the accompanying drawings, illustrates exemplary embodiments of the present invention, including various details to aid understanding. These details should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the invention. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.
[0019] Relying solely on manual analysis of local government financing vehicle (LGFV) policy texts can easily introduce subjective biases. Furthermore, due to the uncertainty of policy impacts, objectively and accurately classifying policies at the time of their release is often quite challenging.
[0020] To address the significant biases in the analysis of local government financing vehicle (LGFV) policy texts, the following technical solutions from the embodiments of the present invention can be adopted.
[0021] See Figure 1 , Figure 1 This is a schematic diagram of the main process of the method for analyzing local government financing vehicle (LGFV) policy texts according to an embodiment of the present invention, specifically including the following steps: S101. Obtain the local government financing vehicle (LGFV) policy text to be analyzed and the market time series data corresponding to the release time of the LGFV policy text, and convert the market time series data into the corresponding market trend chart to generate image data.
[0022] In embodiments of this invention, currently available information is used to analyze and classify local government financing vehicle (LGFV) policy documents. When an LGFV policy document is published, the available information at that point in time includes: the LGFV policy text corresponding to the policy document, historical LGFV policy documents, and market data. For example, market data includes the LGFV bond market trend. Market data belongs to market time series data. Market time series data covers technical indicators such as fundamentals, market conditions, and valuations. For example, LGFV interest-bearing debt: from 2012 to 2022, LGFV interest-bearing debt showed a continuous growth trend, increasing from approximately 10 trillion yuan to approximately 60 trillion yuan.
[0023] In one embodiment of the present invention, the local government financing vehicle (LGFV) policy text to be analyzed can be directly obtained. For example, the LGFV policy text to be analyzed can be obtained based on paper documents using image recognition technology. The release time of the LGFV policy text to be analyzed is obtained, and the corresponding market time series data is retrieved from the database using the release time.
[0024] To analyze the aforementioned market time series data, the market time series data can be converted into corresponding market trend charts, thereby generating image data.
[0025] S102. Input the urban investment policy text and image data into the pre-trained first multimodal large model. The first multimodal large model parses the semantics of the urban investment policy text and the visual trend information in the image data, establishes the correlation between the semantics and the visual trend information, and generates the fusion analysis results.
[0026] The text and image data of local government financing vehicle (LGFV) policies are input into a pre-trained first multimodal large-scale model. Utilizing the multimodal processing capabilities of this large-scale model, the semantics of the LGFV policy text and the visual trend information in the image data are analyzed. A correlation is established between the semantic and visual trend information, generating a fusion analysis result. This fusion analysis result characterizes the correspondence between the LGFV policy text and market time-series data. Furthermore, market time-series data can be incorporated into the analysis of LGFV policy texts.
[0027] S103. Input the fusion analysis results and the preset historical labeled text into the pre-trained second model. The second model outputs market trend analysis and commentary on market time series data, text commentary on urban investment policy texts, and prediction categories of urban investment policy texts.
[0028] The fusion analysis results and the preset historical labeled text are input into the pre-trained second model. The second model learns from the historical labeled text and, based on the fusion analysis results similar to the historical labeled text, outputs market trend analysis and commentary on market time series data, text commentary on local government financing vehicle (LGFV) policy texts, and prediction categories for LGFV policy texts.
[0029] In the above embodiments, for the local government financing vehicle (LGFV) policy document to be analyzed, the document and pre-defined historical annotated text are input into a multimodal large-scale model. After processing by the multimodal large-scale model, market trend analysis and commentary, textual commentary, and prediction categories can be output. The prediction categories include three types: loose, neutral, and tight, representing the degree of impact on the subsequent looseness or tightness of LGFV bonds, respectively. The above process requires no manual operation, thus improving the objectivity and accuracy of the analysis.
[0030] See Figure 2 , Figure 2 This is a schematic diagram of the process for generating image data according to an embodiment of the present invention. Specifically, it includes the following steps: S201. Based on the release time of the local government financing vehicle (LGFV) policy document, extract market time-series data from the market database within a preset time window up to the release time. The market time-series data includes one or more of the following: LGFV bond interest rate, interest-bearing debt scale, net financing amount, and interest rate spread.
[0031] In embodiments of the present invention, the acquisition of market time-series data is based on the release time of the local government financing vehicle (LGFV) policy text and a preset time window. The end point of the preset time window is the release time of the LGFV policy text, and the start point of the preset time window can be determined based on the type of market time-series data to improve the accuracy of the market time-series data. This allows for a thorough determination of the market trend within the corresponding preset time window.
[0032] As an example, for local government financing vehicle (LGFV) bond interest rates, the preset time window starts one month before the release date; for the scale of interest-bearing debt, the preset time window starts two months before the release date; for net financing, the preset time window starts six months before the release date; and for interest rate spreads, the preset time window starts six months before the release date.
[0033] S202. Generate a corresponding time series trend chart for each type of market time series data according to a preset time series, and merge the time series trend charts corresponding to the market time series data to construct image data.
[0034] Each type of market time series data can generate a corresponding time series trend chart according to a preset time series. For example, the preset time series includes a preset time period, namely, January 1st to January 15th. To comprehensively represent market time series data, the time series trend charts corresponding to multiple types of market time series data are merged to obtain image data. For example, under the same time axis, the time series trend charts corresponding to multiple types of market time series data are displayed as image data.
[0035] exist Figure 2 In this embodiment, time-series data of corresponding duration is determined using the release time of the local government financing vehicle (LGFV) policy text to generate image data. The market time-series data is then represented through the image.
[0036] See Figure 3 , Figure 3 This is a flowchart illustrating the process of establishing the association between semantic and visual trend information and generating fusion analysis results according to an embodiment of the present invention. Specifically, it includes the following steps: S301. Input the urban investment policy text into the text encoder of the first multimodal large model to extract the semantic feature vector sequence of the urban investment policy text, and input the image data into the visual encoder of the first multimodal large model to extract visual trend information.
[0037] The purpose of inputting the local government financing vehicle (LGFV) policy text into the text encoder of the first multimodal large-scale model is to extract deep semantic features from the LGFV policy text and generate a text feature vector sequence. The deep semantic features are used to identify keywords in the LGFV policy text, such as: encouraging corporate bond issuance, regulating P projects, and prohibiting the creation of new implicit debt. Based on these keywords, a text feature vector sequence is generated.
[0038] Image data is input into the visual encoder of the first multimodal large model to extract visual trend information from the image data. Visual trend information includes long-term trend direction and inflection points.
[0039] Specifically, image data is input into the visual encoder of the first multimodal large model to identify data curves, indicator change curves, and statistical characteristic curves in the market trend chart. These curves reflect the long-term trend direction, such as rising, falling, or stable. The curves also represent turning points, such as peaks and troughs.
[0040] The data curves, indicator change curves, and statistical characteristic curves all have corresponding text text. For example: "The data shows a phased increase." The text encoder of the first multimodal large model is used to recognize the above text.
[0041] Specifically, the text of the image data is input into the text encoder of the first multimodal large model to generate market trend analysis summaries corresponding to data curves, indicator change curves, and statistical feature curves. That is, each curve has a corresponding market trend analysis summary, such as curve volatility and cyclical patterns. Visual trend information is constructed based on the data curves, indicator change curves, statistical feature curves, and corresponding market trend analysis summaries. Using market trend analysis summaries improves the speed of visual trend information analysis. The visual trend information can be represented using vectors.
[0042] S302. Based on the text feature vector sequence, perform cross-attention calculation on visual trend information according to the attention mechanism to associate semantics with the corresponding visual trend information and generate fusion analysis results.
[0043] The attention mechanism of the first multimodal large model is invoked to perform cross-attention calculation on the text feature vector sequence, enabling the association of semantic and visual trend information to generate fusion analysis results. For example, when the semantics involve "financing support" or "relaxed conditions," the attention mechanism will associate images in the visual trend information that show trends such as "interest rate decline," "narrowing interest rate spread," or "net financing growth." Similarly, when the semantics involve "debt control" or "standardized financing," the attention mechanism will associate images in the visual trend information that show trends such as "rising debt ratio," "excessive proportion of non-standard assets," or "rapid growth in interest payment scale."
[0044] As an example, the fusion analysis results include: policy intent, market status, and correlations. Policy intent is extracted from local government financing vehicle (LGFV) policy texts. Market status is identified from the image data as being related to policy intent. For example, debt levels are high, but interest rates have begun to decline. Correlations are the relationships between policy intent and market status. For example, confirmation, hedging, or deviation.
[0045] exist Figure 3In one embodiment, the first multimodal large model realizes the association and fusion of urban investment policy text and image data.
[0046] See Figure 4 , Figure 4 This is a schematic diagram illustrating the process of training the second large model according to an embodiment of the present invention. Specifically, it includes the following steps: S401. Using the Few-shot method, establish text category example pairs based on historical policy texts and their corresponding category labels, and establish trend category example pairs based on historical market trends and their category labels. Use the definitions of text category example pairs, trend category example pairs, and category labels to build training data.
[0047] The Few-shot approach is a machine learning method whose core idea is to train the model using only a very small number of examples. Specifically, it creates text category example pairs based on historical policy texts and their corresponding category labels, and creates trend category example pairs based on historical market trends and their category labels.
[0048] For example: Text category example pair: Local government financing vehicle (LGFV) policy text 1; Category label: Tight. Trend category example pair: Market trend description B, high debt ratio, rising interest rates; Category label: Tight. Definition of category label: Loose, Neutral, and Tight.
[0049] S402. The second model is trained based on training data, local government financing vehicle (LGFV) policy texts, and corresponding market trend charts. The category labels include loose, neutral, and tight.
[0050] The training data, local government financing vehicle (LGFV) policy texts, and corresponding market trend charts are combined to obtain a prompt, which is then input into a multimodal large model that already possesses general knowledge.
[0051] The second major model learns and summarizes rules when processing prompts. For example, when local government financing vehicles (LGFVs) are mentioned in policy texts containing words like "regulated" or "strictly prohibited," and the market background description indicates high debt levels, the model is marked as "tightening." After training, the second major model applies these summarized rules to the fusion analysis results and pre-defined historical labeled texts.
[0052] exist Figure 4 In one embodiment, training data enables a general multimodal large model to quickly switch and adapt to new tasks, and a second large model can be quickly built.
[0053] See Figure 5 , Figure 5 This is a flowchart illustrating the output prediction content of the second major model according to an embodiment of the present invention. Specifically, it includes the following steps: S501. The second major model, based on the urban investment policy texts and market time-series data in the fusion analysis results, determines the corresponding historical labeled texts. The historical labeled texts include historical policy texts, descriptions of historical market data corresponding to the historical policy texts, category labels of the historical policy texts, analysis and commentary on the historical policy texts, and analysis and commentary on the descriptions of historical market data.
[0054] The fusion analysis results generated by the first multimodal large model are obtained, including local government financing vehicle (LGFV) policy texts and market time-series data. There is a correspondence between the aforementioned LGFV policy texts and market time-series data.
[0055] The fusion analysis results and historical labeled text are combined in a formatted way according to a preset prompt template to construct the input sequence for the second major model. This input sequence is then fed into the second major model.
[0056] The historical case database stores multiple historical annotated texts. These historical annotated texts are structured data. As an example, the historical annotated text includes historical policy texts, corresponding historical market data descriptions, category tags, analytical comments, and analytical comments on the historical market data descriptions.
[0057] In one embodiment of the present invention, the historical annotation text corresponding to the urban investment policy text is determined by utilizing the similarity between vectors.
[0058] First, based on the fusion analysis results of local government financing vehicle (LGFV) policy texts, market time-series data, and the correlation between LGFV policy texts and market time-series data, an analysis vector for LGFV policy texts is established. For example, LGFV policy text vectors and market time-series data vectors are constructed using terms from the LGFV policy texts and market time-series data, respectively. The correlation between LGFV policy texts and market time-series data includes three types: complementary, hedging, and neutral. Different numerical values are used to represent these correlations to establish a correlation vector. The analysis vector for LGFV policy texts is then constructed by sequentially arranging the LGFV policy text vector, market time-series data vector, and correlation vector.
[0059] For the historical case database storing multiple historical annotated texts, a vector for each historical annotated text is pre-built. By calculating the similarity between the vector of the urban investment policy text and the vectors of multiple historical annotated texts, the historical annotated text corresponding to the highest similarity is taken as the historical annotated text corresponding to the urban investment policy text.
[0060] S502. The second major model uses historical market data to describe and analyze market trends based on time-series data, and uses historical policy texts to analyze and analyze local government financing vehicle (LGFV) policy texts. The category labels of historical policy texts are used as the prediction categories for LGFV policy texts.
[0061] After identifying the historical annotation text corresponding to the local government financing vehicle (LGFV) policy documents, market trend analysis and commentary based on historical market data is determined using analytical commentary. For example, market trend analysis and commentary based on market time-series data can not only describe the current market trend, but also include the direction in which the current market trend evolves under the influence of LGFV policies.
[0062] Textual commentary on local government financing vehicle (LGFV) policies is determined by analyzing and commenting on historical policy texts. For example, textual commentary on LGFV policies includes an analysis of the LGFV policy itself and its potential impact, its similarities and differences with historical policies, and a prediction of its possible actual effects.
[0063] Use the category labels of historical policy texts as the predicted categories for local government financing vehicle (LGFV) policy texts.
[0064] exist Figure 5 In this embodiment, the second major model is used to analyze the urban investment policy text based on historically labeled text, thereby improving the reliability of the analysis.
[0065] See Figure 6 , Figure 6 This is a schematic diagram of the original policy text classification model according to an embodiment of the present invention. The input data includes two parts. The first part is directly relevant information about the current policy itself, and the second part is existing historical case information. The output includes prediction categories, market trend analysis and commentary, and policy commentary.
[0066] Figure 6 The multimodal LLM-based trend analyzer is the first large multimodal model, and the LLM-based text classifier is the second large model.
[0067] In the first part of the input data, market data trend charts are plotted using the local government financing vehicle (LGFV) policy text to be analyzed and the corresponding market time-series data at the time of the policy text's release. These market data trend charts are then input into a multimodal LLM-based trend chart analyzer to generate image data, i.e., market trend analysis. This image data is then used as input data to construct an LLM-based text classifier according to a preset prompt template.
[0068] In the second part of the input data, the Few-shot method is used to construct the input data for an LLM-based text classifier by combining the category labels defined by the researchers, the text classification few-shot cases (analysis and commentary on historical policy texts), and the market trend classification few-shot cases (analysis and commentary on market trends) according to the preset prompt template.
[0069] The following examples illustrate the technical solutions in the embodiments of the present invention.
[0070] Example 1: The local government financing vehicle (LGFV) policy text to be analyzed is "Notice 1" issued on January 1, 2020.
[0071] Market time series data corresponding to the release time of the local government financing vehicle (LGFV) policy documents: Moving average of the ratio of land sales to local government financing vehicle (LGFV) interest payments: From 2015 to 2020, this ratio showed a continuous upward trend, indicating that the ratio was increasing year by year.
[0072] Moving average of the difference between local government financing vehicle (LGFV) bond interest rates and credit ratings: From 2011 to 2016, this interest rate rose year by year, reached its peak, and then declined year by year from 2016 to 2020.
[0073] The aforementioned market time-series data demonstrates the changing trends of different financial indicators over different time periods, reflecting the volatility and cyclical changes in the market.
[0074] The input data for the second largest pre-trained model includes: <Objective> Based on the definitions of the three categories—loose, neutral, and tight—in the <Category Definitions>, provide a concise and accurate answer to the <Question>. You must directly output the content of the original text and avoid any modifications to the original text; otherwise, return empty. Do not return evidence not mentioned in the original text.
[0075] <Category Definition>
[0076] Loose definition; Neutral definition; Tight definition.
[0077] <Example of text category>
[0078] 1. This local government financing vehicle (LGFV) policy regulates the development of P projects, strictly prohibiting new projects from being added to the project database in areas where more than 10% of the projects are located. This is considered a relatively tight policy.
[0079] 2. The local government financing vehicle (LGFV) policy stipulates that no new loans may be issued in violation of regulations. This is considered a tightening policy.
[0080] The output data of the second pre-trained model includes: market trend analysis and commentary on market time series data, textual commentary on local government financing vehicle (LGFV) policy texts, and prediction categories for LGFV policy texts.
[0081] Example 2: Analyzing the same policy text in parallel with credit rating researchers yields more diverse analytical perspectives, further supplementing the researchers' subjective analysis. When conclusions are the same, this increases the credibility of the researchers' findings; when conclusions differ, it provides researchers with more perspectives and may even correct their subjective biases.
[0082] The local government financing vehicle (LGFV) policy text to be analyzed is "Notice 2" issued on January 1, 2010.
[0083] Market time series data corresponding to the release time of the local government financing vehicle (LGFV) policy documents: Regional Debt Ratio - Broad-based Moving Average: From 2015 to 2019, the regional debt ratio continued to decline, indicating that the regional debt burden was easing and was relatively loose.
[0084] Regional Debt Ratio - Narrow-caliber Moving Average of Divergence: From 2015 to 2016, the regional debt ratio rose slightly, then began to decline, reaching a relatively low level in 2019, indicating a looser ratio.
[0085] The input data for the second largest pre-trained model includes: <Objective> Based on the definitions of the three categories—loose, neutral, and tight—in the <Category Definitions>, provide a concise and accurate answer to the <Question>. You must directly output the content of the original text and avoid any modifications to the original text; otherwise, return empty. Do not return evidence not mentioned in the original text.
[0086] <Category Definition>
[0087] Loose definition; Neutral definition; Tight definition.
[0088] The output data of the second pre-trained model includes: market trend analysis and commentary on market time series data, textual commentary on local government financing vehicle (LGFV) policy texts, and prediction categories for LGFV policy texts.
[0089] See Table 1, which presents the analysis results of credit rating analysts and the analysis results using the present invention. The objective analysis results of the present invention are consistent with the subjective analysis results of credit rating analysts, further increasing the credibility of the credit rating analysts' analysis results. It also supplements the analysis with market trend analysis information to further improve the interpretability of the analysis conclusions.
[0090] Table 1
[0091] See Figure 7 , Figure 7 This is a schematic diagram of the main structure of an apparatus for analyzing urban investment policy texts according to an embodiment of the present invention. The apparatus for analyzing urban investment policy texts can implement a method for analyzing urban investment policy texts. The apparatus for analyzing urban investment policy texts specifically includes: Image module 701 is used to acquire the local government financing vehicle (LGFV) policy text to be analyzed and the market time series data corresponding to the release time of the LGFV policy text, and convert the market time series data into the corresponding market trend chart to generate image data; The fusion module 702 is used to input the urban investment policy text and the image data into a pre-trained first multimodal large model. The first multimodal large model parses the semantics of the urban investment policy text and the visual trend information in the image data, establishes the association between the semantics and the visual trend information, and generates a fusion analysis result. The analysis module 703 is used to input the fusion analysis results and preset historical labeled text into the pre-trained second large model. The second large model outputs market trend analysis and commentary on the market time series data, text commentary on the urban investment policy text, and prediction category of the urban investment policy text.
[0092] In one embodiment of the present invention, the image module 701 is used to extract market time series data within a preset time window up to the release time of the local government financing vehicle (LGFV) policy text from the market database according to the release time of the LGFV policy text. The market time series data includes one or more of the following: LGFV bond interest rate, interest-bearing debt scale, net financing amount and interest rate spread. For each type of market time series data, a corresponding time series trend chart is generated according to a preset time series, and the time series trend charts corresponding to the market time series data are merged to construct the image data.
[0093] In one embodiment of the present invention, the fusion module 702 is used to input the urban investment policy text into the text encoder of the first multimodal large model to extract the semantics of the urban investment policy text to generate a text feature vector sequence, and to input the image data into the visual encoder of the first multimodal large model to extract visual trend information. Based on the text feature vector sequence, cross-attention calculation is performed on the visual trend information according to the attention mechanism to associate the semantics with the corresponding visual trend information, thereby generating the fusion analysis result.
[0094] In one embodiment of the present invention, the fusion module 702 is used to input the image data into the visual encoder of the first multimodal large model to identify the data curve, indicator change curve and statistical feature curve in the market trend chart; The text of the image data is input into the text encoder of the first multimodal large model to generate a market trend analysis summary corresponding to the data curve, indicator change curve, and statistical feature curve, so as to construct the visual trend information based on the data curve, indicator change curve, statistical feature curve, and the corresponding market trend analysis summary.
[0095] In one embodiment of the present invention, the analysis module 703 is used to establish text category example pairs based on historical policy texts and corresponding category labels using a Few-shot method, establish trend category example pairs based on historical market trends and their category labels, and establish training data using the definitions of the text category example pairs, the trend category example pairs, and the category labels. The second major model is trained based on the training data, and the category labels include loose, neutral, and tight.
[0096] In one embodiment of the present invention, the analysis module 703 is used to determine the corresponding historical annotation text based on the urban investment policy text and market time series data in the fusion analysis results of the second large model. The historical annotation text includes historical policy text, historical market data description corresponding to the historical policy text, category label of the historical policy text, analysis and commentary of the historical policy text, and analysis and commentary of the historical market data description. The second major model uses historical market data to describe analysis and commentary to determine market trend analysis and commentary for the market time series data, uses historical policy text analysis and commentary to determine textual commentary for the local government financing vehicle (LGFV) policy text, and uses the category labels of the historical policy text as the prediction category for the LGFV policy text.
[0097] In one embodiment of the present invention, the analysis module 703 is used to establish an urban investment policy text vector based on the urban investment policy text, market time series data, and the correlation between the urban investment policy text and the market time series data in the fusion analysis results; By utilizing the similarity between the vector of the urban investment policy text and the vectors of multiple historical annotated texts, the historical annotated text corresponding to the urban investment policy text can be determined.
[0098] Figure 8 An exemplary system architecture 800 is shown, which can be applied to a method or apparatus for analyzing urban investment policy texts according to embodiments of the present invention.
[0099] like Figure 8 As shown, system architecture 800 may include terminal devices 801, 802, and 803, a network 804, and a server 805. Network 804 serves as the medium for providing communication links between terminal devices 801, 802, and 803 and server 805. Network 804 may include various connection types, such as wired or wireless communication links or fiber optic cables, etc.
[0100] Users can use terminal devices 801, 802, and 803 to interact with server 805 via network 804 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 801, 802, and 803, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social media platform software, etc. (for example only).
[0101] Terminal devices 801, 802, and 803 can be various electronic devices with displays and web browsing capabilities, including but not limited to smartphones, tablets, laptops, and desktop computers.
[0102] Server 805 can be a server providing various services, such as a backend management server supporting shopping websites browsed by users using terminal devices 801, 802, and 803 (for example only). The backend management server can analyze and process data such as received product information query requests, and feed back the processing results (such as target push information and product information - for example only) to the terminal devices.
[0103] It should be noted that the method for analyzing urban investment policy texts provided in this embodiment of the invention is generally executed by server 805, and correspondingly, the device for analyzing urban investment policy texts is generally located in server 805.
[0104] It should be understood that Figure 8 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.
[0105] The following is for reference. Figure 9 It shows a schematic diagram of the structure of a computer system 900 suitable for implementing a terminal device of the present invention. Figure 9 The terminal device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the present invention.
[0106] like Figure 9 As shown, the computer system 900 includes a central processing unit (CPU) 901, which can perform various appropriate actions and processes based on programs stored in read-only memory (ROM) 902 or programs loaded from storage section 908 into random access memory (RAM) 903. The RAM 903 also stores various programs and data required for the operation of the system 900. The CPU 901, ROM 902, and RAM 903 are interconnected via a bus 904. An input / output (I / O) interface 905 is also connected to the bus 904.
[0107] The following components are connected to I / O interface 905: an input section 906 including a keyboard, mouse, etc.; an output section 907 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 908 including a hard disk, etc.; and a communication section 909 including a network interface card such as a LAN card, modem, etc. The communication section 909 performs communication processing via a network such as the Internet. A drive 910 is also connected to I / O interface 905 as needed. A removable medium 911, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on drive 910 as needed so that computer programs read from it can be installed into storage section 908 as needed.
[0108] In particular, according to the embodiments disclosed in this invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 909, and / or installed from removable medium 911. When the computer program is executed by central processing unit (CPU) 901, it performs the functions defined above in the system of this invention.
[0109] It should be noted that the computer-readable medium shown in this invention can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this invention, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media can also be any computer-readable medium other than computer-readable storage media, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.
[0110] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.
[0111] The modules described in the embodiments of the present invention can be implemented in software or hardware. The described modules can also be housed in a processor; for example, a processor can be described as including an image module, a fusion module, and an analysis module. The names of these modules do not necessarily limit the module itself; for example, the image module can also be described as "used to acquire the local government financing vehicle (LGFV) policy text to be analyzed and the market time series data corresponding to the release time of the LGFV policy text, and convert the market time series data into a corresponding market trend chart to generate image data."
[0112] In another aspect, the present invention also provides a computer-readable medium, which may be included in the device described in the above embodiments; or it may exist independently and not assembled into the device. The computer-readable medium carries one or more programs, which, when executed by the device, cause the device to include: Obtain the local government financing vehicle (LGFV) policy text to be analyzed and the market time series data corresponding to the release time of the LGFV policy text, and convert the market time series data into the corresponding market trend chart to generate image data; The urban investment policy text and the image data are input into a pre-trained first multimodal large model. The first multimodal large model parses the semantics of the urban investment policy text and the visual trend information in the image data, establishes the association between the semantics and the visual trend information, and generates a fusion analysis result. The fusion analysis results and the preset historical labeled text are input into the pre-trained second model. The second model outputs market trend analysis and commentary on the market time series data, text commentary on the urban investment policy text, and prediction category of the urban investment policy text.
[0113] According to the technical solution of this invention, the following steps are taken: First, the local government financing vehicle (LGFV) policy text to be analyzed and the market time series data corresponding to the release time of the LGFV policy text are obtained. The market time series data is then converted into a corresponding market trend chart to generate image data. Next, the LGFV policy text and the image data are input into a pre-trained first multimodal large-scale model. The first multimodal large-scale model parses the semantics of the LGFV policy text and the visual trend information in the image data, establishes the correlation between the semantics and the visual trend information, and generates a fusion analysis result. Finally, the fusion analysis result and preset historical labeled text are input into a pre-trained second large-scale model. The second large-scale model outputs a market trend analysis commentary on the market time series data, a textual commentary on the LGFV policy text, and a predicted category for the LGFV policy text. By using the first multimodal large-scale model to fuse the LGFV policy text and market time series data, and then using the second large-scale model to output the corresponding commentary and category, the above process requires no manual operation, thus improving the objectivity and accuracy of the analysis.
[0114] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can occur depending on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention. It should be noted that the acquisition, storage, and application of user personal information involved in the technical solutions of this disclosure comply with relevant laws and regulations and do not violate public order and good morals.
Claims
1. A method for analyzing local government financing vehicle (LGFV) policy texts, characterized in that, include: Obtain the local government financing vehicle (LGFV) policy text to be analyzed and the market time series data corresponding to the release time of the LGFV policy text, and convert the market time series data into the corresponding market trend chart to generate image data; The urban investment policy text and the image data are input into a pre-trained first multimodal large model. The first multimodal large model parses the semantics of the urban investment policy text and the visual trend information in the image data, establishes the association between the semantics and the visual trend information, and generates a fusion analysis result. The fusion analysis results and the preset historical labeled text are input into the pre-trained second model. The second model outputs market trend analysis and commentary on the market time series data, text commentary on the urban investment policy text, and prediction category of the urban investment policy text.
2. The method for analyzing local government financing vehicle (LGFV) policy texts according to claim 1, characterized in that, The process of acquiring the local government financing vehicle (LGFV) policy text to be analyzed and the market time series data corresponding to the release time of the LGFV policy text, and converting the market time series data into a corresponding market trend chart to generate image data, includes: Based on the release time of the aforementioned local government financing vehicle (LGFV) policy text, market time-series data within a preset time window up to the release time is extracted from the market database. The market time-series data includes one or more of the following: LGFV bond interest rates, interest-bearing debt size, net financing amount, and interest rate spread. For each type of market time series data, a corresponding time series trend chart is generated according to a preset time series, and the time series trend charts corresponding to the market time series data are merged to construct the image data.
3. The method for analyzing local government financing vehicle (LGFV) policy texts according to claim 1, characterized in that, The first multimodal large model parses the semantics of the text and the visual trend information in the image data, establishes the correlation between the semantics and the visual trend information, and generates a fusion analysis result, including: The urban investment policy text is input into the text encoder of the first multimodal large model to extract the semantics of the urban investment policy text to generate a text feature vector sequence, and the image data is input into the visual encoder of the first multimodal large model to extract visual trend information. Based on the text feature vector sequence, cross-attention calculation is performed on the visual trend information according to the attention mechanism to associate the semantics with the corresponding visual trend information, thereby generating the fusion analysis result.
4. The method for analyzing local government financing vehicle (LGFV) policy texts according to claim 3, characterized in that, The step of feeding the image data into the visual encoder of the first multimodal large model to extract visual trend information includes: The image data is input into the visual encoder of the first multimodal large model to identify the data curves, indicator change curves, and statistical feature curves in the market trend chart; The text of the image data is input into the text encoder of the first multimodal large model to generate a market trend analysis summary corresponding to the data curve, indicator change curve, and statistical feature curve, so as to construct the visual trend information based on the data curve, indicator change curve, statistical feature curve, and the corresponding market trend analysis summary.
5. The method for analyzing local government financing vehicle (LGFV) policy texts according to claim 1, characterized in that, The second major model outputs market trend analysis and commentary on the market time series data, textual commentary on the local government financing vehicle (LGFV) policy text, and prediction categories for the LGFV policy text, including: Using the Few-shot method, text category example pairs are established based on historical policy texts and their corresponding category labels, and trend category example pairs are established based on historical market trends and their category labels. Training data is then established using the definitions of the text category example pairs, the trend category example pairs, and the category labels. The second major model is trained based on the training data, and the category labels include loose, neutral, and tight.
6. The method for analyzing local government financing vehicle (LGFV) policy texts according to claim 1, characterized in that, The second major model outputs market trend analysis and commentary on the market time series data, textual commentary on the local government financing vehicle (LGFV) policy text, and prediction categories for the LGFV policy text, including: The second major model, based on the urban investment policy text and market time series data in the fusion analysis results, determines the corresponding historical annotation text. The historical annotation text includes the historical policy text, the historical market data description corresponding to the historical policy text, the category label of the historical policy text, the analysis and commentary of the historical policy text, and the analysis and commentary of the historical market data description. The second major model uses historical market data to describe analysis and commentary to determine market trend analysis and commentary for the market time series data, uses historical policy text analysis and commentary to determine textual commentary for the local government financing vehicle (LGFV) policy text, and uses the category labels of the historical policy text as the prediction category for the LGFV policy text.
7. The method for analyzing local government financing vehicle (LGFV) policy texts according to claim 6, characterized in that, Based on the urban investment policy texts and market time-series data in the fusion analysis results, the corresponding historical annotation texts are determined, including: Based on the urban investment policy texts, market time series data, and the correlation between the urban investment policy texts and the market time series data in the fusion analysis results, an urban investment policy text vector is established. By utilizing the similarity between the vector of the urban investment policy text and the vectors of multiple historical annotated texts, the historical annotated text corresponding to the urban investment policy text can be determined.
8. An apparatus for analyzing local government financing vehicle (LGFV) policy texts, characterized in that, include: The image module is used to acquire the local government financing vehicle (LGFV) policy text to be analyzed and the market time series data corresponding to the release time of the LGFV policy text, and convert the market time series data into the corresponding market trend chart to generate image data; The fusion module is used to input the urban investment policy text and the image data into a pre-trained first multimodal large model. The first multimodal large model parses the semantics of the urban investment policy text and the visual trend information in the image data, establishes the correlation between the semantics and the visual trend information, and generates fusion analysis results. The analysis module is used to input the fusion analysis results and preset historical labeled text into the pre-trained second large model. The second large model outputs market trend analysis and commentary on the market time series data, text commentary on the urban investment policy text, and prediction category of the urban investment policy text.
9. An electronic device for analyzing local government financing vehicle (LGFV) policy texts, characterized in that, include: One or more processors; Storage device for storing one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1-7.
10. A computer-readable medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1-7.