Financial chart data generation and multi-modal analysis method

Abstract

The invention provides a financial chart data generation and multi-modal analysis method, which comprises the following steps of: acquiring financial event information in a target financial scene based on a preset financial event identification model, and determining the category, the influence direction and the influence intensity of a financial event according to indexes such as a price change value and a price change proportion. The method comprises the following steps: performing adaptive adjustment on an initial data generation rule according to the influence intensity of a financial event, constructing a longitudinal axis data generation function capable of reflecting event characteristics through a trend component, a fluctuation component and an event local impact component, generating time-continuous horizontal axis categories and longitudinal axis data of event significant characteristics, and packaging to form structured chart data. And performing visualization processing on the structured data based on the adjusted rendering parameters to generate a chart image containing trend, fluctuation and impact visual features, and executing trend consistency, fluctuation consistency and impact position consistency check to ensure that the image visual features are consistent with the event data.

Description

Technical Field

[0001] This invention relates to the field of intelligent processing technology for financial data, and in particular to a method for generating and parsing financial chart data in a multimodal manner. Background Technology

[0002] In existing financial chart intelligent processing technologies, mainstream methods typically only perform visual analysis on existing charts or construct training data based on general random generation rules. They lack the ability to model the data patterns, trend disturbances, and local shock characteristics caused by real financial events, and cannot reflect complex dynamic phenomena commonly found in financial markets such as announcement effects, sudden fluctuations, and asynchronous reactions. Furthermore, existing chart understanding models generally rely on weakly correlated or modally fragmented data samples during training, lacking a quantifiable mapping relationship between chart images and structured data, making it difficult to achieve intermodal alignment learning.

[0003] Furthermore, traditional generation methods neglect the impact of financial events' position on the timeline, their impact decay patterns, and multidimensional fluctuation structures. This leads to significant discrepancies between the generated data and real market performance, causing the trained model to perform unstably when processing charts with event features, and even failing to correctly interpret the semantic meaning of events. Therefore, we propose a new method for financial chart data generation and multimodal parsing. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by providing a method for generating and parsing financial chart data in a multimodal manner, thereby resolving the technical problems mentioned in the background section.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A method for generating and parsing financial chart data includes the following steps: S1. Financial Event Identification Steps: Analyze the target financial scenario based on the preset event identification model to obtain financial event information related to the chart to be generated; financial event information includes market volatility events, financial announcement events, interest rate change events, industry emergencies, and the direction and intensity of the impact of the events, which are used to indicate the impact characteristics of financial events on the subsequent chart data generation. S2. Adaptive Adjustment Steps for Event-Driven Data Generation Rules: Based on financial event information, construct data generation rules for generating structured chart data, and adaptively adjust the data generation rules according to the direction and intensity of the event's impact. Adaptive adjustment includes adjusting the data's fluctuation range, trend pattern, amplitude changes, key node positions, and text label descriptions so that the generated structured data can reflect the characteristic changes caused by the financial event. S3. Structured chart data generation steps: Based on adaptively adjusted data generation rules, generate structured data to describe the target chart; the structured data includes horizontal axis categories, multiple vertical axis data series, legend labels, and title information, and maintains the continuity of time sequence, consistency of data correlation, and integrity of event feature representation during the generation process; S4. Chart rendering steps to preserve event characteristics: Input structured data into the rendering tool and generate the target chart image according to visual parameters consistent with the adaptive adjustment results; The rendering tool sets the chart type, color combination, legend position, text annotation method and visual emphasis elements so that the chart image can visually preserve the data characteristics caused by the financial event. S5. Multimodal parsing model training and output steps: Align the chart image with the structured data in a multimodal manner, and perform joint training through the multimodal parsing model to enable the model to learn the correspondence between financial event features and structured data; After training, use the multimodal parsing model to parse the input financial chart to obtain a structured parsing result that reflects the features of financial events.

[0006] S1 specifically involves: acquiring financial scenario information related to the target financial chart generation requirements, including the market, involved entities, and time range, to determine the input range for subsequent financial event identification; analyzing the financial scenario information based on a pre-set event identification model to identify financial event information related to the scenario; financial event information includes event category, event time, and the direction of the event's impact; and further determining the impact intensity of the financial event based on the financial event information to characterize the level of impact of the event on the degree of data fluctuation during subsequent data generation.

[0007] S2 specifically involves: constructing initial data generation rules based on general financial chart generation specifications, including horizontal axis category format, vertical axis data generation method, and legend text structure; associating financial event information with the initial data generation rules to determine the impact range of financial events on data trends, data distribution ranges, and key node positions; and adaptively adjusting the initial data generation rules based on the event feature mapping results, so that the adjusted data generation rules can reflect the characteristics of the financial event in terms of trend patterns, fluctuation amplitudes, and text label descriptions.

[0008] S3 specifically refers to: generating horizontal axis categories that are continuous in time series, consistent in order, and able to reflect the location of financial events, based on the adaptively adjusted data generation rules; Multiple vertical axis data series are generated according to the adjusted data generation rules, so that each data series conforms to the characteristics of financial events in terms of numerical variation, trend and key node performance. The horizontal axis categories and vertical axis data are packaged into structured data according to a unified structure, and title text, legend text, and auxiliary explanatory text related to financial events are added.

[0009] S4 specifically involves: determining rendering parameters to reflect the characteristics of financial events based on the adaptively adjusted data generation rules, including chart type, color combination, legend position, and visual emphasis; generating chart images based on structured data and rendering parameters to visually reflect the trend changes, fluctuation characteristics, and key nodes caused by the financial event; performing quality checks on the chart images to ensure that the trends and data forms in the chart images are consistent with the financial event information, and re-executing the chart image generation.

[0010] S5 specifically involves: establishing a one-to-one correspondence between chart images and structured data to construct training sample pairs for the multimodal analytical model; using the training samples to jointly train the multimodal analytical model, enabling the model to learn the correspondence between chart images and structured data, as well as the cross-modal performance of financial event features; and after training, using the multimodal analytical model to analyze the input financial charts and output structured analytical results that reflect the features of financial events.

[0011] The beneficial effects of this invention are as follows: This invention constructs quantitative indicators for events such as price change value, price change ratio, and impact intensity index, and adaptively adjusts fluctuation amplitude, trend slope, and local impact based on event intensity. This enables the generated chart data to accurately represent real market scenarios such as announcement effects, sudden fluctuations, and event reversals, significantly improving the authenticity and interpretability of the synthesized data. By establishing a chain-like mapping relationship of "event parameters → data rules → data form," data generation no longer relies on static rules but can dynamically adjust trend components, fluctuation components, and impact components according to event intensity, fluctuation characteristics, and time position. This solves the technical problems of existing technologies that generate data with a single pattern and lack of event sensitivity.

[0012] This invention uses rendering controls such as trend color intensity parameters, volatility transparency parameters, and impact emphasis weights to ensure that charts maintain consistency with the event-augmented data model in terms of visual elements such as color, shadows, and annotations. It also achieves rigorous visual verification through trend consistency, volatility consistency, and impact location consistency checks, solving the industry problem of "image-data inconsistency" in traditional methods. By simultaneously extracting feature vectors from images and data using visual and structured coding functions, and designing modality alignment loss functions and event-augmentation loss functions, the model learns the correspondence between event features such as trends, volatility, and impacts across different modalities during training. This significantly improves the model's accuracy, stability, and generalization ability when interpreting complex financial charts.

[0013] This invention constructs an end-to-end chart generation and parsing system driven by financial events. This system ensures the clear and traceable logic of training data generation, and provides explicit mathematical expressions for the impact of events on both data and images. It achieves consistency and verifiability among data generation, image presentation, and model parsing results, avoiding the performance instability issues caused by fragmented data sources in traditional synthetic data training. The generated data possesses event sensitivity, multi-scale volatility, and time-location dependence. The multimodal parsing model can learn cross-modal expressions of event features, thus demonstrating stable performance when handling complex charts with sharp jumps, multi-peak fluctuations, and trend reversals. It can effectively support various intelligent financial tasks such as chart interpretation, intelligent analysis, and investment sentiment recognition. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of a financial chart data generation and multimodal parsing method according to the present invention. Detailed Implementation

[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] Example: Figure 1 As shown in the figure, this embodiment provides a method for generating and parsing financial chart data, including the following steps: S1. Financial Event Identification Steps: Analyze the target financial scenario based on the preset event identification model to obtain financial event information related to the chart to be generated; financial event information includes market volatility events, financial announcement events, interest rate change events, industry emergencies, and the direction and intensity of the impact of the events, which are used to indicate the impact characteristics of financial events on the subsequent chart data generation. S2. Adaptive Adjustment Steps for Event-Driven Data Generation Rules: Based on financial event information, construct data generation rules for generating structured chart data, and adaptively adjust the data generation rules according to the direction and intensity of the event's impact. Adaptive adjustment includes adjusting the data's fluctuation range, trend pattern, amplitude changes, key node positions, and text label descriptions so that the generated structured data can reflect the characteristic changes caused by the financial event. S3. Structured chart data generation steps: Based on adaptively adjusted data generation rules, generate structured data to describe the target chart; the structured data includes horizontal axis categories, multiple vertical axis data series, legend labels, and title information, and maintains the continuity of time sequence, consistency of data correlation, and integrity of event feature representation during the generation process; S4. Chart rendering steps to preserve event characteristics: Input structured data into the rendering tool and generate the target chart image according to visual parameters consistent with the adaptive adjustment results; The rendering tool sets the chart type, color combination, legend position, text annotation method and visual emphasis elements so that the chart image can visually preserve the data characteristics caused by the financial event. S5. Multimodal parsing model training and output steps: Align the chart image with the structured data in a multimodal manner, and perform joint training through the multimodal parsing model to enable the model to learn the correspondence between financial event features and structured data; After training, use the multimodal parsing model to parse the input financial chart to obtain a structured parsing result that reflects the features of financial events.

[0017] S1. Financial Event Identification Steps: Based on a pre-defined event identification model, the target financial scenario is analyzed to obtain financial event information related to the chart to be generated. This information includes categories such as significant market fluctuations, financial announcements, interest rate changes, and industry emergencies, as well as the direction and intensity of the event's impact, used to indicate the impact characteristics of the financial event on subsequent chart data generation. Specifically, this includes the following sub-steps: S110, Financial Scenario Acquisition Sub-step: Obtain financial scenario information related to the target financial chart generation requirement. This information includes the target market, the analysis subject, and the time range, providing an input basis for subsequent financial event identification. The time range is used to determine the start and end times of the price series, ensuring that subsequent analysis is confined to a specific time interval.

[0018] S120, Financial Event Analysis Sub-step: Based on the preset financial event identification model, analyze the financial scenario information obtained in step S110 to identify the financial event information corresponding to the scenario; the financial event information includes the financial event category, the time of occurrence of the financial event, the direction of the impact of the financial event, and the basic values ​​used to calculate the intensity of the impact.

[0019] When performing event identification, the price change value is calculated based on the price series within the time range. The calculation method is as follows: ; in, This represents the price data at the start of the event window, indicating the "starting price". This represents the price data at the end of the event window, indicating the "end price". This refers to the "price change value".

[0020] The price change percentage "R" is further calculated as follows: ; Where R stands for "price change percentage", which reflects the relative degree of change of the financial event within the event window.

[0021] Based on the sign and magnitude of the price change ratio R, the type of financial event and the direction of its impact can be determined: When R > 0.03, it is determined to be an "upward event" with a "positive" impact direction; When R < -0.03, it is determined as a "downward event" with a "negative" impact direction; when When the value is greater than 0.05, it is judged as a "violent fluctuation event", and the direction of influence is "two-way fluctuation".

[0022] The above rules make the identification of financial event categories and the determination of their impact direction both executable and reproducible.

[0023] S130, Sub-step for extracting the impact intensity of a financial event: Based on the price change ratio R obtained in step S120, calculate the impact intensity index used to characterize the degree of influence of the financial event on subsequent data generation. ": ; in, This is the "impact intensity index," used to represent the overall impact intensity of financial events on data fluctuations. This is the "intensity weighting coefficient," used to adjust the sensitivity of the impact intensity in different financial scenarios. Its value can be set within the range of 1.0 to 3.0 according to the market type. R is the absolute value of the aforementioned price change ratio.

[0024] To ensure that the influence intensity index can be applied to the generation of subsequent chart data, a fluctuation amplitude parameter "A" needs to be defined to control the overall fluctuation amplitude of the subsequent data. Its calculation method is as follows: ; Where A is the "adjusted fluctuation amplitude parameter", which is used to control the fluctuation magnitude of the vertical axis data; "Base volatility" refers to the default volatility range when not affected by financial events.

[0025] Similarly, to ensure that trend patterns reflect the characteristics of financial events, a trend slope parameter "k" is defined, and its calculation method is as follows: ; Where k is the "adjusted trend slope parameter", used to control the overall upward or downward trend of the data; "Basic trend slope" represents the default data trend; This is the "trend weighting coefficient," used to adjust the degree of influence of financial events on trend changes.

[0026] By calculating the aforementioned impact intensity index, volatility amplitude parameter, and trend slope parameter, the impact of financial events on subsequent data generation rules becomes calculable, traceable, and reproducible.

[0027] S2. Adaptive Adjustment Steps for Event-Driven Data Generation Rules: Based on financial event information, construct data generation rules for generating structured chart data, and adaptively adjust these rules according to the direction and intensity of the event's impact. Adaptive adjustment includes adjusting the data's fluctuation range, trend pattern, amplitude changes, key node positions, and text label descriptions to ensure the generated structured data reflects the characteristic changes caused by the financial event. Specifically, this includes the following sub-steps: S210, Initial Data Generation Rule Construction Sub-step: Based on the data structure specifications of general financial charts, initial data generation rules are constructed. These rules are used to subsequently generate horizontal axis categories, vertical axis data series, and legend text content. The initial data generation rules include the following basic parameters: Basic volatility : This is the "default fluctuation range when not affected by financial events," used to set the upward and downward fluctuation range of the vertical axis data.

[0028] Basic trend slope : This is the "Default Trend Change Rate", used to set the overall upward or downward trend of the vertical axis data in the absence of events.

[0029] Basic random disturbance term : This is the "basic random disturbance term", used to add random noise to the vertical axis data to simulate normal small fluctuations in the market. Its value can come from a random distribution with a mean of 0.

[0030] The initial data generation rules can be represented as the basic generation function for the vertical axis: ; in, This is the "basic vertical axis data," used as the default value at time point t; t is the "time series index," representing the position number of the horizontal axis category.

[0031] The aforementioned initial data generation rules provide a basic framework for subsequently incorporating the impact of financial events.

[0032] S220, Event Feature Mapping Sub-step: Based on the influence intensity index obtained in step S130 This process maps the impact of financial events on data fluctuations, trends, and key node locations into the initial data generation rules. The mapping process includes the following: Volatility Adjustment Mapping: Mapping the influence intensity index to the volatility amplitude expands the volatility A affected by the event to: ; Where A is the "event-adjusted fluctuation range parameter".

[0033] Trend Adjustment Mapping: Mapping the influence strength index to the trend slope, adjusting the trend slope as follows: ; Where k is the "trend slope after event adjustment"; β is the "trend weight coefficient", used to set the sensitivity of the event to the trend.

[0034] Event location mapping: Define the time index of the location where the event occurred. And define the scope of how event characteristics are represented in the data sequence. For example: ; in, For "event location index"; T is "total length of time series"; The "relative position coefficient of the event" is used to indicate the relative position of the event in the time series (e.g., 0.3 indicates earlier, 0.7 indicates later).

[0035] Through the above mapping, the volatility, trend direction, and time position of financial events are accurately mapped into the data generation rules.

[0036] S230, Adaptive Adjustment Sub-step for Data Generation Rules: Based on the mapping result of step S220, the initial data generation rules are adaptively adjusted to reflect the characteristic changes caused by financial events. The adjusted data generation rules include: Event-enhanced volatility model: ; in, To enhance the volatility component of the event; This is the "fluctuation frequency coefficient," used to control the periodicity of the waveform.

[0037] Event-driven trend model: ; in, This is used to "enhance the trend component of the event".

[0038] Local shock model for events with violent fluctuations: ; in, This refers to the "local impact component of the event"; This refers to the "impact amplitude coefficient". This is the "attenuation coefficient," used to control the rate at which the impact effect decays over time.

[0039] Finally, the adaptively adjusted vertical axis data generation function can be expressed as: ; in, This is the "adaptively adjusted vertical axis data," used for generating subsequent structured chart data.

[0040] Through the aforementioned adaptive adjustments, the data generation rules can accurately reflect the impact of financial events in terms of trend slope, volatility, and local shocks, thereby enabling the subsequently generated chart data to have significant event characteristics and interpretability.

[0041] S3. Structured Chart Data Generation Steps: Based on adaptively adjusted data generation rules, generate structured data to describe the target chart. The structured data includes horizontal axis categories, multiple vertical axis data series, legend labels, and title information, maintaining temporal continuity, consistency in data relationships, and completeness of event characteristic representation during the generation process. Specifically, it includes the following sub-steps: S310, Horizontal Axis Category Generation Sub-step: Based on the adaptively adjusted data generation rules in step S230, generate horizontal axis categories that are continuous in time series, consistent in order, and reflect the location of financial events. The horizontal axis categories are denoted as: ; Where X is the "set of categories on the horizontal axis"; The value is the category value at index t on the horizontal axis, which can correspond to a date, quarter, or month; T is the total number of categories on the horizontal axis, corresponding to the length of the time series.

[0042] To highlight the temporal location of financial events, the category values ​​corresponding to the event locations on the horizontal axis are adjusted. Add event identifier text, such as using text like "event occurrence date" or "announcement date," to enhance the semantic consistency between structured data and events.

[0043] S320, Vertical Axis Data Generation Sub-step: Based on the final vertical axis generation function obtained in step S230... For each horizontal axis category Generate the corresponding vertical axis data. The vertical axis data sequence is represented as follows: ; Where Y is the "vertical axis data sequence"; The vertical axis data value corresponding to the serial number t is composed of the event-enhanced trend component, the event-enhanced fluctuation component, the local shock component, and the basic random disturbance.

[0044] The final data function for the vertical axis is as follows: ; in, The trend component is defined as follows: ; K is the "event-adjusted trend slope parameter".

[0045] To enhance the fluctuation component of an event, it is defined as: ; Where A is the event-adjusted fluctuation range parameter. This refers to the "fluctuation frequency coefficient".

[0046] The local impact component of the event is defined as: ; in, This refers to the impact amplitude coefficient. The attenuation coefficient; For event location index; This is the "basic random disturbance term," used to simulate normal random fluctuations in the market.

[0047] Through the aforementioned vertical axis data generation mechanism, the generated data strictly corresponds to the characteristics of real financial events in terms of trend patterns, volatility, and event impact.

[0048] S330, Structured Data Encapsulation Sub-step: Combine the horizontal axis category set X generated in step S310 with the vertical axis data sequence Y generated in step S320, and encapsulate them into a structured chart data object. The structured data includes the following fields: Horizontal axis field: ; Used to represent all horizontal axis category values.

[0049] Vertical axis field: ; Used to represent the vertical axis value corresponding to each horizontal axis.

[0050] Event identifier field: ; Used to pinpoint the location of the financial event within a time series.

[0051] Text label field: This includes: title text (containing an event description), legend text (corresponding to the meaning of the three components: trend, fluctuation, and impact), and supplementary explanatory text (such as "fluctuations intensified during the event's impact period").

[0052] The overall definition of structured data can be expressed as: ; in, This is a "structured chart data set" used to drive subsequent chart rendering steps.

[0053] By using the above encapsulation method, the structured chart data can maintain consistency with the characteristics of financial events at the numerical level, while also possessing the ability to interpret events at the semantic level, thereby providing high-quality training data for model training.

[0054] S4. Chart rendering steps for preserving event characteristics: Input structured data into the rendering tool and generate the target chart image according to visual parameters consistent with the adaptive adjustment results; the rendering tool ensures that the chart image visually retains the data characteristics caused by the financial event by setting the chart type, color combination, legend position, text annotation method, and visual emphasis elements; specifically, it includes the following sub-steps: S410, Rendering Parameter Determination Sub-step: Based on the adaptive adjustment data generation rules and structured data obtained in steps S230 and S330, determine the rendering parameters used to maintain the characteristics of financial events. The rendering parameters include chart type, color scheme, visual emphasis weight, and legend position.

[0055] To ensure that the rendering is consistent with the trends, fluctuations, and impact characteristics of financial events, the following rendering controls are introduced: Trend color intensity parameters : ; in, This is the "Trend Color Intensity Parameter," used to automatically enhance the color of the polyline based on the trend slope k. "Base color intensity"; k is the trend slope; This is the "trend sensitivity coefficient," used to control the degree to which a trend affects color intensity.

[0056] Fluctuation transparency parameter : ; in, This is the "fluctuation transparency parameter," used to adjust the transparency or shadow depth of the fluctuation segment. A represents the "basic transparency"; A represents the fluctuation range after event adjustment. This is the "transparency sensitivity coefficient".

[0057] Event impact emphasizes weight : ; in, "Emphasizing the impact of events"; The emphasis is on "basic weighting"; This refers to the impact amplitude coefficient. This refers to the "impact sensitivity coefficient".

[0058] By using the above rendering parameters, the chart can be visually presented in a way that is consistent with the trend changes, fluctuations, and local impacts caused by the event.

[0059] S420, Chart Image Generation Sub-step: The structured data obtained in step S330 and the rendering parameters in step S410 are input into the rendering engine to generate a chart image that reflects the characteristics of the financial event. Chart rendering includes processing such as line drawing, fluctuation shadow drawing, and local impact annotation.

[0060] To ensure that the image accurately reflects the data features, the following rendering function is introduced: Trend line rendering function: ; in, This is the "trend line rendering function," which visualizes the trend components as a line; the color intensity is determined by... control.

[0061] Rendering function for fluctuating regions: ; in, This is the "Wave Area Rendering Function," used to draw the shadows of the upper and lower wave areas; the shadow opacity is determined by... control.

[0062] Event impact annotation function: ; in, This is an "event impact annotation function" used to add visual emphasis (such as a red dot or radial halo) to the event location; the degree of emphasis is determined by... control.

[0063] The final rendered image is generated by the following combination function: ; in, The "Final Chart Image Matrix" represents the visual result drawn by the rendering engine.

[0064] S430, Image Quality Inspection Sub-step: Perform a quality inspection on the chart image generated in step S420 to ensure that the image accurately reflects the characteristics of the financial event in visual presentation. The quality inspection includes trend consistency inspection, volatility consistency inspection, and impact location consistency inspection.

[0065] Introduce the following quality check functions: Trend consistency check function: ; in, This is the "Trend Consistency Index," with a value of 1 indicating that the trend rendering is correct.

[0066] Fluctuation consistency check function: ; in, It is a "volatility consistency index".

[0067] Impact position consistency check function: ; in, This is the "impact position consistency index".

[0068] Whether the final quality is up to standard is determined by the following comprehensive decision function: ; in, For "final image quality index", when When the time is right, it means the image rendering is completely correct; if If this happens, a re-render will be triggered.

[0069] S5. Multimodal Parsing Model Training and Output Steps: Align the chart images with the structured data using a multimodal approach. Perform joint training using the multimodal parsing model to enable the model to learn the correspondence between financial event features and the structured data. After training, use the multimodal parsing model to parse the input financial charts to obtain structured parsing results that reflect the characteristics of financial events. This includes the following sub-steps: S510, Training Sample Construction Sub-step: Based on the structured data object generated in step S330 and the chart image generated in step S420 The two are combined in a one-to-one correspondence to form multimodal training samples, each of which consists of image modality and structured modality.

[0070] To ensure the model can learn the correspondence between event features across different modalities, the following training sample representation is introduced: ; in, For "the "Multimodal training samples"; For "the "A matrix of chart images"; For "the A structured chart data object.

[0071] Structured data objects Includes: ; All the symbols have the same meaning as those in the steps described above.

[0072] The training samples constructed in the above manner ensure that the image content and structured data strictly correspond in terms of event characteristics such as trends, fluctuations, and shocks.

[0073] S520, Multimodal Joint Training Sub-step: Use the sample set constructed in step S510 to perform joint training on the multimodal analytical model, so that the model can simultaneously learn the correspondence between visual features in chart images and numerical features in structured data.

[0074] Introducing a multimodal feature extraction function: Visual encoding function: ; in, For "the "Visual feature vectors of a chart image"; For "visual encoding functions", such as those based on convolutional neural networks or visual Transformers.

[0075] Structured data encoding functions: ; in, For "the "Semantic feature vectors of structured data"; For "structured data encoding function".

[0076] To enable the model to learn the alignment relationship between two modalities, a multimodal alignment loss function is introduced: ; in, This is the "modal alignment loss function," used to constrain the spatial distance between visual features and structured features at corresponding event locations.

[0077] Meanwhile, to enhance the model's ability to learn event features, an event augmentation loss term is introduced: ; in, For "the model predicts the first "Influence intensity index of each sample"; The term refers to the "true impact strength index in structured data"; This is the "event reinforcement coefficient," used to enhance the model's ability to learn event features.

[0078] The final training loss function is: ; in, This is the "total loss function," used to guide gradient updates in multimodal analytical models.

[0079] The model minimizes This is to achieve consistent learning of visual and structural features.

[0080] S530, Outputting Analysis Results: After training is complete, input the chart image to be analyzed into the trained multimodal analysis model, and output the structured analysis result corresponding to the chart.

[0081] The analysis process is illustrated below: Visual encoding: ; Structured prediction: ; in, "Structured data feature vectors predicted by the model"; This is a "feature decoding function" used to restore visual features into a structured data format.

[0082] Reconstruction of structured parsing results: Transform the predicted feature vectors into the final structured output: ; Each symbol has the same meaning as described above, but adding "^" indicates a model prediction.

[0083] The final structured analysis result should simultaneously include trend features, volatility features, event impact features, and their corresponding text descriptions, enabling multimodal analysis of financial charts.

[0084] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0085] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0086] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0087] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and modules described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0088] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0089] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0090] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0091] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0092] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0093] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for generating and multimodal parsing financial chart data, characterized in that, Includes the following steps: S1. Analyze the target financial scenario based on the preset event recognition model to obtain financial event information related to the chart to be generated; the financial event information includes market volatility events, financial announcement events, interest rate change events, industry emergency event category information, as well as the direction and intensity of the impact of the event, which are used to indicate the impact characteristics of the financial event on the subsequent chart data generation. S2. Based on financial event information, construct data generation rules for generating structured chart data, and adaptively adjust the data generation rules according to the direction and intensity of the impact of the event. The adaptive adjustment includes adjusting the fluctuation range, trend pattern, amplitude change, key node position and text label description of the data, so that the generated structured data can reflect the characteristic changes caused by the financial event. S3. Generate structured data to describe the target chart based on adaptively adjusted data generation rules. The structured data includes horizontal axis categories, multiple vertical axis data series, legend labels, and title information, while maintaining the continuity of time sequence, consistency of data relationships, and integrity of event feature representation during the generation process. S4. Input the structured data into the rendering tool and generate the target chart image according to the visual parameters consistent with the adaptive adjustment results; Rendering tools enable charts to retain the data characteristics resulting from financial events in their visual presentation by setting chart type, color combinations, legend position, text annotation methods, and visual emphasis elements.

2. The method for generating and multimodal parsing financial chart data according to claim 1, characterized in that, It also includes S5, which performs multimodal alignment between chart images and structured data, performs joint training through a multimodal parsing model, so that the model learns the correspondence between financial event features and structured data; after training, the multimodal parsing model is used to parse the input financial charts to obtain structured parsing results that reflect the features of financial events.

3. The method for generating and multimodal parsing financial chart data according to claim 1, characterized in that, S1 specifically refers to: Obtain financial scenario information related to the target financial chart generation needs, including the market, involved entities, and time range, to determine the input scope for subsequent identification of financial events; Based on a pre-defined event recognition model, financial scenario information is analyzed to identify financial event information related to the scenario. Financial event information includes event category, event time, and the direction of the event's impact. Based on the financial event information, the impact intensity of the financial event is further determined to characterize the level of impact of the event on the degree of data volatility during the subsequent data generation process.

4. The method for generating and multimodal parsing financial chart data according to claim 1, characterized in that, S2 specifically refers to: The initial data generation rules are constructed based on general financial chart generation specifications. The rules include the horizontal axis category format, the vertical axis data generation method, and the legend text structure. By associating financial event information with the initial data generation rules, the impact of financial events on data trends, data distribution ranges, and key node locations can be determined. Based on the event feature mapping results, the initial data generation rules are adaptively adjusted so that the adjusted data generation rules can reflect the characteristics of financial events in terms of trend patterns, volatility, and text label descriptions.

5. The method for generating and multimodal parsing financial chart data according to claim 1, characterized in that, S3 specifically refers to: Based on the adaptively adjusted data generation rules, horizontal axis categories are generated that are continuous in time series, consistent in order, and can reflect the location of financial events. Multiple vertical axis data series are generated according to the adjusted data generation rules, so that each data series conforms to the characteristics of financial events in terms of numerical variation, trend and key node performance. The horizontal axis categories and vertical axis data are packaged into structured data according to a unified structure, and title text, legend text, and auxiliary explanatory text related to financial events are added.

6. The method for generating and multimodal parsing financial chart data according to claim 1, characterized in that, S4 specifically refers to: Based on the adaptively adjusted data generation rules, the rendering parameters used to reflect the characteristics of financial events are determined, including chart type, color combination, legend position, and visual emphasis. Charts and images are generated based on structured data and rendering parameters, so that the charts and images can visually reflect the trend changes, fluctuation characteristics and key nodes caused by financial events.

7. The method for generating and multimodal parsing financial chart data according to claim 6, characterized in that, S4 also includes: performing a quality check on the chart image to ensure that the trends and data patterns in the chart image are consistent with the financial event information, and re-performing the chart image generation.

8. The method for generating and multimodal parsing financial chart data according to claim 2, characterized in that, S5 specifically refers to: By mapping charts and images one-to-one with structured data, training sample pairs for multimodal analytical models are constructed. By using training samples to jointly train a multimodal analytical model, the model learns the correspondence between charts and images and structured data, as well as the cross-modal performance of financial event features.

9. The method for generating and multimodal parsing financial chart data according to claim 8, characterized in that, S5 also includes: after training, using a multimodal parsing model to parse the input financial charts and output structured parsing results that reflect the characteristics of financial events.

Images