Determination method and device of virtual resource investment research report, equipment and storage medium

Abstract

The invention provides a virtual resource investment and research report determination method and device, equipment and a storage medium, and the method comprises the steps: obtaining multi-modal data related to a target virtual resource, the multi-modal data comprising data of at least two modal types; acquiring key information data of the multi-modal data corresponding to the different modal types according to the information extraction modes of the different modal types and the multi-modal data; determining a preset display format; and fusing the key information data of the multi-modal data corresponding to the different modal types to obtain a target investment and research report which corresponds to the target virtual resource and is based on a preset display format. The investment research report determination efficiency and the investment return rate are improved, and the investment risk is reduced.

Description

Technical Field

[0001] The embodiments of the present application relate to the technical field of financial planning and analysis, and in particular to a method, apparatus, device and storage medium for determining a virtual resource investment research report. Background Art

[0002] With the development of globalization and the advancement of information technology, financial markets are becoming increasingly complex and dynamic. At present, when investing in virtual resources such as stocks, investors and investment analysts need to interpret and analyze a large amount of market data, such as the financial data of joint-stock companies, news reports on related stocks, social media discussions, and stock trends, and determine investment research reports based on these market data. However, due to the large amount of market data and the need to rely on personal subjective experience to manually determine investment research reports, there are problems such as low efficiency and high risk in determining investment research reports. Summary of the invention

[0003] The present application provides a method, device, equipment and storage medium for determining a virtual resource investment research report to improve the efficiency of determining the investment research report and the rate of return on investment and reduce investment risks.

[0004] In a first aspect, the present application provides a method for determining a virtual resource investment and research report, comprising: obtaining multimodal data related to a target virtual resource, the multimodal data comprising data of at least two modal types; obtaining key information data of the multimodal data corresponding to different modal types according to information extraction methods of different modal types and multimodal data; determining a preset display format; fusing the key information data of the multimodal data corresponding to different modal types to obtain a target investment and research report corresponding to the target virtual resource based on a preset display format.

[0005] In a second aspect, the present application provides a device for determining a virtual resource investment and research report, comprising: a first acquisition module, used to: obtain multimodal data related to a target virtual resource, the multimodal data including data of at least two modal types; a second acquisition module, used to: obtain key information data of the multimodal data corresponding to different modal types according to information extraction methods of different modal types and multimodal data; a determination module, used to: determine a preset display format; a fusion module, used to: fuse the key information data of the multimodal data corresponding to different modal types to obtain a target investment and research report corresponding to the target virtual resource based on a preset display format.

[0006] In a third aspect, the present application provides an electronic device, comprising: a processor and a memory, the memory being used to store a computer program, the processor being used to call and run the computer program stored in the memory, and executing the method in the first aspect or its various implementations.

[0007] In a fourth aspect, the present application provides a computer-readable storage medium for storing a computer program, wherein the computer program enables a computer to execute the method in the first aspect or its various implementations.

[0008] In a fifth aspect, the present application provides a computer program product, comprising computer program instructions, which enable a computer to execute the method in the first aspect or its various implementations.

[0009] In a sixth aspect, the present application provides a computer program, which enables a computer to execute the method in the first aspect or its various implementations.

[0010] In the above process, the electronic device can determine multiple key information data through multimodal data related to the target virtual resource, fuse the multiple key information data, and obtain the target investment research report corresponding to the target virtual resource based on the preset display format according to the preset display format. This realizes the automation and intelligent determination method of the investment research report, which can improve the efficiency of investment research report determination and investment return rate, and reduce investment risks. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0012] Figure 1 An application scenario diagram provided for an embodiment of the present application;

[0013] Figure 2 A flowchart of a method for determining a virtual resource investment research report provided in an embodiment of the present application;

[0014] Figure 3 A schematic diagram of a method for determining a virtual resource investment research report provided in an embodiment of the present application;

[0015] Figure 4 A schematic diagram of a device 400 for determining a virtual resource investment research report provided in an embodiment of the present application;

[0016] Figure 5 A schematic block diagram of an electronic device 500 provided in an embodiment of the present application. DETAILED DESCRIPTION

[0017] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions 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 making creative work are within the scope of protection of this application.

[0018] 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 server that includes 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.

[0019] As mentioned above, the existing technology has the problem of low efficiency and high risk in determining investment research reports.

[0020] In order to solve the above technical problems, this application determines multiple key information data through multimodal data related to the target virtual resource, fuses the multiple key information data, and obtains the target investment research report corresponding to the target virtual resource based on the preset display format according to the preset display format. This realizes the automation and intelligent determination method of the investment research report, which can improve the efficiency of investment research report determination and investment return rate, and reduce investment risks.

[0021] It should be understood that the technical solution of the present application can be applied to the following scenarios, but is not limited to:

[0022] In some possible implementations, the technical solution of the present application can be used in scenarios of virtual resource investment, for example, it can be applied to scenarios of generating stock investment research reports, i.e., investment research reports, and the present application does not impose any restrictions on this.

[0023] In some possible implementations, Figure 1 An application scenario diagram provided for an embodiment of the present application, such as Figure 1 As shown, the application scenario may include a terminal device 110 and an electronic device 120. The terminal device 110 and the electronic device 120 may be connected via a wired network or a wireless network.

[0024] Among them, the terminal device 110 can be installed with a financial client, and the electronic device 120 can generate an investment research report corresponding to the virtual resources based on a preset display format according to the multimodal data related to the virtual resources in a preset display format. Afterwards, the electronic device 120 can send the investment research report to the terminal device 110, so that the terminal device 110 can display the investment research report and make corresponding investments based on the investment research report.

[0025] Exemplarily, the terminal device 110 may be a mobile phone, a tablet computer, a desktop, a laptop, a handheld computer, a notebook computer, a vehicle-mounted device, an ultra-mobile personal computer (UMPC), a netbook, a cellular phone, a personal digital assistant (PDA), an augmented reality (AR) or virtual reality (VR) device, etc., but is not limited thereto. The electronic device 120 may be a server, a server cluster consisting of multiple servers, or a cloud platform control center, which is not limited in the embodiments of the present application.

[0026] also, Figure 1 The number of terminal devices and electronic devices shown is only exemplary, and may actually include other numbers of terminal devices and electronic devices, which is not limited in this application.

[0027] In some other possible implementations, the technical solution of the present application can be executed by the above-mentioned terminal device 110, and the present application does not impose any limitation on this.

[0028] It should be noted that in the specific implementation of this application, multimodal data, original text data, historical market data and other related data are involved. When the embodiments of this application are applied to specific products or technologies, user permission, consent or authorization is required, and the collection, use and processing of relevant data need to comply with relevant laws, regulations and standards of relevant countries and regions.

[0029] After introducing the application scenarios of the embodiments of the present application, the technical solution of the present application will be described in detail below:

[0030] Figure 2 A flowchart of a method for determining a virtual resource investment research report provided in an embodiment of the present application, the method can be performed as follows Figure 1 The electronic device 120 shown in the figure performs, but is not limited to this. Figure 2 As shown, the method may include the following steps:

[0031] S210: Acquire multimodal data related to the target virtual resource, where the multimodal data includes data of at least two modal types;

[0032] S220: Acquire key information data of multimodal data corresponding to different modal types according to information extraction methods of different modal types and multimodal data;

[0033] S230: Determine a preset display format;

[0034] S240: The key information data of the multimodal data corresponding to different modal types are integrated to obtain a target investment research report corresponding to the target virtual resource based on a preset display format.

[0035] In some possible implementations, the target virtual resource may be any of the following, but is not limited thereto: stocks, funds, foreign exchange, futures, and options.

[0036] The modality type may include any of the following types: text type, numerical type, image type, audio type, video type, but not limited thereto. Accordingly, multimodal data may include at least two of the following types of data, but not limited thereto: text type data, image type data, numerical type data, audio type data, video type data.

[0037] Specifically, text-type data related to the target virtual resources include but are not limited to: profile information and news information of the target virtual resources; numerical-type data related to the target virtual resources include but are not limited to: market data such as resource prices of the target virtual resources, financial data, etc.; image-type data related to the target virtual resources include but are not limited to: image data extracted from different data sources, for example, image data that can be extracted from news information, image data generated based on numerical-type data (such as resource price data), etc.; audio-type data related to the target virtual resources include but are not limited to: audio data extracted from different data sources, for example, audio data related to the target virtual resources that can be extracted from broadcasts (such as audio data related to the target virtual resources corresponding to the target virtual resources or market changes); video-type data related to the target virtual resources include but are not limited to: video data extracted from different data sources, for example, video data related to the target virtual resources that can be extracted from news (such as video data related to the target virtual resources corresponding to the target virtual resources or market changes).

[0038] In some implementations, the electronic device may obtain raw data (raw_data) such as raw text data through a data interface service application programming interface (Application Programming Interface, API), and then pre-process the raw data through a data processing service API to obtain multimodal data.

[0039] Specifically, the above-mentioned acquisition of multimodal data related to the target virtual resource may include at least the following two situations, but is not limited thereto:

[0040] Case 1, in response to the modality type being a text type, obtaining text type data related to the target virtual resource, including: obtaining the resource identifier of the target virtual resource; obtaining the original text data of the target object corresponding to the target virtual resource based on the resource identifier; constructing structured text data and unstructured text data based on the original text data and the attribute fields carried by the original text data, the unstructured text data including the unstructured text in the original text data and the fields in the attribute fields corresponding to the unstructured text; determining the unstructured text data and the structured text data as text type data in the multimodal data.

[0041] Among them, the original text data includes but is not limited to: profile information and news information of the target virtual resource; taking the target virtual resource as stocks as an example, the original text data includes but is not limited to: profile information and news information of the stocks; for example, the profile information includes but is not limited to: stock name, place of registration, listing date, industry and sector, and company shareholders and other text data, and the information information includes but is not limited to: date, title, text content, summary, etc. of stock-related information.

[0042] Exemplarily, the data interface service API includes a target object profile data interface and an information data interface. The electronic device can input the resource identifier and the preset language into the target object profile data interface, and obtain the profile information of the target object corresponding to the target virtual resource based on the preset language through the target object profile data interface; input the resource identifier, the preset language and the preset time period into the information data interface, and obtain the information information related to the target virtual resource based on the preset language within the preset time period through the information data interface, and determine the above profile information and information information as original text data. Among them, the target object corresponding to the target virtual resource may refer to the company to which the virtual resource belongs. The information information related to the target virtual resource may include dynamic information information of the target virtual resource and industry analysis information. For example, the above profile information may be an overview of the company to which the stock belongs, and the information information may be individual stock dynamics and industry analysis information.

[0043] For example, assuming that the target virtual resource is a stock, the electronic device can first obtain the stock identifier (stock_id) and the preset language (language), and then pass the stock identifier and the preset language to the target object profile data interface, thereby obtaining detailed company information corresponding to the stock, that is, the profile information corresponding to the target, including but not limited to: company name (company_name), place of registration (place of registration), listing date (listData), industry and sector (business and contentPlate), and company shareholders (chair_name, shares, directorBrief). Because this information is rich and comprehensive, it is easy to gain an in-depth understanding of the basic situation of the company behind the individual stock, and provide a reliable reference for investment decisions from a fundamental perspective.

[0044] For example, assuming that the target virtual resource is a stock, the electronic device can first obtain the stock identifier (stock_id), the preset language (language) and the preset time period (day_range), and then pass the stock identifier, the preset language and the preset time period to the information data interface, thereby obtaining the information related to the individual stock within the preset time period, that is, the information information, including but not limited to: the information release date (time), the information title (title), the information content (content), the information abstract (abstract), the information tag (important_tag), the language used in the information content (language), and the information source (source). Because this information is rich and comprehensive, it can provide an in-depth understanding of the relevant information of the company behind the individual stock, and provide a reliable reference for investment decisions from the news perspective.

[0045] In order to enhance the accuracy of the key information data of the subsequent extracted text type data, the original text data can be preprocessed to obtain the corresponding structured text data and unstructured text data. The unstructured text data carries all the original text data, which can provide comprehensive text content and related attribute information for the subsequent extraction and fusion of key information data. The structured text has clearer fields, which can help to understand and associate the contextual information of the original text data during the subsequent extraction and fusion of key information data. Considering both structured text and unstructured text, comprehensive and accurate analysis and decision-making can be carried out from a multi-dimensional perspective.

[0046] Specifically, after obtaining the original text data, the original text data can be used as unstructured text data. At the same time, text recognition can be performed on the original text data based on the preset attribute field to identify the target attribute field that matches the preset attribute field, and then the field value corresponding to the target attribute field is obtained in the original text data, and finally a structured text is constructed based on the target attribute field value and the field value corresponding to the target attribute field. Among them, the preset attribute field can be an attribute field used to describe important characteristic information of the target virtual resource. For example, taking the target virtual resource as a stock as an example, the preset attribute field can be a growth rate, a profit margin, a price-earnings ratio, etc.

[0047] For example, for the original text data of stocks, "In 2021, Company 1 performed beyond expectations, with total revenue of approximately US$16.68 billion, a year-on-year increase of 53%, operating profit margin of approximately 25%, and price-to-earnings ratio of 40%." The original text data can be identified as unstructured text data. At the same time, by performing structured integration on the text fields in the original text data, structured text data [total revenue: US$16.68 billion], [year-on-year growth rate: 53%], [operating profit margin: 25%], [price-to-earnings ratio: 40%], etc. can be obtained.

[0048] It is understandable that tools such as public opinion analysis that rely on manual or simple natural language cannot effectively obtain and mine the large amount of information contained in unstructured data, and the speed of information processing and integration is slow, generally with delays of hours to days. The present application can automatically divide the original text data into structured and unstructured types, and determine the modal data in combination with the corresponding attribute fields to generate high-quality input content, thereby effectively obtaining and mining a large amount of information contained in unstructured data and improving the comprehensiveness and accuracy of investment research reports.

[0049] Furthermore, the electronic device can also first classify the original text data into structured text and unstructured text, and determine the attribute fields carried by the original text data. Since structured text has fixed attributes and fields, the structured text can be directly determined as structured text data, and the unstructured text and the fields in the attribute fields corresponding to the unstructured text can be determined as unstructured text data, so that the model can simultaneously consider the text content of the unstructured text and the fields corresponding to the unstructured text, thereby performing more comprehensive analysis and decision-making.

[0050] Among them, assuming that the unstructured text is "In 2021, Company 1's performance exceeded expectations, with total revenue of approximately US$16.68 billion, a year-on-year increase of 53%, an operating profit margin of approximately 25%, and a price-to-earnings ratio of 40%", then the fields corresponding to the unstructured text are: company, total revenue, operating profit margin, and price-to-earnings ratio.

[0051] For unstructured text, the electronic device can structure and integrate the text content of the unstructured text and the corresponding fields, such as title, content, plate, tag, source, language, company name, industry classification, financial indicators and other related fields, to obtain structured text, so that the model can process it in the form of processing structured data, and it can facilitate the model to better understand the relationship between the context of text information. Alternatively, the text content of the unstructured text and the corresponding attribute fields are directly input into the model, so that the model can consider the text content and related attributes at the same time, so as to make more comprehensive analysis and decision-making.

[0052] Exemplarily, the structured text data and the unstructured text data may be in JSON (JavaScript Object Notation) format, for example, in string form.

[0053] Case 2, in response to the modality type being an image type, image type data related to the target virtual resource is obtained, including: obtaining the resource identifier of the target virtual resource; obtaining historical market data corresponding to the target virtual resource based on the resource identifier; rendering the historical market data to obtain image type data in the multimodal data; wherein the historical market data at least includes: financial report information and K-line chart data of the target object corresponding to the target virtual resource; the image type data at least includes: a market trend chart of the target virtual resource and a capital flow chart of the target object corresponding to the target virtual resource.

[0054] Exemplarily, the data interface service API includes a financial data interface and a market data interface. The electronic device can input the resource identifier and the preset financial type into the financial data interface to obtain the financial report information of the target object corresponding to the target virtual resource based on the preset financial type; input the resource identifier, K-line chart type, start time and end time into the market data interface to obtain the K-line chart data of the target virtual resource based on the K-line chart type from the start time to the end time; input the resource identifier into the stock trend chart interface, render the financial report information and K-line chart data, and obtain the market trend chart of the target virtual resource and the capital flow chart of the target object.

[0055] For example, assuming that the target virtual resource is a stock, the electronic device can first obtain the stock identifier (stock_id), the preset financial type (type) and the time period (date), and pass them into the financial data interface, thereby obtaining the financial report information of the target object corresponding to the stock based on the preset financial type, including but not limited to: currency unit, operating income (revenue), net profit (net_income), gross profit margin (gross_margin), net asset return (roe), debt ratio (debt_ratio), cash flow (cash_flow). Since these data can be used for fundamental analysis to help evaluate the profitability and financial stability of the target object, at the same time, these data can also be used as auxiliary indicators for technical analysis to help more accurately judge the price trend and future development potential of virtual resources, so that by comprehensively utilizing fundamental and technical analysis, the potential value and risk of virtual resources can be more comprehensively understood, providing a more reliable reference for investment decisions.

[0056] For example, assuming that the target virtual resource is a stock, the electronic device can first obtain the stock identifier (stock_id), K-line chart type (kline_type), start time (begin_time) and end time (end_time), and pass them into the market data interface, thereby obtaining K-line chart data based on the K-line chart type, including but not limited to: currency unit (currency), opening price (open_price), closing price (close_price), turnover rate (turnover), trading volume (pe), price-earnings ratio. Based on these data, a comprehensive evaluation of virtual resources can be conducted from multiple dimensions such as technical analysis, market sentiment, valuation analysis and risk management, so as to more accurately understand the market performance and potential risks of virtual resources and provide a more reliable reference for investment decisions.

[0057] For example, assuming that the target virtual resource is a stock, the electronic device can first obtain the stock identifier (stock_id), and combine the historical market data obtained by the above-mentioned financial data interface and market data interface, call the front-end rendering code of the stock trend chart interface, render the historical market data, and obtain the market trend chart of the stock (for example, daily k, weekly k, monthly k, quarterly k, annual k, stock transaction, etc.) and the capital flow chart of the target object corresponding to the stock), and determine it as image type data related to the stock. Because these images provide an intuitive display of market data, they are helpful for market sentiment analysis, trend identification and capital flow analysis, so as to formulate more effective investment strategies and decisions to optimize the performance of the investment portfolio and manage risks.

[0058] It is understandable that image data has higher information density and stronger information expression ability than text or numerical data, so it can help identify market trends and price patterns more accurately. However, the source of image data is very wide, that is, the source of historical market data is very wide, including: PDF files such as financial reports and annual reports, and images generated by numerical data, WORD files or information content. Therefore, it is necessary to integrate historical market data from various different data sources in a standardized format, that is, to perform the above-mentioned rendering process. During the integration process, different historical market data and their sources can be marked to maintain the integrity and accuracy of the data.

[0059] Further, the electronic device can first determine the source of the historical market data, and then extract relevant images from the historical market data from different sources, for example, images can be extracted from PDF files. Then, the extracted relevant images can be identified and classified to clarify their sources and image types, wherein the types of images can include: technical analysis charts, expression of commentary and opinion charts, emoticons and emoticon packs, voting and survey charts, and daily sharing charts. Then the above-mentioned relevant images are standardized and unified in the same format to ensure the uniformity and consistency of the image type data, so that they are suitable for subsequent processing and analysis. For example, the relevant images can be processed as: size 384*384, format JPEG format. Finally, the electronic device can perform data annotation on the format-integrated images, that is, annotate and explain them, including: the content, source and any other relevant information of the picture. When processing image data based on the model in the future, the image data (i.e., the format-integrated images) and the data obtained in the above-mentioned integration process (such as image source, image type, annotation and description) can be input into the model together for the use and understanding of the model.

[0060] Case 3: In response to the modality type being a numerical type, obtaining numerical type data related to the target virtual resource includes: obtaining a resource identifier of the target virtual resource; obtaining initial numerical data corresponding to the target virtual resource according to the resource identifier; processing the initial numerical data according to multiple numerical presentation forms to obtain numerical type data in multiple numerical presentation forms, wherein the multiple numerical presentation forms include at least two of the following: array, list, table, and chart. Among them, the chart may include a pie chart, a bar chart, a curve chart, etc.

[0061] For example, stock price data can be processed into numerical data in the form of a list, short selling data analysis can be processed into numerical data in the form of a bar chart, capital distribution can be processed into numerical data in the form of a pie chart, and capital flow can be processed into numerical data in the form of a curve chart.

[0062] Exemplarily, referring to the embodiment corresponding to the above situation 2, the initial numerical data may include: financial report information, market trend information, and technical indicator information of the target object corresponding to the target virtual resource of numerical type. Among them, the initial numerical data can be presented in the form of a dictionary (dict), including an array (array) or a list (list). In order to generate high-quality input content, the electronic device can organize the initial numerical data into a table (table) or a graph (graph), i.e., a chart.

[0063] It is understandable that numerical data is objective and quantifiable, which can help to understand the company's financial status and market performance more clearly and intuitively. Since numerical data can be expressed in many different ways, such as structured fields, tables, and various graphics such as pie charts and bar charts, the initial numerical data can be integrated in the following forms:

[0064] For example, the above-mentioned multiple numerical presentation forms may include structured input forms and unstructured output forms. For structured input forms, the electronic device may integrate the initial numerical data in a standardized field form to make it easy to understand and process, and obtain numerical type data in a structured input form, for example, numerical type data in a presentation form such as a list or array. For structured input forms, the electronic device may initially organize the numerical data into a table form to more intuitively display the relationship and trend between the data, or may use various graphics such as pie charts, bar charts, etc. to visualize the numerical data and obtain numerical type data in a graphical form, so as to more intuitively present the characteristics and changing trends of the data.

[0065] Case four, in response to the modal type being an audio type, obtaining audio type data related to the target virtual resource, including: obtaining a resource identifier of the target virtual resource; obtaining initial audio data related to the target virtual resource based on the resource identifier; extracting key information from the initial audio data to obtain audio type data in the multimodal data, wherein the electronic device can obtain the initial audio data from a news program or information.

[0066] Exemplarily, the above-mentioned extraction of key information from the initial audio data includes: performing text recognition and extraction on the initial audio data to obtain an audio text; performing key information extraction on the audio text to obtain a key audio text; performing text-to-speech (TTS) processing on the key audio text to obtain audio type data in the multimodal data.

[0067] Case five, in response to the modality type being a video type, obtaining video type data related to the target virtual resource, including: decomposing the video type data into audio type data and image type data, wherein the decomposed image type data is an image corresponding to all frames in the video type data; processing the decomposed image type data and audio type data separately according to the technical solutions in the above cases two and four; combining the processed image type data and audio type data to obtain the video type data in the multimodal data.

[0068] In the above embodiment, the resource identifier of the target virtual resource may be the resource name or resource code of the target virtual resource, but is not limited thereto. For example, assuming that the target virtual resource is a stock, the resource identifier may be the stock name or stock code.

[0069] Through the above embodiments, fragmented information from scattered data sources and information sources can be integrated through a variety of interfaces and preprocessing methods to obtain multimodal data, so as to obtain comprehensive market insights, reduce collection time, avoid missing key data, and ensure the comprehensiveness and accuracy of investment decisions.

[0070] In some implementations, the electronic device may use a large language model (LLM) to obtain key information data of multimodal data. Specifically, the key information data of text type data may be obtained using a text key information extraction model, the key information data of numerical type data may be obtained using a numerical key information extraction model, and the key information data of image type data may be obtained using an image key information extraction model.

[0071] Among them, the text key information extraction model, the numerical key information extraction model or the image key information extraction model can all be models trained based on the large language model. The electronic device can fine-tune the LLM to obtain the above text key information extraction model, the numerical key information extraction model or the image key information extraction model to save costs and improve performance.

[0072] At the same time, electronic devices can use prompt word engineering technology (i.e., text prompt instructions, numerical prompt instructions, and image prompt instructions in the following embodiments) to stimulate the model to learn relevant knowledge about financial investment research, conduct deeper or richer feature mining and processing of data in various fields, so as to adapt to market and technological changes and generate in-depth stock analysis reports.

[0073] In addition, electronic devices can use Agent technology and DataTool to implement the technical solution of this application to stimulate the potential of key information extraction models. For example, electronic devices can use DataTool to perform pre-processing such as data cleaning on input data such as training data and multimodal data of key information extraction models; Agent technology can be used to build an intelligent agent, and based on the intelligent agent, automatically perform tasks corresponding to the key information extraction model, such as collecting model input data, model training, and parameter adjustment. Among them, DataTool is a data processing tool that integrates multiple batch processing functions, and Agent technology is a computer system encapsulated in a specific environment. It not only has autonomy, sociality, responsiveness, and initiative, but also has some characteristics that only humans have, such as knowledge, beliefs, intentions, etc.

[0074] This application will introduce the process of obtaining key information data according to the key information extraction model in the following embodiments:

[0075] In some implementable embodiments, the above-mentioned information extraction method according to different modal types and multimodal data obtains key information data of multimodal data corresponding to different modal types, including: in response to the modal type being a text type, writing the text type data in the multimodal data into a text prompt instruction to obtain an updated text prompt instruction; based on the text type data in the multimodal data and the updated text prompt instruction, extracting text key information through a text key information extraction model to obtain fundamental key information and message key information; wherein, the fundamental key information includes profile information, news information and industry dynamics information of the target object corresponding to the target virtual resource, and the message key information includes event information associated with the target virtual resource.

[0076] Among them, the text prompt instruction includes: a message side key information data template, a fundamental side key information data template and a task instruction for a text key information extraction model; among them, the fundamental side key information data template is used to indicate the data structure of the fundamental side key information data output by the text key information extraction model, the message side key information data template is used to indicate the data structure of the message side key information data output by the text key information extraction model, and the task instruction is used to indicate the text key information extraction model to output the corresponding message side key information and fundamental side key information data.

[0077] For example, the fundamental key information data template can be presented in JSON format, including at least the following data fields: Company Name, Company History and Encyclopedia, Industry Sector, Market Value, Listing Date, Major Shareholder Information, Management Team Composition, Industry Sector and Headquarters Location.

[0078] The key information data template of the message surface can be presented in JSON format, including at least the following data fields: event occurrence time (Date), event title (Event Title), event details (Event Details), event subject (Entity), event location (Location), event cause (Cause of Event), event course (Course ofEvent), event result (Event Outcome), and emotional expression information generated by the event (Event Sentiment Polarity);

[0079] The task instructions may be: As a professional information extraction expert, perform a company profile analysis task from the input [text type data], extract the fundamental key information data related to the target virtual resources in JSON format [fundamental key information data template], and extract the message key information data related to the target virtual resources in JSON format [message key information data template].

[0080] After obtaining the text type data associated with the target virtual resource, the text type data can be written into the text prompt instruction to update the text prompt instruction, and then the updated text prompt instruction is input into the text key information extraction model. The text key information extraction model is used to extract key information from the text type data to obtain the message surface key information data.

[0081] Furthermore, in one embodiment, based on the text type data in the multimodal data and the updated text prompt instructions, text key information is extracted through a text key information extraction model to obtain basic key information data and message key information data, including: performing text segmentation processing on the text type data through the text key information extraction model to obtain multiple text paragraphs; performing entity recognition and event extraction on each text paragraph through the text key information extraction model to obtain paragraph key information of each text paragraph, wherein the paragraph key information includes entity text and event information; extracting key information from the entity text in the paragraph key information to generate basic key information data; extracting key information from the event information in the paragraph key information to generate message key information data.

[0082] It is understandable that by first segmenting and then processing each text paragraph separately, that is, through secondary key information extraction, it is possible to fully mine the text type data, thereby ensuring that the generated key information data (key_points) is more comprehensive. In addition, in order to ensure the professionalism and accuracy of information extraction, multiple types of information can be extracted from each text paragraph: profile information, news information and industry dynamics information, and structured, clear and specific information can be extracted from it, further ensuring that the generated key information data is more comprehensive, clear and accurate.

[0083] Specifically, the text type data can first be segmented through the text key information extraction model to obtain multiple text paragraphs, and then the text key information extraction model can be used to perform language processing tasks such as entity recognition and event extraction on each text paragraph to obtain the entity text of each text paragraph and paragraph key information such as event information. Then, the basic key information data template is used to perform secondary key information extraction from the entity text in the paragraph key information to generate basic key information data, and the message key information data template is used to perform secondary key information extraction from the event information in the paragraph key information to generate message key information data.

[0084] For example, the fundamental key information data can be as follows:

[0085] ["Company Name":"Sample Company";

[0086] "Main business area": ​​"Information technology";

[0087] "Company Market Value": "300,000";

[0088] "Company's founding history": "The sample company was founded in 2000, and its main business is cloud service business...";

[0089] "Listing time": "2005-04-01";

[0090] "Company Headquarters Location":"Location 1";

[0091] "Shareholder information": "Shareholder A holds 30% of the shares, Shareholder B holds 20% of the shares, and other shareholders...";

[0092] "Management Team": ["CEO: Person 1", "CTO: Person 2", "CFO: Person 3"]}.

[0093] The fundamental key information data is presented in a fixed structure format to ensure the standardization, structuring and ease of use of the fundamental key information data, provide detailed archival materials for the subsequent generation of target investment research reports on target virtual resources, and provide a solid foundation for evaluating the value and market position of the target object.

[0094] For example, key information data on the message side can be arranged in the order of event occurrence and presented in a timeline format to ensure the logic and orderliness of the information, as shown below:

[0095] {"Event Timeline":[

[0096] {"Event time":"2023-01-10",

[0097] "Event Title": "Sample Company Launches New Product",

[0098] "Event Details":{"Occurrence Location":"Location 1",

[0099] "Event Subject":"Sample Company",

[0100] "Cause of the event": "Expand market share",

[0101] "Event Process": "Sample Company holds a launch event at location 1...",

[0102] "Event Result":"The new product was well received by the market, causing the stock price to rise",

[0103] "Emotional information generated by the event": "Positive"}},

[0104] {"Event time":"2023-02-20",

[0105] "Event Title":"Changes in senior management of the sample company",

[0106] "Event Details":{"Occurrence Location":"Headquarters",

[0107] "Event Subject":"Sample Company",

[0108] "Cause of the event":"Company's strategic adjustment",

[0109] "Event Process": "CEO announces resignation, search for successor begins...",

[0110] "Event Result":"Market reaction was mixed, stock price fluctuated",

[0111] "Emotional information generated by the event": "Neutral"}}]}

[0112] The key information data on the message surface is presented in a fixed structure format, so that the event information corresponding to the key information data on the message surface is arranged in chronological order and presented in the form of a timeline (Event Timeline), and there are no overlapping events between multiple event information, thereby ensuring the logic and orderliness of the event information and the independence of the events, providing detailed message surface data for the subsequent generation of the target investment research report on the target virtual resources, and providing a solid foundation for evaluating the value and market position of the target object.

[0113] In one embodiment, based on the text type data in the multimodal data and the updated text prompt instructions, text key information is extracted through a text key information extraction model to obtain message surface key information, including: extracting multiple event information related to the target virtual resource from the text type data in the multimodal data through the text key information extraction model; generating message surface key information corresponding to the text type data based on the event information.

[0114] Specifically, the updated text prompt instruction is input into the text key information extraction model, and the event extraction task is performed on the text type data through the text key information extraction model to obtain the event information in the text type data. The event information is used to describe the event content related to the target virtual resource. For example, individual stock related events can be extracted from the news information; then, the key information data corresponding to each event information, such as the event time, event title, event details, event subject, event location, event cause, event process, event result, and emotional expression information generated by the event, are extracted from the event information in combination with the message side key information data template, and the message side key information data is generated in a timeline manner according to the chronological order of the event occurrence.

[0115] In some implementable embodiments, the above-mentioned information extraction method according to different modal types and multimodal data obtains key information data of multimodal data corresponding to different modal types, including: in response to the modal type being a numerical type, inputting the numerical type data and the numerical prompt instruction in the multimodal data into the numerical key information extraction model, classifying the numerical type data through the numerical key information extraction model, and obtaining technical indicator numerical data and market trend numerical data corresponding to the target virtual resources; extracting the statistical characteristics of the technical indicator numerical data through the numerical key information extraction model to obtain the statistical characteristic analysis results; analyzing the asset inflow and outflow data, price trend data and abnormal movement data in the market trend numerical data through the numerical key information extraction model to obtain the current market evaluation results; determining the key information data corresponding to the numerical type data according to the statistical characteristic analysis results and the current market evaluation results; wherein the numerical prompt instruction is used to control the numerical key information extraction model to determine the key information data corresponding to the numerical type data.

[0116] Exemplarily, the numerical prompt instructions may include: task instructions for the numerical key information extraction model, the output format of the output statistical feature analysis results, and the output format of the current market evaluation results, but are not limited thereto.

[0117] Among them, the task instruction for the numerical key information extraction model is used to instruct the data key information extraction model to output technical key information data based on numerical type data. For example, the task instruction for the numerical key information extraction model can be as follows: As an experienced digital data analysis expert, from the input [numerical type data], perform precise analysis of the time series technical data related to virtual resources to combine the JSON format [output format of statistical feature analysis results] to output the statistical feature analysis results related to the target virtual resources, and combine the JSON format [output format of current market evaluation results] to output the current market evaluation results related to the target virtual resources, so as to realize technical indicator analysis and market trend analysis, and reveal the deep market dynamics behind stocks or investment assets.

[0118] Among them, the output format of the statistical feature analysis results can be output in JSON format, including at least the statistical characteristics of technical indicator numerical data, trend labels, interpretation labels, analysis result information, etc.

[0119] For example, the output format of the statistical feature analysis results can be as follows:

[0120] "Technical indicator numerical data analysis":{"Moving Average":{"30-day moving average":,"Trend":","Analysis":""};

[0121] "RSI":{"Current Value":,"Interpretation":","Analysis":""};

[0122] "Bollinger Bands":{"Stock Price Position":"","Analysis":""}}

[0123] Among them, the output format of the current market evaluation results can be output in JSON format, including at least the statistical characteristics of market trend numerical data, trend change labels, capital flow change labels, etc.

[0124] For example, the output format of the pre-market assessment results can be as follows:

[0125] "Market Trend Analysis":{"Short-Term Trend":{"Description":"","Stock Price Change":"","Net Fund Flow Change":""};

[0126] "Long-Term Trend":{"Description":"","Stock Price Change":"","Net Fund Flow Change":""}}.

[0127] Exemplarily, technical indicator numerical data may include at least one of the following, but not limited to: moving average, relative strength index (RSI), Bollinger Bands, which can reflect the technical situation of the target virtual resources and provide accurate and valuable reference for determining investment research reports.

[0128] Accordingly, the statistical feature analysis results can be presented in JSON format, including the analysis results of each technical indicator numerical data, such as the current value, indicator interpretation, trend and specific analysis results. The statistical feature analysis results can be as follows:

[0129] "Technical indicator numerical data analysis":{

[0130] "Moving Average":{"30-day moving average":25.5,"Trend":"Upward","Analysis":"In the past 30 days, the moving average of the stock price has risen from 22.8 to 25.5, indicating a clear upward trend."},

[0131] "RSI":{"Current Value":68,"Interpretation":"Slightly Overbought","Analysis":"The current RSI value is 68, exceeding the traditional overbought threshold of 65, which suggests that the stock may need a period of price correction."},

[0132] "Bollinger Bands":{"Stock Price Position":"Near Upper Band","Analysis":"The stock price has touched the upper Bollinger Band (27.3), which based on historical data could indicate a potential short-term price correction."}}

[0133] For example, market trend numerical data may include asset inflow and outflow data, price trend data, and abnormal movement data, which can reflect the short-term and long-term market trends of virtual resources such as securities or assets. By analyzing them, it is possible to evaluate current market dynamics by comparing historical data trends, thereby providing technical support for investment decisions.

[0134] The current market assessment results may include short-term market trends (Short-Term Trend) and long-term market trends (Long-Term Trend), which can be shown as follows:

[0135] "Market Trend Analysis":{

[0136] "Short-Term Trend":{"Description":"Observing the data of the past 7 days, it shows that the stock price has risen from 24.9 to 25.7, and the net capital flow has increased, reflecting the market's active participation in the stock in the short term.","Stock Price Change":"From 24.9 to 25.7","Net Capital Flow Change":"Increase"},

[0137] "Long-Term Trend":{"Description":"Data from the past 6 months show that the stock price rose from 20.0 to 25.5, and the net fund flow continued to increase, indicating that the market continued to pay attention to and participate in the stock over a longer period of time.","Stock Price Change":"From 20.0 to 25.5","Net Fund Flow Change":"Continued to Increase"}}

[0138] After obtaining the numerical type data associated with the target virtual resource, the numerical type data can be written into the numerical prompt instruction to update the numerical prompt instruction, and then the updated numerical prompt instruction can be input into the numerical key information extraction model, and the key information can be extracted from the numerical type data through the numerical key information extraction model to obtain the first technical key information data; specifically, the numerical type data can be first classified and processed through the text key information extraction model to obtain technical indicator numerical data and market trend numerical data, and then the text key information extraction model can be used to perform data analysis on the technical indicator numerical data and the market trend numerical data, respectively. Specifically, the statistical features of the technical indicator numerical data are extracted through the numerical key information extraction model, and combined with the preset output format and statistical features to generate statistical feature analysis results. At the same time, the statistical features of the market trend numerical data are extracted through the numerical key information extraction model, and combined with the preset output format and statistical features to generate the current market evaluation results. The statistical feature analysis results and the current market evaluation results are used as the first technical key information data, and output as the key information data corresponding to the numerical type data.

[0139] In addition, in combination with the above embodiments, for numerical type data in a structured input form and numerical type data in an unstructured output form, the electronic device can write numerical prompt instructions respectively, and extract them respectively through the numerical key information extraction model.

[0140] In some implementable embodiments, the above-mentioned information extraction method according to different modal types and multimodal data obtains key information data of multimodal data corresponding to different modal types, including: in response to the modal type being image type data, inputting the image type data and image prompt instructions in the multimodal data into the image key information extraction model, classifying the image type data through the image key information extraction model, and obtaining the K-line chart data and technical indicator image data of the target virtual resource; performing numerical analysis on the K-line chart data through the image key information extraction model to obtain price fluctuation information of the target virtual resource; identifying line changes in the technical indicator image data through the image key information extraction model to obtain market trend information; determining the key information data corresponding to the image type data based on the price fluctuation information and the market trend information; wherein the image prompt instruction is used to control the image key information extraction model to determine the key information data corresponding to the image type data.

[0141] For example, since the K-line chart data can intuitively display the dynamic changes in the price of virtual resources, it can facilitate the identification of price trends. Price fluctuation information can include: opening price (OpeningPrice), closing price (ClosingPrice), highest price (HighPrice), lowest price (LowPrice), price fluctuations (PriceFluctuations), trend identification (IdentifiedTrends) and chart pattern analysis (ChartPatterns).

[0142] For example, price fluctuation information can look like this:

[0143] "Price Fluctuation Information":{

[0144] "Opening Price":"$150",

[0145] "Closing Price":"$155",

[0146] "Maximum Price":"$160",

[0147] "Lowest Price":"$148",

[0148] "Price Volatility": "The stock price of XYZ Technology Company experienced fluctuations between $148 and $160 during the selected period, closing higher than the opening price.",

[0149] "Trend Identification":"The stock gradually rises over a period of time, with the closing price slightly higher than the opening price.",

[0150] "Chart Pattern Analysis":"The appearance of certain chart patterns indicates the possibility of continued price fluctuations, rather than a definite trend."}

[0151] Exemplarily, the technical indicator image data may be image data corresponding to the following indicators: Moving Averages, Relative Strength Index (RSI) and Moving Average Convergence Divergence (MACD). Since the indicators can form intuitive lines and graphics in the image, the changes in these lines and graphics not only reflect the historical performance of virtual resources, but also predict the possible future price trends. For example, the crossing and divergence of the indicator lines provide instant signals of price momentum and trend changes. Therefore, by identifying the line changes of the technical indicator image data, more comprehensive and accurate market trend information can be determined. Market trend information may include: the signs reflected by each technical indicator image data (Indicators), the information predicted by the convergence and divergence of the lines in the indicator (ConvergenceDivergence), and the market price momentum impacts shown by the technical indicators (MomentumImplications).

[0152] For example, price fluctuation information can look like this:

[0153] "Technical indicator evaluation":{

[0154] "Technical Indicator Image Data Reflection":{"MovingAverages":"The 50-day moving average shows an upward trajectory, and the stock price is currently above this level.",

[0155] "RSI":"RSI is 60, indicating an increased level of trading activity during this period.",

[0156] "MACD":"The MACD line is above the signal line, indicating a change in price momentum."},

[0157] "Indicator lines divergence message":"Divergence between stock price and MACD may indicate a potential change in price direction.",

[0158] "Technical indicators show the impact of market price momentum": "Current technical indicators show that the market is active."}

[0159] Exemplarily, the image prompt instruction may include: task instructions for the image key information extraction model, the output format of the output price fluctuation information and the output format of the market trend information, but is not limited thereto.

[0160] The task instruction for the image key information extraction model is used to instruct the image key information extraction model to output technical key information data based on the image type data. For example, the task instruction for the numerical key information extraction model may be as follows: As an expert proficient in chart interpretation, interpret the chart from the input [image type data] to output the price fluctuation information related to the target virtual resource in combination with the JSON format [output format of price fluctuation information], and output the market trend information related to the target virtual resource in combination with the JSON format [output format of market trend information], extract valuable insights, and provide a comprehensive understanding of market dynamics and investment opportunities. The output format of the price fluctuation information and the output format of the market trend information may be presented in JSON format.

[0161] In some achievable embodiments, the above-mentioned key information data may be in text form, so that the investment research report can be presented to investors in an intuitive and easy-to-understand manner, providing comprehensive market analysis and investment decision support.

[0162] In some possible implementations, the preset display format is used to indicate the structural format of the investment research report to be output; for example, the preset display format may be a JSON format structure, which defines various types of information data to be represented by the investment research report.

[0163] Specifically, the target investment research report may include but is not limited to: the target virtual resource's target object name, i.e., company name, virtual resource code (such as stock code) and report content, and the report content may include but is not limited to: basic overview, technical summary, news and overall investment advice. The preset display format may be a JSON format structure, which may be specifically expressed as:

[0164] {"Company Name":"Name of the target object corresponding to the target virtual resource",

[0165] "Stock Code": "The trading stock code of the target object name corresponding to the target virtual resource",

[0166] "Report content":{

[0167] "Basic Overview": "Explanation of the company's financial performance and business development",

[0168] "Technical Summary":"Company Stock Price Trend and Technical Indicator Analysis",

[0169] "News Watch":"The company's latest news and strategic initiatives",

[0170] "Overall investment advice": "Based on the above analysis, comprehensive investment advice considering market trends and potential risks"}}

[0171] In some implementations, the preset display format can also be used to indicate the file format of the investment research report to be output, which can specifically include any of the following: file display formats such as Excel and PDF, chart display formats, and graphic display formats. The client can display multiple display format options, and in response to the target user's selection operation for a preset display format in a plurality of display formats, determine and send the preset display format to the electronic device, so that the electronic device can determine the preset display format. In this way, it is possible to support interactive visualization of investment research reports, enhance user experience, and facilitate users to perform data analysis and report sharing.

[0172] Specifically, after obtaining the key information data corresponding to different modal types, the target investment research report can be generated based on the structure format and / or file format indicated by the preset display format in combination with the key information data. Specifically, for the fundamental key information data, the message key information data, and the technical key information data composed of the key information data corresponding to the numerical type data and the key information data corresponding to the image type data, the corresponding filling positions can be determined one by one in the preset display format, and then the above-mentioned fundamental key information data, message key information data and technical key information data can be filled into the filling positions indicated by the preset display format.

[0173] Furthermore, considering the correlation between different types of key information data, further text summary processing can be performed on the technical key information data composed of fundamental key information data, message key information data, key information data corresponding to numerical type data, and key information data corresponding to image type data, so as to obtain summary text for the target virtual resources in terms of fundamentals, message and technology, and fill the summary text into the filling position preset by the preset display format to obtain the final target investment research report.

[0174] In some achievable embodiments, the above-mentioned fusion of key information data of multimodal data corresponding to different modality types to obtain a target investment research report corresponding to the target virtual resource and based on a preset display format includes: determining a fusion instruction, the fusion instruction is used to characterize the processing tasks and processing criteria of the comprehensive analysis model, and the fusion instruction includes a preset display format; inputting multiple key information data and the fusion instruction into the comprehensive analysis model, controlling the comprehensive analysis model according to the fusion instruction, fusing the multiple key information data, and obtaining a target investment research report corresponding to the target virtual resource and based on a preset display format;

[0175] Among them, the fusion instruction is used to instruct the key information data of the multimodal data corresponding to different modal types to be processed to output an investment research report. The comprehensive analysis model can specifically be a model obtained by training based on a large language model. The electronic device can train the LLM according to the sample key information data, so that the LLM can perform a text summary on the key information data corresponding to the input data for characterizing the target virtual resources under different modal types, and combine the key information data corresponding to the data under different modal types and the text summary information obtained by the text summary with the preset display format indicated by the fusion instruction to generate an investment research report on the target virtual resources.

[0176] Specifically, multiple key information data and fusion instructions are input into the comprehensive analysis model. Through the fusion instructions, the comprehensive analysis model is triggered to determine the corresponding filling positions in the preset display format for the technical key information data composed of the fundamental key information data, the message key information data, and the key information data corresponding to the numerical type data and the key information data corresponding to the image type data. At the same time, the fundamental key information data, the message key information data, and the technical key information data composed of the key information data corresponding to the numerical type data and the key information data corresponding to the image type data are further summarized in text to obtain the text summary information, and the corresponding filling positions are determined in the preset display format. Finally, the fundamental key information data, the message key information data, the technical key information data and the text summary information are respectively written into the filling positions indicated by the preset display format to obtain the investment research report.

[0177] The target virtual resource may be a type of virtual resource, and virtual resource 1 may be a category of the target virtual resource. For example, if the target virtual resource is stock, virtual resource 1 may be stock 1.

[0178] Exemplarily, the fusion instruction may be as follows:

[0179] The comprehensive analysis model can be an expert in information extraction, integration, association and reorganization, extracting, integrating and reorganizing the content of investment research reports to create professional, accurate and expert analysis documents. It is necessary to follow the standardized JSON format [preset display format] to organize information [target investment research report] to ensure the clarity, accuracy and professionalism of the content.

[0180] The comprehensive analysis model generates content according to the following processing criteria:

[0181] Accuracy, ensuring all data and information [multiple key information data] are correct and up-to-date, reflecting the latest market conditions and company financial statements.

[0182] Professionalism, using specific industry terminology related to finance and investing, ensures content is familiar and trusted by professionals and knowledgeable investors.

[0183] Organization: Organize the report content in a logical and coherent manner so that readers can easily understand and follow the information.

[0184] Comprehensiveness: investment research reports should cover fundamental analysis, technical analysis, news observation and overall investment advice to provide readers with a comprehensive investment perspective.

[0185] Objectivity, maintaining objectivity and neutrality while providing advice, showing the expected impact under different market conditions and possibilities.

[0186] Based on this, we create investment research reports that meet professional standards, provide valuable insights, and assist investors in making wise investment decisions.

[0187] In addition, the fusion instructions can include the associations between multiple key information data so that the comprehensive analysis model can accurately understand and execute the fusion tasks.

[0188] Correspondingly, the target investment research report can be as follows:

[0189] {"Company Name":"Company 1",

[0190] "Stock Code":"111",

[0191] "Report content":{

[0192] "Fundamentals Overview":"Company 1, a leading manufacturer of graphics processing units, has shown strong financial performance in the past quarter. The company reported revenue of $5 billion, up 22% year-on-year, net profit of $1.7 billion, up 28% year-on-year, and gross profit margin remained stable at 65%. This growth was mainly due to its strong sales in the artificial intelligence and gaming markets, as well as its continued expansion in the data center field. R&D investment remained stable, ensuring that the company maintains its leading position in the highly competitive technology market.",

[0193] "Technical Summary":"Company 1's share price has performed well over the past year, showing a steady upward trend and is currently trading around $281. The stock is holding steadily above its 200-day moving average and the RSI reading of 63 suggests that the share price may be near a short-term top, but the positive MACD suggests that the bullish trend remains solid. The increase in trading volume reflects investors' positive sentiment towards the company.",

[0194] "News Observation":"Company 1's recent strategic initiatives, including the opening of an AI R&D center, demonstrate its determination to expand its technology and market globally. Analysts continue to give positive comments on Company 1's future and expect the company to further consolidate its leadership in the high-performance computing market.",

[0195] "Overall investment advice":"Considering the solid performance of Company 1 at the financial and technical levels, as well as the market's positive expectations for its future development prospects, we believe that Company 1's stock 1 is a high-quality stock suitable for long-term growth investment. Investors should consider including or keeping it in their portfolios based on their own risk preferences and investment strategies. In view of the volatility of the market, investors are advised to keep an eye on market dynamics and adjust their investment positions in a timely manner when the market provides suitable opportunities. Company 1's long-term growth potential and market position make it attractive among technology stocks, but we still recommend that investors adopt a cautious investment strategy to cope with potential macroeconomic and market risks."}}

[0196] Furthermore, the fusion instruction may also include preference information for characterizing different types of key information data such as fundamentals, news, and technical aspects, so as to instruct the comprehensive analysis model to make further textual summaries of the fundamental key information data, the news key information data, and the technical key information data composed of the key information data corresponding to the numerical type data and the key information data corresponding to the image type data, and to adjust the attention weight values ​​of the comprehensive analysis model to the fundamental key information data, the news key information data, and the technical key information data, so as to adjust the emphasis of the textual summary information on different types of key information data, so as to improve the click-through rate of the target users on the final investment research report.

[0197] Specifically, the financial client can send a request message for obtaining a target investment research report on a target virtual resource to an electronic device. The electronic device obtains the request message and, based on the page identifier carried in the request message, obtains the page feature information corresponding to the page identifier. The electronic device can then match the page feature information with data feature information of different types of information data such as fundamentals, news, and technology, generate a weight sequence based on the matching results, and finally construct a fusion instruction based on the weight sequence.

[0198] Among them, the page identifier is used to indicate the application page that triggers the request information. Based on the page identifier, the page content of the application page that triggers the request information for obtaining the target investment research report of the target virtual resource can be obtained to further extract page feature information from the page content; for example, the text information in the page content can be identified, and the text information can be converted into a text vector to construct page feature information, and the image information in the page content can also be feature extracted to obtain page feature information. It can be understood that when a user triggers the request information for obtaining the target investment research report of the target virtual resource on the news information page, it can indicate that the user currently has a higher preference for news-type information data. For example, when a user triggers the request information for obtaining the target investment research report of the target virtual resource on the market data page of the target virtual information (such as the individual stock page), it can indicate that the user currently has a higher preference for fundamental-type and technical-type information data.

[0199] The data feature information of different types of information data such as fundamentals, news and technical aspects can obtain common word texts of different types of information data such as fundamentals, news and technical aspects, so as to use the text vector of the common word text as the data feature information of the corresponding type of information data; for example, the common word text of fundamental type information data includes but is not limited to industry, sector, shareholder, revenue, growth rate, profit margin, etc., the common word text of news type information data includes but is not limited to information, report, news, strategic measures, etc., and the common word text of technical type information data includes but is not limited to market trend, K-line, RSI, Bollinger Bands, price direction, etc.

[0200] After obtaining the page feature information, the page feature information can be matched with data feature information of different types of information data such as fundamentals, news, and technology based on the following formula (1):

[0201]

[0202] in, (j=1, 2, 3; where j=1 corresponds to the fundamental aspect, j=2 corresponds to the message aspect, and j=3 corresponds to the technical aspect) The page feature information represents the data feature information of different types of information data on the fundamental aspect (j=1), the message aspect (j=2), and the technical aspect (j=3), respectively, h′ i Indicates the feature information of the i-th page, s j (j=1, 2, 3) represents the matching results between the page feature information and the data feature information of different types of information data such as fundamentals, news, and technology.

[0203] In this way, investment research reports with different focuses can be provided to investors with different investment experience and different category preferences to adapt to the needs of different users and provide technical support for each individual.

[0204] In addition, if the number of features between the data feature information of the page feature information and the data feature information of the information data of the fundamental type (or message type, technical type) is inconsistent, the matching result between the data feature information of the page feature information and the data feature information of the information data of the fundamental type (or message type, technical type) can also be calculated by deep learning feature matching.

[0205] After obtaining the matching results between the page feature information and the data feature information of different types of information data such as fundamentals, news, and technology, the matching results can be normalized to construct a weight sequence with the normalized matching results. For example, the weight sequence can be in the form of [s1, s2, s3]. The subsequent electronic device can write the weight sequence into the fusion instruction. When the comprehensive analysis model is triggered to perform text summarization on the fundamental key information data, the message key information data, and the technical key information data, the fundamental key information data, the message key information data, and the technical key information data can be encoded first to obtain the feature vector representation for representing the semantic information corresponding to the corresponding key information data, and then generate the attention weights corresponding to the fundamental key information data, the message key information data, and the technical key information data through the weight values ​​of the fundamental key information data, the message key information data, and the technical key information data indicated in the weight sequence, and then input the feature vector representation of the corresponding key information data to the decoder after weighting by adjusting the attention weights respectively, so as to obtain the text summary information through the decoder; it can be understood that the above encoder and decoder are pre-trained neural network models for text summarization. Through the weight sequences of different types of key information data, the comprehensive analysis model pays more attention to the corresponding key information data, so that the generated text summary information is more accurate and more in line with user expectations.

[0206] Exemplarily, the electronic device may create a result fusion and storage API based on the comprehensive analysis model, and then input multiple key information data into the result fusion and storage API in the form of a dictionary, and directly output a target investment research report based on the comprehensive analysis model.

[0207] Exemplarily, the electronic device can return the target investment research report to the corresponding client of the target user, and write the target investment research report into a database such as MySQL for subsequent query and analysis, and ensure the security and traceability of the data.

[0208] Through the above embodiments, a comprehensive analysis model can be used to integrate the analysis results of text, numerical and image data, that is, key information data, and comprehensively consider various factors to provide investors with comprehensive and multi-dimensional market analysis reports, and provide users with clear investment advice and market prospects.

[0209] In some implementations, the electronic device can update data and key information data in real time to provide users with the latest market trends.

[0210] In some possible implementations, electronic devices can analyze multimodal data in other fields such as commodity trading and market research according to the technical solution of the present application, but are not limited thereto.

[0211] In some possible implementations, with the user's authorization, the electronic device can perform a combination analysis on the user's holdings (i.e., multiple virtual resources) according to the technical solution of this application, provide opinions for the overall investment portfolio (e.g., return and risk analysis opinions), and provide technical support for implementing a personal financial assistant.

[0212] In some possible implementations, in combination with the above embodiments, such as Figure 3 As shown, the electronic device can create a TextSummary function, a DataSummary function, and an ImageSummary function, where:

[0213] The TextSummary function can call the object profile data interface, information data interface, data processing service API, and model call service API. Among them, the object profile data interface and the information data interface can be used to obtain text type data such as company profile, individual stock dynamics, industry analysis, etc. The data processing service API can preprocess the text type data. The model call service API can write text-related prompt sentences, i.e., text prompt instructions, for company profiles, information, and industry information, and perform natural language technology, semantic analysis, topic extraction, and information summary on the preprocessed text type data based on the text key information extraction model in combination with the text-related prompt sentences, and return the fundamental and news related analysis results, i.e., the key information data corresponding to the text type data in the key information data.

[0214] The DataSummary function can call the financial data interface, market data interface, data processing service API, and model calling service API. The financial data interface and market data interface can be used to obtain numerical data such as financial data, market conditions, and special indicators. The data processing service API can preprocess numerical data. The model calling service API can write numerical prompt statements, i.e., numerical prompt instructions, for numerical data such as statistical features and price movement information. It can also extract statistical features, trend analysis, anomaly detection, data analysis, etc. from the preprocessed numerical data based on the numerical key information extraction model in combination with the numerical prompt statements, and return the technical analysis results, i.e., the key information data corresponding to the numerical data in the key information data.

[0215] The ImageSummary function can call the financial data interface, market data interface, data processing service API, and model calling service API. Among them, the financial data interface and market data interface can be used to obtain image type data such as stock trend charts, financial statements, short selling data charts, technical indicator charts, etc. The data processing service API can preprocess the image type data. The model calling service API can write image-related prompt statements, i.e., image prompt instructions, for image type data such as image information, and perform computer vision technology, image preprocessing, etc. on the preprocessed image type data based on the image key information extraction model in combination with the image-related prompt statements, and return the technical related analysis results, i.e., the key information data corresponding to the image type data in the key information data.

[0216] Next, the electronic device can write a fusion instruction, i.e., a fusion prompt statement, call a large model such as a comprehensive analysis model to obtain the analysis results, parse the analysis results (adjust the format according to a preset display format), obtain the target investment research report, and store the target investment research report in the database, and return it to the user through the client.

[0217] Through the technical solution of this application, advanced large language models and big data processing technologies can be used to analyze the multimodal data (images, texts, and numerical data) of virtual resources to achieve automated production of investment research reports that are both timely and professional. This can solve the current problems of information fragmentation, low processing efficiency, and lack of real-time performance in the virtual resource market, and achieve the goal of inclusive investment research, that is, enabling the majority of individual investors to obtain high-quality investment research capabilities at a low cost.

[0218] It should be noted that all the above technical solutions can be combined in any way to form optional embodiments of the present application, which will not be described one by one here.

[0219] Figure 4 Schematic diagram of a device 400 for determining a virtual resource investment research report provided in an embodiment of the present application. Figure 4 As shown, the device 400 for determining the virtual resource investment research report includes: a first acquisition module 401, a second acquisition module 402, a determination module 403, and a fusion module 404.

[0220] In some implementable embodiments, the first acquisition module 401 is used to: acquire multimodal data related to the target virtual resource, the multimodal data including data of at least two modal types; the second acquisition module 402 is used to: acquire key information data of the multimodal data corresponding to the different modal types according to the information extraction method of the different modal types and the multimodal data; the determination module 403 is used to: determine the preset display format; the fusion module 404 is used to: fuse the key information data of the multimodal data corresponding to the different modal types to obtain a target investment research report corresponding to the target virtual resource and based on the preset display format.

[0221] In some implementable embodiments, the second acquisition module 402 is specifically used to: in response to the modality type being a text type, write the text type data in the multimodal data into the text prompt instruction to obtain an updated text prompt instruction; based on the text type data in the multimodal data and the updated text prompt instruction, extract text key information through a text key information extraction model to obtain fundamental key information data and message key information data; wherein, the fundamental key information data includes profile information, news information and industry dynamics information of the target object corresponding to the target virtual resource, and the message key information data includes event information associated with the target virtual resource.

[0222] In some implementable embodiments, the second acquisition module 402 is specifically used to: perform text segmentation processing on text type data in multimodal data through a text key information extraction model to obtain multiple text paragraphs; perform entity recognition and event extraction on multiple text paragraphs respectively through a text key information extraction model to obtain paragraph key information of each of the multiple text paragraphs, wherein the paragraph key information includes entity text and event information; perform key information extraction on the entity text in the multiple paragraph key information to generate basic key information data; perform key information extraction on the event information in the multiple paragraph key information to generate message key information data.

[0223] In some implementable embodiments, the second acquisition module 402 is specifically used to: in response to the modal type being a numerical type, input the numerical type data and the numerical prompt instruction in the multimodal data into the numerical key information extraction model, classify the numerical type data through the numerical key information extraction model, and obtain the technical indicator type numerical data and market trend type numerical data corresponding to the target virtual resources; extract the statistical characteristics of the technical indicator type numerical data through the numerical key information extraction model to obtain the statistical characteristic analysis results; analyze the asset inflow and outflow data, price trend data and abnormal movement data in the market trend type numerical data through the numerical key information extraction model to obtain the current market evaluation results; determine the key information data corresponding to the numerical type data according to the statistical characteristic analysis results and the current market evaluation results; wherein the numerical prompt instruction is used to control the numerical key information extraction model to determine the key information data corresponding to the numerical type data.

[0224] In some implementable embodiments, the second acquisition module 402 is specifically used to: in response to the modal type being image type data, input the image type data and image prompt instructions in the multimodal data into the image key information extraction model, classify the image type data through the image key information extraction model, and obtain the K-line chart data and technical indicator image data of the target virtual resource; perform numerical analysis on the K-line chart data through the image key information extraction model to obtain price fluctuation information of the target virtual resource; identify line changes in the technical indicator image data through the image key information extraction model to obtain market trend information; determine the key information data corresponding to the image type data based on the price fluctuation information and the market trend information; wherein the image prompt instruction is used to control the image key information extraction model to determine the key information data corresponding to the image type data.

[0225] In some implementable embodiments, the first acquisition module 401 is specifically used to: obtain a resource identifier of a target virtual resource; obtain original text data of a target object corresponding to the target virtual resource based on the resource identifier; construct structured text data and unstructured text data based on the original text data and attribute fields carried by the original text data, wherein the unstructured text data includes unstructured text in the original text data and fields in the attribute fields corresponding to the unstructured text; and determine the unstructured text data and the structured text data as text type data in multimodal data.

[0226] In some implementable methods, the first acquisition module 401 is specifically used to: obtain the resource identifier of the target virtual resource; obtain the historical market data corresponding to the target virtual resource based on the resource identifier; render the historical market data to obtain image type data in the multimodal data; wherein the historical market data at least includes: the financial report information and K-line chart data of the target object corresponding to the target virtual resource; the image type data at least includes: the market trend chart of the target virtual resource and the capital flow chart of the target object corresponding to the target virtual resource.

[0227] In some implementations, the fusion module 404 is specifically used to: determine a fusion instruction, the fusion instruction is used to characterize the processing tasks and processing criteria of the comprehensive analysis model, and the fusion instruction includes a preset display format; input multiple key information data and the fusion instruction into the comprehensive analysis model, control the comprehensive analysis model according to the fusion instruction, fuse the multiple key information data, and obtain a target investment research report corresponding to the target virtual resource and based on the preset display format;

[0228] It should be understood that the device embodiment and the method embodiment may correspond to each other, and similar descriptions may refer to the method embodiment. To avoid repetition, they will not be described here. Specifically, Figure 4 The device 400 shown can execute the above method embodiments, and the above and other operations and / or functions of each module in the device 400 are respectively for implementing the corresponding processes in the above methods, which will not be repeated here for the sake of brevity.

[0229] The above describes the device 400 of the embodiment of the present application from the perspective of the functional module in conjunction with the accompanying drawings. It should be understood that the functional module can be implemented in hardware form, can be implemented by instructions in software form, and can also be implemented by a combination of hardware and software modules. Specifically, the steps of the method embodiment in the embodiment of the present application can be completed by the hardware integrated logic circuit and / or software form instructions in the processor, and the steps of the method disclosed in the embodiment of the present application can be directly embodied as a hardware decoding processor to perform, or a combination of hardware and software modules in the decoding processor to perform. Optionally, the software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, an electrically erasable programmable memory, a register, etc. The storage medium is located in a memory, and the processor reads the information in the memory, and completes the steps in the above method embodiment in conjunction with its hardware.

[0230] Figure 5 A schematic block diagram of an electronic device 500 provided in an embodiment of the present application.

[0231] like Figure 5 As shown, the electronic device 500 may include:

[0232] The memory 510 and the processor 520, the memory 510 is used to store the computer program and transmit the program code to the processor 520. In other words, the processor 520 can call and run the computer program from the memory 510 to implement the method in the embodiment of the present application.

[0233] For example, the processor 520 may be configured to execute the above method embodiments according to instructions in the computer program.

[0234] In some embodiments of the present application, the processor 520 may include but is not limited to:

[0235] General-purpose processor, digital signal processor (DSP), application-specific integrated circuit (ASIC), field programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic device, discrete hardware components, etc.

[0236] In some embodiments of the present application, the memory 510 includes but is not limited to:

[0237] Volatile memory and / or non-volatile memory. Among them, the non-volatile memory can be read-only memory (ROM), programmable ROM (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM) or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct RAMbus RAM (DR RAM).

[0238] In some embodiments of the present application, the computer program may be divided into one or more modules, which are stored in the memory 510 and executed by the processor 520 to complete the method provided by the present application. The one or more modules may be a series of computer program instruction segments capable of completing specific functions, and the instruction segments are used to describe the execution process of the computer program in the electronic device.

[0239] like Figure 5 As shown, the electronic device may also include:

[0240] The transceiver 530 may be connected to the processor 520 or the memory 510 .

[0241] The processor 520 may control the transceiver 530 to communicate with other devices, specifically, to send information or data to other devices, or to receive information or data sent by other devices. The transceiver 530 may include a transmitter and a receiver. The transceiver 530 may further include an antenna, and the number of antennas may be one or more.

[0242] It should be understood that the various components in the electronic device are connected via a bus system, wherein the bus system includes not only a data bus but also a power bus, a control bus and a status signal bus.

[0243] The present application also provides a computer storage medium on which a computer program is stored, and when the computer program is executed by a computer, the computer can perform the method of the above method embodiment. In other words, the present application embodiment also provides a computer program product containing instructions, and when the instructions are executed by a computer, the computer can perform the method of the above method embodiment.

[0244] When software is used for implementation, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instruction is loaded and executed on a computer, the computer can be made to perform the corresponding flow in each method in the embodiment of the present application in whole or in part, and generate the functions that can be realized by each method in the embodiment of the present application. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instruction can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instruction can be transmitted from a website site, computer, server or data center by wired (such as coaxial cable, optical fiber, digital subscriber line (Digital Subscriber Line, DSL)) or wireless (such as infrared, wireless, microwave, etc.) mode to another website site, computer, server or data center. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server, a data center, etc. that contains one or more available media integrations. The available medium may be a magnetic medium (eg, a floppy disk, a hard disk, a magnetic tape), an optical medium (eg, a digital video disc (DVD)), or a semiconductor medium (eg, a solid state disk (SSD)).

[0245] Those of ordinary skill in the art will appreciate that the modules and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0246] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the module is only a logical function division. There may be other division methods in actual implementation, such as multiple modules 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 devices or modules, which can be electrical, mechanical or other forms.

[0247] The modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical modules, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. For example, each functional module in each embodiment of the present application may be integrated into a processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0248] The above are only specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for determining a virtual resource investment research report, characterized in that: include: Acquire multimodal data related to the target virtual resource, wherein the multimodal data includes data of at least two modal types; According to the information extraction methods of different modal types and the multimodal data, key information data of the multimodal data corresponding to the different modal types are obtained; Determine the preset display format; The key information data of the multimodal data corresponding to the different modal types are integrated to obtain a target investment research report corresponding to the target virtual resource and based on the preset display format.

2. The method according to claim 1, characterized in that The step of obtaining key information data of the multimodal data corresponding to the different modal types according to the information extraction methods of the different modal types and the multimodal data includes: In response to the modality type being a text type, writing text type data in the multimodal data into a text prompt instruction to obtain an updated text prompt instruction; Based on the text type data in the multimodal data and the updated text prompt instruction, extracting text key information through a text key information extraction model to obtain fundamental key information data and message key information data; Among them, the fundamental key information data includes the profile information, news information and industry dynamic information of the target object corresponding to the target virtual resource, and the message key information data includes event information associated with the target virtual resource.

3. The method according to claim 2, characterized in that The text key information extraction is performed based on the text type data in the multimodal data and the updated text prompt instruction by using a text key information extraction model to obtain basic key information data and message key information data, including: Performing text segmentation processing on the text type data in the multimodal data by using the text key information extraction model to obtain multiple text paragraphs; Performing entity recognition and event extraction on the multiple text paragraphs respectively through the text key information extraction model to obtain paragraph key information of each of the multiple text paragraphs, wherein the paragraph key information includes entity text and event information; Extract key information from entity texts in the multiple paragraph key information to generate the basic key information data; Key information is extracted from the event information in the plurality of paragraph key information to generate the message face key information data.

4. The method according to claim 1, characterized in that: The step of obtaining key information data of the multimodal data corresponding to the different modal types according to the information extraction methods of the different modal types and the multimodal data includes: In response to the modality type being a numerical type, inputting the numerical type data and the numerical prompt instruction in the multimodal data into a numerical key information extraction model, classifying the numerical type data through the numerical key information extraction model, and obtaining technical indicator numerical data and market trend numerical data corresponding to the target virtual resource; Extracting statistical features of the technical indicator numerical data through the numerical key information extraction model to obtain statistical feature analysis results; The asset inflow and outflow data, price trend data and abnormal movement data in the market trend numerical data are analyzed by the numerical key information extraction model to obtain the current market evaluation result; Determine the key information data corresponding to the numerical type data according to the statistical feature analysis result and the current market evaluation result; The numerical prompt instruction is used to control the numerical key information extraction model to determine the key information data corresponding to the numerical type data.

5. The method according to claim 1, characterized in that The step of obtaining key information data of the multimodal data corresponding to the different modal types according to the information extraction methods of the different modal types and the multimodal data includes: In response to the modality type being an image type, inputting the image type data and the image prompt instruction in the multimodal data into an image key information extraction model, classifying the image type data through the image key information extraction model, and obtaining the K-line chart data and technical indicator image data of the target virtual resource; Performing numerical analysis on the K-line chart data through the image key information extraction model to obtain price fluctuation information of the target virtual resource; Identify line changes in the technical indicator image data using the image key information extraction model to obtain market trend information; Determining key information data corresponding to the image type data according to the price fluctuation information and the market trend information; The image prompt instruction is used to control the image key information extraction model to determine the key information data corresponding to the image type data.

6. The method according to any one of claims 1 to 5, characterized in that: The acquiring of multimodal data related to the target virtual resource includes: Obtaining a resource identifier of the target virtual resource; According to the resource identifier, obtaining original text data of the target object corresponding to the target virtual resource; Based on the original text data and the attribute fields carried by the original text data, constructing structured text data and unstructured text data, wherein the unstructured text data includes the unstructured text in the original text data and the fields in the attribute fields corresponding to the unstructured text; The unstructured text data and the structured text data are used to determine text type data in the multimodal data.

7. The method according to any one of claims 1 to 5, characterized in that: The acquiring of multimodal data related to the target virtual resource includes: Obtaining a resource identifier of the target virtual resource; According to the resource identifier, obtaining historical market data corresponding to the target virtual resource; Rendering the historical market data to obtain image type data in the multimodal data; Among them, the historical market data at least includes: financial report information and K-line chart data of the target object corresponding to the target virtual resource; the image type data at least includes: the market trend chart of the target virtual resource and the capital flow chart of the target object corresponding to the target virtual resource.

8. The method according to any one of claims 1 to 5, characterized in that: The step of fusing the key information data of the multimodal data corresponding to the different modal types to obtain a target investment research report corresponding to the target virtual resource and based on the preset display format includes: Determining a fusion instruction, the fusion instruction is used to characterize the processing tasks and processing criteria of the comprehensive analysis model, and the fusion instruction includes the preset display format; Input the multiple key information data and the fusion instructions into the comprehensive analysis model, control the comprehensive analysis model according to the fusion instructions, fuse the multiple key information data, and obtain the target investment research report corresponding to the target virtual resource and based on the preset display format.

9. A device for determining a virtual resource investment research report, characterized in that: include: A first acquisition module is used to: acquire multimodal data related to the target virtual resource, wherein the multimodal data includes data of at least two modal types; A second acquisition module is used to: acquire key information data of the multimodal data corresponding to the different modal types according to the information extraction methods of the different modal types and the multimodal data; A determination module, used to: determine a preset display format; The fusion module is used to: fuse the key information data of the multimodal data corresponding to the different modality types to obtain a target investment research report corresponding to the target virtual resource and based on the preset display format.

10. An electronic device, characterized in that: include: processor; as well as A memory, configured to store executable instructions of the processor; The processor is configured to perform the method of any one of claims 1 to 8 by executing the executable instructions.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.

Images