Method for intelligently and rapidly generating analysis report based on market subject

Abstract

The invention discloses a method for intelligently and rapidly generating an analysis report based on a market subject. The method comprises the following steps: constructing a data warehouse by adopting multi-source data; obtaining to-be-analyzed information in the data warehouse according to the query demand; performing double-branch processing on the information to be analyzed, namely processing the structured data to obtain basic information, performing data conversion on the unstructured data to obtain unstructured text data, and performing data analysis on the unstructured text data to obtain unstructured processing information; according to the report format, a preset template is used for matching the basic information and the unstructured processing information to generate a preliminary report, and after data insight is correspondingly generated at the numerical characters in the preliminary report, the report is output. The market emotion thermodynamic diagram is generated through numeric field reminding marks of structured data and multi-dimensional semantic analysis of unstructured data, information is obtained from the market emotion thermodynamic diagram, the limitation of a fixed report template can be broken through, and self-adaptive generation of report content and format is achieved.

Description

Technical Field

[0001] This invention relates to the field of report generation technology, and more specifically, to a method for rapidly generating analytical reports based on the intelligence of market entities. Background Technology

[0002] In today's digital age, market entity analysis reports play a crucial role in corporate decision-making, industry research, and government regulation. Accurate, timely, and comprehensive analysis reports can provide strong support to relevant entities, helping them grasp market dynamics and formulate appropriate strategies.

[0003] Currently, the generation of market entity analysis reports mainly relies on hardcoding market entity information into the code and generating reports based on fixed templates. Specifically, developers pre-define the data acquisition methods and processing logic in the code, programming for specific data structures and report templates. When an analysis report needs to be generated, the system retrieves data from a fixed data source according to the preset code logic, then populates the data into a fixed template to form the final analysis report.

[0004] However, this technology has obvious drawbacks:

[0005] Hard-coded logic is rigid: In traditional software development practices, hard-coded implementations lead to serious maintainability and scalability issues. When using a hard-coded approach, data structure definitions and data acquisition logic are directly embedded in business code, resulting in high code coupling. This architecture violates the Single Responsibility Principle and Open / Closed Principle in software engineering, exposing significant vulnerabilities when facing changes in data structure or expansion of data sources. Specifically, changes to data structures require modifications to all code modules involved in data processing, including but not limited to data parsing, storage, transmission, and presentation layers. Due to the lack of abstraction layers, modifying one piece of code often triggers a chain reaction, affecting other functional modules. In enterprise-level systems, this cascading modification pattern easily leads to system compatibility issues. Furthermore, the need for new data sources requires developers to rewrite the data access logic, adapting not only to new protocol standards and data formats but also integrating it into existing business processes. This process involves complex interface design, data conversion and verification, and integration testing with existing systems, significantly increasing development time and manpower costs. In addition, the hard-coded approach weakens the system's testability and reusability. The lack of modular design makes it difficult to isolate unit tests, resulting in insufficient test coverage; duplicated code implementations reduce the reusability of software assets. In the long run, the accumulation of this technical debt will significantly increase system maintenance costs, reduce delivery efficiency, and ultimately affect the company's business agility and market competitiveness.

[0006] Poor Report Adaptability: In the field of enterprise-level data analysis and report generation, static template-based report generation mechanisms suffer from significant adaptability deficiencies. These templates typically employ hard-coded or predefined structures, strongly binding data presentation logic to business requirements, violating the dependency inversion principle and interface isolation principle in software design. Due to the lack of dynamic configuration and scalability design, the system exposes architectural limitations when dealing with diverse analysis scenarios. From a technical implementation perspective, fixed templates solidify data presentation dimensions, indicator calculation rules, and visualization methods in code or configuration files, lacking an abstract metadata-driven layer. When facing the analysis needs of different market players in various industries, significant differences exist in data characteristics (such as risk indicators in the financial industry and supply chain data in the retail industry) and analysis dimensions (such as time series analysis and regional distribution analysis), requiring customized modifications to the templates. This modification involves not only the reconstruction of the front-end presentation layer but also the simultaneous adjustment of the back-end data query logic, indicator calculation algorithms, and data mapping relationships. Due to the lack of modular design and layered architecture, any modification may trigger data consistency issues, leading to system compatibility failures. Furthermore, fixed template mechanisms cannot achieve adaptive adjustments based on user behavior. In practical applications, different user groups have significantly different reading habits and focuses when it comes to reports, and static templates cannot dynamically adapt to these personalized needs. This design flaw not only reduces the practical value of reports but also makes it difficult for enterprises to fully leverage the value of data in the process of data-driven decision-making, ultimately affecting business insights and market responsiveness.

[0007] Poor Unstructured Data Processing Capabilities: In the fields of big data analytics and market entity research, existing technologies face significant bottlenecks in processing unstructured data. Unstructured data (such as PDF documents, images, and audio) lacks a unified data model and standard format, making it impossible to directly apply traditional relational database storage and query technologies. This leads to multiple challenges in data processing, including semantic understanding, feature extraction, and information structuring. Current mainstream data processing frameworks still focus on structured data, relying on shallow techniques like traditional regular expression matching and keyword extraction for unstructured data processing, which struggles to overcome the semantic gap caused by data heterogeneity. From a technical perspective, PDF documents contain complex layouts and non-standardized text structures. While existing OCR (Optical Character Recognition) technology can extract text, it suffers from high recognition error rates when handling multilingual text, handwritten text, and graphical formulas. Furthermore, the visualization information in PDF documents, such as charts and flowcharts, lacks effective parsing methods, making it difficult to convert them into quantifiable analytical data. In image data processing, while deep learning-based computer vision technology has made progress in object detection and image classification, it lacks a deep understanding of the semantic information contained in images (such as data trends in business charts and the market value of trademarks), failing to effectively transform visual information into structured data required for market entity analysis. Furthermore, unstructured data processing lacks end-to-end automation. From data collection, cleaning, and feature engineering to model training, each step requires manual intervention and customized development, resulting in low processing efficiency and difficulty in scaling. This technological limitation prevents market entity analysis from fully integrating multi-source heterogeneous data, leading to missing analytical dimensions and loss of data value, ultimately affecting the completeness, depth, and decision-making reference value of the analysis report.

[0008] Therefore, this invention provides a method for rapidly generating analysis reports based on the intelligence of market entities. Summary of the Invention

[0009] In view of this, the present invention provides a method for rapidly generating analysis reports based on the intelligence of market entities, comprising:

[0010] Acquire multi-source data;

[0011] A data warehouse is constructed using the multi-source data. The data warehouse includes multiple data groups, each of which includes a unique identifier, a market entity name, and business information. The unique identifier, the market entity name, and the business information are in one-to-one correspondence, and the business information includes structured data and unstructured data.

[0012] Obtain the query requirements, which include query identifiers and report formats;

[0013] Based on the query identifier, query the data warehouse for the unique identifier that is the same as the query identifier, and obtain the business information corresponding to the unique identifier that is the same as the query identifier as the information to be analyzed;

[0014] The information to be analyzed undergoes a two-branch processing method, including: identifying numerical fields in the structured data of the information to be analyzed, formatting the numerical fields using thousands separators, and marking the numerical fields with reminders to obtain basic information; simultaneously, converting the unstructured data of the information to be analyzed into unstructured text data, and parsing the unstructured text data to obtain unstructured processed information, including: preprocessing the unstructured data, including: performing at least one of image denoising, tilt correction, and layout analysis on the unstructured data, wherein the image denoising uses... The filtering method includes at least one of mean filtering, median filtering, Gaussian filtering, bilateral filtering, nonlocal mean filtering, Fourier transform filtering, and wavelet transform filtering. The tilt correction includes: determining a tilt angle based on the text line direction, line direction, edge direction, or contour direction in the unstructured data, and performing rotation correction based on the tilt angle. The layout analysis includes: performing region segmentation on the unstructured data or constructing the hierarchical relationship of the unstructured data; performing text recognition on the preprocessed unstructured data to extract information text; filtering negative comments in the information text and desensitizing private data in the information text. The process includes: filtering negative comments in the information text by using convolutional neural networks and recurrent neural networks to classify statements in the information text into negative and positive information, and filtering out the negative information; desensitizing privacy data in the information text by scanning the information text, using regular expressions to match the privacy data in the information text according to preset rules, and replacing or partially hiding the privacy data; performing logical verification and format standardization conversion on the information text to obtain the unstructured text data; and performing word segmentation on the unstructured text data using natural language processing technology stacks. Multiple words are obtained; feature processing is performed on each of the multiple words to obtain multi-dimensional semantic vectors corresponding to the words; all the multi-dimensional semantic vectors are input into a sentiment classification model, and the sentiment classification model labels each multi-dimensional semantic vector with a label, which is positive, negative, or neutral; a sentiment trend heatmap is constructed using a sliding time window based on all the multi-dimensional semantic vectors labeled with the label, and the sentiment polarity distribution, sentiment intensity index, and sentiment entropy value are obtained from the sentiment trend heatmap. The sentiment trend heatmap, the sentiment polarity distribution, the sentiment intensity index, and the sentiment entropy value constitute the unstructured processing information.

[0015] The basic information and the unstructured processed information are merged to obtain merged data;

[0016] According to the report format, a preliminary report is generated by matching the merged data with a preset template. After generating data insights corresponding to the numeric characters in the preliminary report, the report is output. The process includes: the numeric characters have corresponding data types; querying the baseline value of the data type corresponding to the numeric characters; comparing the numeric characters with the baseline value of the data type corresponding to the numeric characters to obtain a comparison result; using the comparison result as the data insight corresponding to the numeric characters and marking it at the numeric characters; and then outputting the report.

[0017] Optionally, text recognition is performed on the preprocessed unstructured data to extract the information text, including:

[0018] Convolutional neural networks and recurrent neural networks are used to identify the preprocessed unstructured data and extract the information text.

[0019] Optionally, the unstructured text data is obtained by performing logical verification and format standardization on the information text, including:

[0020] The information text is scanned, and when legal terms or numerical formulas are scanned, the legal terms or numerical formulas are subjected to contextual logic verification processing, and then the format is standardized and converted to obtain the unstructured text data.

[0021] Optionally, obtaining the query requirement includes:

[0022] The query request is standardized by removing special characters and converting it into a preset format. Compared with existing technologies, the method for rapidly generating analysis reports based on market entity intelligence provided by this invention achieves at least the following beneficial effects:

[0023] This invention provides a method for rapidly generating analysis reports based on market entity intelligence. The method involves acquiring multi-source data and constructing a data warehouse using this data. Based on query requirements, the method retrieves the information to be analyzed from the data warehouse. The information is then processed in a two-branch manner: structured data is processed to obtain basic information, and unstructured data is preprocessed. Text recognition is performed on the preprocessed unstructured data to extract information text. Logical verification and format standardization are then performed on the information text to obtain unstructured text data. Natural language processing (NLP) technology is used to segment the unstructured text data into multiple words. Feature processing is then performed on each word to obtain the corresponding multidimensional semantic data. Semantic vectors are input into a sentiment classification model. The model labels each semantic vector as positive, negative, or neutral. A sentiment trend heatmap is constructed using a sliding time window based on all labeled semantic vectors. Sentiment polarity distribution, sentiment intensity index, and sentiment entropy are obtained from the heatmap. These three data constitute unstructured information. The basic information and unstructured information are merged to obtain merged data. A preliminary report is generated by matching the merged data with a pre-set template according to the report format. Data insights are generated at the corresponding numeric characters in the preliminary report, and then the final report is output. By efficiently utilizing data from multiple sources, combining structured and unstructured data, we can better understand information relevant to market participants, thereby improving the accuracy and intelligence of data processing. The information to be analyzed is processed in two branches: numerical fields in structured data are used to highlight key information, while multidimensional semantic analysis of unstructured data generates a market sentiment heatmap. Information extracted from this heatmap breaks through the limitations of fixed report templates, enabling adaptive generation of report content and format. Subsequent data insights are then applied to the initial report to generate targeted and practical analytical content, enhancing readability.

[0024] Of course, any product implementing this invention does not necessarily need to achieve all of the technical effects described above at the same time.

[0025] Other features and advantages of the invention will become clear from the following detailed description of exemplary embodiments of the invention with reference to the accompanying drawings. Attached Figure Description

[0026] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments of the invention and, together with their description, serve to explain the principles of the invention.

[0027] Figure 1 This is a flowchart illustrating a method for rapidly generating analysis reports based on the intelligence of market entities, provided by this invention.

[0028] Figure 2 This is a flowchart illustrating a process for acquiring data from multiple sources.

[0029] Figure 3 This is another flowchart illustrating the method for rapidly generating analysis reports based on the intelligence of market entities provided by this invention.

[0030] Figure 4 This is another flowchart illustrating the method for rapidly generating analysis reports based on the intelligence of market entities provided by this invention. Detailed Implementation

[0031] Various exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be noted that, unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps set forth in these embodiments do not limit the scope of the invention.

[0032] The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the invention or its application or use.

[0033] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and equipment should be considered part of the specification.

[0034] In all the examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values.

[0035] It should be noted that similar labels and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be discussed further in subsequent figures.

[0036] Example 1

[0037] Reference Figure 1 and Figure 2 , Figure 1 This is a flowchart illustrating a method for rapidly generating analysis reports based on the intelligence of market entities, provided by the present invention. Figure 2 This is a flowchart illustrating a process for acquiring multi-source data, and is a specific embodiment of the method for rapidly generating analysis reports based on market entity intelligence provided in this embodiment, including:

[0038] S101: Acquire multi-source data;

[0039] S102: A data warehouse is built using multi-source data. The data warehouse includes multiple data groups. Each data group includes a unique identifier, a market entity name, and business information. The unique identifier, market entity name, and business information are in one-to-one correspondence. The business information includes structured data and unstructured data.

[0040] S103: Obtain query requirements, which include query identifiers and report formats;

[0041] S104: Based on the query identifier, query the data warehouse for a unique identifier that is the same as the query identifier, and obtain the business information corresponding to the unique identifier that is the same as the query identifier as the information to be analyzed;

[0042] S105: Perform dual-branch processing on the information to be analyzed, including: identifying numerical fields in the structured data of the information to be analyzed, performing thousands separator formatting on the numerical fields, and marking the numerical fields with reminders to obtain basic information; simultaneously, performing data transformation on the unstructured data of the information to be analyzed to obtain unstructured text data, and performing data parsing on the unstructured text data to obtain unstructured processed information, including: preprocessing the unstructured data, including: performing at least one of image noise reduction, tilt correction, and layout analysis on the unstructured data, image... Noise reduction employs at least one of the following: mean filtering, median filtering, Gaussian filtering, bilateral filtering, nonlocal mean filtering, Fourier transform filtering, and wavelet transform filtering. Tilt correction includes determining the tilt angle based on the text line direction, line direction, edge direction, or contour direction in the unstructured data, and performing rotation correction based on the tilt angle. Layout analysis includes region segmentation of the unstructured data or constructing hierarchical relationships within the unstructured data. Text recognition is performed on the preprocessed unstructured data to extract information text. Negative comments and hidden information in the information text are filtered out. The process of desensitizing private data includes: filtering negative comments in the information text by using convolutional neural networks and recurrent neural networks to classify statements in the information text into negative and positive information, and filtering out negative information; desensitizing private data in the information text by scanning the information text, using regular expressions to match the private data in the information text according to preset rules, and replacing or partially hiding the private data; performing logical verification and format standardization on the information text to obtain unstructured text data; using natural language processing technology stack to segment the unstructured text data to obtain multiple words; performing feature processing on multiple words to obtain multi-dimensional semantic vectors corresponding to the words; inputting all multi-dimensional semantic vectors into a sentiment classification model, which labels each multi-dimensional semantic vector as positive, negative, or neutral; constructing a sentiment trend heatmap using a sliding time window based on all labeled multi-dimensional semantic vectors, and obtaining sentiment polarity distribution, sentiment intensity index, and sentiment entropy value from the sentiment trend heatmap. The sentiment trend heatmap, sentiment polarity distribution, sentiment intensity index, and sentiment entropy value constitute unstructured processed information.

[0043] S106: Merge basic information and unstructured processed information to obtain merged data;

[0044] S107: Based on the report format, use a preset template to match and merge data to generate a preliminary report. After generating data insights corresponding to the numeric characters in the preliminary report, output the report, including: the numeric characters have corresponding data types, query the baseline value of the data type corresponding to the numeric characters, compare the numeric characters with the baseline value of the data type corresponding to the numeric characters to obtain the comparison result, use the comparison result as the data insight corresponding to the numeric characters and mark it at the numeric characters, and then output the report.

[0045] It should be noted that in step S102, the unique identifier can be the Unified Social Credit Code. Setting the Unified Social Credit Code and the market entity name as a composite primary key ensures the uniqueness and accuracy of the data set, laying a solid foundation for subsequent data integration work. (Refer to...) Figure 2Multi-source data can be downloaded from the official websites of various bureaus or commissions, supplementing the data of market entities through data sources from multiple departments. This enriches the information of market entities, which is essentially business information. Business information includes not only basic information about the market entity but also business attribute fields specific to each department, making the content of market entities more comprehensive and detailed. This improves the integrity and consistency of the market entity data and provides strong support for subsequent data analysis and decision support. In step S105, to improve query efficiency, a dual-branch parallel processing mechanism is adopted. During the resource allocation phase, independent memory space and computing resources are allocated to the two threads for basic information processing and tag information processing, ensuring that both can run efficiently simultaneously. This parallel processing method not only shortens the overall query time but also improves resource utilization. Through the dual-branch query mechanism, user needs can be responded to quickly while ensuring the integrity and accuracy of data processing. In step S105, negative comments in the information text are filtered, and private data in the information text is desensitized. Specifically, in the field of text information processing, using a toxic text classifier to filter negative comments is a key means to ensure the positivity and effectiveness of text information such as public opinion data. The toxic text classifier, leveraging advanced Natural Language Processing (NLP) technology, can accurately identify negative elements in text. It first preprocesses the input text through word segmentation, part-of-speech tagging, and semantic analysis to extract key features. Then, using trained machine learning models such as logistic regression and support vector machines, or deep learning models such as recurrent neural networks (RNNs) and convolutional neural networks (CNNs), it classifies the text to determine whether it contains negative language such as insults, attacks, falsehoods, and hate speech. For language judged as negative, the classifier filters it out, retaining only positive, healthy, and valuable text information, providing a reliable foundation for subsequent public opinion analysis and data mining. Regarding data security protection, identifying and anonymizing private data such as phone numbers and email addresses using regular expressions is a crucial step. Regular expressions are powerful pattern matching tools that can accurately locate private data in text according to preset rules; for example, for email addresses, it can match strings containing the @ symbol and conforming to a specific format. After identifying private data, the system will process it according to established de-identification rules, such as replacing the middle four digits of the phone number with asterisks and hiding some characters in the email address. This ensures data security, complies with relevant regulations on data privacy protection, prevents the leakage of private information, and safeguards the legitimate rights and interests of users.In step S105, the unstructured text data is parsed, specifically, sentiment analysis is performed on the unstructured text data. Natural language processing (NLP) techniques are used to segment the unstructured text data into multiple words. This can be based on bidirectional word segmentation algorithms (such as jieba segmentation combined with HMM hidden Markov models), part-of-speech tagging (POS tagging), named entity recognition (NER), and stop word filtering. Regular expressions are used to handle special symbols and punctuation noise. Feature processing is then performed on each word to obtain its corresponding multidimensional semantic vector. This can be achieved by fusing bag-of-words (BOW) models, TF-IDF weight calculation, and pre-trained word vectors (such as Word2Vec and GloVe), combined with dynamic representation techniques based on context (pre-trained models such as ELMo and BERT). All multidimensional semantic vectors are then input into a sentiment classification model. This model can employ a hybrid neural network structure: the bottom layer uses Bi-LSTM to capture long-distance dependencies, the middle layer introduces an attention mechanism to focus on key sentiment words, and the top layer combines a Softmax classifier to achieve a three-class classification task: positive, negative, and neutral. To address the specific characteristics of domain-specific data, transfer learning techniques (such as RoBERTa-wwm) can be used to fine-tune pre-trained models, and model performance can be evaluated using metrics such as F1-score and AUC-ROC. Sentiment polarity distribution, sentiment intensity index, and sentiment entropy can be obtained from sentiment trend heatmaps. Sentiment polarity distribution involves calculating the proportions of multidimensional semantic vectors labeled positive, negative / positive, and neutral. The sentiment intensity index can be calculated using degree adverb weights. These factors contribute to the analysis of market sentiment fluctuations.

[0046] Understandably, by efficiently utilizing data from multiple sources, combining structured and unstructured data, it is possible to better understand information relevant to market participants, thereby improving the accuracy and intelligence of data processing. By employing a dual-branch processing approach for the information to be analyzed, using numerical fields from structured data to highlight key information and conducting multi-dimensional semantic analysis of unstructured data to generate a market sentiment heatmap, information can be extracted from this heatmap. This approach breaks through the limitations of fixed report templates, enabling adaptive generation of report content and format. Subsequent data insights can be applied to the initial report to generate targeted and practical analytical content, enhancing the report's readability.

[0047] Example 2

[0048] Combination Figure 3 and Figure 4 , Figure 3 This is another flowchart illustrating the method for rapidly generating analysis reports based on the intelligence of market entities provided by the present invention. Figure 4 This is another flowchart illustrating a method for rapidly generating analysis reports based on market entity intelligence provided by the present invention, used to explain another specific embodiment of the method for rapidly generating analysis reports based on market entity intelligence provided by the present invention, including:

[0049] S201: Acquire multi-source data;

[0050] S202: A data warehouse is built using multi-source data. The data warehouse includes multiple data groups. Each data group includes a unique identifier, a market entity name, and business information. The unique identifier, market entity name, and business information are in one-to-one correspondence. The business information includes structured data and unstructured data.

[0051] S203: Obtain query requirements, including: standardizing query requirements, including: removing special characters from query requirements and converting query requirements into a preset format. Query requirements include query identifiers and report formats.

[0052] S204: Based on the query identifier, query the data warehouse for a unique identifier that is the same as the query identifier, and obtain the business information corresponding to the unique identifier that is the same as the query identifier as the information to be analyzed;

[0053] S205: Perform dual-branch processing on the information to be analyzed, including: identifying numerical fields in the structured data of the information to be analyzed, performing thousands separator formatting on the numerical fields, and marking the numerical fields with reminders to obtain basic information; simultaneously, performing data transformation on the unstructured data of the information to be analyzed to obtain unstructured text data, and performing data parsing on the unstructured text data to obtain unstructured processed information, including: preprocessing the unstructured data, including: performing at least one of image denoising, tilt correction, and layout analysis on the unstructured data, with image denoising using mean filtering, median filtering, Gaussian filtering, bilateral filtering, and nonlocal filtering. The system includes at least one of mean filtering, Fourier transform filtering, and wavelet transform filtering; tilt correction, including: determining the tilt angle based on the text line direction, line direction, edge direction, or contour direction in the unstructured data, and performing rotation correction based on the tilt angle; layout analysis, including: segmenting unstructured data into regions or constructing hierarchical relationships for unstructured data; text recognition of preprocessed unstructured data to extract information text, including: using convolutional neural networks and recurrent neural networks to recognize preprocessed unstructured data and extract information text; filtering negative comments in information text and desensitizing private data in information text to filter information. Negative statements in the text include: using convolutional neural networks and recurrent neural networks to classify statements in the information text into negative and positive information, filtering out negative information; desensitizing private data in the information text, including: scanning the information text, using regular expressions to match private data in the information text according to preset rules, and replacing or partially hiding the private data; performing logical verification and format standardization conversion on the information text to obtain unstructured text data, including: scanning the information text, and when legal terms or numerical formulas are scanned, performing contextual logical verification on the legal terms or numerical formulas, and then performing format standardization conversion to obtain unstructured text data. Structured text data is processed; unstructured text data is segmented using natural language processing techniques to obtain multiple words; feature processing is performed on each word to obtain multi-dimensional semantic vectors corresponding to the words; all multi-dimensional semantic vectors are input into a sentiment classification model, which labels each multi-dimensional semantic vector as positive, negative, or neutral; a sentiment trend heatmap is constructed using a sliding time window based on all labeled multi-dimensional semantic vectors, and sentiment polarity distribution, sentiment intensity index, and sentiment entropy value are obtained from the sentiment trend heatmap. The sentiment trend heatmap, sentiment polarity distribution, sentiment intensity index, and sentiment entropy value constitute unstructured processed information.

[0054] S206: Merge basic information and unstructured processed information to obtain merged data;

[0055] S207: Based on the report format, use a pre-set template to match and merge data to generate a preliminary report. After generating data insights corresponding to the numeric characters in the preliminary report, output the report, including: the numeric characters have corresponding data types, query the baseline value of the data type corresponding to the numeric characters, compare the numeric characters with the baseline value of the data type corresponding to the numeric characters to obtain the comparison result, use the comparison result as the data insight corresponding to the numeric characters and mark it at the numeric characters, and then output the report.

[0056] It should be noted that in step S203, the query request is standardized, including removing special characters and converting the query request into a preset format, such as "Yonghui Supermarket Enterprise Report". This first step standardizes the natural language input by the user, including removing special characters and unifying the format, to ensure the accuracy of subsequent processing. This step not only improves the standardization of the query statement but also lays the foundation for subsequent semantic parsing and data processing. Through input standardization, the system can effectively avoid parsing errors caused by differences in user input, thereby improving overall query efficiency.

[0057] In step S205, the unstructured data undergoes at least one of image denoising, tilt correction, and layout analysis to optimize the quality of the input unstructured data and ensure accurate parsing of complex layouts (such as multi-column layouts, nested tables, and mixed handwriting). Of course, the specific techniques for image denoising, tilt correction, and layout analysis can be adjusted according to actual needs, and this embodiment does not impose specific limitations on them. Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) are used to identify the preprocessed unstructured data, extracting information text and achieving accurate recognition of printed text, handwritten text, and special symbols. Especially in complex scenarios such as low-resolution images and text covered by seals, the recognition accuracy is improved by more than 30% compared to traditional OCR. After performing contextual logic verification on legal terms or data formulas, format standardization conversion is performed to obtain unstructured text data. Specifically, post-processing verification is performed through semantic analysis and a rule engine. Contextual logic verification and format standardization conversion are performed on key information such as legal terms in contract clauses and numerical formulas in financial statements. Taking the financial industry as an example, when processing massive amounts of scanned financial statements, OCR enhancement technology can quickly locate key areas such as balance sheets and cash flow statements, automatically extract account values, and generate structured data reports. Compared to manual data entry, efficiency is improved by more than 80%, with an error rate controlled below 0.1%. In the legal field, for PDF contracts containing complex clauses, the system can accurately identify core information such as the rights and obligations of both parties, breach of contract clauses, and validity periods, automatically generating risk point summaries to provide data support for contract review.

[0058] In step S206, basic information and unstructured data are merged using unique identifiers such as the unified social credit code. Subsequently, the unstructured data is added to the basic information record as a new field, and issues such as field naming conflicts and inconsistent data types are resolved. This data fusion method not only expands the dimensions of the data but also improves the readability and analytical value of the data. Through conflict handling, the integrity and consistency of the merged data are ensured.

[0059] In step S207, a pre-set template is used to generate a summary description based on the merged data, and data insights are dynamically generated, such as "exceeding the investment benchmark value by 10%". This natural language generation function makes the output more intuitive and user-friendly, helping users quickly understand the insights behind the data. Through formatted output, the system not only improves the readability of the data but also enhances the user experience. (Refer to...) Figure 3 After obtaining the query requirements, a preset template can be matched first, and then the preset template can be used directly in step S207. Of course, a preset template can also be matched and used in step S207. This embodiment does not impose specific restrictions on this.

[0060] Understandably, semantic transformation capabilities allow for flexible processing of data of different types and structures, dynamically adjusting report content and format based on user-defined analysis dimensions and needs. No longer limited by fixed code logic and templates, it can quickly adapt to changes in data structures and diverse analysis scenarios, meeting the personalized needs of different users. By aggregating and processing data from multiple sources, including parsing unstructured data, data cleaning, deduplication, and anonymization, it can comprehensively and accurately acquire and process relevant information about market entities. Combined with sentiment analysis and trend prediction technologies, it delves deeper into the potential value of the data, providing richer and more valuable content for analysis reports. The dynamic generation of analysis reports reduces the workload of manual intervention and code modification, enabling rapid response to data changes and timely generation of the latest analysis reports. Simultaneously, model-based intelligent processing and accurate semantic transformation improve the accuracy of data processing and the logical consistency of report content, making the generated analysis reports more scientific and reliable, providing stronger support for decision-making. Through data cleaning, deduplication, anonymization, and toxic text filtering technologies, it ensures the quality and security of input data, preventing low-quality data and privacy information from affecting analysis results, thus improving data processing reliability and user trust.

[0061] In summary, the method for rapidly generating analysis reports based on market entity intelligence provided by this invention constructs a cross-departmental data fusion architecture, ensures uniqueness through joint primary keys, and integrates multi-dimensional business data. It develops a full-link technology of "preprocessing-deep learning recognition-semantic verification," improving the accuracy of OCR technology in complex scenarios by over 30% and increasing financial report processing efficiency by 80% compared to manual methods. Utilizing a dual-branch parallel processing mechanism, it achieves simultaneous parsing of basic information and tag information, improving response efficiency by 50%. It constructs a multi-level semantic analysis model (specifically, Bi-LSTM + attention mechanism), supporting sentiment classification and trend prediction, and dynamically generating market sentiment heatmaps. Through semantic conversion, it breaks through the limitations of fixed templates, supporting adaptive generation of report content and format. Adopting a layered architecture of "data source-intelligent agent-application," it achieves loose coupling of modules through APIs, eliminating the need to modify the underlying code when adding new data sources or analysis dimensions, significantly improving maintainability and scalability. It constructs an end-to-end automated process, reducing report generation time to 1 / 6 of traditional solutions, meeting diverse scenario needs.

[0062] As can be seen from the above embodiments, the method for rapidly generating analysis reports based on the intelligence of market entities provided by the present invention achieves at least the following beneficial effects:

[0063] This invention provides a method for rapidly generating analysis reports based on market entity intelligence. The method involves acquiring multi-source data and constructing a data warehouse using this data. Based on query requirements, the method retrieves the information to be analyzed from the data warehouse. The information is then processed in a two-branch manner: structured data is processed to obtain basic information, and unstructured data is preprocessed. Text recognition is performed on the preprocessed unstructured data to extract information text. Logical verification and format standardization are then performed on the information text to obtain unstructured text data. Natural language processing (NLP) technology is used to segment the unstructured text data into multiple words. Feature processing is then performed on each word to obtain the corresponding multidimensional semantic data. Semantic vectors are input into a sentiment classification model. The model labels each semantic vector as positive, negative, or neutral. A sentiment trend heatmap is constructed using a sliding time window based on all labeled semantic vectors. Sentiment polarity distribution, sentiment intensity index, and sentiment entropy are obtained from the heatmap. These three data constitute unstructured information. The basic information and unstructured information are merged to obtain merged data. A preliminary report is generated by matching the merged data with a pre-set template according to the report format. Data insights are generated at the corresponding numeric characters in the preliminary report, and then the final report is output. By efficiently utilizing data from multiple sources, combining structured and unstructured data, we can better understand information relevant to market participants, thereby improving the accuracy and intelligence of data processing. The information to be analyzed is processed in two branches: numerical fields in structured data are used to highlight key information, while multidimensional semantic analysis of unstructured data generates a market sentiment heatmap. Information extracted from this heatmap breaks through the limitations of fixed report templates, enabling adaptive generation of report content and format. Subsequent data insights are then applied to the initial report to generate targeted and practical analytical content, enhancing readability.

[0064] While specific embodiments of the invention have been described in detail by way of examples, those skilled in the art should understand that the examples are for illustrative purposes only and not intended to limit the scope of the invention. Those skilled in the art should understand that modifications can be made to the above embodiments without departing from the scope and spirit of the invention. The scope of the invention is defined by the appended claims.

Claims

1. A method for rapidly generating analytical reports based on the intelligence of market entities, characterized in that, include: Acquire multi-source data; A data warehouse is constructed using the multi-source data. The data warehouse includes multiple data groups, each of which includes a unique identifier, a market entity name, and business information. The unique identifier, the market entity name, and the business information are in one-to-one correspondence, and the business information includes structured data and unstructured data. Obtain the query requirements, which include query identifiers and report formats; Based on the query identifier, query the data warehouse for the unique identifier that is the same as the query identifier, and obtain the business information corresponding to the unique identifier that is the same as the query identifier as the information to be analyzed; The information to be analyzed undergoes a two-branch processing method, including: identifying numerical fields in the structured data of the information to be analyzed, formatting the numerical fields using thousands separators, and marking the numerical fields with reminders to obtain basic information; simultaneously, converting the unstructured data of the information to be analyzed into unstructured text data, and parsing the unstructured text data to obtain unstructured processed information, including: preprocessing the unstructured data, including: performing at least one of image denoising, tilt correction, and layout analysis on the unstructured data, wherein the image denoising uses... The filtering method includes at least one of mean filtering, median filtering, Gaussian filtering, bilateral filtering, nonlocal mean filtering, Fourier transform filtering, and wavelet transform filtering. The tilt correction includes: determining a tilt angle based on the text line direction, line direction, edge direction, or contour direction in the unstructured data, and performing rotation correction based on the tilt angle. The layout analysis includes: performing region segmentation on the unstructured data or constructing the hierarchical relationship of the unstructured data; performing text recognition on the preprocessed unstructured data to extract information text; filtering negative comments in the information text and desensitizing private data in the information text. The process includes: filtering negative comments in the information text by using convolutional neural networks and recurrent neural networks to classify statements in the information text into negative and positive information, and filtering out the negative information; desensitizing privacy data in the information text by scanning the information text, using regular expressions to match the privacy data in the information text according to preset rules, and replacing or partially hiding the privacy data; performing logical verification and format standardization conversion on the information text to obtain the unstructured text data; and performing word segmentation on the unstructured text data using natural language processing technology stacks. Multiple words are obtained; feature processing is performed on each of the multiple words to obtain multi-dimensional semantic vectors corresponding to the words; all the multi-dimensional semantic vectors are input into a sentiment classification model, and the sentiment classification model labels each multi-dimensional semantic vector with a label, which is positive, negative, or neutral; a sentiment trend heatmap is constructed using a sliding time window based on all the multi-dimensional semantic vectors labeled with the label, and the sentiment polarity distribution, sentiment intensity index, and sentiment entropy value are obtained from the sentiment trend heatmap. The sentiment trend heatmap, the sentiment polarity distribution, the sentiment intensity index, and the sentiment entropy value constitute the unstructured processing information. The basic information and the unstructured processed information are merged to obtain merged data; According to the report format, a preliminary report is generated by matching the merged data with a preset template. After generating data insights corresponding to the numeric characters in the preliminary report, the report is output. The process includes: the numeric characters have corresponding data types; querying the baseline value of the data type corresponding to the numeric characters; comparing the numeric characters with the baseline value of the data type corresponding to the numeric characters to obtain a comparison result; using the comparison result as the data insight corresponding to the numeric characters and marking it at the numeric characters; and then outputting the report.

2. The method for rapidly generating analysis reports based on the intelligence of market entities according to claim 1, characterized in that, Text recognition is performed on the preprocessed unstructured data to extract the information text, including: Convolutional neural networks and recurrent neural networks are used to identify the preprocessed unstructured data and extract the information text.

3. The method for rapidly generating analysis reports based on the intelligence of market entities according to claim 1, characterized in that, The unstructured text data is obtained by performing logical verification and format standardization on the information text, including: The information text is scanned, and when legal terms or numerical formulas are scanned, the legal terms or numerical formulas are subjected to contextual logic verification processing, and then the format is standardized and converted to obtain the unstructured text data.

4. The method for rapidly generating analysis reports based on the intelligence of market entities according to claim 1, characterized in that, Obtaining the query requirement includes: The query request is standardized by removing special characters from the query request and converting the query request into a preset format.

Images