A financial event prediction system, method and server

Abstract

The present application discloses a financial event prediction system, including a data processing module, an event extraction module, a relationship extraction module, an event generalization module, an event alignment module, a conditional mining module, a graph construction module, a user query module and a graph retrieval module. The system automatically constructs a financial event graph through knowledge documents containing financial events, and provides investment references and decision-making basis for financial workers through logical reasoning of existing financial events and accurate prediction of future financial events, solving the problem of weak logical reasoning ability of large language models in financial event reasoning in the prior art.

Description

Technical Field

[0001] The present application belongs to the technical field of large language models, and in particular, relates to a financial event prediction system, method and server. Background Art

[0002] With the development of technology, RAG technology has been used to automatically predict financial events. RAG technology retrieves relevant information from a wide range of financial document databases before the large language model generates answers, and then uses this information to guide the generation process, greatly improving the accuracy and relevance of the content. RAG effectively alleviates the "hallucination" problem that traditional large models are prone to, improves the speed of financial knowledge updates, and enhances the traceability of content generation, making large language models more practical and credible in practical applications.

[0003] However, since RAG technology first cuts financial document data into several text segments, when answering global questions about the entire text corpus, such as "Based on this document library, which are the three stocks most likely to make money in the next six months", the context involved in this question may be distributed throughout the entire large financial document library. Traditional RAG vector retrieval methods find it difficult to obtain such scattered and fine-grained document information, resulting in incomplete information retrieval.

[0004] To solve the above problems, a technology called RAG has emerged that combines knowledge graphs or fact graphs to predict financial events. However, RAG technology combined with knowledge graphs still faces the problem of weak logical reasoning ability. The big model essentially predicts the "next word" through the statistical relationship between words in the document, which makes it difficult to distinguish the causal relationship between events and is prone to making incorrect logical inferences and decisions. The knowledge graph focuses on representing the static relationship between entities, which essentially only expands the knowledge scope of the big model and does not significantly improve the logical reasoning ability of the big model.

[0005] At the same time, the automated construction method of the causal graph also faces the problem of difficulty in representing and constructing the causal graph. Since the relationship between financial events is often conditional and constantly changing, the traditional triple knowledge representation method is difficult to accurately express the uncertain relationship between financial events. In addition, the causal graph also faces the problem of difficulty in construction. For example, the machine learning-based method requires a large amount of annotated corpus, and the deep learning-based method also requires event extraction, relationship extraction, event generalization, transition probability and causal strength calculation processes. There is currently no relatively simple method for constructing the causal graph.

[0006] Therefore, there is an urgent need for a technical solution for financial event prediction that can solve the above problems. Summary of the invention

[0007] In order to address the deficiencies of the prior art, the present application provides a financial event prediction system, method and server to solve the problem of weak logical reasoning ability of large language models in the prior art in financial event reasoning.

[0008] The technical effects to be achieved by this application are achieved through the following solutions:

[0009] According to a first aspect of the present application, a financial event prediction system is provided, including a data processing module, an event extraction module, a relationship extraction module, an event generalization module, an event alignment module, a conditional mining module, a graph construction module, a user query module and a graph retrieval module, wherein:

[0010] The data processing module is used to obtain financial report data and pre-process the data;

[0011] The event extraction module is used to extract financial events from financial report data through a large language model and construct a financial event dataset;

[0012] The relationship extraction module is used to extract event relationships from the financial event data set through a large language model to construct an original financial event graph;

[0013] The event generalization module is used to vectorize each entity in the external corpus to build an external knowledge base, build a financial event list through the external knowledge base, and obtain generalized financial events through a large language model and a financial event generalization template;

[0014] The event alignment module is used to filter the triples containing the generalized financial event from the original financial event graph with the generalized financial event as the center, and merge them into a local subgraph of the generalized financial event containing background text information;

[0015] The conditional mining module is used to obtain local causal text sequences, causal quadruple acquisition, and sequential quadruple acquisition;

[0016] The graph construction module is used to construct a financial event graph based on the causal quadruple, sequential quadruple, financial events and condition-relationship;

[0017] The user query module is used to extract financial events from questions raised by users through a large language model and event extraction template;

[0018] The graph retrieval module is used to perform event retrieval and link retrieval based on the extracted financial events to form a financial event prediction list, and generate a final event prediction for the user's question and return it to the user.

[0019] Preferably, in the event extraction module, financial events are extracted from financial report data through a large language model and an event extraction template to construct a financial event data set containing context, wherein each piece of data in the financial event data set contains a text-financial event pair, in which the financial event has a subject-predicate-object structure.

[0020] Preferably, in the relationship extraction module, event relationships are extracted from the financial event data set through a large language model and an event relationship extraction template to construct an original financial event graph, wherein each piece of data in the original financial event graph contains a triple of financial event A-relationship-financial event B, and background text information of the triple, wherein the relationship includes causality, sequence, and hierarchical relationships.

[0021] Preferably, in the event generalization module, each entity is vectorized, and the external knowledge base is constructed specifically by: using the "Da Ci Lin" to vectorize each entity, retaining the upper and lower encirclement relationship between the entity and the entity vector, and forming the external knowledge base;

[0022] Constructing a list of financial events through an external knowledge base specifically includes: searching for hypernyms of the subject, predicate, and object of financial events in the external knowledge base, and combining them into a generalized list of financial events;

[0023] The generalized financial events are obtained by generalizing the events through the large language model as follows: the generalized financial event list and the text-financial event pairs are input into the large language model, and the generalized financial events are obtained by extracting the templates through the generalized events.

[0024] Preferably, in the conditional mining module, the method for obtaining the local causal text sequence is: screening out the causal relationship from each generalized financial event local subgraph, and converting the financial event A-causal relationship-financial event B triple and background text information into the generalized financial event local causal text sequence;

[0025] The method for obtaining the causal quadruple is as follows: inputting the local causal text sequence into a large language model, mining the conditions behind the same cause but different results through a financial event causal condition mining template, inserting the conditions into a financial event A-relationship-financial event B triple to form a financial event A-condition-causal relationship-financial event B quadruple, and obtaining a causal quadruple;

[0026] The method for obtaining the sequential quadruple is: through the financial event sequential condition mining template, the conditions behind different sequential events are mined, and the financial event A-condition-sequential relationship-financial event B quadruple is constructed to obtain the sequential quadruple.

[0027] Preferably, in the graph retrieval module, event retrieval based on the extracted financial event is specifically as follows: based on the financial event, a financial event list formed by the most matching Top_K financial events is retrieved from the financial event graph to complete the event retrieval.

[0028] Preferably, link retrieval is specifically performed as follows: traversing the financial event list, for each financial event, along the causal relationship and the sequential relationship, finding a list of financial event reasoning links with a length not exceeding 3;

[0029] The financial event reasoning link list is traversed, and each financial event reasoning link is converted into a financial event reasoning text according to a financial event template to form a financial event reasoning text list.

[0030] Preferably, the financial event reasoning text list is traversed, the financial event reasoning text and the user question are input into the large language model together, and a financial event prediction list for the user question is formed under each financial event reasoning text through a financial event prediction template;

[0031] The financial event prediction list is input into a large language model, and a final event prediction for the user's question is generated through a financial event prediction comprehensive analysis template and returned to the user.

[0032] According to a second aspect of the present application, there is provided a financial event prediction method using the above-mentioned financial event prediction system, comprising the following steps:

[0033] Step 1: Obtain financial report data and preprocess the data;

[0034] Step 2: Extract financial events from financial report data through a large language model and construct a financial event dataset;

[0035] Step 3: Extract event relationships from the financial event dataset through a large language model and construct an original financial event graph;

[0036] Step 4: Vectorize each entity to build an external knowledge base, build a list of financial events through the external knowledge base, and generalize the events through the large language model to obtain generalized financial events;

[0037] Step 5: Taking the generalized financial event as the center, filter the triples containing the generalized financial event from the original financial event graph and merge them into a local subgraph of the generalized financial event containing background text information;

[0038] Step 6: Acquire local causal text sequence, causal quadruple, and sequential quadruple;

[0039] Step 7: Construct a financial event graph based on the causal quadruple, sequential quadruple, financial events, and condition-relationships;

[0040] Step 8: Extract financial events from the questions raised by users through the large language model and event extraction template;

[0041] Step 9: Perform event retrieval and link retrieval based on the extracted financial events to form a financial event prediction list, and generate the final event prediction for the user's question and return it to the user.

[0042] According to a third aspect of the present application, there is provided a server, comprising: a memory and at least one processor;

[0043] The memory stores a computer program, and the at least one processor executes the computer program stored in the memory to implement the above-mentioned financial event prediction method.

[0044] According to one embodiment of the present application, the beneficial effect of adopting the present financial event prediction system is that it automatically constructs a financial event graph through knowledge documents containing financial events, and provides investment references and decision-making basis for financial workers through logical reasoning of existing financial events and accurate prediction of future financial events, thereby solving the problem of weak logical reasoning ability encountered by large language models in the prior art in financial event reasoning. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the embodiments of the present application or the existing technical solutions, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0046] Figure 1 This is a structural block diagram of a financial event prediction system in one embodiment of the present application;

[0047] Figure 2 This is a flow chart of a financial event prediction method in one embodiment of the present application;

[0048] Figure 3 This is a structural block diagram of a server in an embodiment of the present application. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical solution and advantages of the present application clearer, the technical solution of the present application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application.

[0050] like Figure 1 As shown, the financial event prediction system in one embodiment of the present application includes a data processing module, an event extraction module, a relationship extraction module, an event generalization module, an event alignment module, a conditional mining module, a graph construction module, a user query module and a graph retrieval module, wherein:

[0051] The data processing module is used to obtain financial report data and pre-process the data. The pre-processing process includes: data conversion, text cleaning, and paragraph division;

[0052] The event extraction module extracts financial events from financial report data through a large language model and event extraction template, and constructs a financial event dataset containing context. Each piece of data in the financial event dataset contains a text-financial event pair, where the financial event is a subject-predicate-object structure.

[0053] The event extraction template includes prompt instructions, specific cases, output formats, etc., which are used to guide the large model to directly extract the subject-predicate-object structure of events from natural text. The extraction instructions are as follows:

[0054] {Extract subject-predicate-object triple events from the text content; the extracted events are objective and have a certain influence, comments, opinions, and daily behaviors of individuals cannot be called events; the extracted events are subject-predicate-object triples, and their format is (subject, predicate, object);

[0055] For example:

[0056] Task: Extract event information from text content.

[0057] Text content: The Social Sports Guidance Center of the State Sports General Administration has joined forces with many well-known athletes

[0058] Output: "think": ["step1: This text contains key predicate conjunctions, the subject is the Social Sports Guidance Center of the General Administration of Sports of China, and the object is a number of well-known athletes", "step2: End!"],

[0059] "output": [("Social Sports Guidance Center of the State Sports General Administration", "United", "Many well-known athletes")].}

[0060] The relation extraction module extracts event relations from the financial event data set through a large language model and event relation extraction template, and constructs an original financial event graph. Each piece of data in the event graph contains a triple of financial event A-relationship-financial event B, as well as the background text information of the triple. The relations include causal, sequential, and hierarchical relations.

[0061] The extraction instructions of the financial event relationship extraction template are as follows:

[0062] {Extract financial event relations from text content;

[0063] The extracted relations should not only include relational connectives, but also relational types, such as causal relations, sequential relations, and hierarchical relations;

[0064] For example, the task is to extract event relationship information from text content.

[0065] Output: [("Carbon Dioxide","Emissions",""),("Cause-Cause"),("Global Climate","Warming","")]}

[0066] The event generalization module is used to vectorize each entity to build an external knowledge base, build a list of financial events through the external knowledge base, and generalize events through a large language model to obtain generalized financial events. Specifically, it includes:

[0067] Financial event generalization - external knowledge base construction, using the "Da Ci Lin" to vectorize each entity, while retaining the hierarchical relationship between entities and entity vectors as the external knowledge base;

[0068] Financial event generalization - generalized financial event retrieval, searching for hypernyms in the subject, predicate, and object of financial events in the above external knowledge base, and combining them into a generalized financial event list;

[0069] Financial event generalization - generalized financial event generation: input the above generalized financial event list and text-financial event pairs into the large language model, extract templates through generalized events, and obtain generalized financial events;

[0070] Generalized events are a higher level abstraction of original events and can represent events in a wider range. The instructions for extracting financial generalized events are:

[0071] {You are a human assistant for key information extraction. Your task is to extract subject-verb-object triple events from the above text content.

[0072] The extracted events must reflect the content of the original events and cannot be fabricated.

[0073] For example: Input text: "Someone is setting off fireworks on the north side of Building 4 in Xingshengyuan Community"

[0074] Output: [(, set off, fireworks)].}

[0075] The event alignment module is used to filter the triples containing the generalized financial event from the original financial event graph, centering on the generalized financial event, and merge them into a local subgraph of the generalized financial event containing background text information;

[0076] The conditional mining module is used to obtain local causal text sequences, causal quadruple acquisition, and sequential quadruple acquisition; specifically:

[0077] Financial event conditional mining - local causal text sequence acquisition: From each generalized financial event local subgraph, filter out the causal relationship, and convert its financial event A-causal relationship-financial event B triples and background text information into a generalized financial event local causal text sequence;

[0078] Financial event condition mining - causal quadruple acquisition: The above generalized financial event local causal text sequence is input into the large language model, and the financial event causal condition mining template is used to mine the conditions behind the same cause but different results. The conditions are inserted into the financial event A-relationship-financial event B triple, and the financial event A-condition-causal relationship-financial event B quadruple is constructed to form a causal quadruple;

[0079] Financial event condition mining - sequential quadruple acquisition: Input the above generalized financial event local causal text sequence into the large language model, and use the financial event sequential condition mining template to mine the conditions behind different sequential events, insert the conditions into the financial event A-relationship-financial event B triple, and construct the financial event A-condition-sequential relationship-financial event B quadruple to form a sequential quadruple;

[0080] The graph construction module is used to construct a financial event graph based on the causal quadruple and the sequential quadruple, taking financial events as nodes and condition-relationships as composite edges;

[0081] User query module, used to extract financial events from questions raised by users through large language models and event extraction templates;

[0082] The graph retrieval module is used to perform event retrieval and link retrieval based on the extracted financial events to form a financial event prediction list, and generate the final event prediction for the user's question and return it to the user. Specifically:

[0083] The event retrieval based on the extracted financial events is specifically as follows: based on the financial events, a list of financial events formed by the most matching Top_K financial events is retrieved from the financial event graph to complete the event retrieval;

[0084] The most suitable method for retrieving is semantic similarity matching, that is, embedding financial events and event entities in the event graph into the vector semantic space, and finding the k financial events with the highest similarity through the semantic similarity matching algorithm.

[0085] The link retrieval is specifically as follows: traverse the list of financial events, and for each financial event, find a list of financial event reasoning links with a length not exceeding 3 along the causal relationship and sequential relationship;

[0086] Then, the financial event graph retrieval-link retrieval is performed, and the list of financial event reasoning links is traversed. According to the financial event template, each financial event reasoning link is converted into a financial event reasoning text to form a list of financial event reasoning texts.

[0087] Traverse the list of financial event reasoning texts, input the financial event reasoning texts and user questions into the large language model, and form a list of financial event predictions for user questions under each financial event reasoning text through the financial event prediction template;

[0088] The instructions for financial event prediction are:

[0089] {Based on the general financial event reasoning logic, predict the future development trend of existing financial events; the predicted financial event trend is a subject-predicate-object event triple, and its format is (subject, predicate, object);

[0090] Based on the applicability between existing financial events and general financial event reasoning logic, and your own logical judgment, predict the probability of the predicted financial event, divide it into three levels of "high, medium, and low", and give your own basis;

[0091] The elements in the financial event prediction list include predicted events, probabilities and bases, i.e. [{"Event Name": Financial Event A, "Occurrence Probability": High / Medium / Low, "Prediction Basis": Specific Prediction Basis A},{"Event Name": Financial Event B, "Occurrence Probability": High / Medium / Low, "Prediction Basis": Specific Prediction Basis B},...].}

[0092] The financial event prediction list is input into the large language model, and the final event prediction for the user's question is generated and returned to the user through the financial event prediction comprehensive analysis template.

[0093] The comprehensive analysis of financial event prediction is as follows: based on the above financial events and possible financial event prediction links, a reliable comprehensive analysis and prediction report of financial events is given. The comprehensive analysis report includes existing financial events, possible trends of financial events, the probability of such trends occurring, and the basis for prediction.

[0094] like Figure 2 As shown, in one embodiment of the present application, a financial event prediction method using the above financial event prediction system is disclosed, comprising the following steps:

[0095] Step 1: Data cleaning: Obtain financial report data and preprocess the data. The preprocessing process includes: data conversion, text cleaning, and paragraph division;

[0096] Step 2: Financial event extraction: extract financial events from financial report data through a large language model and build a financial event dataset containing context;

[0097] Step 3: Financial event relationship extraction: Extract event relationships from the financial event dataset through a large language model to construct an original financial event graph;

[0098] Step 4: Financial event generalization: vectorize each entity to build an external knowledge base, build a list of financial events through the external knowledge base, and generalize the events through the large language model to obtain generalized financial events;

[0099] Step 5: Financial event alignment: With the generalized financial event as the center, the triples containing the generalized financial event are selected from the original financial event graph and merged into a generalized financial event local subgraph containing background text information;

[0100] Step 6: Financial event condition mining: obtaining local causal text sequences, causal quadruple acquisition, and sequential quadruple acquisition;

[0101] Step 7: Construction of financial event graph: Constructing a financial event graph based on the causal quadruple, sequential quadruple, financial events and condition-relationships;

[0102] Step 8: User query event extraction: Extract financial events from user questions using a large language model and event extraction template;

[0103] Step 9: Financial event graph retrieval: event retrieval based on the extracted financial events: based on the extracted financial events, retrieve a financial event list consisting of the most matching Top_K financial events from the financial event graph;

[0104] And link retrieval: traverse the above financial event list, for each financial event, along the causal relationship and sequential relationship, find a list of financial event reasoning links with a length not exceeding 3;

[0105] Traverse the list of financial event reasoning links, and according to the financial event template, convert each financial event reasoning link into a financial event reasoning text to form a list of financial event reasoning texts. Through the financial event prediction template, form a list of financial event predictions for user questions under each financial event reasoning text; input the above financial event prediction list into the large language model, and generate the final event prediction for the user question through the financial event prediction comprehensive analysis template, and return it to the user.

[0106] like Figure 3 As shown, a server in an embodiment of the present application includes: a memory 301 and at least one processor 302;

[0107] The memory 301 stores computer programs, and the at least one processor 302 executes the computer programs stored in the memory 301 to implement the above-mentioned financial event prediction method.

[0108] According to the above embodiments, the beneficial effect of adopting the present financial event prediction system is that a financial event graph is automatically constructed through knowledge documents containing financial events, and through logical reasoning of existing financial events and accurate prediction of future financial events, investment references and decision-making basis are provided to financial workers, thereby solving the problem of weak logical reasoning ability of large language models in the prior art in financial event reasoning, and solving the problem that the existing triple knowledge representation method is difficult to accurately express the uncertain relationship between financial events, resulting in difficulty in constructing the event graph.

[0109] It should be noted that the above detailed descriptions are exemplary and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the art to which the present application belongs.

[0110] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should also be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0111] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the terms used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein.

[0112] In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units that are not explicitly listed or inherent to these processes, methods, products, or apparatuses.

[0113] For ease of description, spatially relative terms, such as "above", "above", "on the upper surface of", "above", etc., may be used herein to describe the spatial positional relationship between a device or feature and other devices or features as shown in the figure. It should be understood that spatially relative terms are intended to include different orientations of the device in use or operation in addition to the orientation described in the figure. For example, if the device in the accompanying drawings is inverted, the device described as "above other devices or structures" or "above other devices or structures" will be positioned as "below other devices or structures" or "below other devices or structures". Thus, the exemplary term "above" may include both "above" and "below". The device may also be positioned in other different ways, such as rotated 90 degrees or in other orientations, and the spatially relative descriptions used herein are interpreted accordingly.

[0114] In the above detailed description, reference is made to the accompanying drawings, which form a part of this document. In the accompanying drawings, similar symbols typically identify similar components unless the context indicates otherwise. The illustrated embodiments described in the detailed description, drawings, and claims are not meant to be limiting. Other embodiments may be used, and other changes may be made, without departing from the spirit or scope of the subject matter presented herein.

[0115] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A financial event prediction system, characterized in that: It includes data processing module, event extraction module, relationship extraction module, event generalization module, event alignment module, conditional mining module, graph construction module, user query module and graph retrieval module, among which: The data processing module is used to obtain financial report data and pre-process the data; The event extraction module is used to extract financial events from financial report data through a large language model and construct a financial event data set; in the event extraction module, financial events are extracted from financial report data through a large language model and an event extraction template to construct a financial event data set containing context, wherein each piece of data in the financial event data set contains a text-financial event pair, wherein the financial event is a subject-predicate-object structure; The relationship extraction module is used to extract event relationships from the financial event data set through a large language model to construct an original financial event graph; in the relationship extraction module, event relationships are extracted from the financial event data set through a large language model and an event relationship extraction template to construct an original financial event graph, wherein each piece of data in the original financial event graph contains a triple of financial event A-relationship-financial event B, and background text information of the triple, wherein the relationship includes causal, sequential, and hierarchical relationships; The event generalization module is used to vectorize each entity to build an external knowledge base, build a financial event list through the external knowledge base, and obtain generalized financial events through the large language model and the financial event generalization template; in the event generalization module, each entity is vectorized to build an external knowledge base by: using the "Big Dictionary" to vectorize each entity, retaining the upper and lower encirclement relationship between the entity and the entity vector, and forming an external knowledge base; Constructing a list of financial events through an external knowledge base specifically includes: searching for hypernyms of the subject, predicate, and object of financial events in the external knowledge base, and combining them into a generalized list of financial events; The generalized financial event is obtained by generalizing the event through the large language model. Specifically, the generalized financial event list and the text-financial event pair are input into the large language model, and the generalized financial event is obtained by extracting the template through the generalized event; The event alignment module is used to filter the triples containing the generalized financial event from the original financial event graph with the generalized financial event as the center, and merge them into a local subgraph of the generalized financial event containing background text information; The conditional mining module is used to obtain local causal text sequences, causal quadruplets and sequential quadruplets; in the conditional mining module, the method for obtaining the local causal text sequence is: screening out the causal relationship from each generalized financial event local subgraph, and converting the financial event A-causal relationship-financial event B triple and background text information into a generalized financial event local causal text sequence; The method for obtaining the causal quadruple is as follows: inputting the local causal text sequence into a large language model, mining the conditions behind the same cause but different results through a financial event causal condition mining template, inserting the conditions into a financial event A-relationship-financial event B triple to form a financial event A-condition-causal relationship-financial event B quadruple, and obtaining a causal quadruple; The method for obtaining the sequential quadruple is: through the financial event sequential condition mining template, the conditions behind different sequential events are mined, and the financial event A-condition-sequential relationship-financial event B quadruple is constructed to obtain the sequential quadruple; The graph construction module is used to construct a financial event graph based on the causal quadruple, sequential quadruple, financial events and condition-relationship; The user query module is used to extract financial events from questions raised by users through a large language model and event extraction template; The graph retrieval module is used to perform event retrieval and link retrieval based on the extracted financial events to form a financial event prediction list, and generate a final event prediction for the user's question and return it to the user; in the graph retrieval module, event retrieval based on the extracted financial events is specifically as follows: based on the financial events, a financial event list formed by the most matching Top_K financial events is retrieved from the financial event graph to complete the event retrieval; The link search is specifically performed as follows: traversing the financial event list, for each financial event, along the causal relationship and the sequential relationship, finding a list of financial event reasoning links with a length not exceeding 3; The financial event reasoning link list is traversed, and each financial event reasoning link is converted into a financial event reasoning text according to a financial event template to form a financial event reasoning text list.

2. The financial event prediction system according to claim 1, characterized in that: Traversing the financial event reasoning text list, inputting the financial event reasoning text and the user question into the large language model, and forming a financial event prediction list for the user question under each financial event reasoning text through a financial event prediction template; The financial event prediction list is input into a large language model, and a final event prediction for the user's question is generated through a financial event prediction comprehensive analysis template and returned to the user.

3. A financial event prediction method using the financial event prediction system according to claim 1 or 2, characterized in that: The steps include: Step 1: Obtain financial report data and preprocess the data; Step 2: Extract financial events from financial report data through a large language model and construct a financial event dataset; Step 3: Extract event relationships from the financial event dataset through a large language model and construct an original financial event graph; Step 4: Vectorize each entity to build an external knowledge base, build a list of financial events through the external knowledge base, and generalize the events through the large language model to obtain generalized financial events; Step 5: Taking the generalized financial event as the center, filter the triples containing the generalized financial event from the original financial event graph and merge them into a local subgraph of the generalized financial event containing background text information; Step 6: Acquire local causal text sequence, causal quadruple, and sequential quadruple; Step 7: Construct a financial event graph based on the causal quadruple, sequential quadruple, financial events, and condition-relationships; Step 8: Extract financial events from the questions raised by users through the large language model and event extraction template; Step 9: Perform event retrieval and link retrieval based on the extracted financial events to form a financial event prediction list, and generate the final event prediction for the user's question and return it to the user.

4. A server, characterized in that: include: memory and at least one processor; The memory stores a computer program, and the at least one processor executes the computer program stored in the memory to implement the financial event prediction method described in claim 3.

Images