Virtual digital human automatic writing method and device, computer device, and storage medium

Abstract

Embodiments of the present application disclose a virtual digital person automatic writing method and device, computer equipment and a storage medium. The method comprises: collecting and collating financial information materials; performing semantic structuring processing on the financial information materials to obtain structured data; analyzing the structured data to determine an automatic writing topic; and generating an automatic writing text according to the automatic writing topic. Through the method of the embodiments of the present application, automatic anchor commentary topic planning and content creation can be realized according to financial business materials and financial information news, and a commentary text conforming to industry specifications is generated.

Description

Technical Field

[0001] This invention relates to artificial intelligence, and more specifically to a method, apparatus, computer equipment, and storage medium for automated writing of virtual digital humans. Background Technology

[0002] In the financial sector, virtual digital humans are increasingly being used in various scenarios, such as virtual anchors. However, most of the narration content for virtual digital humans is currently prepared manually in advance and cannot be automatically written based on actual situations. Furthermore, it is impossible to automatically plan and create anchor narration topics based on financial business materials and financial news, which can easily lead to the generated narration texts not conforming to industry standards.

[0003] Therefore, it is necessary to design a new method to automate the planning and content creation of anchor commentary topics based on financial business materials and financial news, and generate commentary texts that conform to industry standards. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method, apparatus, computer equipment and storage medium for automated writing of virtual digital humans.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a virtual digital human automated writing method, comprising: Collect and organize financial information materials; The financial information material is semantically structured to obtain structured data; The structured data is analyzed to determine topics for automated writing. Automated writing text is generated based on the selected topics for automated writing.

[0006] The further technical solution is as follows: the semantic structuring processing of the financial information material to obtain structured data includes: The financial information material is subjected to text segmentation and entity recognition to obtain the recognition results; The identification results are then classified to obtain classification results; The classification results are subjected to relational structure extraction to obtain structured data.

[0007] The further technical solution is as follows: the text segmentation and entity recognition of the financial information material to obtain the recognition result includes: The financial information material was processed by combining a pre-trained deep model with a CRF and a BI sequence labeling system to perform text segmentation and entity recognition, thereby obtaining the recognition results.

[0008] The further technical solution is as follows: classifying the identification result to obtain a classification result includes: The recognition results are classified using a BERT pre-trained model based on Transformer architecture to obtain the classification results.

[0009] The further technical solution is as follows: the relationship structure extraction of the classification results to obtain structured data includes: The classification results are structured and extracted using a combination of trigger words and matching templates to obtain the first extraction result. An event and relation extraction processing model is used to extract the relation structure from the classification results that cannot be extracted, so as to obtain a second extraction result; The first extraction result and the second extraction result are integrated to obtain structured data.

[0010] The further technical solution is as follows: the analysis of the structured data to determine automated writing topics includes: Events within the structured data are aggregated according to key elements to obtain the aggregation result; Identify the current subject or time-based hot events within the aggregated results; Analyze the transmission chain of the aforementioned hot events; By integrating topic selection elements based on the aforementioned transmission chain, automated writing topics can be determined.

[0011] The further technical solution is as follows: the generation of automated writing text based on the automated writing topic includes: Extract text summaries from the automated writing topics; Query and populate a visual chart showing the current market trend or indicator movement; Generate a text summary of the hotspot event's correlation logic with the transmission link to obtain the target text summary; Automated writing text is generated based on the visualization charts and the target text summary.

[0012] The present invention also provides a virtual digital human automated writing device, comprising: The data collection unit is used to collect and organize financial information materials. The processing unit is used to perform semantic structuring processing on the financial information material to obtain structured data; An analysis unit is used to analyze the structured data to determine topics for automated writing. The text generation unit is used to generate automated writing text based on the automated writing topic.

[0013] The present invention also provides a computer device, the computer device including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the above-described method.

[0014] The present invention also provides a storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0015] The beneficial effects of this invention compared with the prior art are as follows: This invention performs semantic structuring processing on collected and organized financial information materials to form structured data, and analyzes the data to determine automated writing topics. Based on these topics, it generates automated writing text, realizing automated anchor commentary topic planning and content creation based on financial business materials and financial news, and generating commentary text that conforms to industry standards.

[0016] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram illustrating an application scenario of the virtual digital human automated writing method provided in an embodiment of the present invention. Figure 2 A flowchart illustrating the automated writing method for virtual digital humans provided in an embodiment of the present invention; Figure 3 A schematic diagram of a sub-process of the virtual digital human automated writing method provided in an embodiment of the present invention; Figure 4 A schematic diagram of a sub-process of the virtual digital human automated writing method provided in an embodiment of the present invention; Figure 5 A schematic diagram of a sub-process of the virtual digital human automated writing method provided in an embodiment of the present invention; Figure 6 A schematic diagram of a sub-process of the virtual digital human automated writing method provided in an embodiment of the present invention; Figure 7 This is a schematic diagram of the event and relationship extraction processing model provided in an embodiment of the present invention; Figure 8 A schematic block diagram of a virtual digital human automated writing device provided in an embodiment of the present invention; Figure 9A schematic block diagram of the processing unit of the virtual digital human automated writing device provided in an embodiment of the present invention; Figure 10 A schematic block diagram of the extraction subunit of the virtual digital human automated writing device provided in an embodiment of the present invention; Figure 11 A schematic block diagram of the analysis unit of the virtual digital human automated writing device provided in an embodiment of the present invention; Figure 12 A schematic block diagram of a computer device provided for an embodiment of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0021] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0022] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0023] Please see Figure 1 and Figure 2 , Figure 1 This is a schematic diagram illustrating an application scenario of the virtual digital human automated writing method provided in an embodiment of the present invention. Figure 2This is a schematic flowchart illustrating the automated writing method for virtual digital humans provided in an embodiment of the present invention. The automated writing method for virtual digital humans is applied in a server. The server interacts with the terminal, acquires financial information materials input by the terminal, performs semantic structuring processing on the materials to obtain structured data, analyzes this data to determine automated writing topics, and generates automated writing text based on these topics. This enables automated planning and content creation for anchor commentary based on financial business materials and financial news, generating commentary text that conforms to industry standards.

[0024] Figure 2 This is a flowchart illustrating the automated writing method for virtual digital humans provided in an embodiment of the present invention. Figure 2 As shown, the method includes the following steps S110 to S140.

[0025] S110. Collect and organize financial information materials.

[0026] In this embodiment, financial information materials refer to information materials for the financial industry, mainly including: financial business and product materials and financial news materials. Among them, financial business materials mainly include credit cards, large-amount deposits and loans, wealth management products, etc. Therefore, it is necessary to organize promotional text materials for each marketing product, such as: basic product information, product term, profitability introduction, risk introduction, target customer group, purchase process, maturity or early redemption, etc.

[0027] The first step is to organize the financial product materials, specifically the text and image materials introducing the products to be marketed. This involves organizing the basic information of the product according to these materials, including the product name, issuing entity, issuance scope, product term, description of profitability, description of risks, target customer group, purchase process, maturity or early redemption, and other basic information. Generally, basic product characteristics are stored in PDF documents or product pages on the homepage, and some are presented in tables or images. This information needs to be organized according to the basic product information items mentioned above. Tools such as pdfbox should be used to extract and recognize the text information from the PDF documents, converting it into text corpus data.

[0028] Next, news information is collected and organized. Besides the basic information on financial products mentioned above, the most important aspect is collecting relevant news information about current financial products, primarily from financial industry websites and major departmental websites. This targeted data collection uses XPath extraction templates to extract data, transforming tables and other data into storable structured data. For unstructured data such as news articles, extraction is mainly based on HTML webpage text extraction methods. Specific methods generally employ XPath fixed template extraction technology and statistical methods based on character tag features. For text data such as news articles, the main focus is on identifying data information such as title, date, body text, author, and publication channel.

[0029] S120. Perform semantic structuring processing on the financial information material to obtain structured data.

[0030] In this embodiment, structured data refers to data such as text segmentation, tag classification, and event information.

[0031] When automated writing by robots, or automated writing by virtual digital humans, is required, it is necessary to establish structured knowledge data, or structured data, by quantifying and structuring unstructured data. For example, analyzing the trends of financial products from news and information can provide basic data for automated writing by digital humans.

[0032] In one embodiment, please refer to Figure 3 The above-mentioned step S120 may include steps S121 to S123.

[0033] S121. Perform text segmentation and entity recognition on the financial information material to obtain the recognition results.

[0034] In this embodiment, the recognition result refers to the entity recognition result of each word after text segmentation of the financial information material.

[0035] Specifically, a pre-trained deep model combined with CRF and a BI sequence labeling system are used to perform text segmentation and entity recognition on the financial information material to obtain recognition results.

[0036] In this embodiment, text word segmentation and entity recognition are the basis of semantic understanding. Common entity categories in the financial field include: Person, Time, Num, Local, Org, Product, Price, Index, etc. Non-entity word categories include basic parts of speech such as Norm, Verb, Adj, Adv, etc. This embodiment adopts a method combining a pre-trained deep model and CRF, and uses a BI sequence labeling system, that is, the algorithm model for word segmentation and entity recognition. For example, for the phrase "Australian iron ore plummets", "Australia" is labeled as: Aus / B-Local tralia / I-Local; "iron ore" is labeled as: iron / B-Product ore / I-Product stone / I-Product, etc.

[0037] In this embodiment, financial information materials are input character by character. The Word Embedding of each character is pre-trained using a language model such as Word2Vec, and information such as the pinyin of the character and the stroke structure of the character are spliced and fused to obtain the final Word Embedding vectorized representation of the character. The word sequence label classifications such as B-Local and I-Local are represented by One-Hot encoding, and the Viterbi algorithm is used for decoding at the CRF level.

[0038] S122. Classify the recognition result to obtain a classification result.

[0039] In this embodiment, the classification result refers to the category corresponding to the recognition result, that is, the category of the entity.

[0040] In this embodiment, the BERT pre-trained model with Transformer as the basic structure is used to classify the recognition result to obtain the classification result.

[0041] Specifically, the text classification labels in the financial field, that is, the labels of the classification results, are mainly divided into: product labels, event labels, indicator noun labels, sentiment labels, etc., as follows: Financial products are categorized as follows: 1) Financial trading instruments (wealth management products, stock indices, etc.); 2) Financial industry sectors (finance, non-ferrous metals, agricultural products, chemicals, etc.); 3) Financial regions (China, the United States, Singapore, Japan, Australia, etc.), (Southeast Asia, the Middle East, Europe, etc.), (Shanghai, London, New York, etc.); 4) Financial market institutions; 5) National government agencies (Ministry of Finance, People's Bank of China, China Securities Regulatory Commission, etc.); and 6) Foreign government agencies or international economic organizations (Federal Reserve, German Central Bank, World Bank, etc.). Events are categorized and labeled as follows: 1) Market-related events (market surges and crashes, record highs, etc.); 2) Supply chain events (rising raw material prices, changes in raw material production, import and export restrictions, etc.); 3) Policy-related events (industrial encouragement, industrial restrictions, production rate adjustments, market matching policies for similar products, etc.); 4) Natural disaster events (earthquakes, floods, insect infestations, high temperatures, etc.); 5) Macroeconomic and financial events (rising unemployment, continued economic strength, etc.). The indicator tags are categorized as follows: 1) Macroeconomic indices (GDP, CPI data indicators, etc.); 2) Microeconomic product data indicators (output, transaction amount, inventory, etc.). Emotional tags are categorized as follows: 1) Different degrees of impact (positive, negative, neutral); 2) Importance factor tags (very significant, significant, average, minor, etc.).

[0042] In this paper, a BERT pre-trained model based on the Transformer architecture is used for classification. The BERT pre-trained model is jointly trained for the MLM task and the next sentence prediction task, and uses a Transformer module based entirely on the attention mechanism, which can capture the semantic relationships between words in long texts. This allows the model's output vector matrix to accurately and comprehensively depict the overall information of the original text, providing a good model input for downstream fine-tuning tasks. In specific classification applications, a fine-tuned layer is added to the BERT output layer for classification prediction. First, the word segmentation structure is used based on the financial text corpus, then Google's BERT open-source training tool is used for word semantic representation learning, and finally, multi-sub-task training is performed according to dimensions such as subject, event, etc.

[0043] S123. Perform relational structure extraction on the classification results to obtain structured data.

[0044] In this embodiment, the classification results are extracted in a structured manner by combining trigger words and matching templates to obtain structured data; Alternatively, an event and relation extraction processing model can be used to extract the relation structure from the classification results to obtain structured data.

[0045] The structured extraction of event relationships mainly involves extracting two parts: event type and event elements. Financial event types are primarily categorized into market trends, industry chains, macroeconomic factors, and natural disasters, while event elements mainly include structured information such as the event subject and event time. Therefore, for event relationship extraction from news information, event type identification and specific event element extraction are generally separated into two stages, using either heuristic rule-based extraction methods or model-predictive extraction methods. Rule-based extraction is precise and facilitates human intervention, while model-based extraction has stronger generalization capabilities. Therefore, this embodiment uses a combination of rule-based and model-based methods for structured relationship extraction.

[0046] In one embodiment, please refer to Figure 4 The above step S123 may include steps S1231 to S1233.

[0047] S1231. Use a combination of trigger words and matching templates to extract the relationship structure from the classification results to obtain the first extraction result.

[0048] In this embodiment, the first extraction result refers to the result obtained by using a combination of trigger words and matching templates to perform relational structure extraction on the classification results.

[0049] S1232. Using an event and relation extraction processing model, perform relation structure extraction on the classification results that cannot be extracted to obtain a second extraction result.

[0050] In this embodiment, the second extraction result refers to the result obtained by using an event and relation extraction processing model to perform relational structure extraction on the classification results that cannot be extracted by combining trigger words and matching templates.

[0051] S1233. Integrate the first extraction result and the second extraction result to obtain structured data.

[0052] Specifically, a rule-based approach is preferred, which involves using trigger words and matching templates to extract relationships in a structured manner. Trigger words are generally event keywords, such as "plummeting," "surging," and "hitting new lows," while the matching template consists of entity words, parts of speech, generalization matchers (*), overlap counts, and logical operators. For example, the rule template for a financial instrument decline event is shown in Table 1.

[0053] Table 1. Rule Template Fall | Plunge | Drop | More Than ?$(DATE) &$(PRODUCT) ~k(Decline|Plunge|Decline|More Than) Secondly, an event and relation extraction processing model is used for extraction, as follows: Figure 7 As shown, while rule matching is accurate, its generalization ability is insufficient. For text that cannot be matched by rules, a model-based extraction method is needed. Generally, news information is first classified to determine the event type, and then BiLSTM+CRF is used to label the event element sequence based on the event type, similar to the previous CRF entity recognition. The difference from entity recognition is that the event joint extraction model integrates event category and event element extraction for joint modeling and training; the event element sequence prediction uses the intermediate semantic information of Bi-LSTM in the original text, the optimal serialization label embedding of NER-CRF, and the event type embedding.

[0054] This allows for the acquisition of text segmentation, tag classification, and event information. This data provides structured semantic material for intelligent investment research, thus providing a basic structured corpus for automated writing of virtual digital humans.

[0055] S130. Analyze the structured data to determine topics for automated writing.

[0056] In this embodiment, the topic selection for automated writing refers to the subject matter corresponding to automated writing, namely, factors such as financial entities, events, causes, and impacts.

[0057] In one embodiment, please refer to Figure 5 The above step S130 may include steps S131 to S134.

[0058] S131. Aggregate the events in the structured data according to key elements to obtain the aggregation result.

[0059] In this embodiment, the aggregation result refers to the result formed after aggregating the events in the structured data according to key elements such as subject, event, and event type.

[0060] Specifically, aggregation algorithms such as Canopy and Kmeans can be used to aggregate events in the structured data according to key elements, while the similarity measurement between events can be achieved by converting events into vectors and using the Cosine similarity algorithm to calculate the similarity.

[0061] S132. Determine the current subject or time hotspot events within the aggregation result.

[0062] In this embodiment, the aggregation results are used to identify hot (key) events, mainly based on various information such as the time of occurrence of the event, the degree of dissemination, the reporting channels, and market event rules (such as the main stock price fluctuation exceeding 8%), to determine the hot (key) events of the current subject or time.

[0063] S133. Analyze the transmission chain of the hot events.

[0064] In this embodiment, the transmission link refers to the transmission path of hot events.

[0065] Specifically, the analysis of the transmission chain of hot events mainly involves integrating various dimensions such as financial entities, event tags, relational elements, occurrence time, and related impacts. In practice, a financial product tag map can be used to deduce the cause chain of events using a breadth-first algorithm with limited depth.

[0066] S134. Integrate the topic selection elements according to the transmission link to determine the topic for automated writing.

[0067] Specifically, the elements of the topic selection are integrated, summarized, and the content of the topic is determined, thereby identifying the financial subjects, events, causes, and impacts of the automated writing.

[0068] S140. Generate automated writing text based on the automated writing topic selection.

[0069] In this embodiment, the automated written text refers to the explanatory text.

[0070] The topics for automated writing were determined, including information on various content elements such as the subject, event, time, cause, and impact. Based on these selections, automated writing text was generated. In this embodiment, financial automated writing is structured according to a three-part framework: a section describing the current situation or event of the financial entity; a section analyzing the causes of the current situation or event; and a section outlining potential subsequent events or conclusions related to the current situation or event. Each section is essentially composed of a content summary and charts illustrating relevant data indicators.

[0071] In one embodiment, please refer to Figure 6 The above-mentioned step S140 may include steps S141 to S144.

[0072] S141. Extract text summaries from the automated writing topics.

[0073] In this embodiment, the text summary refers to the summary generated using a template.

[0074] Text summarization is performed on the selected financial entities and events, i.e., the topics for automated writing. The summarization can be done using templates or deep learning techniques, such as Seq2Seq. For accurate, standardized, and controllable language, a template approach is used, for example, for "thermal coal price crash," specifying key information such as the event subject, time, crash percentage, trading volume, and factory inventory. A text summary is then generated using the corresponding template.

[0075] S142. Query and populate the visualization chart of the current main market conditions or indicator trends.

[0076] In this embodiment, a visual chart refers to a chart generated by rendering it in the form of an Echarts chart, or a chart generated by converting it into template variable text filling.

[0077] Specifically, it queries and populates the visualization chart of the current main market trend or indicator trend. Based on the current main theme and time, it retrieves market / indicator data from relevant databases such as real-time market data, daily candlestick charts, and spot warehouses, and renders the data into a chart in the Echarts chart format or converts it into template variable text for population.

[0078] S143. Generate a text summary of the hotspot event pair transmission link association logic to obtain the target text summary.

[0079] In this embodiment, the target text digest refers to the text digest of the current financial hot events and their correlation with the transmission chain logic. Generate a text summary of the transmission chain logic of current financial hot events. Specifically, it obtains the transmission path and related logic of <causal event, main current event> from the event transmission association knowledge base, and converts it into a specific text description using a template summary method. The main functions are to generate the main market or indicator chart of the causal event, generate a text summary and transmission logic of the subsequent transmission affecting products or events, and generate relevant market and indicator data of the main body of the subsequent transmission affecting events.

[0080] S144. Generate automated writing text based on the visualization chart and the target text summary.

[0081] After the above processing, it becomes possible to automate the writing process for hot products and events in the financial field.

[0082] The aforementioned virtual digital human automated writing method performs semantic structuring on collected and organized financial information materials to form structured data, analyzes this data to determine automated writing topics, and generates automated writing text based on these topics. This enables automated anchor commentary topic planning and content creation based on financial business materials and financial news, generating commentary texts that conform to industry standards.

[0083] Figure 8 This is a schematic block diagram of a virtual digital human automated writing device 300 provided in an embodiment of the present invention. Figure 8 As shown, corresponding to the above-described automated writing method for virtual digital humans, the present invention also provides an automated writing device 300 for virtual digital humans. This automated writing device 300 includes a unit for executing the above-described automated writing method for virtual digital humans, and the device can be configured in a server. Specifically, please refer to... Figure 8 The virtual digital human automated writing device 300 includes a data acquisition unit 301, a processing unit 302, an analysis unit 303, and a text generation unit 304.

[0084] The data acquisition unit 301 is used to collect and organize financial information materials; the processing unit 302 is used to perform semantic structuring processing on the financial information materials to obtain structured data; the analysis unit 303 is used to analyze the structured data to determine the topics for automated writing; and the text generation unit 304 is used to generate automated writing text based on the topics for automated writing.

[0085] In one embodiment, such as Figure 9 As shown, the processing unit 302 includes an identification subunit 3021, a classification subunit 3022, and an extraction subunit 3023.

[0086] The identification subunit 3021 is used to perform text segmentation and entity recognition on the financial information material to obtain the identification result; the classification subunit 3022 is used to classify the identification result to obtain the classification result; and the extraction subunit 3023 is used to extract the relationship structure from the classification result to obtain structured data.

[0087] In one embodiment, the recognition subunit 3021 is used to perform text segmentation and entity recognition on the financial information material by combining a pre-trained deep model with CRF and a BI sequence labeling system, so as to obtain recognition results.

[0088] In one embodiment, the classification subunit 3022 is used to classify the recognition result using a BERT pre-trained model based on Transformer to obtain a classification result.

[0089] In one embodiment, such as Figure 10 As shown, the extraction subunit 3023 includes a first extraction module 30231, a second extraction module 30232, and an integration module 30233.

[0090] The first extraction module 30231 is used to extract the relationship structure from the classification results by combining trigger words and matching templates to obtain a first extraction result; the second extraction module 30232 is used to extract the relationship structure from the classification results that cannot be extracted by using an event and relationship extraction processing model to obtain a second extraction result; the integration module 30233 is used to integrate the first extraction result and the second extraction result to obtain structured data.

[0091] In one embodiment, such as Figure 11 As shown, the analysis unit 303 includes an aggregation subunit 3031, an event determination subunit 3032, an analysis subunit 3033, and an element integration subunit 3034.

[0092] Aggregation subunit 3031 is used to aggregate events in the structured data according to key elements to obtain aggregation results; event determination subunit 3032 is used to determine the current subject or time hot events in the aggregation results; analysis subunit 3033 is used to analyze the transmission link of the hot events; element integration subunit 3034 is used to integrate topic selection elements according to the transmission link to determine automated writing topics.

[0093] In one embodiment, the text generation unit 304 includes an extraction subunit, an icon generation subunit, a summary generation subunit, and a written text generation subunit.

[0094] The extraction subunit is used to extract text summaries from the automated writing topics; the icon generation subunit is used to query and populate the visualization charts of the current main market conditions or indicator trends; the summary generation subunit is used to generate text summaries of the hot events' correlation logic to obtain the target text summary; and the writing text generation subunit is used to generate automated writing text based on the visualization charts and the target text summary.

[0095] It should be noted that those skilled in the art can clearly understand that the specific implementation process of the above-mentioned virtual digital human automated writing device 300 and each unit can be referred to the corresponding description in the foregoing method embodiments. For the sake of convenience and brevity, it will not be repeated here.

[0096] The aforementioned virtual digital human automated writing device 300 can be implemented as a computer program, which can, for example... Figure 12 It runs on the computer device shown.

[0097] Please see Figure 12 , Figure 12This is a schematic block diagram of a computer device provided in an embodiment of this application. The computer device 500 can be a server, wherein the server can be a standalone server or a server cluster composed of multiple servers.

[0098] See Figure 12 The computer device 500 includes a processor 502, a memory, and a network interface 505 connected via a system bus 501. The memory may include a non-volatile storage medium 503 and internal memory 504.

[0099] The non-volatile storage medium 503 may store an operating system 5031 and a computer program 5032. The computer program 5032 includes program instructions that, when executed, cause the processor 502 to perform a virtual digital human automated writing method.

[0100] The processor 502 provides computing and control capabilities to support the operation of the entire computer device 500.

[0101] The internal memory 504 provides an environment for the operation of the computer program 5032 in the non-volatile storage medium 503. When the computer program 5032 is executed by the processor 502, the processor 502 can execute a virtual digital human automated writing method.

[0102] This network interface 505 is used for network communication with other devices. Those skilled in the art will understand that... Figure 12 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device 500 to which the present application is applied. The specific computer device 500 may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0103] The processor 502 is used to run a computer program 5032 stored in the memory to perform the following steps: Collect and organize financial information materials; perform semantic structuring processing on the financial information materials to obtain structured data; analyze the structured data to determine automated writing topics; and generate automated writing text based on the automated writing topics.

[0104] In one embodiment, when the processor 502 performs semantic structuring processing on the financial information material to obtain structured data, it specifically implements the following steps: The financial information material is subjected to text segmentation and entity recognition to obtain recognition results; the recognition results are classified to obtain classification results; and the classification results are subjected to relational structure extraction to obtain structured data.

[0105] In one embodiment, when the processor 502 performs text segmentation and entity recognition on the financial information material to obtain the recognition result, it specifically implements the following steps: The financial information material was processed by combining a pre-trained deep model with a CRF and a BI sequence labeling system to perform text segmentation and entity recognition, thereby obtaining the recognition results.

[0106] In one embodiment, when the processor 502 performs the step of classifying the recognition result to obtain the classification result, it specifically implements the following steps: The recognition results are classified using a BERT pre-trained model based on Transformer architecture to obtain the classification results.

[0107] In one embodiment, when the processor 502 performs the step of extracting the relationship structure from the classification result to obtain structured data, it specifically implements the following steps: The classification results are structured and extracted using a combination of trigger words and matching templates to obtain a first extraction result; the classification results that cannot be extracted are structured and extracted using an event and relation extraction processing model to obtain a second extraction result; the first extraction result and the second extraction result are integrated to obtain structured data.

[0108] In one embodiment, when the processor 502 performs the step of analyzing the structured data to determine the automated writing topic, it specifically implements the following steps: Events within the structured data are aggregated according to key elements to obtain aggregation results; hot events of the current subject or time within the aggregation results are identified; the transmission links of the hot events are analyzed; and topic selection elements are integrated based on the transmission links to determine topics for automated writing.

[0109] In one embodiment, when implementing the step of generating automated writing text based on the automated writing topic, the processor 502 specifically implements the following steps: Extract text summaries from the automated writing topics; query and populate the visualization charts of the current main market conditions or indicator trends; generate text summaries of the correlation logic between the hot events and the transmission links to obtain the target text summary; generate automated writing text based on the visualization charts and the target text summary.

[0110] It should be understood that in the embodiments of this application, the processor 502 may be a central processing unit (CPU) 302, or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0111] It will be understood by those skilled in the art that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program includes program instructions and can be stored in a storage medium, which is a computer-readable storage medium. The program instructions are executed by at least one processor in the computer system to implement the process steps of the embodiments of the above methods.

[0112] Therefore, the present invention also provides a storage medium. This storage medium can be a computer-readable storage medium. The storage medium stores a computer program, wherein when executed by a processor, the computer program causes the processor to perform the following steps: Collect and organize financial information materials; perform semantic structuring processing on the financial information materials to obtain structured data; analyze the structured data to determine automated writing topics; and generate automated writing text based on the automated writing topics.

[0113] In one embodiment, when the processor executes the computer program to perform semantic structuring processing on the financial information material to obtain structured data, it specifically implements the following steps: The financial information material is subjected to text segmentation and entity recognition to obtain recognition results; the recognition results are classified to obtain classification results; and the classification results are subjected to relational structure extraction to obtain structured data.

[0114] In one embodiment, when the processor executes the computer program to perform text segmentation and entity recognition on the financial information material to obtain recognition results, it specifically implements the following steps: The financial information material was processed by combining a pre-trained deep model with a CRF and a BI sequence labeling system to perform text segmentation and entity recognition, thereby obtaining the recognition results.

[0115] In one embodiment, when the processor executes the computer program to classify the recognition result to obtain a classification result, it specifically implements the following steps: The recognition results are classified using a BERT pre-trained model based on Transformer architecture to obtain the classification results.

[0116] In one embodiment, when the processor executes the computer program to perform the step of extracting the structured relationships from the classification results to obtain structured data, it specifically implements the following steps: The classification results are structured and extracted using a combination of trigger words and matching templates to obtain a first extraction result; the classification results that cannot be extracted are structured and extracted using an event and relation extraction processing model to obtain a second extraction result; the first extraction result and the second extraction result are integrated to obtain structured data.

[0117] In one embodiment, when the processor executes the computer program to perform the analysis of the structured data to determine the automated writing topic selection step, it specifically implements the following steps: Events within the structured data are aggregated according to key elements to obtain aggregation results; hot events of the current subject or time within the aggregation results are identified; the transmission links of the hot events are analyzed; and topic selection elements are integrated based on the transmission links to determine topics for automated writing.

[0118] In one embodiment, when the processor executes the computer program to implement the step of generating automated writing text based on the automated writing topic, it specifically implements the following steps: Extract text summaries from the automated writing topics; query and populate the visualization charts of the current main market conditions or indicator trends; generate text summaries of the correlation logic between the hot events and the transmission links to obtain the target text summary; generate automated writing text based on the visualization charts and the target text summary.

[0119] The storage medium can be any computer-readable storage medium capable of storing program code, such as a USB flash drive, portable hard drive, read-only memory (ROM), magnetic disk, or optical disk.

[0120] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0121] In the several embodiments provided by this invention, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For example, the division of each unit is merely a logical functional division, and there may be other division methods in actual implementation. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed.

[0122] The steps in the method of this invention can be adjusted, merged, or reduced in order according to actual needs. The units in the device of this invention can be merged, divided, or reduced according to actual needs. Furthermore, the functional units in the various embodiments of this invention can be integrated into one processing unit 302, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0123] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a terminal, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention.

[0124] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A virtual digital human automated writing method, characterized in that, include: Collect and organize financial information materials; The financial information material is semantically structured to obtain structured data; The structured data is analyzed to determine topics for automated writing. Automated writing text is generated based on the aforementioned automated writing topics; The semantic structuring of the financial information material to obtain structured data includes: The financial information material is subjected to text segmentation and entity recognition to obtain the recognition results; The identification results are then classified to obtain classification results; The classification results are subjected to relational structure extraction to obtain structured data; The process of performing text segmentation and entity recognition on the financial information material to obtain recognition results includes: The financial information material was processed by combining a pre-trained deep model with a CRF and a BI sequence labeling system to perform text segmentation and entity recognition, so as to obtain the recognition results. The analysis of the structured data to determine automated writing topics includes: Events within the structured data are aggregated according to key elements to obtain the aggregation result; Identify the current subject or time-based hot events within the aggregated results; Analyze the transmission chain of the aforementioned hot events; Based on the aforementioned transmission chain, topic selection elements are integrated to determine automated writing topics; The process of generating automated writing text based on the automated writing topic selection includes: Extract text summaries from the automated writing topics; Query and populate a visual chart showing the current market trend or indicator movement; Generate a text summary of the hotspot event's correlation logic with the transmission link to obtain the target text summary; Automated writing text is generated based on the visualization charts and the target text summary.

2. The virtual digital human automated writing method according to claim 1, characterized in that, The process of classifying the identification results to obtain classification results includes: The recognition results are classified using a BERT pre-trained model based on Transformer architecture to obtain the classification results.

3. The virtual digital human automated writing method according to claim 1, characterized in that, The step of extracting structured relationships from the classification results to obtain structured data includes: The classification results are structured and extracted using a combination of trigger words and matching templates to obtain the first extraction result. An event and relation extraction processing model is used to extract the relation structure from the classification results that cannot be extracted, so as to obtain a second extraction result; The first extraction result and the second extraction result are integrated to obtain structured data.

4. A virtual digital human automated writing device, characterized in that, The device uses the virtual digital human automated writing method as described in any one of claims 1 to 3, comprising: The data collection unit is used to collect and organize financial information materials. The processing unit is used to perform semantic structuring processing on the financial information material to obtain structured data; An analysis unit is used to analyze the structured data to determine topics for automated writing. The text generation unit is used to generate automated writing text based on the automated writing topic.

5. A computer device, characterized in that, The computer device includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method as described in any one of claims 1 to 3.

6. A storage medium, characterized in that, The storage medium stores a computer program that, when executed by a processor, implements the method as described in any one of claims 1 to 3.

Images