Digital reading system based on knowledge graph and text image recognition technology

Abstract

The present application discloses a digital reading system based on knowledge graph and text image recognition technology. The system includes: a collection module for collecting financial data and a first financial document from multiple financial data sources; a construction module for identifying target entities in the financial field in the financial data and constructing a graph structure of the financial knowledge graph; a first parsing module for receiving a second financial document uploaded by a user, identifying target keywords in the second financial document, and generating a mind map corresponding to the target keywords according to a preset hierarchical structure; a second parsing module for identifying document elements in the first financial document; a typesetting module for acquiring screen information of a mobile terminal device; a watermark adding module for generating a digital watermark based on copyright information and custom information; and a question processing module for receiving query questions. The present application solves the technical problem of low efficiency in financial knowledge processing in related digital reading systems.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence, and more specifically, to a digital reading system based on knowledge graph and text image recognition technology. Background Art

[0002] With the rapid development of information technology, digital reading has become an important way for modern people to obtain information, learn knowledge, and relax and entertain.

[0003] In the field of digital reading, especially in the financial industry, traditional reading methods can no longer meet the growing demand for information processing. With the massive accumulation of internal data and external market information in financial institutions, employees are facing challenges such as information overload, inefficient knowledge absorption, and unmet personalized reading needs. Although digital reading platforms on the current market have tried to introduce artificial intelligence technology, they still have shortcomings in automated text understanding, personalized recommendations, and immersive reading experience.

[0004] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0005] The present application provides a digital reading system based on knowledge graph and text image recognition technology to at least solve the technical problem of low efficiency in financial knowledge processing of related digital reading systems.

[0006] According to one aspect of the present application, a digital reading system based on knowledge graph and text image recognition technology is provided, including: a collection module for collecting financial data and a first financial document from multiple financial data sources; a construction module for identifying target entities in the financial field in the financial data, extracting semantic relationships between the target entities, and constructing a graph structure of the financial knowledge graph with the target entities as nodes and the semantic relationships as edges; a first parsing module for receiving a second financial document uploaded by a user, identifying target keywords in the second financial document based on text image recognition technology, and generating a mind map corresponding to the target keywords according to a preset hierarchical structure; a second parsing module for identifying document elements in the first financial document, and performing structured parsing on the document elements to obtain parsing results, wherein the document elements include at least one of the following: text paragraphs, tables, and text elements. Format, graphics; a typesetting module, used to obtain screen information of the mobile terminal device, determine the typesetting strategy according to the parsing results and the screen information, and adapt the format of the document element according to the typesetting strategy to obtain a first element; a watermark adding module, used to generate a digital watermark according to copyright information and custom information, add the digital watermark to the first element, and obtain a target element, wherein the custom information includes: a reading platform identifier and a document purpose description; a question processing module, used to receive a query question, if the query question includes a first preset keyword related to a mind map, display the mind map, if the query question includes a second preset keyword related to a knowledge graph, determine the knowledge graph subgraph corresponding to the query question in the graph structure, and display the knowledge graph subgraph, if the query question includes a third preset keyword related to a document element, display the target element.

[0007] Optionally, a construction module is used to identify target entities in the financial field in financial data, extract semantic relationships between target entities, and construct a graph structure of a financial knowledge graph with target entities as nodes and semantic relationships as edges, including the following steps: preprocessing text data in financial data, wherein the preprocessing includes: removing noise characters, unifying text formats, processing abbreviations and synonym replacements; using a named entity recognition algorithm to scan the preprocessed text data to identify target entities in the financial field, wherein the target entities include: financial institutions, financial products, financial market participants, and economic indicators; using preset financial field knowledge rules to extract semantic relationships between target entities, wherein semantic relationships include: belonging relationships, influence relationships, and transaction relationships, wherein belonging relationships include: a certain financial product belongs to a certain financial institution, influence relationships include: a certain economic indicator affects a certain financial market, and transaction relationships include: transaction relationships between financial market participants.

[0008] Optionally, a question processing module receives a query question, and if the query question includes a second preset keyword related to the knowledge graph, determines a knowledge graph subgraph corresponding to the query question in the graph structure, and displays the knowledge graph subgraph, including the following steps: receiving a query question input by a user, and if the query question includes a second preset keyword related to the knowledge graph, converting the query question into a query statement for the graph structure; executing the query statement, determining a knowledge graph subgraph corresponding to the query question in the graph structure, and displaying the knowledge graph subgraph through a graphical interface.

[0009] Optionally, the parsing module performs structured parsing on the document elements to obtain parsing results, including the following steps: for tables in the document elements, parsing the structural information of the table, wherein the structural information includes: the number of rows, the number of columns, the row width, the column width, extracting the text content in the table, and determining the importance index of the text content in the table; for graphics in the document elements, identifying the type of graphics and data identifiers.

[0010] Optionally, the typesetting module determines the typesetting strategy based on the parsing results and the screen information, including the following steps: for text paragraphs in the document element, adjusting the line spacing and character spacing according to the screen width; for tables in the document element, when the screen width is less than a first preset threshold, if the table width exceeds the screen, a scroll bar is used to display the table content, or columns are selectively hidden based on the importance index of the text content in the table, and an operation button for expanding all columns is provided; for graphics in the document element, the graphic size is scaled proportionally according to the screen space and the graphic type.

[0011] Optionally, the watermark adding module adds a digital watermark to the first element, including the following steps: for a text paragraph in the first element, obtain the document theme, determine the theme keyword with the highest correlation index with the document theme, and determine the importance score of the text paragraph according to the frequency of occurrence of the theme keyword in the text paragraph; add a first number of digital watermarks at a first preset position of a text paragraph whose importance score is less than a third preset threshold, and add a second number of digital watermarks at a second preset position of a text paragraph whose importance score is greater than the third preset threshold, wherein the second number is greater than the first number; for a table in the first element, determine the table length and table width of the table, and When the table length is greater than a fourth preset threshold and / or the table width is greater than a fifth preset threshold, a digital watermark is added to the first row, the first column, the last row and the last column; when the table length is not greater than the fourth preset threshold and / or the table width is not greater than the fifth preset threshold, a digital watermark is added at the third preset position of the table; for the graphics in the first element, the graphics are divided into multiple sub-blocks, each sub-block is subjected to a discrete cosine transform to obtain multiple frequency domain coefficients, a target area whose frequency domain coefficients are within a preset interval is selected, the frequency domain coefficients within the target area are modified according to the digital watermark, and after the modification is completed, the graphics are converted back to the spatial domain through an inverse discrete cosine transform.

[0012] Optionally, the electronic reading system based on knowledge graph and intelligent typesetting also includes: a detection module, which is used to perform the following steps after displaying the target element: determine whether the mobile terminal device performs screenshot and / or copy behavior on the target data that meets the preset rules through a pre-trained machine learning model, wherein the pre-trained machine learning model is obtained by training in the following method: collecting screenshot event data and copy event data, performing feature extraction on the screenshot event data to obtain a first feature, wherein the first feature includes: screenshot time, screenshot area, and screenshot method; performing feature extraction on the copy event data to obtain a second feature, wherein the second feature Including: the data type, copy source, and copy time of the copied content; use the first feature and the second feature to train the machine learning model. During the training process, the difference between the predicted result and the true label is calculated according to the loss function of the model, and the parameters of the model are adjusted through the stochastic gradient descent algorithm to minimize the loss function. When the loss function meets the preset conditions, a machine learning model that has completed the training is obtained; when it is determined that the mobile terminal device performs screenshot and / or copy behavior on the target data, the digital watermark of the document where the target data is located is determined, the copyright information in the digital watermark is extracted, and a target alarm message including the copyright information is generated.

[0013] Optionally, the electronic reading system based on knowledge graph and intelligent typesetting also includes: a personalized recommendation module, which is used to perform the following steps: collecting the reading behavior data of the user on the digital reading platform on the front-end interactive interface, wherein the reading behavior data includes: reading history data, collection preference data and search record data, the reading history data is used to record the identification information, reading start time and reading end time of the First Financial documents read by the user, the collection preference data is used to record the document category information, identification information and collection time of the user's collection, and the search record data is used to record the search keywords and search time entered by the user; the reading behavior data is transmitted to the back-end data storage center through an encrypted data transmission channel; the reading behavior data is stored in the back-end data storage center; The data is cleaned and normalized, and features are extracted from the processed reading behavior data to obtain a target feature vector, and a clustering algorithm is used to divide the target feature vector into different interest groups, wherein the feature extraction includes: encoding the document according to the subject and field dimensions, calculating the user's preference scores for documents of different categories based on the collection preference data, and extracting keywords with an appearance frequency greater than a second preset threshold from the search record data; analyzing the time series of the user's reading behavior data to predict the user's future reading interests; generating recommended content based on the user's reading behavior data, future reading interests, and the reading behavior data of other users in the interest group, and displaying the recommended content on the user's personal interface.

[0014] Optionally, the first parsing module is also used to perform the following steps: determining keywords in the second financial document whose correlation index with the document topic is greater than a preset threshold, summarizing the article theme of the second financial document based on natural language processing technology, and generating an article summary including the article theme; receiving questions about the second financial document input by the user, and determining the answer corresponding to the question using semantic understanding and knowledge retrieval technology; displaying an editing area, wherein the editing area is used to receive text input information; receiving and responding to target instructions, and translating the second financial document into the target language indicated by the target instructions.

[0015] In the present application, a collection module is used to collect financial data and a first financial document from multiple financial data sources; a construction module is used to identify target entities in the financial field in the financial data, extract semantic relationships between target entities, and use target entities as nodes and semantic relationships as edges to construct a graph structure of a financial knowledge graph; a first parsing module is used to receive a second financial document uploaded by a user, identify target keywords in the second financial document based on text image recognition technology, and generate a mind map corresponding to the target keywords according to a preset hierarchical structure; a second parsing module is used to identify document elements in the first financial document, and perform structured parsing on the document elements to obtain parsing results, wherein the document elements include at least one of the following: text paragraphs, tables, and graphics; a typesetting module is used to obtain screen information of a mobile terminal device, determine a typesetting strategy based on the parsing results and screen information, and generate a mind map corresponding to the target keywords according to the typesetting strategy. Strategy, adapts and adjusts the format of the document element to obtain the first element; a watermark adding module is used to generate a digital watermark according to the copyright information and the custom information, and add the digital watermark to the first element to obtain the target element, wherein the custom information includes: a reading platform identifier and a document purpose description; a question processing module is used to receive a query question, and if the query question includes a first preset keyword related to a mind map, the mind map is displayed; if the query question includes a second preset keyword related to a knowledge graph, the knowledge graph subgraph corresponding to the query question is determined in the graph structure, and the knowledge graph subgraph is displayed; if the query question includes a third preset keyword related to the document element, the target element is displayed, thereby achieving the purpose of improving the processing efficiency of financial knowledge, thereby realizing the technical effect of improving user experience, and further solving the technical problem of low efficiency in financial knowledge processing in related digital reading systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0017] Figure 1 It is a structural diagram of a digital reading system based on knowledge graph and text image recognition technology according to an embodiment of the present application. DETAILED DESCRIPTION

[0018] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0019] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0020] Figure 1 is a structural diagram of a digital reading system based on knowledge graph and text image recognition technology according to an embodiment of the present application, such as Figure 1 As shown, the system includes:

[0021] The collecting module 11 is used to collect financial data and a first financial document from multiple financial data sources.

[0022] Specifically, the collection module 11 integrates API interfaces with multiple financial data sources to automatically capture and synchronize financial data, including but not limited to real-time market data, macroeconomic indicators, industry reports, etc. A document upload platform is built to support users to upload First Financial documents in PDF, Word and other formats, and to use web crawler technology to capture public First Financial document resources. Financial data sources are regularly updated to ensure the timeliness and accuracy of the knowledge graph.

[0023] The construction module 12 is used to identify target entities in the financial field in the financial data, extract the semantic relationships between the target entities, and construct a graph structure of the financial knowledge graph with the target entities as nodes and the semantic relationships as edges.

[0024] In the above-mentioned construction module 12, key entities such as company names and financial indicators are identified from the collected financial data. Then, the relationship extraction algorithm is used to analyze the relationship between entities, such as "Company A acquires Company B", "Rule X affects financial indicator Y", etc. The identified entities are used as nodes and semantic relationships as edges to construct a knowledge graph in the financial field, and graph database technology is used for storage and management. The entity recognition and relationship extraction algorithms are continuously improved to improve the accuracy and coverage of knowledge graph construction.

[0025] The first parsing module 13 is used to receive the second financial document uploaded by the user, identify the target keywords in the second financial document based on the text image recognition technology, and generate a mind map corresponding to the target keywords according to a preset hierarchical structure.

[0026] The second parsing module 14 is used to identify document elements in the first financial document and perform structural parsing on the document elements to obtain parsing results, wherein the document elements include at least one of the following: text paragraphs, tables, and graphics.

[0027] Specifically, the second parsing module 14 can use optical character recognition technology to identify text paragraphs, table recognition technology to analyze table structures, and image recognition technology to analyze graphic content. It performs structured parsing on the identified document elements, extracts key information, such as financial data, analysis conclusions, chart data, etc., and forms parsing results.

[0028] The typesetting module 15 is used to obtain screen information of the mobile terminal device, determine a typesetting strategy according to the parsing result and the screen information, and adapt the format of the document element according to the typesetting strategy to obtain the first element.

[0029] The typesetting module 15 can obtain information such as the screen size, resolution, and orientation of the user's mobile terminal device, and automatically generate a typesetting strategy adapted to the user's device based on the screen information and parsing results, such as adjusting the font size and layout method. The format of the document element is adjusted to obtain the first element to ensure that it can be displayed clearly and beautifully on different devices. Based on user feedback, the document parsing effect, typesetting strategy, and query response speed are optimized to improve the overall user experience.

[0030] The watermark adding module 16 is used to generate a digital watermark according to the copyright information and the custom information, and add the digital watermark to the first element to obtain the target element, wherein the custom information includes: a reading platform identifier and a document purpose description.

[0031] The watermark adding module 16 generates a digital watermark based on the copyright information and the custom information (such as the reading platform logo and the document purpose description), and adds the digital watermark to the document elements after the format adjustment to protect the copyright and the integrity of the document.

[0032] The question processing module 17 is used to receive a query question. If the query question includes a first preset keyword related to a mind map, the mind map is displayed. If the query question includes a second preset keyword related to a knowledge graph, a knowledge graph subgraph corresponding to the query question is determined in the graph structure and the knowledge graph subgraph is displayed. If the query question includes a third preset keyword related to a document element, the target element is displayed.

[0033] The question processing module 17 can receive user queries, analyze the keywords in the questions, and determine whether they are related to the knowledge graph or document elements. If the query contains keywords related to the knowledge graph, the system searches for and displays subgraphs related to the query in the knowledge graph, providing deep insights into the financial field. If the query contains keywords related to document elements, the system displays target documents containing these elements, supporting users to read and analyze them in detail. The query results are displayed through an intuitive interface, such as using graphical displays of knowledge graph subgraphs, or providing detailed content of document elements, while supporting users to conduct further searches and ask questions.

[0034] The following Figure 1 The modules shown are illustrative and explanatory.

[0035] According to some optional embodiments of the present application, the construction module 12 is also used to perform the following steps: preprocessing the text data in the financial data, wherein the preprocessing includes: removing noise characters, unifying the text format, processing abbreviations and synonym replacement; using a named entity recognition algorithm to scan the preprocessed text data to identify the target entities in the financial field, wherein the target entities include: financial institutions, financial products, financial market participants, and economic indicators; using preset financial field knowledge rules to extract the semantic relationships between target entities, wherein the semantic relationships include: belonging relationships, influence relationships, and transaction relationships, wherein the belonging relationships include: a certain financial product belongs to a certain financial institution, the influence relationships include: a certain economic indicator affects a certain financial market, and the transaction relationships include: transaction relationships between financial market participants.

[0036] In the above embodiment, the financial data is first preprocessed with text data. Specifically, regular expressions are used to identify and remove useless symbols and blank characters in the text, such as punctuation marks (except necessary commas and periods), spaces, line breaks, etc., to ensure the purity of the text data. Then, a natural language processing toolkit is applied to clean the text, including removing non-text elements such as HTML tags, numbers, special characters, etc. The text data is then converted to a unified format, such as a plain text file encoded in UTF-8, to eliminate parsing problems caused by inconsistent encoding. The document is standardized, including adjusting the font size, line spacing, and paragraph format to make it suitable for data analysis and model training. Finally, abbreviations and synonyms are replaced, and a special abbreviation dictionary for the financial field is constructed. The abbreviations in the document are converted to their full form, such as replacing "CPI" with "Consumer Price Index", and word sense disambiguation technology is applied to understand the exact meaning of synonyms based on the context to ensure the accuracy and consistency of text semantics.

[0037] Then, the named entity recognition algorithm is used to scan the preprocessed text data to identify the target entities in the financial field. Specifically, the pre-trained NLP model is fine-tuned to adapt to the financial field to achieve accurate recognition of entities such as financial institutions, financial products, financial market participants, and economic indicators. A customized financial field named entity recognition model is used and trained using a large amount of financial field data to improve the accuracy and coverage of entity recognition.

[0038] Finally, the semantic relationships between target entities are extracted using preset financial domain knowledge rules. Specifically, the semantic relationship rules in the financial field are determined, which are based on financial knowledge graphs, industry standards, and expert experience. The relationship extraction algorithm is determined to automatically identify the relationships between entities in the text according to the rules, including "belonging relationships", "influence relationships", and "transaction relationships".

[0039] Among them, belonging relationship refers to identifying financial institutions and financial products in the text, and analyzing the belonging relationship between products and institutions, such as "XX Fund" belongs to "YY Bank"; influence relationship refers to identifying the mutual influence between economic indicators and financial markets by analyzing the causal relationship and associated words in the text, such as "GDP growth" affects "stock market fluctuations"; transaction relationship refers to using natural language understanding and deep learning technology to identify the transaction relationship between financial market participants, such as the buying and selling behavior between "investor A" and "institution B".

[0040] Furthermore, the extracted entities and relationships are integrated into the financial knowledge graph to form a structured data representation, which is convenient for subsequent query and analysis. By comparing the extraction results with the manually annotated data, the accuracy and recall rate of the model are evaluated, and parameter adjustments and model optimizations are continuously performed to improve the recognition and extraction performance. The named entity recognition and semantic relationship extraction functions are integrated into the digital reading platform to provide users with intelligent document parsing and deep information extraction services.

[0041] Through the above specific implementation steps, the construction module can provide in-depth text understanding and relationship analysis for the financial industry, helping users quickly obtain key information and improve decision-making efficiency and business insights.

[0042] According to other optional embodiments of the present application, the question processing module 17 is also used to perform the following steps: receiving a query question input by a user, and if the query question includes a second preset keyword related to the knowledge graph, converting the query question into a query statement for the graph structure; executing the query statement, determining the knowledge graph subgraph corresponding to the query question in the graph structure, and displaying the knowledge graph subgraph through a graphical interface.

[0043] In the above embodiment, the question processing module 17 receives the query question input by the user and performs keyword recognition. Specifically, the user can ask a query question through a text box or voice input, such as "How to understand the impact of asset securitization on bank liquidity" or "What is the impact of the recent credit policy on the loan interest rate of small and micro enterprises?" The question processing module analyzes and pre-processes the query question input by the user, including steps such as removing stop words, stem extraction and part-of-speech tagging, and identifies keywords related to the knowledge graph, such as "asset securitization", "bank liquidity", "credit policy" and "small and micro enterprise loan interest rate".

[0044] After identifying the second preset keyword, the semantic structure of the query question is further parsed and converted into a query statement for the graph structure (i.e., the financial knowledge graph), such as using the query language of the graph database. After the query statement is generated, the question processing module uses the graph database technology to search for entities and their relationships associated with the keyword in the financial knowledge graph and execute the query. The query process follows the topological relationship of the graph structure, finds the nodes and edges (relationships) that match the query question, and finally determines one or more subgraphs as the answer.

[0045] After the query is executed, the question processing module displays the knowledge graph subgraph corresponding to the query question through a graphical interface. The subgraph will be presented in the form of nodes and edges, where nodes represent entities (such as financial institutions, financial products, economic indicators, etc.) and edges represent the relationships between entities (such as belonging relationships, influence relationships, transaction relationships, etc.). At the same time, a brief explanation can be provided for the knowledge graph subgraph to help users understand the specific meaning of each entity and relationship in the subgraph. The explanation includes the definition of the entity, the description of the relationship, and specific examples related to the query question. Users can further explore and ask questions about the knowledge graph subgraph in the graphical display interface, such as clicking on a node to view detailed information or asking new related questions. By continuously collecting and analyzing user feedback on knowledge graph queries, the system can continuously optimize the query algorithm and knowledge graph model to improve the accuracy and efficiency of queries. For example, when a user is dissatisfied with a query result, the system can automatically adjust the query statement or update the relationship definition in the knowledge graph to be closer to the user's real needs.

[0046] Through the above steps, the query questions entered by the user can be intelligently converted into query statements for the knowledge graph, and the query results can be displayed through a graphical interface to provide in-depth and intuitive knowledge insights. This function greatly improves the learning efficiency and depth of understanding of users in the financial field, and promotes the rapid acquisition and application of knowledge.

[0047] In some optional embodiments of the present application, the second parsing module 14 is also used to perform the following steps: for tables in document elements, parse the structural information of the table, wherein the structural information includes: number of rows, number of columns, row width, column width, extract the text content in the table, and determine the importance index of the text content in the table; for graphics in document elements, identify the type of graphics and data identifier.

[0048] In the above embodiment, the table elements in the document are first automatically identified by OCR technology or PDF parser, and separated from the text stream to prepare for structural analysis and content extraction. Then, the table parsing algorithm is used to accurately calculate the structural information such as the number of rows, columns, row width and column width of the table, which is specifically achieved by analyzing the table border, cell spacing and table style to ensure accurate restoration of the table structure. For formatted electronic documents, such as Excel or Word, their contents are directly read without OCR processing.

[0049] The extracted text content is then analyzed to determine the importance index of the text content. Specifically, TF-IDF in natural language processing technology is used to assign an importance weight to the text content of each cell, which reflects the importance of the content in the entire document. For example, if the content of a cell rarely appears in other documents, it has a higher TF-IDF weight, indicating that the content is more critical in the current document.

[0050] Furthermore, for graphics in document elements, the type of graphics and data identifiers are identified. Specifically, the types of graphic elements in the document, including line graphs, bar graphs, pie charts, flow charts, architecture diagrams, etc., are automatically identified by analyzing the shape, color and layout characteristics of the graphics to ensure accurate determination of the graphic type. For graphic elements containing data, such as line graphs, bar graphs, etc., image recognition algorithms are used to locate data identifiers in the graphics, such as axis labels, data points, legends, etc., and specific text information is identified through OCR technology. Then, the graphics and table contents are analyzed in association with the surrounding text to understand their specific role and meaning in the document. For example, a table can be used to show the asset status of multiple financial institutions in a specific period, while a line graph shows the trend of a certain economic indicator over time. The relevance of these graphics and tables to the document theme can be determined through context analysis to improve the accuracy of parsing.

[0051] Furthermore, the parsed tables and graphic information are connected to the financial knowledge graph to build a more comprehensive data association. For example, by identifying the names of financial institutions and economic indicators in the table and matching them with the entities in the knowledge graph, the semantic understanding and application of the data are enhanced. The parsed tables and graphic information are displayed through a graphical interface, so that users can intuitively view the data structure and importance indicators. At the same time, interactive functions such as zooming in, zooming out, and data export are provided to meet users' in-depth analysis and use needs for graphic and table information.

[0052] Through the above steps, it is possible to intelligently parse the tables and graphic elements in the document, extract key structural information and text content, provide users with comprehensive document parsing services, and promote the reading efficiency and depth of understanding of employees in the financial industry when faced with complex data.

[0053] As some optional embodiments of the present application, the typesetting module 15 is also used to perform the following steps: for text paragraphs in the document element, adjusting the line spacing and character spacing according to the screen width; for tables in the document elements, when the screen width is less than a first preset threshold, if the table width exceeds the screen, a scroll bar is used to display the table content, or columns are selectively hidden according to the importance index of the text content in the table, and an operation button for expanding all columns is provided; for graphics in the document element, the graphic size is scaled proportionally according to the screen space and the graphic type.

[0054] In the above embodiment, the layout module 15 monitors the screen width of the user device in real time to perform dynamic layout adjustment to ensure that it can respond to changes in screen size immediately. When a change in screen width is detected, the layout module 15 collaborates with the front-end layout engine to dynamically adjust the line spacing and character spacing according to the new screen width.

[0055] For the screen-adaptive display of tables in document elements, first read the original width of the table and compare it with the current screen width to determine whether it exceeds the first preset threshold (e.g., 900px). When the table width exceeds the screen width, the overflow-x: auto attribute is set in CSS to automatically display a horizontal scroll bar, allowing the user to drag left and right to view the complete table content, ensuring that the table can still be fully displayed when the width is limited. And evaluate the importance of the text content in the table based on text frequency, relevance to the topic, or importance level marked by experts. For example, when the screen width is less than the first preset threshold, hide non-critical columns according to the importance index, and display columns with high importance first. An operation button can be provided, and when the user clicks it, all hidden columns can be displayed again to restore the full view of the table.

[0056] For screen adaptive scaling of graphics in documents, identify the type of graphic (line chart, pie chart, flow chart, etc.) and calculate the ratio of available screen space to the original size of the graphic to determine the degree of scaling. Then, based on the matching degree of screen space and graphic type, the graphic elements in scalable vector graphics format are automatically scaled proportionally to ensure that the graphic can be fully displayed without distortion on different screen sizes. Specifically, calculate the ratio of the current screen width to the original width of the graphic, set the scaling factor, and dynamically adjust the width and height attributes of the graphic based on the scaling factor, while ensuring that the aspect ratio of the graphic remains unchanged to avoid graphic deformation. For small-screen devices, additional optimization is required, such as reducing the level of detail in the graphic and hiding non-critical legends or data points to ensure clear presentation of core information.

[0057] Through the above specific implementation methods, it is possible to effectively adapt to different screen sizes, provide users with a clear and complete document reading experience, and ensure the readability and usability of text, tables and graphic elements. This screen-adaptive layout not only improves the user experience, but also ensures the accurate transmission of information, and promotes the knowledge learning and business processing efficiency of financial industry employees.

[0058] In some optional embodiments of the present application, the watermark adding module 16 is further used to perform the following steps: for the text paragraph in the first element, obtain the document theme, determine the theme keyword with the highest correlation index with the document theme, and determine the importance score of the text paragraph according to the frequency of occurrence of the theme keyword in the text paragraph; add a first number of digital watermarks at a first preset position of the text paragraph whose importance score is less than a third preset threshold, and add a second number of digital watermarks at a second preset position of the text paragraph whose importance score is greater than the third preset threshold, wherein the second number is greater than the first number; for the table in the first element, determine the table length and table width of the table, When the table length is greater than a fourth preset threshold and / or the table width is greater than a fifth preset threshold, a digital watermark is added to the first row, the first column, the last row and the last column; when the table length is not greater than the fourth preset threshold and / or the table width is not greater than the fifth preset threshold, a digital watermark is added at the third preset position of the table; for the graphics in the first element, the graphics are divided into multiple sub-blocks, each sub-block is subjected to discrete cosine transform to obtain multiple frequency domain coefficients, a target area whose frequency domain coefficients are within a preset interval is selected, the frequency domain coefficients in the target area are modified according to the digital watermark, and after the modification is completed, the graphics are converted back to the spatial domain through an inverse discrete cosine transform.

[0059] In the above embodiment, the watermark adding module 16 can be used for importance scoring and digital watermark adding of text paragraphs. Specifically, importance scoring of text paragraphs includes the following steps: first, subject analysis is performed on the entire document to extract the main discussed areas or concepts as the document subject. According to the document subject, a list of keywords most relevant to the subject is extracted. These keywords can be concepts repeatedly mentioned in the document, or terms closely related to the subject. At the same time, the correlation index between each keyword and the document subject is calculated, which can be the TF-IDF value of the keyword, or the weight of the keyword in the subject model. Then, each text paragraph is traversed to count the frequency of occurrence of the subject keywords therein. According to the frequency and correlation index, the importance score of each paragraph is calculated, and a weighted function can be used to combine the keyword frequency with the correlation index to obtain a comprehensive score.

[0060] Furthermore, according to the requirements of document content and watermark visibility, a third preset threshold is set to distinguish the importance of text paragraphs. For paragraphs with importance scores less than the third preset threshold, a first number (fewer) of digital watermarks are added to a first preset position (such as the end of the paragraph); for paragraphs with importance scores greater than the third preset threshold, a second number (more) of digital watermarks are added to a second preset position (such as the beginning of the paragraph or a prominent position).

[0061] Furthermore, for adding digital watermarks to table elements, the structure of the table is analyzed, the number of rows and columns is obtained, and the length and width of the table, that is, the total number of rows and columns of the table, are calculated. Then, a fourth preset threshold (table length threshold) and a fifth preset threshold (table width threshold) are set to determine whether the table is a large-size table. If the length or width of the table exceeds the preset threshold, in order to ensure that the watermark is not easily noticeable and cover the entire table at the same time, a digital watermark is added to the first row, first column, last row, and last column of the table to ensure that the watermark information is retained even if the table is cropped. If the size of the table is below the preset threshold, a watermark is added to the third preset position of the table (such as the center or lower right corner of the table). In this case, the watermark is more concentrated and has less impact on the table data.

[0062] Furthermore, for the digital watermark embedding of graphic elements, the graphic is divided into a plurality of sub-blocks, and the size of each sub-block can be set according to the graphic resolution and the watermark embedding requirements. Discrete cosine transform is performed on each sub-block, and the graphic in the spatial domain is converted into a frequency domain representation to obtain a frequency domain coefficient matrix. Then, according to the embedding requirements of the digital watermark, a preset interval is set, and the medium and low frequency areas with less visual impact on the image can be selected. By analyzing the frequency domain coefficient matrix, the target area falling within the preset interval is selected as the carrier for watermark embedding. And according to the content of the digital watermark, the frequency domain coefficients in the target area are adjusted. The watermark information can be encoded as a series of binary numbers, and the watermark embedding function is used to calculate the modification amount of the frequency domain coefficients, and the embedding of the watermark information is realized by adjusting the value of the frequency domain coefficients. After the modification of the frequency domain coefficients is completed, the inverse discrete cosine transform is used to convert the graphic back to the spatial domain to restore the visual form of the graphic. At this point, the watermark information has been embedded in the graphic, and it is visually consistent with the original graphic and is not easy to be detected.

[0063] Preferably, the strength of watermark embedding can be adjusted to balance the visibility and security of the watermark. A watermark that is too weak will be lost during image processing or compression, while a watermark that is too strong will affect the image quality.

[0064] Through the above steps, digital watermarks can be intelligently added or embedded according to the content and structure of text paragraphs, tables and graphics to protect the copyright of documents and the security of data without affecting the user's reading and understanding of documents. This technology is particularly important in fields such as the financial industry that have strict requirements on data security, and can effectively prevent the illegal copying and dissemination of sensitive information.

[0065] As other optional embodiments of the present application, the above-mentioned electronic reading system based on knowledge graph and intelligent typesetting also includes: a detection module, which is used to perform the following steps after displaying the target element: determine whether the mobile terminal device performs a screenshot and / or copy behavior on the target data that meets the preset rules through a pre-trained machine learning model, wherein the pre-trained machine learning model is obtained by training in the following method: collecting screenshot event data and copy event data, performing feature extraction on the screenshot event data, and obtaining a first feature, wherein the first feature includes: screenshot time, screenshot area, and screenshot method; performing feature extraction on the copy event data to obtain a second feature, Among them, the second feature includes: the data type, copy source, and copy time of the copied content; the first feature and the second feature are used to train the machine learning model. During the training process, the difference between the predicted result and the true label is calculated according to the loss function of the model, and the parameters of the model are adjusted through the stochastic gradient descent algorithm to minimize the loss function, and when the loss function meets the preset conditions, a machine learning model that has completed the training is obtained; when it is determined that the mobile terminal device performs a screenshot and / or copy behavior on the target data, the digital watermark of the document where the target data is located is determined, the copyright information in the digital watermark is extracted, and a target alarm message including the copyright information is generated.

[0066] In the above embodiment, the detection module first records the screenshot event by embedding an event listener, including the exact time when the screenshot occurs, the screenshot area selected by the user, and the mode of the screenshot operation (such as full-screen screenshot, selected area screenshot), as the input of the model training, which is used to analyze the pattern and characteristics of the screenshot behavior. Similarly, the copy event is recorded through the event listener, including the data type of the copied content (such as text, table, graphic), the source of the copied content (such as document ID, chapter title), and the timestamp of the copy operation, to understand the user's interest in a specific type or source of data. Using data preprocessing techniques such as data cleaning, normalization, and feature selection, the screenshot event and copy event data are converted into digital features that can be understood by the machine learning model. For example, the "screenshot time" is encoded as the number of hours in a day, the "screenshot area" is encoded as a percentage of the screen resolution, and the "data type of the copied content" is converted into a numerical code, etc.

[0067] Then, NLP and computer vision technologies are combined to further extract features and process the screenshot and copy event data, such as extracting keywords from text, identifying the type and content of graphics, etc., to enhance the model's prediction ability. Specifically, a suitable machine learning algorithm is selected to train the model using the extracted first feature (screenshot event feature) and the second feature (copy event feature). During the training process, the data set is divided into a training set and a validation set, the model parameters are adjusted using the training set data, and the model performance is evaluated using the validation set data. A loss function is defined to measure the gap between the model prediction results and the actual labels. Loss functions include cross entropy loss, mean square error loss, etc. Stochastic gradient descent or its variants (such as Adam, RMSprop, etc.) are used as optimization algorithms to continuously adjust model parameters to minimize the loss function and improve the accuracy and generalization ability of the model. Preset conditions are set, such as the convergence threshold of the loss function, the maximum number of iterations, or performance indicators on the validation set (such as accuracy, F1 score, etc.). When any of the preset conditions is met, the model training is terminated and the trained model is saved.

[0068] Finally, when it is determined that the mobile terminal device performs screenshot and / or copy behavior on the target data, the digital watermark of the document where the target data is located is determined, the copyright information in the digital watermark is extracted, and a target alarm message including the copyright information is generated. Specifically, the trained pre-trained machine learning model is used to predict the user's screenshot and copy behavior on the target data. If the model prediction result shows that the user will perform a screenshot or copy operation on the target data, and the behavior meets the preset rules (such as too high frequency, involving sensitive information, etc.), the subsequent alarm and copyright protection process is triggered. The digital watermark of the document where the target data is located is determined. After locating the watermark, a watermark extraction algorithm (such as DCT-based watermark extraction) is used to extract copyright information from the digital watermark, such as the copyright ownership of the document, the authorization method, etc. Based on the extracted copyright information, an alarm message is generated, including a copyright warning, a reminder of the consequences of illegal copying, and a guide to the legal use of documents, etc., to remind users to respect copyright and legally use the resources on the digital reading platform. The target alarm message is pushed to the user through the platform's instant messaging service or email system. At the same time, the alarm event is recorded in the database for subsequent audit tracking and copyright protection strategy optimization.

[0069] Through the above steps, it is possible to intelligently predict the user's screenshot and copy behavior of the target data, and when the predicted behavior meets the preset rules, extract the digital watermark in the document, protect the copyright information, and generate an alarm message to remind the user, forming a complete digital copyright protection and user behavior monitoring mechanism. This mechanism helps banks and financial institutions protect their intellectual assets, prevent unauthorized copying and dissemination, and maintain the security and compliance of financial information.

[0070] In some optional embodiments, the electronic reading system based on knowledge graph and intelligent typesetting further includes: a personalized recommendation module, which is used to perform the following steps: collect the reading behavior data of the user on the digital reading platform on the front-end interactive interface, wherein the reading behavior data includes: reading history data, collection preference data and search record data, the reading history data is used to record the identification information, reading start time and reading end time of the first financial documents read by the user, the collection preference data is used to record the document category information, identification information and collection time collected by the user, and the search record data is used to record the search keywords and search time entered by the user; transmit the reading behavior data to the back-end data storage center through an encrypted data transmission channel; The reading behavior data is cleaned and normalized, and the processed reading behavior data is subjected to feature extraction to obtain a target feature vector, and a clustering algorithm is used to divide the target feature vector into different interest groups, wherein the feature extraction includes: encoding the document according to the subject and field dimensions, calculating the user's preference scores for documents of different categories based on the collection preference data, and extracting keywords whose occurrence frequency is greater than a second preset threshold from the search record data; analyzing the time series of the user's reading behavior data to predict the user's future reading interest; generating recommended content based on the user's reading behavior data, future reading interest, and the reading behavior data of other users in the interest group, and displaying the recommended content on the user's personal interface.

[0071] In the above embodiment, the personalized recommendation module monitors and records the user's reading behavior in the front-end interface, including the identification information (such as document ID) of the first financial document that the user has read, the start time and the end time of reading. At the same time, the document category information, identification information and collection time of the user's collection, as well as the search keywords and search time entered by the user are recorded. Then, the HTTPS protocol or other secure transmission mechanism is used to ensure that the user's reading behavior data on the digital reading platform is not intercepted or tampered with by a third party during the transmission process. Before the data is sent, encryption algorithms such as AES and RSA can be used to encrypt the data to improve data security.

[0072] Furthermore, the reading behavior data transmitted by the front end is received and pre-processed in the back-end data storage center, including removing outliers (such as obviously wrong reading time), filling missing values ​​(such as the collection time not filled in by the user), and normalizing the data (such as converting the timestamp to a unified time format for easy analysis). This ensures the consistency and accuracy of the data and lays the foundation for subsequent analysis.

[0073] Furthermore, feature extraction is performed on the processed reading behavior data to generate a target feature vector. Specifically, the following steps are performed: encoding the document subject and field to convert the text content into numerical features; calculating the user's preference scores for different document categories based on the collection preference data by counting the frequency and time interval of the user's collection of a certain category of documents; extracting high-frequency keywords (with an occurrence frequency greater than a second preset threshold) from the search record data as the user's interest keywords; using a clustering algorithm (such as K-means, DBSCAN, or a clustering method based on deep learning) to analyze the extracted target feature vector, and dividing the users into different interest groups, each of which represents a specific reading preference and behavior pattern.

[0074] Finally, the user's reading interest is predicted, recommended content is generated, and the recommended content is displayed on the user's personal interface. Specifically, time series analysis techniques such as ARIMA model, state space model or recurrent neural network and long short-term memory network in deep learning are used to analyze the user's reading behavior data, capture the changing trend of the user's reading history, collection behavior and search habits over time, and predict the user's future reading interest.

[0075] Based on the user's time series analysis results, future reading interests, and reading behavior data of other users in the interest group, the recommendation system algorithm (such as collaborative filtering, content-based recommendation, or deep learning recommendation model) is used to generate recommended content, including but not limited to predicting documents, books, or industry reports that the user may be interested in, and optimizing the sorting of the recommendation list to improve the accuracy and personalization of the recommendation. The recommended content is displayed on the user's personal interface, such as home page recommendations, reading history pages, or search results interfaces. It should be noted that when displaying content, a responsive layout is used to take into account the user's reading habits and screen size to ensure the display effect of the recommended content on different devices. In addition, methods such as A / B testing can be used to continuously optimize the recommendation algorithm and display strategy to improve user satisfaction and platform usage.

[0076] Through the above steps, users' reading behavior data can be intelligently collected and analyzed, their future reading interests can be predicted, and personalized content recommendations can be generated, greatly improving user experience and reading efficiency. This recommendation system based on user behavior analysis has important application value in knowledge-intensive fields such as the financial industry, which can help employees quickly obtain the required information and promote knowledge sharing and business innovation. At the same time, the platform's secure encryption of data transmission and back-end data processing and analysis ensure user privacy and data security, and meet the compliance requirements of the financial industry.

[0077] In some optional embodiments, the first parsing module 13 is further used to perform the following steps: determining keywords in the second financial document whose relevance index to the document topic is greater than a preset threshold, summarizing the article theme of the second financial document based on natural language processing technology, and generating an article summary including the article theme; receiving a question about the second financial document input by a user, and determining an answer corresponding to the question using semantic understanding and knowledge retrieval technology; displaying an editing area, wherein the editing area is used to receive text input information; receiving and responding to a target instruction, and translating the second financial document into a target language indicated by the target instruction.

[0078] The first analysis module 13 can realize the following functions: 1. Quick reading. The analysis operation area on the right first extracts the keywords, summary and main content of the article in the form of an introduction to help users quickly grasp the essence of the article. 2. In-depth dialogue. Through the dialogue module, users can obtain the main content of the full text, and support quick questions to answer doubts in reading immediately. 3. Mind map generation. Users can generate mind maps according to their needs, set generation requirements and node keywords, intuitively present the structure of the article, and deepen their understanding. 4. Reading note recording. Provides text editing function, users can easily record reading experience and form personalized reading notes. 5. Multi-language translation. Supports one-click translation of the full text and underline translation of book content, breaking down language barriers and broadening reading horizons.

[0079] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0080] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0081] In the above-mentioned embodiments of the present application, the collected information is information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data comply with relevant laws, regulations and standards, take necessary protection measures, do not violate public order and good morals, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0082] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0083] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0084] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0085] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the relevant technology or all or part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, server or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk, etc. Various media that can store program codes.

[0086] The above is only a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A digital reading system based on knowledge graph and text image recognition technology, characterized in that: include: A collection module, configured to collect financial data and a first financial document from a plurality of financial data sources; A construction module is used to identify target entities in the financial field in the financial data, extract semantic relationships between the target entities, and construct a graph structure of a financial knowledge graph with the target entities as nodes and the semantic relationships as edges; A first parsing module, configured to receive a second financial document uploaded by a user, identify target keywords in the second financial document based on a text image recognition technology, and generate a mind map corresponding to the target keywords according to a preset hierarchical structure; A second parsing module is used to identify document elements in the first financial document and perform structural parsing on the document elements to obtain parsing results, wherein the document elements include at least one of the following: text paragraphs, tables, and graphics; A typesetting module, used to obtain screen information of a mobile terminal device, determine a typesetting strategy according to the parsing result and the screen information, and adapt and adjust the format of the document element according to the typesetting strategy to obtain a first element; A watermark adding module, used to generate a digital watermark according to the copyright information and the custom information, and add the digital watermark to the first element to obtain a target element, wherein the custom information includes: a reading platform identifier and a document purpose description; A question processing module, configured to receive a query question, and if the query question includes a first preset keyword related to a mind map, display the mind map; if the query question includes a second preset keyword related to a knowledge graph, determine a knowledge graph subgraph corresponding to the query question in the graph structure, and display the knowledge graph subgraph; if the query question includes a third preset keyword related to the document element, display the target element; A watermark adding module, adding a digital watermark to the first element, comprising the following steps: for a text paragraph in the first element, obtaining a document theme, determining a theme keyword with the highest correlation index with the document theme, and determining an importance score of the text paragraph according to the frequency of occurrence of the theme keyword in the text paragraph; adding a first number of digital watermarks at a first preset position of a text paragraph whose importance score is less than a third preset threshold, and adding a second number of digital watermarks at a second preset position of a text paragraph whose importance score is greater than the third preset threshold, wherein the second number is greater than the first number; for a table in the first element, determining a table length and a table width of the table, and adding a first number of digital watermarks at a second preset position of a text paragraph whose importance score is greater than the third preset threshold; When the table length is greater than a fourth preset threshold and / or the table width is greater than a fifth preset threshold, the digital watermark is added in the first row, the first column, the last row and the last column; when the table length is not greater than the fourth preset threshold and / or the table width is not greater than the fifth preset threshold, the digital watermark is added at the third preset position of the table; for the graphics in the first element, the graphics are divided into multiple sub-blocks, each sub-block is discrete cosine transformed to obtain multiple frequency domain coefficients, a target area of ​​the frequency domain coefficients within a preset interval is selected, the frequency domain coefficients within the target area are modified according to the digital watermark, and after the modification is completed, the graphics are converted back to the spatial domain through an inverse discrete cosine transform.

2. The system according to claim 1, characterized in that The construction module identifies the target entities in the financial field in the financial data, extracts the semantic relationships between the target entities, and constructs the graph structure of the financial knowledge graph with the target entities as nodes and the semantic relationships as edges, including the following steps: Preprocessing the text data in the financial data, wherein the preprocessing includes: removing noise characters, unifying text formats, processing abbreviations and replacing synonyms; Use named entity recognition algorithm to scan the preprocessed text data and identify the target entities in the financial field, where the target entities include: financial institutions, financial products, financial market participants, and economic indicators; By using preset financial domain knowledge rules, the semantic relationships between the target entities are extracted, wherein the semantic relationships include: belonging relationships, influence relationships, and transaction relationships. The belonging relationships include: a certain financial product belongs to a certain financial institution, the influence relationships include: a certain economic indicator affects a certain financial market, and the transaction relationships include: transaction relationships between financial market participants.

3. The system according to claim 2, characterized in that The question processing module receives a query question, and if the query question includes a second preset keyword related to the knowledge graph, determines a knowledge graph subgraph corresponding to the query question in the graph structure, and displays the knowledge graph subgraph, including the following steps: receiving the query question input by the user, and if the query question includes a second preset keyword related to the knowledge graph, converting the query question into a query statement for the graph structure; Execute the query statement, determine the knowledge graph subgraph corresponding to the query question in the graph structure, and display the knowledge graph subgraph through a graphical interface.

4. The system according to claim 1, characterized in that The parsing module performs structural parsing on the document elements to obtain parsing results, including the following steps: For the table in the document element, parse the structural information of the table, wherein the structural information includes: the number of rows, the number of columns, the row width, the column width, extract the text content in the table, and determine the importance index of the text content in the table; for the graphics in the document element, identify the type of the graphic and the data identifier.

5. The system according to claim 4, characterized in that The typesetting module determines the typesetting strategy according to the analysis result and the screen information, including the following steps: For a text paragraph in the document element, adjusting line spacing and word spacing according to screen width; For a table in the document element, when the screen width is less than a first preset threshold, if the table width exceeds the screen, a scroll bar is used to display the table content, or columns are selectively hidden according to an importance index of the text content in the table, and an operation button for expanding all columns is provided; For graphics in the document element, the size of the graphics is scaled proportionally according to the screen space and the type of graphics.

6. The system according to claim 5, characterized in that The electronic reading system based on knowledge graph and intelligent typesetting further includes: a detection module, which is used to perform the following steps after displaying the target element: Determine whether the mobile terminal device performs screenshot and / or copy behavior on target data that meets preset rules through a pre-trained machine learning model, wherein the pre-trained machine learning model is obtained by training in the following method: collect screenshot event data and copy event data, perform feature extraction on the screenshot event data to obtain a first feature, wherein the first feature includes: screenshot time, screenshot area, and screenshot method; perform feature extraction on the copy event data to obtain a second feature, wherein the second feature includes: data type of the copied content, copy source, and copy time; use the first feature and the second feature to train the machine learning model, during the training process, calculate the difference between the predicted result and the true label according to the loss function of the model, adjust the parameters of the model through the stochastic gradient descent algorithm to minimize the loss function, and obtain a trained machine learning model when the loss function meets the preset conditions; When it is determined that the mobile terminal device has performed a screenshot and / or copying action on the target data, a digital watermark of the document where the target data is located is determined, copyright information in the digital watermark is extracted, and a target warning message including the copyright information is generated.

7. The system according to claim 1, characterized in that The electronic reading system based on knowledge graph and intelligent typesetting also includes: a personalized recommendation module, which is used to perform the following steps: Collecting the reading behavior data of the user on the digital reading platform on the front-end interactive interface, wherein the reading behavior data includes: reading history data, collection preference data and search record data, wherein the reading history data is used to record the identification information, reading start time and reading end time of the first financial document read by the user, the collection preference data is used to record the category information, identification information and collection time of the document collected by the user, and the search record data is used to record the search keyword input by the user and the search time; Transmitting the reading behavior data to a backend data storage center through an encrypted data transmission channel; The reading behavior data is cleaned and normalized, and the processed reading behavior data is subjected to feature extraction to obtain a target feature vector, and the target feature vector is divided into different interest groups using a clustering algorithm, wherein the feature extraction includes: encoding the document according to the subject and field dimensions, calculating the user's preference scores for documents of different categories according to the collection preference data, and extracting keywords whose occurrence frequency is greater than a second preset threshold from the search record data; Analyze the time series of the user's reading behavior data to predict the user's future reading interests; generate recommended content based on the user's reading behavior data, the future reading interests, and the reading behavior data of other users in the interest group, and display the recommended content on the user's personal interface.

8. The system according to claim 1, characterized in that The first parsing module is further configured to perform the following steps: Determine in the second financial document a keyword whose relevance index to the document subject is greater than a preset threshold, summarize the article subject of the second financial document based on natural language processing technology, and generate an article summary including the article subject; receiving a question about the second financial document input by a user, and determining an answer corresponding to the question using semantic understanding and knowledge retrieval technology; Displaying an editing area, wherein the editing area is used to receive text input information; The target instruction is received and, in response to the target instruction, the second financial document is translated into a target language indicated by the target instruction.

Images