A method, apparatus and device for constructing a financial event graph

Abstract

The present application discloses a method, apparatus, and device for constructing a financial event graph. Specifically, when implementing the method for constructing a financial event graph provided by the embodiments of the present application, first, the abstract association relationship of financial news public opinion can be obtained. Then, the known events in the abstract association relationship are merged with the existing events through correlation calculation to obtain a merged link, and the causal relationship between the known events and the existing events in the merged link is extracted, where the existing events are the events existing in the retrieval database. Next, the new events in the abstract association relationship are subjected to element extraction from the financial news public opinion to obtain element information, and the element information is integrated to obtain specific event nodes. Then, the event association targets of the search data are obtained. Finally, the causal relationship, specific event nodes, and event association targets are input into the graph database to generate a knowledge graph. The financial event graph provided by the present application improves the accuracy of financial event analysis in the current market.

Description

Technical Field

[0001] This application relates to the technical field of relationship graph construction, and particularly to a method, apparatus, and device for constructing a financial event graph. Background Art

[0002] The changes in the financial market are mainly affected by the changes in the emotions of participants, financial policies, etc. Financial practitioners use statistical inference methods or function fitting methods to describe and infer the financial market from financial-related information. For the financial market, financial data is usually the performance after a certain financial event occurs, and financial data is a way of describing the financial market.

[0003] However, for the manifestation forms of financial events in the financial market, in addition to the event descriptions obtained from some structured data (for example: a certain index skyrockets; the trading volume of a certain futures suddenly decreases), more are described in the form of text (for example: a country issues a certain policy; a senior executive of a certain enterprise suddenly leaves, etc.). These text-based descriptions cannot be directly processed using conventional methods. The most common practice in the industry currently is to classify events in financial news sentiment, announcements, etc., and then use the statistical values of the classification results as analysis indicators. This approach has a relatively coarse granularity and it is difficult to make a reasonable judgment on the evolution of events.

[0004] Therefore, how to provide a more fine-grained way of describing financial events is a technical problem that needs to be urgently solved by those skilled in the art. Summary of the Invention

[0005] Based on the above problems, this application provides a method, apparatus, and device for constructing a financial event graph to improve the usability of icons in the terminal device interface for different users. The embodiments of this application disclose the following technical solutions:

[0006] A method for constructing a financial event graph, the method further includes:

[0007] Obtain the abstract association relationship of financial news sentiment;

[0008] Use correlation calculation to merge the known events in the abstract association relationship with the existing events to obtain a merged link, and extract the causal relationship in the merged link, where the causal relationship is the abstract relationship between the known event and the existing event; the existing event is an event existing in the retrieval database;

[0009] Extract element information from the financial news sentiment according to the new events in the abstract association relationship, and integrate the element information to obtain specific event nodes;

[0010] Obtain the event-related targets of the search data;

[0011] Input the causal relationship, the specific event nodes, and the event-related targets into a graph database to generate a knowledge graph.

[0012] In some possible implementation manners, obtaining the abstract association relationship of the financial news public opinion includes:

[0013] Perform sentence splitting on the financial news public opinion to obtain first target sentences;

[0014] Input the first target sentences into a causal relationship processing model to obtain multiple causal triples and multiple probability scores corresponding to the multiple causal triples;

[0015] Use the causal triples with probability scores greater than a score threshold as candidate triples;

[0016] Use the candidate triples whose event description lengths conform to the length rule as the abstract association relationship.

[0017] In some possible implementation manners, the method further includes:

[0018] Calculate the correlation between each causal event in the abstract association relationship and the existing events;

[0019] Use the causal events with a correlation greater than a correlation threshold as the known events; use the causal events with a correlation less than or equal to the correlation threshold as the new events.

[0020] In some possible implementation manners, the method further includes:

[0021] Add argument definitions to the known events that cannot be merged with the existing events through a new event generation module.

[0022] In some possible implementation manners, extracting element information from the financial news public opinion according to the new events in the abstract association relationship includes:

[0023] Perform sentence splitting on the new events to obtain second target sentences;

[0024] Input the second target sentences into an argument extraction model to obtain multiple candidate argument groups;

[0025] Obtain multiple argument confidence levels corresponding to the multiple candidate argument groups, and use the candidate argument groups with argument confidence levels greater than a confidence threshold as target argument groups;

[0026] Use a greedy algorithm to combine each target argument group to obtain the element information.

[0027] In some possible implementation manners, obtaining the event-associated target of the search data includes:

[0028] Identifying the event name and resource of the search data;

[0029] Grouping the event name and the resource according to time and user unique identifier to obtain a plurality of candidate binary tuples;

[0030] Counting the occurrence frequency of each candidate binary tuple;

[0031] Taking the candidate binary tuples with the occurrence frequency higher than the frequency threshold as target binary tuples;

[0032] Performing cumulative information entropy filtering on the target binary tuples to obtain the event-associated target.

[0033] A device for constructing a financial event graph, the device includes:

[0034] A first obtaining unit, configured to obtain the abstract association relationship of financial news public opinions;

[0035] A calculation and extraction unit, configured to use correlation calculation to merge the known events in the abstract association relationship with the existing events to obtain a merged link, and extract the causal relationship in the merged link, where the causal relationship is the abstract relationship between the known event and the existing event; the existing event is an event existing in the retrieval database;

[0036] An element extraction unit, configured to extract element information from the financial news public opinions according to the new events in the abstract association relationship, and integrate the element information to obtain specific event nodes;

[0037] A second obtaining unit, configured to obtain the event-associated target of the search data;

[0038] A generating unit, configured to input the causal relationship, the specific event nodes, and the event-associated target into a graph database to generate a knowledge graph.

[0039] In some possible implementation manners, the device further includes:

[0040] A first clause processing unit, configured to perform clause processing on the financial news public opinions to obtain a first target clause;

[0041] A first input unit, configured to input the first target clause into a causal relationship processing model to obtain a plurality of causal triples and a plurality of probability scores corresponding to the plurality of causal triples;

[0042] A first setting unit, configured to take the causal triples with the probability scores greater than the score threshold as candidate triples;

[0043] A second setting unit, configured to use the candidate triple whose event description length conforms to the length rule as the abstract association relationship.

[0044] An electronic device, comprising: a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein when the processor executes the computer program, the method for constructing a financial event graph as described above is implemented.

[0045] A vehicle, comprising a control module, and the control module is configured to execute the method for constructing a financial event graph as described above.

[0046] Compared with the prior art, the present application has the following beneficial effects: The present application discloses a method, device and equipment for constructing a financial event graph. Specifically, when implementing the method for constructing a financial event graph provided in the embodiments of the present application, first, a disease hierarchy divided based on medical codes can be obtained. Then, a medical graph is constructed according to disease names and drug names, and the medical graph includes a first entity layer, which includes disease names and drug names. Then, according to the correspondence between the disease names in the first entity layer and the disease hierarchy, the disease hierarchy is integrated with the medical graph to obtain a first integrated graph. Next, an underwriting and claims settlement graph is constructed according to insurance names and disease names, and the underwriting and claims settlement graph includes a second entity layer, which includes disease names and insurance names. Finally, according to the correspondence between the disease names in the second entity layer and the first integrated graph, the first integrated graph and the underwriting and claims settlement graph are integrated to obtain a second integrated graph. The financial event graph of the present application can describe financial events in a finer granularity. At the same time, the financial event graph of the present application can organically combine the association relationships of events in the form of a graph network. The financial event graph of the present application records the general laws of event changes and the specific content of financial events, so that the financial event graph can effectively support the attribution and expected inference of financial events. Description of the Drawings

[0047] To more clearly illustrate the technical solutions in the embodiments or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0048] Figure 1 It is a flowchart of a method for constructing a financial event graph provided by an embodiment of the present application;

[0049] Figure 2 It is a schematic structural diagram of a device for constructing a financial event graph provided by an embodiment of the present application. Detailed implementation manners

[0050] In order to enable those skilled in the art to better understand the solution of this application, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of this application. Obviously, the described embodiments are only a part of the embodiments of this application, rather than all the embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without making creative efforts shall fall within the protection scope of this application.

[0051] To facilitate the understanding of the technical solutions provided in the embodiments of this application, the background art related to the embodiments of this application will be described first below.

[0052] The changes in the financial market are mainly affected by the changes in the emotions of participants, financial policies, etc. Financial practitioners use statistical inference methods or function fitting methods to describe and infer the financial market from financial-related information. For the financial market, financial data is usually the performance after a certain financial event occurs, and financial data is a way of describing the financial market.

[0053] Among them, the financial market refers to the place and platform for trading financial assets, including the stock market, bond market, foreign exchange market, futures market, etc. The main function of the financial market is to provide investors with opportunities for financing and investment, and to promote the flow and allocation of funds. In the financial market, individuals and institutions can achieve investment appreciation or financing needs by buying and selling financial assets.

[0054] The characteristics of the financial market include high liquidity, diverse risks, transparent prices, rich information, etc. Different types of financial markets have different scales and trading methods, and investors can choose a suitable market for trading according to their investment goals and risk tolerance.

[0055] The financial market plays an important role in the development of the economy. It provides services such as capital raising, resource allocation, and risk management, and promotes the effective operation of the economy. At the same time, the financial market also faces some challenges, such as market fluctuations, information asymmetry, regulatory risks and other issues, which require regulatory and risk management measures to ensure the normal operation of the market.

[0056] However, in addition to the event descriptions obtained from certain structured data (e.g., a certain index surges; the trading volume of a certain futures contract sharply decreases), the manifestations of financial events in the financial market are more described in text form (e.g., a country issues a certain policy; a senior executive of a certain enterprise suddenly leaves). These text-based descriptions cannot be directly processed using conventional methods. Currently, the most common practice in the industry is to classify events in financial news sentiment, announcements, etc., and then use the statistical values of the classification results as analysis indicators. This approach has a relatively coarse granularity and it is difficult to make a reasonable judgment on the evolution of events.

[0057] To solve this problem, in the embodiments of the present application, a method, device, and equipment for constructing a financial event graph are provided. In the present application, first, the abstract association relationships of financial news sentiment are obtained. Then, using correlation calculation, the known events in the abstract association relationships are merged with the existing events to obtain a merged link, and the causal relationships in the merged link are extracted, where the causal relationship is an abstract relationship between the known event and the existing event; the existing event is an event existing in the retrieval database. Next, according to the new events in the abstract association relationships, element extraction is performed from the financial news sentiment to obtain element information, and the element information is integrated to obtain specific event nodes. After obtaining the specific event nodes, it is also necessary to obtain the event-related targets of the search data. Finally, the causal relationships, specific event nodes, and event-related targets are input into the graph database to generate a knowledge graph. The financial event graph of the present application can describe financial events with a finer granularity. At the same time, the financial event graph of the present application can organically combine the association relationships of events in the form of a graph network. The financial event graph of the present application records the general laws of event changes and the specific content of financial events, so that the financial event graph can effectively support the attribution and expected inference of financial events, so as to save a large amount of manpower and time in underwriting and claims settlement to determine whether a customer meets the underwriting and claims settlement conditions.

[0058] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0059] See Figure 1 , which is a flowchart of a method for constructing a financial event graph provided by an embodiment of the present application. As Figure 1 shown, the method for constructing the financial event graph may include steps S101 - S105:

[0060] S101: Obtain the abstract association relationships of financial news sentiment.

[0061] In order to construct a financial event graph, the financial event graph construction system can first obtain the abstract correlation relationship of financial news and public opinion.

[0062] News and public opinion refers to the collective collective of social attitudes, opinions, and emotions regarding a particular event or issue. It reflects the public's level of attention to a particular event, the direction of public opinion, and the potential impact on relevant parties. News and public opinion can be captured through analysis and monitoring of media reports, social media content, online discussions, and other information. In the financial sector, news and public opinion have a significant influence. Financial markets are often influenced by the emotions and expectations of market participants, and news and public opinion plays a significant role in shaping these emotions and expectations. When an event or news is widely reported or discussed, market participants are often affected, leading to market volatility.

[0063] Modern technological advancements have enabled more sophisticated and comprehensive analysis and monitoring of news and public opinion. Leveraging technologies such as natural language processing, machine learning, and data mining, we can automatically process and analyze large amounts of news text and social media data, thereby deriving key information and insights. These tools can help financial institutions and industry practitioners better understand and address the impact of news and public opinion on the market.

[0064] In some possible implementations, the abstract association relationship is displayed in the form of a triple.

[0065] In some possible implementations, the abstract association relationship of obtaining financial news public opinion includes A1-A4:

[0066] A1: Segment the financial news and public opinion to obtain a first target sentence.

[0067] When obtaining the abstract correlation relationship of financial news public opinion, it is first necessary to segment the financial news public opinion into sentences to obtain the first target sentence.

[0068] A2: Input the first target sentence into a causal relationship processing model to obtain a plurality of causal triples and a plurality of probability scores corresponding to the plurality of causal triples.

[0069] After obtaining the first target sentence, it is also necessary to input the first target sentence into the causal relationship processing model to obtain multiple causal triples and multiple probability scores corresponding to the multiple causal triples.

[0070] Among them, the causal relationship processing model is a tool or method for analyzing and understanding causal relationships. It can help users determine the causal connections between events, behaviors, or variables, as well as the mechanism of action between them. The causal relationship processing model can identify the causal relationships in the text and generate corresponding causal triples (such as causal events, correlation words, and result events). At the same time, these models can usually provide probability scores or confidence levels for the generated causal triples.

[0071] When training the causal relationship processing model, supervised learning can be used with annotated causal relationship data. The model learns the mapping relationship from the input text to the output causal triples to predict the causal relationship as accurately as possible and assign a probability score to it.

[0072] It should be noted that different causal relationship processing models may adopt different architectures and algorithms, so the way of providing probability scores or confidence levels may be different. According to the selected model and task requirements, an appropriate model can be selected and the probability score of the causal triple can be evaluated based on its output.

[0073] The following are several common causal relationship processing models: ① Traditional statistical models: Traditional statistical models (such as linear regression, logistic regression, etc.) can be used to analyze causal relationships. By controlling other variables, we can obtain the degree of influence of one variable on another. ② Experimental design models: Experimental design models determine causal relationships by comparing the experimental group and the control group and controlling and eliminating other interfering factors. For example, a randomized controlled trial can be used to evaluate the treatment effect of a certain drug on disease symptoms. ③ Causal graph models: Causal graph models are methods used to describe and visualize causal relationships. They use arrows to represent the causal relationships between variables and can show direct and indirect causal paths. Causal graph models help to better understand complex causal relationship networks. ④ Structural equation models: Structural equation models (SEM) combine measurement models and structural models to evaluate the causal relationships between multiple variables. It can quantify the direct and indirect effects between variables and provide a comprehensive model to explain the data. These models are applied in different situations and research fields. The selection of a suitable causal relationship processing model depends on the nature of the specific problem, the type of available data, and the research purpose.

[0074] A3: Use the causal triples with probability scores greater than the score threshold as candidate triples.

[0075] After obtaining the causal triples and the corresponding probability scores of the causal triples, the construction system of the financial event graph can use the causal triples with probability scores greater than the score threshold among multiple causal triples as candidate triples.

[0076] In some possible implementation manners, the score threshold may be, but is not limited to, 0.75. In the present application, the score threshold can be adjusted according to specific circumstances, and the present application does not make specific limitations on the score threshold.

[0077] A4: Use the candidate triples whose event descriptions conform to the length rule as the abstract association relationships.

[0078] After obtaining the candidate triples, the financial event graph construction system can use the candidate triples whose event descriptions conform to the length rule as the abstract association relationships.

[0079] Among them, the length of the event description can be determined according to specific requirements and application scenarios. Generally, the length of the event description can range from a few words to hundreds of words.

[0080] For short event descriptions, they may only contain the main key information and elements, such as the theme, time, location, and basic background of the event, etc. Such descriptions are usually used to quickly understand the general situation of the event and provide a concise summary.

[0081] For more detailed event descriptions, they may contain more details and context information to comprehensively understand the occurrence and impact of the event. These descriptions may involve aspects such as the cause, process, participants, motivation, and consequences of the event. Such descriptions are usually used for in-depth analysis and research of the event and provide a more comprehensive perspective and insight.

[0082] In some possible implementation manners, conforming to the length rule may be, but is not limited to, that the length of the event description is greater than 3 characters and less than 20 characters, which all conform to the length rule.

[0083] In some possible implementation manners, for the candidate triples whose event descriptions do not conform to the length rule (that is, the length of the event description is less than 3 characters and the length of the event description is greater than 20 characters, both do not conform to the length rule), they need to be deleted.

[0084] S102: Use correlation calculation to merge the known events in the abstract association relationships with the existing events to obtain a merged link, and extract the causal relationship in the merged link, where the causal relationship is the abstract relationship between the known event and the existing event; the existing event is the event existing in the retrieval database.

[0085] After obtaining the abstract association relationships of financial news public opinions, the known events in the abstract association relationships can be merged with the existing events using correlation calculation to obtain a merged link, and the causal relationships between the known events and the existing events in the merged link can be extracted. Here, the existing events are the events existing in the retrieval database. For example, if there is a relationship in the database: 'Project winning the bid ---(causal relationship)---> Enterprise revenue increase', and if a new relationship 'Revenue growth ---(causal relationship)---> Stock price increase' is found from the financial news public opinion. By calculating the correlation, it is found that the two events 'Enterprise revenue increase' and 'Revenue growth' have a high correlation. Therefore, these two events can be merged into one. Finally, the causal link 'Project winning the bid ---(causal relationship)---> Enterprise revenue increase ---(causal relationship)---> Stock price increase' is constructed in the database.

[0086] Among them, correlation calculation is a method used to measure the similarity or correlation degree between documents or data. In database retrieval and search engines,

[0087] In some possible implementation manners, the correlation calculation methods include but are not limited to the following methods: ① Cosine Similarity: Cosine Similarity measures the similarity degree between vectors by calculating the included angle between vectors. Represent the documents as vectors and calculate the cosine similarity between them. The value ranges from -1 to 1, and the closer to 1, the more similar. ② TF-IDF (Term Frequency-Inverse Document Frequency): TF-IDF is an index to measure the importance of words in a document collection. It is based on two factors: term frequency (the number of times a word appears in a document) and inverse document frequency (the rarity of a word in the entire document collection). By calculating the TF-IDF value of each word in a document, correlation ranking and matching can be performed. ③ BM25 (Best Matching 25): BM25 is a popular information retrieval algorithm used to calculate the correlation between a document and a query. It considers factors such as term frequency, document length, and inverse document frequency of query terms, and obtains the final correlation score through weighted calculation of these factors. ④ Jaccard Similarity: Jaccard Similarity is used to compare the similarity degree between two sets. It measures the similarity by calculating the ratio of the size of the intersection of two sets to the size of the union, and the value ranges from 0 to 1. The closer to 1, the more similar. These correlation calculation methods can select appropriate methods for calculation according to specific application scenarios and data types. In database retrieval and search engines, multiple methods are often combined to comprehensively evaluate and rank the correlation of documents to provide more accurate search results.

[0088] In some possible implementation manners, the retrieval database includes a vector database and Elasticsearch. Among them, the vector database is a database specifically used for storing, managing, and retrieving vector data. It treats vectors as the main data type and can effectively perform similarity search and matching. The vector database uses measurement methods such as cosine similarity to calculate the similarity between vectors and optimizes the search performance through indexing and efficient query algorithms. This makes the vector database very suitable for application scenarios that require large-scale similarity search, such as face recognition, image search, recommendation systems, etc.

[0089] Elasticsearch is a full-text search and analysis engine with powerful search and query functions. It can be used to store and index various types of data, including text, structured data, and vector data. Although Elasticsearch is not specifically designed to handle vector data, vector indexing and similarity search can be achieved through extended plugins (such as Elasticsearch VectorScoring Plugin) or custom scripts. The distributed architecture and high scalability of Elasticsearch make it an excellent choice for handling large amounts of text data and complex queries.

[0090] In some possible implementation manners, the method further includes B1 - B2:

[0091] B1: Calculate the correlation between each causal event in the abstract association relationship and the existing events.

[0092] To obtain known events and new events, the financial event graph construction system first needs to calculate the correlation between each causal event in the abstract association relationship of financial news sentiment and the existing events.

[0093] In some possible implementation manners, similarity or correlation measurement methods (such as cosine similarity, Pearson correlation coefficient, etc.) can be used to calculate the similarity or correlation score between each causal event and the existing events. The specific method can be selected according to the data type and requirements.

[0094] B2: Use the causal events with a correlation greater than the correlation threshold as the known events; use the causal events with a correlation less than or equal to the correlation threshold as the new events.

[0095] After calculating the correlation between each causal event in the abstract association relationship and the existing events, the financial event graph construction system can use the causal events with a correlation greater than the correlation threshold in the causal events as known events; use the causal events with a correlation less than or equal to the correlation threshold in the causal events as new events.

[0096] In some possible implementation manners, the relevant threshold can be determined according to the actual situation. If a greater recall is expected, the threshold can be appropriately reduced, for example: 0.4. If better precision is desired, the threshold can be increased, for example: 0.8.

[0097] In some possible implementation manners, the method further includes:

[0098] Adding argument definitions to the known events that cannot be merged with the existing events through the new event generation module.

[0099] If an event that cannot be merged is encountered, argument definitions can be added by generating a new event. This can be used to expand the model labels when iteratively training the event element extraction model next time.

[0100] In some possible implementation manners, the argument definitions can be added to the new event module through the following steps: (1) Define a new event: Select a clear and descriptive name to define the new event based on the existing event elements and relevant information. (2) Determine the arguments: Analyze the relationships between the existing events and elements, and determine the arguments required for the new event. Arguments are the key elements or attributes of an event, such as participants, time, location, etc. (3) Add argument definitions: Add definitions for each argument in the new event. The definitions should clearly and accurately describe the meaning and role of the arguments. (4) Data annotation: Using the method of manual annotation, add the arguments of the new event as labels to the training data. You can match the existing event text data with the definition of the new event, and then label the arguments of the new event. (5) Model training: Retrain the event element extraction model using the augmented annotation data. During the training process of the model, the model will learn the argument definitions and recognition methods of the new event. (6) Iterative testing and improvement: Use the newly trained model for testing and evaluation, and make adjustments and improvements according to the actual results. Repeat this process until the model can accurately extract event elements, including the newly added arguments.

[0101] In this way, the ability of the event element extraction model can be gradually improved, enabling it to identify and extract more types of event elements.

[0102] S103: Extracting element information from the financial news sentiment according to the new event in the abstract association relationship, and integrating the element information to obtain a specific event node.

[0103] After obtaining the merged link by merging the known events in the abstract association relationship with the existing events using correlation calculation and extracting the causal relationship in the merged link, the financial event graph construction system can then extract element information from the financial news sentiment according to the new event in the abstract association relationship, and integrate the element information to obtain a specific event node.

[0104] In some possible implementation manners, extracting element information from the financial news public opinion according to the new event in the abstract association relationship includes C1 - C4:

[0105] C1: Performing clause splitting on the new event to obtain a second target clause.

[0106] To extract element information from the financial news public opinion, the construction system of the financial event graph first needs to perform clause splitting on the new event in the abstract association relationship of the financial news public opinion to obtain a second target clause.

[0107] C2: Inputting the second target clause into an argument extraction model to obtain multiple candidate argument groups.

[0108] After performing clause splitting on the new event to obtain a second target clause, the construction system of the financial event graph can input the second target clause into an argument extraction model to obtain multiple candidate argument groups.

[0109] Among them, argument extraction is a task in natural language processing, aiming to identify arguments related to a specific predicate from the given text. The following is a common framework for argument extraction models: ① Feature extraction: First, useful features need to be extracted from the text to represent each word or phrase. These features can include part of speech, syntactic relationship, dependency relationship, context information, etc. A common method is to use pre - trained word vectors (such as Word2Vec, GloVe) to represent words. ② Sequence labeling model: Model the argument extraction problem as a sequence labeling problem, and models such as conditional random field (CRF) or recurrent neural network (such as long short - term memory network LSTM) can be used for modeling. The input of the model is the text sequence after feature extraction, and the output is the label of each word, indicating whether the word is an argument and the type of the argument it belongs to. ③ Labeled data preparation: To train the argument extraction model, corpus data with associated predicates needs to be labeled. Each sentence in the corpus needs to be labeled with the arguments related to the target predicate and the type of the argument is specified. This usually requires manual annotation or using existing labeled datasets. ④ Model training and optimization: Train the argument extraction model using the labeled data. By maximizing predefined performance metrics (such as accuracy, recall, F1 - value, etc.), adjust the parameters and weights of the model to improve the performance of the model. The backpropagation algorithm or other optimization algorithms can be used for model training. ⑤ Inference and evaluation: After training, the model can be used to infer new text and identify the arguments related to the predicate. The performance of the model can be evaluated by calculating metrics such as precision, recall, and F1 - value by comparing with manual annotation or test datasets.

[0110] It should be noted that to implement a complete argument extraction model, multiple aspects such as feature selection, model architecture, labeled data preparation, and model training need to be comprehensively considered, and appropriate debugging and optimization should be carried out. At the same time, the performance of the model is also limited by the quality and scale of the available labeled data.

[0111] C3: Obtain the argument confidence levels corresponding to the multiple candidate argument tuples, and use the candidate argument tuples with argument confidence levels greater than the confidence threshold as the target argument tuples.

[0112] After obtaining multiple candidate argument tuples by inputting the second target clause into the argument extraction model, the financial event graph construction system can obtain the argument confidence levels corresponding to the multiple candidate argument tuples, and use the candidate argument tuples with argument confidence levels greater than the confidence threshold in the candidate argument tuples as the target argument tuples.

[0113] In some possible implementation manners, the confidence threshold can be, but is not limited to, 0.5 or 0.6, and is determined according to the specific scenario. In this application, the confidence threshold can be adjusted according to the specific situation, and this application does not make specific limitations on the confidence threshold.

[0114] In some possible implementation manners, the following methods can be adopted to estimate the confidence level of an argument: ① Prediction probability score: Some machine learning models (such as logistic regression, support vector machine) provide the ability to predict probability scores. These probabilities represent the confidence level of the model for each argument category. You can use these probabilities as an approximation of the confidence level. ② Post-processing threshold setting: By applying a threshold to the scores output by the model, arguments with high confidence levels can be filtered out. A higher score may indicate that the model is more confident in its prediction of this argument. You can set the threshold according to specific requirements and select the arguments with higher confidence levels. ③ Model-based uncertainty estimation: Some models can estimate the uncertainty of their predictions through ensemble methods (such as Bootstrap, Dropout) or Bayesian inference. By considering the uncertainty information of the model, the confidence level of the argument can be better evaluated.

[0115] C4: Use the greedy algorithm to combine each target argument tuple to obtain the element information.

[0116] After obtaining the target argument tuples, the financial event graph construction system can use the greedy algorithm to combine each target argument tuple to obtain the element information.

[0117] Using the greedy algorithm to combine the complete argument tuple information can be an effective method. Among them, the greedy algorithm is a heuristic algorithm that gradually constructs a solution by selecting the current best decision at each step.

[0118] S104: Obtain the event-related targets of the search data.

[0119] After obtaining the specific event node, the financial event graph construction system can obtain the event-related targets of the search data.

[0120] Event-related targets refer to specific objects, entities, or assets associated with a particular event or activity. These targets typically have a certain relevance to the event and may be participants, influencing factors, subjects, or targets. For example, in the financial sector, an event-related target might be a financial instrument such as a company's stock, commodity futures, or foreign exchange. If a company makes a significant announcement or releases financial statements, the prices of these financial instruments may fluctuate, thus becoming associated with the event. Event-related targets are specific objects, entities, or assets associated with an event in the context of the event, describing the impact, dependency, or relationship between the event and the related objects. This correlation can be used for analysis, forecasting, and decision-making.

[0121] In some possible implementations, the event-related targets of the acquired search data include D1-D5:

[0122] D1: Identify the event name and resource of the search data.

[0123] To obtain the event-related targets of the search data, the financial event graph construction system first needs to identify the event name and resource (asset) of each search data.

[0124] In some possible implementations, natural language processing (NLP) techniques and models can be used to identify event names and resources in the search data. That is, deep learning-based NLP models, such as BERT or GPT, can be used to perform semantic understanding and information extraction on the query.

[0125] D2: Group the event name and the resource according to time and user unique identifier to obtain multiple candidate tuples.

[0126] In order to obtain event-related targets for search data, the financial event graph construction system also needs to group event names and resources according to time and user unique identifiers to obtain multiple candidate tuples.

[0127] Among them, the user ID is a unique identifier assigned to each user by the system or platform, which is used to identify the user and manage user information.

[0128] D3: Count the occurrence frequency of each candidate bigram.

[0129] After obtaining the candidate binary groups, the financial event graph construction system can count the frequency of occurrence of each candidate binary group.

[0130] D4: Use the candidate binary tuples with occurrence frequencies higher than the frequency threshold as target binary tuples.

[0131] After obtaining the candidate binary tuples, the financial event graph construction system can use the candidate binary tuples with occurrence frequencies higher than the frequency threshold as target binary tuples.

[0132] In some possible implementation manners, the frequency threshold can be, but is not limited to, a time slice determined by day, and generally the frequency is taken to be greater than 50. In this application, the frequency threshold can be adjusted according to specific situations, and this application does not make specific limitations on the frequency threshold.

[0133] D5: Perform cumulative information entropy filtering on the target binary tuples to obtain the event-related targets.

[0134] After obtaining the target binary tuples, the financial event graph construction system can perform cumulative information entropy filtering on the target binary tuples to obtain the event-related targets.

[0135] S105: Input the causal relationship, the specific event nodes, and the event-related targets into a graph database to generate a knowledge graph.

[0136] After obtaining the causal relationship, the specific event nodes, and the event-related targets, the financial event graph construction system can input the causal relationship, the specific event nodes, and the event-related targets into a graph database to generate a knowledge graph.

[0137] Among them, a graph database is a database system designed specifically for storing and processing graph-structured data. It is different from traditional relational databases or document databases. It organizes and represents data in the form of a graph, where nodes represent entities and edges represent the relationships between nodes. The core of a graph database is a graph model, which consists of nodes and edges. Nodes represent entities or objects, such as people, places, items, etc., and edges represent the relationships between nodes. Both nodes and edges can contain attributes for storing more detailed information. Common graph database products include Neo4j, Amazon Neptune, JanusGraph, etc. These databases provide rich functions and APIs, enabling developers to conveniently process and operate graph data.

[0138] Based on the content of S101-S105, it can be seen that the abstract correlation relationship of financial news public opinion is obtained, and the known events in the abstract correlation relationship are merged with existing events using correlation calculation to form a merged link. The causal relationship in the merged link is then extracted, where the causal relationship is the abstract relationship between the known event and the existing event; the existing event is the event in the search database. Based on the new event in the abstract correlation relationship, element information is extracted from the financial news public opinion to obtain element information, and the element information is integrated to obtain specific event nodes. Next, the event-related targets of the search data are obtained, and finally, the causal relationship, specific event nodes, and event-related targets are input into the graph database to generate a knowledge graph. The financial event graph of this application can describe financial events at a more granular level. At the same time, the financial event graph of this application can organically combine the correlation relationships of events in the form of a graph network. The financial event graph of this application records the general patterns of event changes and the specific content of financial events, making it effective in supporting the attribution and expected inference of financial events.

[0139] See also Figure 2 , which is a schematic diagram of the structure of a device for constructing a financial event graph provided by an embodiment of the present application. Figure 2 As shown, the device for constructing the financial event graph includes:

[0140] The first acquisition unit 201 is used to acquire the abstract correlation relationship of financial news and public opinion.

[0141] News and public opinion refers to the collective collective of social attitudes, opinions, and emotions regarding a particular event or issue. It reflects the public's level of attention to a particular event, the direction of public opinion, and the potential impact on relevant parties. News and public opinion can be captured through analysis and monitoring of media reports, social media content, online discussions, and other information. In the financial sector, news and public opinion have a significant influence. Financial markets are often influenced by the emotions and expectations of market participants, and news and public opinion plays a significant role in shaping these emotions and expectations. When an event or news is widely reported or discussed, market participants are often affected, leading to market volatility.

[0142] Modern technological advancements have enabled more sophisticated and comprehensive analysis and monitoring of news and public opinion. Leveraging technologies such as natural language processing, machine learning, and data mining, we can automatically process and analyze large amounts of news text and social media data, thereby deriving key information and insights. These tools can help financial institutions and industry practitioners better understand and address the impact of news and public opinion on the market.

[0143] In some possible implementations, the abstract association relationship is displayed in the form of a triple.

[0144] The calculation and extraction unit 202 is configured to use correlation calculation to merge the known events in the abstract association relationship with the existing events to obtain a merged link, and extract the causal relationship in the merged link, where the causal relationship is the abstract relationship between the known event and the existing event; the existing event is an event existing in the retrieval database.

[0145] For example, if there is a relationship in the database: 'winning a project bid ---(causal relationship)---> increase in corporate revenue', and a new relationship 'increase in revenue ---(causal relationship)---> increase in stock price' is found from financial news and public opinion. By calculating the correlation, it is found that the correlation between the two events 'increase in corporate revenue' and 'increase in revenue' is very high. Therefore, these two events can be merged into one. Finally, the causal link 'winning a project bid ---(causal relationship)---> increase in corporate revenue ---(causal relationship)---> increase in stock price' is constructed in the database.

[0146] Among them, correlation calculation is a method used to measure the similarity or relevance between documents or data. In database retrieval and search engines,

[0147] In some possible implementation manners, the correlation calculation methods include but are not limited to the following methods: ① Cosine Similarity: Cosine Similarity measures the similarity degree between vectors by calculating the included angle between them. Represent the documents as vectors and calculate the cosine similarity between them. The value ranges from -1 to 1, and the closer it is to 1, the more similar it indicates. ② TF-IDF (Term Frequency-Inverse Document Frequency): TF-IDF is an index to measure the importance of words in a document collection. It is based on two factors: term frequency (the number of times a word appears in a document) and inverse document frequency (the rarity of a word in the entire document collection). By calculating the TF-IDF value of each word in a document, correlation ranking and matching can be performed. ③ BM25 (Best Matching 25): BM25 is a popular information retrieval algorithm used to calculate the correlation between a document and a query. It takes into account factors such as term frequency, document length, and the inverse document frequency of query terms, and obtains the final correlation score through weighted calculation of these factors. ④ Jaccard Similarity: Jaccard Similarity is used to compare the similarity degree between two sets. It measures the similarity by calculating the size of the intersection of two sets divided by the size of the union. The value ranges from 0 to 1, and the closer it is to 1, the more similar it indicates. These correlation calculation methods can select appropriate methods for calculation according to specific application scenarios and data types. In database retrieval and search engines, multiple methods are often combined to comprehensively evaluate and rank the correlation of documents to provide more accurate search results.

[0148] In some possible implementation manners, the retrieval database includes a vector database and Elasticsearch. Among them, the vector database is a database specifically used for storing, managing, and retrieving vector data. It regards vectors as the main data type and can effectively perform similarity search and matching. The vector database uses measurement methods such as cosine similarity to calculate the similarity between vectors, and optimizes the search performance through indexing and efficient query algorithms. This makes the vector database very suitable for application scenarios that require large-scale similarity search, such as face recognition, image search, recommendation systems, etc.

[0149] Elasticsearch is a full-text search and analysis engine with powerful search and query capabilities. It can be used to store and index various types of data, including text, structured data, and vector data. Although Elasticsearch is not specifically designed for handling vector data, vector indexing and similarity search can be achieved through extended plugins (such as Elasticsearch VectorScoring Plugin) or custom scripts. Elasticsearch's distributed architecture and high scalability make it an excellent choice for handling large amounts of text data and complex queries.

[0150] The element extraction unit 203 is configured to extract element information from the financial news sentiment according to the new events in the abstract association relationship, and integrate the element information to obtain a specific event node.

[0151] The second acquisition unit 204 is configured to acquire the event-related targets of the search data.

[0152] Herein, the event-related target refers to the specific object, entity, or asset related thereto in a certain specific event or activity. These targets usually have a certain relevance to the event and may be the participants, influencing factors, objects, or goals of the event, etc. For example, in the financial field, an event-related target can be financial products such as the stocks of a certain company, commodity futures, foreign exchange, etc. If there is an important announcement or release of financial statement data of the company, the prices of these financial products may change, so these financial products are related to this event. The event-related target is the specific object, entity, or asset related thereto in the context of a certain event, and is used to describe the impact, dependence, or relationship of the event on the related objects. This relevance can be used for aspects such as analysis, prediction, and decision-making.

[0153] The generation unit 205 is configured to input the causal relationship, the specific event node, and the event-related target into the graph database to generate a knowledge graph.

[0154] Herein, the graph database is a database system specifically designed for storing and processing graph-structured data. Different from traditional relational databases or document databases, it organizes and represents data in the form of a graph, where nodes represent entities and edges represent the relationships between nodes. The core of the graph database is the graph model, which consists of nodes and edges. Nodes represent entities or objects, such as people, places, items, etc., and edges represent the relationships between nodes. Both nodes and edges can contain attributes for storing more detailed information. Common graph database products include Neo4j, Amazon Neptune, JanusGraph, etc. These databases provide rich functions and APIs, enabling developers to conveniently process and operate graph data.

[0155] In some possible implementations, the apparatus further includes:

[0156] A first clause processing unit, configured to perform clause processing on the financial news public opinion to obtain a first target clause.

[0157] A first input unit, configured to input the first target clause into a causal relationship processing model to obtain a plurality of causal triples and a plurality of probability scores corresponding to the plurality of causal triples.

[0158] Among them, the causal relationship processing model is a tool or method for analyzing and understanding causal relationships. It can help users determine the causal connections between events, behaviors, or variables, as well as the mechanism of action between them. The causal relationship processing model can identify the causal relationships in the text and generate corresponding causal triples (such as causal events, correlation words, and result events). At the same time, these models can usually provide probability scores or confidence levels for the generated causal triples.

[0159] When training the causal relationship processing model, supervised learning can be performed using annotated causal relationship data. The model learns the mapping relationship from the input text to the output causal triples to predict the causal relationships as accurately as possible and assign probability scores to them.

[0160] It should be noted that different causal relationship processing models may adopt different architectures and algorithms, so the way of providing probability scores or confidence levels may be different. According to the selected model and task requirements, an appropriate model can be selected and the probability scores of the causal triples can be evaluated based on its output.

[0161] The following are several common causal relationship processing models: ① Traditional statistical models: Traditional statistical models (such as linear regression, logistic regression, etc.) can be used to analyze causal relationships. By controlling other variables, we can obtain the degree of influence of one variable on another variable. ② Experimental design models: Experimental design models determine causal relationships by comparing experimental groups and control groups and controlling and excluding other interfering factors. For example, randomized controlled trials can be used to evaluate the treatment effect of a certain drug on disease symptoms. ③ Causal graph models: Causal graph models are methods used to describe and visualize causal relationships. They use arrows to represent the causal relationships between variables and can show direct and indirect causal paths. Causal graph models help to better understand complex causal relationship networks. ④ Structural equation models: Structural equation models (SEM) combine measurement models and structural models to evaluate the causal relationships between multiple variables. It can quantify the direct and indirect effects between variables and provide a comprehensive model to explain the data. These models are applied in different scenarios and research fields. The selection of a suitable causal relationship processing model depends on the nature of the specific problem, the type of available data, and the research purpose.

[0162] A first setting unit, configured to use causal triples with probability scores greater than a score threshold as candidate triples.

[0163] In some possible implementation manners, the score threshold may be, but is not limited to, 0.75. In this application, the score threshold can be adjusted according to specific circumstances, and this application does not make specific limitations on the score threshold.

[0164] A second setting unit, configured to use the candidate triples whose event descriptions conform to a length rule as the abstract association relationships.

[0165] Wherein, the length of the event description can be determined according to specific requirements and application scenarios. Generally, the length of the event description can range from a few words to hundreds of words.

[0166] For short event descriptions, they may only contain the main key information and elements, such as the theme, time, location, and basic background of the event, etc. Such descriptions are usually used to quickly understand the general situation of the event and provide a concise summary.

[0167] For more detailed event descriptions, they may contain more details and context information to comprehensively understand the occurrence and impact of the event. These descriptions may cover aspects such as the cause, process, participants, motivation, and consequences of the event. Such descriptions are usually used for in-depth analysis and research of the event, providing a more comprehensive perspective and insight.

[0168] In some possible implementation manners, conforming to the length rule may be, but is not limited to, that the length of the event description is greater than 3 characters and less than 20 characters, which all conform to the length rule.

[0169] In some possible implementation manners, for the candidate triples whose event descriptions do not conform to the length rule (that is, the length of the event description is less than 3 characters and the length of the event description is greater than 20 characters, both of which do not conform to the length rule), they need to be deleted.

[0170] In some possible implementation manners, the device further includes:

[0171] A calculation unit, configured to calculate the correlation between each causal event in the abstract association relationship and the existing events.

[0172] In some possible implementation manners, similarity or correlation measurement methods (such as cosine similarity, Pearson correlation coefficient, etc.) can be used to calculate the similarity or correlation score between each causal event and the existing events. The specific method can be selected according to the data type and requirements.

[0173] A third setting unit, configured to use the causal events with a relevance greater than a correlation threshold as the known events; and use the causal events with a relevance less than or equal to the correlation threshold as the new events.

[0174] In some possible implementation manners, the correlation threshold can be determined according to actual situations. If a greater recall is expected, the threshold can be appropriately reduced, for example: 0.4. If better precision is desired, the threshold can be increased, for example: 0.8.

[0175] In some possible implementation manners, the apparatus further includes:

[0176] A generation and addition unit, configured to add argument definitions to the known events that cannot be merged with the existing events through a new event generation module.

[0177] In some possible implementation manners, the argument definitions of the new event generation module can be generated through the following steps: (1) Define a new event: According to the existing event elements and relevant information, select a clear and descriptive name to define the new event. (2) Determine the arguments: Analyze the relationships between the existing events and elements, and determine the arguments required for the new event. An argument is a key element or attribute of an event, such as a participant, time, location, etc. (3) Add argument definitions: Add definitions for each argument in the new event. The definitions should clearly and accurately describe the meaning and role of the arguments. (4) Data annotation: Using the method of manual annotation, add the arguments of the new event as labels to the training data. You can match the existing event text data with the definition of the new event, and then annotate the arguments of the new event. (5) Model training: Retrain the event element extraction model using the augmented annotated data. During the training process of the model, the model will learn the argument definitions and recognition methods of the new event. (6) Iterative testing and improvement: Use the newly trained model for testing and evaluation, and make adjustments and improvements according to the actual results. Repeat this process until the model can accurately extract event elements, including the newly added arguments.

[0178] In this way, the ability of the event element extraction model can be gradually improved, enabling it to identify and extract more types of event elements.

[0179] In some possible implementation manners, the apparatus further includes:

[0180] A second clause processing unit, configured to perform clause processing on the new event to obtain a second target clause.

[0181] A second input unit, configured to input the second target clause into an argument extraction model to obtain multiple candidate argument groups.

[0182] Among them, argument extraction is a task in natural language processing, aiming to identify the arguments related to a specific predicate from the given text. The following is a common framework for argument extraction models: ① Feature extraction: First, useful features need to be extracted from the text to represent each word or phrase. These features can include part of speech, syntactic relationship, dependency relationship, context information, etc. A common method is to use pre-trained word vectors (such as Word2Vec, GloVe) to represent words. ② Sequence labeling model: Model the argument extraction problem as a sequence labeling problem, and models such as conditional random field (CRF) or recurrent neural network (such as long short-term memory network LSTM) can be used for modeling. The input of the model is the text sequence after feature extraction, and the output is the label of each word, indicating whether the word is an argument and the type of the argument it belongs to. ③ Labeled data preparation: To train the argument extraction model, it is necessary to label the corpus data related to the predicate. Each sentence in the corpus needs to be labeled with the arguments related to the target predicate and the type of the argument is specified. This usually requires manual annotation or using existing labeled datasets. ④ Model training and optimization: Use the labeled data to train the argument extraction model. By maximizing predefined performance metrics (such as accuracy, recall, F1 value, etc.), adjust the parameters and weights of the model to improve the performance of the model. The backpropagation algorithm or other optimization algorithms can be used for model training. ⑤ Inference and evaluation: After training, the model can be used to infer new text and identify the arguments related to the predicate. The performance of the model can be evaluated by calculating metrics such as precision, recall, and F1 value by comparing with manual annotation or test datasets.

[0183] It should be noted that implementing a complete argument extraction model requires comprehensive consideration of multiple aspects such as feature selection, model architecture, labeled data preparation, and model training, and appropriate debugging and optimization. At the same time, the performance of the model is also limited by the quality and scale of the available labeled data.

[0184] A third acquisition unit is used to acquire the argument confidence levels corresponding to the multiple candidate argument groups, and use the candidate argument groups with argument confidence levels greater than the confidence threshold as target argument groups.

[0185] In some possible implementation manners, the confidence threshold can be, but is not limited to, 0.5 or 0.6, which is determined according to the specific scenario. In this application, the confidence threshold can be adjusted according to the specific situation, and this application does not make specific limitations on the confidence threshold.

[0186] In some possible implementation manners, the following methods can be adopted to estimate the confidence of arguments: ① Predictive probability scores: Some machine learning models (such as logistic regression, support vector machines) provide the ability to predict probability scores. These probabilities represent the confidence of the model in each argument category. You can use these probabilities as an approximation of the confidence. ② Post - processing threshold setting: By applying a threshold to the scores output by the model, arguments with high confidence can be filtered out. Higher scores may indicate that the model is more confident in its prediction of the argument. You can set the threshold according to specific requirements to select arguments with higher confidence. ③ Model - based uncertainty estimation: Some models can estimate the uncertainty of their predictions through ensemble methods (such as Bootstrap, Dropout) or Bayesian inference. By considering the uncertainty information of the model, the confidence of the argument can be better evaluated.

[0187] A combination unit, configured to combine each target argument tuple by using a greedy algorithm to obtain the element information.

[0188] Using a greedy algorithm to combine the complete argument tuple information can be an effective method. Among them, the greedy algorithm is a heuristic algorithm that gradually constructs a solution by selecting the current best decision at each step.

[0189] In some possible implementation manners, the device further includes:

[0190] An identification unit, configured to identify the event name and resources of the search data.

[0191] In some possible implementation manners, natural language processing (NLP) techniques and models can be used to identify the event name and resources of the search data. That is to say, NLP models based on deep learning, such as BERT or GPT, can be used to perform semantic understanding and information extraction on the query.

[0192] A grouping unit, configured to group the event name and the resources according to time and user unique identifier to obtain a plurality of candidate binary tuples.

[0193] Among them, the user ID is a unique identifier assigned by the system or platform to each user, used to identify the user identity and manage user information.

[0194] A statistical unit, configured to count the occurrence frequencies of each candidate binary tuple.

[0195] A third setting unit, configured to use the candidate binary tuples with occurrence frequencies higher than the frequency threshold as target binary tuples.

[0196] In some possible implementation manners, the frequency threshold may be, but is not limited to, a time slice determined by day, and generally the frequency is greater than 50. In the present application, the frequency threshold may be adjusted according to specific circumstances, and the present application does not make specific limitations on the frequency threshold.

[0197] A filtering unit, configured to perform cumulative information entropy filtering on the target binary group to obtain the event-associated target.

[0198] In addition, an embodiment of the present application further provides a device for constructing a financial event graph. The device includes a memory and a processor. The memory is used to store a program or code, and the processor is used to run the program or code stored in the memory to implement the above-mentioned method for constructing a financial event graph.

[0199] In addition, an embodiment of the present application further provides a computer-readable storage medium, characterized in that the computer-readable storage medium stores code, and when the code is run, the device running the code implements the above-mentioned method for constructing a financial event graph.

[0200] An embodiment of the present application provides a device for constructing a financial event graph. First, the first acquisition unit 201 may be used to acquire the abstract association relationship of financial news and public opinion. Then, through the calculation and extraction unit 202, the known events in the abstract association relationship are merged with the existing events by using correlation calculation to obtain a merged link, and the causal relationship in the merged link is extracted, where the causal relationship is an abstract relationship between the known event and the existing event; the existing event is an event existing in the retrieval database. Then, the element extraction unit 203 extracts element information from the new events in the abstract association relationship from the financial news and public opinion, and integrates the element information to obtain specific event nodes. The second acquisition unit 204 acquires the event-associated target of the search data. Finally, the generation unit 205 inputs the causal relationship, the specific event nodes, and the event-associated target into the graph database to generate a knowledge graph. The financial event graph of the present application can describe financial events in a finer granularity. At the same time, the financial event graph of the present application can organically combine the association relationships of events in the form of a graph network. The financial event graph of the present application records the general rules of event changes and the specific content of financial events, so that the financial event graph can effectively support the attribution and expected inference of financial events.

[0201] The above has introduced in detail a method, apparatus, device, and storage medium for constructing a financial event graph provided by this application. The various embodiments in the specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. For the same or similar parts among the various embodiments, reference can be made to each other. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple. For the relevant parts, reference can be made to the description in the method section. It should be noted that for those of ordinary skill in the art in this technical field, without departing from the principle of this application, several improvements and modifications can still be made to this application, and these improvements and modifications also fall within the protection scope of the claims of this application.

[0202] It should be understood that in this application, "at least one (item)" means one or more, and "a plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships can exist. For example, "A and / or B" can mean: only A exists, only B exists, and both A and B exist simultaneously. Among them, A and B can be singular or plural. The character " / " generally represents an "or" relationship between the associated objects before and after. "At least one (one) of the following" or its similar expression refers to any combination of these items, including any combination of single item (one) or plural items (ones). For example, at least one (one) of a, b, or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0203] It should also be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise", or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article, or device. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of additional identical elements in the process, method, article, or device including the said element.

[0204] The steps of the methods or algorithms described in connection with the embodiments disclosed herein may be implemented directly in hardware, in a software module executed by a processor, or in a combination thereof. The software module may be placed in a random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0205] The foregoing description of the disclosed embodiments enables those skilled in the art to make or use the present application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Thus, the present application is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for constructing a financial event graph, characterized in that, The method further includes: Obtaining an abstract association relationship of financial news public opinion; Using correlation calculation to merge the known events in the abstract association relationship with the existing events to obtain a merged link, and extracting the causal relationship in the merged link, where the causal relationship is an abstract relationship between the known event and the existing event; the existing event is an event existing in the retrieval database; Performing element extraction on the financial news public opinion according to the new events in the abstract association relationship to obtain element information, and integrating the element information to obtain specific event nodes; Obtaining the event associated target of the search data; Inputting the causal relationship, the specific event nodes, and the event associated target into a graph database to generate a knowledge graph; The performing element extraction on the financial news public opinion according to the new events in the abstract association relationship to obtain element information includes: Performing clause splitting on the new event to obtain a second target clause; Inputting the second target clause into an argument extraction model to obtain a plurality of candidate argument groups; Obtaining a plurality of argument confidence degrees corresponding to the plurality of candidate argument groups, and taking the candidate argument groups with argument confidence degrees greater than the confidence threshold as target argument groups; Using a greedy algorithm to combine each target argument group to obtain the element information.

2. The method according to claim 1, wherein The obtaining an abstract association relationship of financial news public opinion includes: Performing clause splitting on the financial news public opinion to obtain a first target clause; Inputting the first target clause into a causal relationship processing model to obtain a plurality of causal triples and a plurality of probability scores corresponding to the plurality of causal triples; Taking the causal triples with probability scores greater than the score threshold as candidate triples; Taking the candidate triples whose event description lengths conform to the length rule as the abstract association relationship.

3. The method according to claim 1, wherein The method further includes: Calculating the correlation between each causal event in the abstract association relationship and the existing events; Taking the causal events with the correlation greater than the correlation threshold as the known events; taking the causal events with the correlation less than or equal to the correlation threshold as the new events.

4. The method according to claim 1, characterized in that, The method further includes: Adding argument definitions to the known events that cannot be merged with the existing events through a new event generation module.

5. The method according to claim 1, wherein The obtaining the event associated target of the search data includes: Identifying the event name and resources of the search data; Grouping the event name and the resources according to time and user unique identifier to obtain a plurality of candidate binary groups; Counting the occurrence frequencies of each candidate binary group; Taking the candidate binary groups with the occurrence frequencies higher than the frequency threshold as target binary groups; Performing cumulative information entropy filtering on the target binary groups to obtain the event associated target.

6. A device for constructing a financial event graph, characterized in that, The apparatus includes: A first acquisition unit, configured to obtain an abstract association relationship of financial news public opinion; A calculation and extraction unit, configured to use correlation calculation to merge the known events in the abstract association relationship with the existing events to obtain a merged link, and extract the causal relationship in the merged link, where the causal relationship is an abstract relationship between the known event and the existing event; the existing event is an event existing in the retrieval database; An element extraction unit, configured to extract element information from the financial news public opinion according to a new event in the abstract association relationship, and integrate the element information to obtain a specific event node; A second acquisition unit, configured to acquire an event association target of search data; A generation unit, configured to input the causal relationship, the specific event node, and the event association target into a graph database to generate a knowledge graph; The extracting element information from the financial news public opinion according to a new event in the abstract association relationship includes: Performing clause splitting on the new event to obtain a second target clause; Inputting the second target clause into an argument extraction model to obtain a plurality of candidate argument groups; Obtaining a plurality of argument confidence degrees corresponding to the plurality of candidate argument groups, and using the candidate argument groups with argument confidence degrees greater than a confidence threshold as target argument groups; Using a greedy algorithm to combine each target argument group to obtain the element information.

7. The device according to claim 6, wherein The apparatus further includes: A first clause processing unit, configured to perform clause splitting on the financial news public opinion to obtain a first target clause; A first input unit, configured to input the first target clause into a causal relationship processing model to obtain a plurality of causal triples and a plurality of probability scores corresponding to the plurality of causal triples; A first setting unit, configured to use the causal triples with probability scores greater than a score threshold as candidate triples; A second setting unit, configured to use the candidate triples with the length of the event description meeting a length rule as the abstract association relationship.

8. An electronic device, characterized in that, including: A memory, a processor, and a computer program stored on the memory and executable on the processor, wherein when the processor executes the computer program, the method for constructing a financial event graph according to any one of claims 1-5 is implemented.

9. A vehicle, characterized in that, The vehicle includes a control module, and the control module is configured to execute the method for constructing a financial event graph according to any one of claims 1-5.

Images