Financial investment risk management method based on big data

Abstract

The invention provides a financial investment risk management method based on big data, and belongs to the field of big data. Firstly, a first environment analysis result and a first time interval are output through a first environment analysis model; a plurality of first successful enterprises are obtained through matching according to the first time interval; then matching to obtain a second successful enterprise and determining a first financial management behavior; and finally determining a second financial management behavior. According to the method, the financial influence information can be accurately extracted through semantic analysis and the emotion knowledge graph; by establishing the mapping relation between the excellent condition of the economic situation and the time interval and searching the financial management behavior, the successful management experience of the previous enterprise can be fully referred to; and finally, the rationality of the financial management behavior of the enterprise can be further improved through the matching degree analysis of the excellent condition of the economic situation and the financial management behavior.

Description

Technical Field

[0001] The present invention belongs to the field of big data, and in particular relates to a financial investment risk management method based on big data. Background Art

[0002] In the existing technology, the control of corporate financial investment risks often requires manual analysis and calculation, taking into account the company's assets, R&D direction and revenue. This method of consideration is difficult to fully consider all aspects, requires high labor costs, and the risk management decisions made are often not comprehensive and objective in considering the real-time financial environment.

[0003] With the development of big data technology, relevant existing technologies have explored the use of artificial intelligence to make financial decisions for enterprises, but they can often only predict decisions in a single dimension and cannot provide refined financial decision-making recommendations. Summary of the invention

[0004] In order to solve the above problems in the prior art, the present invention proposes a financial investment risk management method based on big data.

[0005] A financial investment risk management method based on big data, characterized in that the method comprises:

[0006] S1: Acquire first financial impact information, input a first environment analysis model, and output a first environment analysis result and a first time interval;

[0007] S2: According to the first time interval, a number of first successful enterprises are matched;

[0008] S3: obtaining first enterprise information of the enterprise to be managed currently, matching a second successful enterprise from the plurality of first successful enterprises according to the first enterprise information, and determining a first financial management behavior;

[0009] S4: inputting the first environmental analysis result and the first financial management behavior into a first rationality analysis model, thereby determining a second financial management behavior; the second financial management behavior is a financial management behavior recommended for the current enterprise to be managed;

[0010] The step S1 specifically includes the following sub-steps:

[0011] S11: Acquire first financial impact information; S11 specifically includes:

[0012] S111: Acquire first information through a web crawler, where keywords used in the crawling operation may be pre-set;

[0013] S112: performing sentiment analysis on the first information, analyzing the text content of the first information by using a pre-trained sentiment analysis model, and integrating the sentiment information in the first information to generate a comprehensive sentiment analysis result;

[0014] S113: obtaining second information through the comprehensive sentiment analysis result, wherein the second information is obtained by screening the first information; the screening criterion is to exclude information without sentiment information in the first information, so as to use the information set with sentiment information in the first information as the second information;

[0015] S114: Generate an emotional knowledge graph according to the content of the second information and the comprehensive emotional analysis result;

[0016] S12: inputting the first financial impact information into the first environmental analysis model, and outputting a first environmental analysis result and a first time interval;

[0017] The step S13 includes the following sub-steps:

[0018] S131: Based on the information provided by the enterprise to be managed, the first enterprise information is obtained; the first enterprise information is in the form of a three-dimensional vector, and the three-dimensional vectors are the industry, asset scale and enterprise scale respectively;

[0019] S132: respectively obtaining the three-dimensional vectors of the current enterprise to be managed and the plurality of first successful enterprises, respectively calculating the cosine similarity of the three-dimensional vectors of the current enterprise to be managed and each of the first successful enterprises, and selecting the first successful enterprises whose cosine similarity is greater than a preset value to obtain the plurality of second successful enterprises;

[0020] S133: Obtain the financial management behavior taken by the second successful enterprise within the first time interval through a lookup table as the first financial management behavior; the lookup table records the adjustment events of investment, management and R&D direction made by the second successful enterprise within the first time interval.

[0021] Preferably, step S4 comprises:

[0022] S41: taking each of the time intervals in step S12 as an analysis object, and classifying a number of the time intervals according to the degree of excellence of the economic situation;

[0023] S42: for each type of time interval, summarizing the financial management behaviors taken by the enterprise in the time interval, and each financial management behavior is represented in the form of a semantic vector;

[0024] S43: clustering the plurality of semantic vectors using a K-MEANS clustering algorithm to obtain a plurality of cluster centers;

[0025] S44: establishing a mapping relationship between the excellent type of the economic situation and the plurality of cluster centers;

[0026] S45: acquiring a preliminary financial management behavior according to the first environment analysis result and the mapping relationship;

[0027] The preliminary financial management behavior is a cluster center corresponding to the first environmental analysis result;

[0028] S46: Calculate the cosine similarity of the semantic expressions of the first financial management behavior and the preliminary financial management behavior, and when the cosine similarity is greater than a preset threshold, determine that the preliminary financial management behavior is the second financial management behavior.

[0029] Preferably, the step 114 includes:

[0030] Extract key entities, relationships and attributes from the second information to construct a primary knowledge graph; based on the primary knowledge graph, use a reinforcement learning strategy to convert unstructured information into structured data, including entity attributes and relationship internal logic, to generate a structured data set; based on the entity attributes and relationship internal logic, link the entities, events and emotional polarity to create emotional level connections and generate an emotional perception map.

[0031] Preferably, the step 114 further includes:

[0032] If there are multiple pieces of the second information, an emotion perception map is extracted for each piece of the second information, and several of the emotion perception maps are integrated to form an integrated first emotion perception map.

[0033] Preferably, the specific training method of the first environment analysis model is as follows:

[0034] The historical time intervals are divided according to the degree of excellence of the economic situation, and at least one financial landmark event is selected in each of the divided historical time intervals; a mapping table is established for the historical time intervals, financial landmark events and the degree of excellence of the economic situation; a convolutional neural network model is trained with the financial landmark events as input and the degree of excellence of the economic situation and the historical time intervals as output, thereby obtaining the first environmental analysis model.

[0035] The beneficial effects of the present invention include: a financial investment risk management method based on big data proposed in the present application. First, important domestic and international information is obtained, and the first financial impact information can be accurately extracted through sentiment analysis, and the first financial impact information is input into the first environmental analysis model, and the first environmental analysis result and the first time interval are output. Next, according to the first time interval, a number of first successful enterprises are matched, the first enterprise information of the current enterprise to be managed is obtained, and according to the first enterprise information, a second successful enterprise is matched among the several first successful enterprises, and the first financial management behavior is determined. Finally, the first environmental analysis result and the first financial management behavior are input into the first rationality analysis model to determine the second financial management behavior. The present invention can accurately extract financial impact information through semantic analysis and sentiment knowledge graph; by establishing a mapping relationship between the excellent situation of the economic situation and the time interval, and searching for financial management behavior, the successful management experience of previous enterprises can be fully learned; finally, through the matching degree analysis of the excellent situation of the economic situation and the financial management behavior, the rationality of the financial management behavior of the enterprise can be further increased. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present application, but do not constitute an improper limitation of the present invention. In the drawings:

[0037] Figure 1 It is an execution flow chart of a financial investment risk management method based on big data of the present invention.

[0038] Figure 2 It is an execution flow chart of obtaining the first environment analysis result and the first time interval in the present invention. DETAILED DESCRIPTION

[0039] The present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments, wherein the illustrative embodiments and descriptions are only used to explain the present invention but are not intended to limit the present invention.

[0040] The following is a detailed description of a financial investment risk management method based on big data of the present invention.

[0041] A financial investment risk management method based on big data, the specific process is as follows Figure 1 As shown, the following steps are included:

[0042] S1: Acquire first financial impact information, input a first environmental analysis model, and output a first environmental analysis result and a first time interval.

[0043] The step S1 is divided into the following sub-steps: Figure 2 As shown:

[0044] S11: Obtain the first financial impact information.

[0045] The first financial impact information includes information that has a relatively large impact on the financial situation, including both domestic information and international information.

[0046] Preferably, domestic information can be crawled through the following steps.

[0047] First, the web crawler crawls the financial sections of major portals and financial websites to obtain the first information. The keywords used in the crawling operation can be set in advance, including keywords related to currency, diplomacy, public health events, etc.

[0048] Secondly, sentiment analysis is performed on the first information, and the text content of the first information is analyzed by a pre-trained sentiment analysis model, and the sentiment information in the first information is integrated to generate a comprehensive sentiment analysis result. For example, for a financial report, sentiment information is extracted from the article, and the sentiment information may be keywords such as "upward", "downward", and "sanctions". After integrating the extracted sentiment keywords, a comprehensive sentiment analysis result can be obtained.

[0049] Thirdly, the second information is obtained through the comprehensive sentiment analysis result. The second information is obtained by screening the first information, and the specific screening criteria are to exclude the information without sentiment information in the first information, so as to use the information set with sentiment information in the first information as the second information.

[0050] Finally, an emotional knowledge graph is generated based on the content of the second information and the comprehensive emotional analysis results. Specifically, key entities, relationships and attributes are extracted from the second information to construct a primary knowledge graph; based on the primary knowledge graph, a reinforcement learning strategy is used to convert unstructured information into structured data, including entity attributes and relationship internal logic, to generate a structured data set; based on the entity attributes and relationship internal logic, the entities and events are linked to emotional polarity to create emotional level connections and generate an emotional perception graph.

[0051] Preferably, international information can be crawled through the following steps.

[0052] First, we screen out countries or regions that have a significant impact on the international financial situation. The screening rule focuses on GDP. Specifically, we sort all countries from high to low according to GDP, and select the top N countries as the first target countries.

[0053] Secondly, the web crawler crawls the financial sections of major portals and financial websites to obtain the first information. The keywords used in the crawling operation can be set in advance, including keywords related to the first target country, currency, diplomacy, public health events, etc.

[0054] Secondly, sentiment analysis is performed on the first information, and the text content of the first information is analyzed by a pre-trained sentiment analysis model, and the sentiment information in the first information is integrated to generate a comprehensive sentiment analysis result. For example, for a financial report, sentiment information is extracted from the article, and the sentiment information may be keywords such as "upward", "downward", and "sanctions". After integrating the extracted sentiment keywords, a comprehensive sentiment analysis result can be obtained.

[0055] Thirdly, the second information is obtained through the comprehensive sentiment analysis result. The second information is obtained by screening the first information, and the specific screening criteria are to exclude the information without sentiment information in the first information, so as to use the information set with sentiment information in the first information as the second information.

[0056] Finally, an emotional knowledge graph is generated based on the content of the second information and the comprehensive emotional analysis results. Specifically, key entities, relationships and attributes are extracted from the second information to construct a primary knowledge graph; based on the primary knowledge graph, a reinforcement learning strategy is used to convert unstructured information into structured data, including entity attributes and relationship internal logic, to generate a structured data set; based on the entity attributes and relationship internal logic, the entities and events are linked to emotional polarity to create emotional level connections and generate an emotional perception graph.

[0057] Preferably, the domestic information and the international information are integrated to obtain the first financial impact information. Specifically, for each piece of the second information in the domestic information and each piece of the second information in the international information, an emotional perception map is extracted respectively, and several of the emotional perception maps are integrated to form an integrated first emotional perception map. For example, if 2 pieces of domestic information and 2 pieces of international information are finally screened out, 4 emotional perception maps can be obtained. After integrating the 4 emotional perception maps, a first emotional perception map can be obtained. In the first emotional perception map, several entities, events and emotional polarities, as well as the connection relationship between the above nodes, can be included. The first financial impact information can be obtained through the first emotional perception map. Exemplarily, the first financial impact information is a binary vector, such as [Country A imposes trade sanctions on Country B, Industry C]. If the first financial impact information involves multiple pieces of information and multiple industries, the first financial impact information can be expanded to a set of multiple binary vectors.

[0058] Preferably, the sentiment analysis model adopts a deep neural network model, selects several financial news information as samples, selects sentiment keywords in the information by manual annotation, and gives sentiment analysis results. The financial news information is used as input, and the sentiment keywords and sentiment analysis results are used as output to train the deep neural network model, thereby obtaining the sentiment analysis model.

[0059] S12: Input the first financial impact information into the first environmental analysis model, and output a first environmental analysis result and a first time interval.

[0060] The first environment analysis result refers to the quality of the economic situation, which can be divided into five types: upward, stable, volatile, downward, and severe.

[0061] The first environment analysis model adopts a convolutional neural network model, which is obtained by training in the following way:

[0062] First, divide the historical time intervals according to the degree of excellence of the economic situation. Specifically, select the time point of the occurrence of the financial landmark event, take the time point of the occurrence of the financial landmark event as the center, select 5 years, 10 years or other time intervals as the time granularity, and thus obtain several of the historical time intervals. For the stable or volatile financial and economic situation, you can directly select the time interval without the occurrence of financial landmark events, and you can appropriately expand the time range of the time interval, and select a financial event at the middle time point of the time interval as the financial landmark event of the time interval. For several of the time intervals, manually mark the degree of excellence of the economic situation.

[0063] Secondly, a mapping table is established for the historical time period, the landmark financial events and the quality of the economic situation.

[0064] Finally, the convolutional neural network model is trained with the landmark financial event as input and the degree of excellence of the economic situation and the historical time interval as output, thereby obtaining the first environmental analysis model.

[0065] S2: According to the first time interval, a number of first successful enterprises are matched.

[0066] Within the first time interval, the top N enterprises in terms of assets and the top N enterprises in terms of asset growth rates are obtained as the first successful enterprises.

[0067] S3: Acquire the first enterprise information of the enterprise to be managed currently, match and obtain a second successful enterprise from the plurality of first successful enterprises according to the first enterprise information, and determine the first financial management behavior.

[0068] The current enterprise to be managed refers to an enterprise that needs financial investment risk management. The first enterprise information is obtained based on the information provided by the enterprise to be managed and the crawled information. The first enterprise information is in the form of a three-dimensional vector, and the three-dimensional vectors are the industry, asset scale and enterprise scale.

[0069] The three-dimensional vectors of the current enterprise to be managed and several of the first successful enterprises are obtained respectively. The cosine similarity of the three-dimensional vectors of the current enterprise to be managed and each of the first successful enterprises is calculated respectively, and the first successful enterprises whose cosine similarity is greater than a preset value are screened out to obtain several of the second successful enterprises.

[0070] Preferably, after determining a number of the second successful enterprises, it is also necessary to obtain the adjustment events of investment, management and R&D direction made by the second successful enterprises within the first time interval. A lookup table is established to store the main decision-making behaviors made by each of the second successful enterprises in the three dimensions of investment, management and R&D direction adjustment.

[0071] Furthermore, the financial management behavior taken by the second successful enterprise within the first time interval is obtained through the lookup table as the first financial management behavior.

[0072] S4: Inputting the first environmental analysis result and the first financial management behavior into a first rationality analysis model to determine a second financial management behavior.

[0073] Generally speaking, the financial management behaviors of enterprises need to be consistent with the overall economic situation, so that the financial risks faced by the enterprises will be relatively low.

[0074] Therefore, it is necessary to analyze the matching degree between the first environmental analysis result and the first financial management behavior through the first rationality analysis model.

[0075] The step S4 includes the following sub-steps:

[0076] S41: taking each of the time intervals in step S12 as an analysis object, and classifying a number of the time intervals according to the quality of the economic situation.

[0077] It can be divided into five types: upward, stable, oscillating, downward and severe.

[0078] S42: For each type of time interval, the financial management behaviors taken by the enterprise within the time interval are summarized, and each financial management behavior is represented in the form of a semantic vector.

[0079] S43: Clustering the plurality of semantic vectors using a K-MEANS clustering algorithm to obtain a plurality of cluster centers.

[0080] S44: Establishing a mapping relationship between the excellent types of the economic situation and the plurality of cluster centers.

[0081] S45: Acquire a preliminary financial management behavior according to the first environment analysis result and the mapping relationship.

[0082] The preliminary financial management behavior is a cluster center corresponding to the first environmental analysis result.

[0083] S46: Calculate the cosine similarity of the semantic expressions of the first financial management behavior and the preliminary financial management behavior, and when the cosine similarity is greater than a preset threshold, determine that the preliminary financial management behavior is the second financial management behavior.

[0084] A financial investment risk management method based on big data is proposed in this application. First, important domestic and international information is obtained, and the first financial impact information can be accurately extracted through sentiment analysis. The first financial impact information is input into the first environmental analysis model, and the first environmental analysis result and the first time interval are output. Next, according to the first time interval, a number of first successful enterprises are matched, the first enterprise information of the current enterprise to be managed is obtained, and according to the first enterprise information, a second successful enterprise is matched among the several first successful enterprises, and the first financial management behavior is determined. Finally, the first environmental analysis result and the first financial management behavior are input into the first rationality analysis model to determine the second financial management behavior. The present invention can accurately extract financial impact information through semantic analysis and sentiment knowledge graph; by establishing a mapping relationship between the excellent situation of the economic situation and the time interval, and searching for financial management behavior, the successful management experience of previous enterprises can be fully learned; finally, through the matching degree analysis of the excellent situation of the economic situation and the financial management behavior, the rationality of the financial management behavior of the enterprise can be further increased.

[0085] The above description is only a preferred embodiment of the present invention, so all equivalent changes or modifications made according to the structure, characteristics and principles described in the scope of the patent application of the present invention are included in the scope of the patent application of the present invention.

Claims

1. A financial investment risk management method based on big data, characterized in that: The method includes: S1: Acquire first financial impact information, input a first environment analysis model, and output a first environment analysis result and a first time interval; S2: According to the first time interval, a number of first successful enterprises are matched; S3: obtaining first enterprise information of the enterprise to be managed currently, matching a second successful enterprise from the plurality of first successful enterprises according to the first enterprise information, and determining a first financial management behavior; S4: inputting the first environmental analysis result and the first financial management behavior into a first rationality analysis model, thereby determining a second financial management behavior; the second financial management behavior is a financial management behavior recommended for the current enterprise to be managed; The step S1 specifically includes the following sub-steps: S11: Acquire first financial impact information; S11 specifically includes: S111: Acquire first information through a web crawler, where keywords used in the crawling operation may be pre-set; S112: performing sentiment analysis on the first information, analyzing the text content of the first information by using a pre-trained sentiment analysis model, and integrating the sentiment information in the first information to generate a comprehensive sentiment analysis result; S113: obtaining second information through the comprehensive sentiment analysis result, wherein the second information is obtained by screening the first information; the screening criterion is to exclude information without sentiment information in the first information, so as to use the information set with sentiment information in the first information as the second information; S114: generating an emotional knowledge graph according to the content of the second information and the comprehensive emotional analysis result; S12: inputting the first financial impact information into the first environmental analysis model, and outputting the first environmental analysis result and the first time interval; The step S13 includes the following sub-steps: S131: Based on the information provided by the enterprise to be managed, the first enterprise information is obtained; the first enterprise information is in the form of a three-dimensional vector, and the three-dimensional vectors are the industry, asset scale and enterprise scale respectively; S132: respectively obtaining the three-dimensional vectors of the current enterprise to be managed and the plurality of first successful enterprises, respectively calculating the cosine similarity of the three-dimensional vectors of the current enterprise to be managed and each of the first successful enterprises, and selecting the first successful enterprises whose cosine similarity is greater than a preset value to obtain the plurality of second successful enterprises; S133: Obtain the financial management behavior taken by the second successful enterprise within the first time interval through a lookup table as the first financial management behavior; the lookup table records the adjustment events of investment, management and R&D direction made by the second successful enterprise within the first time interval.

2. The financial investment risk management method based on big data according to claim 2 is characterized in that: The step S4 comprises: S41: taking each of the time intervals in step S12 as an analysis object, and classifying a number of the time intervals according to the degree of excellence of the economic situation; S42: for each type of time interval, summarizing the financial management behaviors taken by the enterprise in the time interval, and each financial management behavior is represented in the form of a semantic vector; S43: clustering the plurality of semantic vectors using a K-MEANS clustering algorithm to obtain a plurality of cluster centers; S44: establishing a mapping relationship between the excellent type of the economic situation and the plurality of cluster centers; S45: acquiring a preliminary financial management behavior according to the first environment analysis result and the mapping relationship; the preliminary financial management behavior is a cluster center corresponding to the first environment analysis result; S46: Calculate the cosine similarity of the semantic expressions of the first financial management behavior and the preliminary financial management behavior, and when the cosine similarity is greater than a preset threshold, determine that the preliminary financial management behavior is the second financial management behavior.

3. The financial investment risk management method based on big data according to claim 1 is characterized in that: The step 114 includes: Extract key entities, relationships and attributes from the second information to construct a primary knowledge graph; based on the primary knowledge graph, use a reinforcement learning strategy to convert unstructured information into structured data, including entity attributes and relationship internal logic, to generate a structured data set; based on the entity attributes and relationship internal logic, link the entities, events and emotional polarity to create emotional level connections and generate an emotional perception map.

4. The financial investment risk management method based on big data according to claim 3 is characterized in that: The step 114 further includes: If there are multiple pieces of the second information, an emotion perception map is extracted for each piece of the second information, and several of the emotion perception maps are integrated to form an integrated first emotion perception map.

5. The financial investment risk management method based on big data according to claim 1 is characterized in that: The specific training method of the first environment analysis model is as follows: The historical time intervals are divided according to the degree of excellence of the economic situation, and at least one financial landmark event is selected in each of the divided historical time intervals; a mapping table is established for the historical time intervals, financial landmark events and the degree of excellence of the economic situation; a convolutional neural network model is trained with the financial landmark events as input and the degree of excellence of the economic situation and the historical time intervals as output, thereby obtaining the first environmental analysis model.

Images