Financial analysis report generation method and related device

Abstract

The embodiment of the invention provides a financial analysis report generation method and a related device, and relates to the technical field of big data, and the method comprises the steps: obtaining multi-modal financial data of a target object; performing feature extraction on structured data and unstructured data in the financial data to obtain a feature vector set; the feature vector set comprises numerical features, text features and graph features; fusing the numerical value features, the text features and the graph features, and determining a financial evaluation result according to a fused context vector; obtaining contribution degrees of the numerical value features, the text features and the graph features to the financial evaluation result, and generating a decision path report according to the contribution degrees; and generating the financial analysis report according to the financial evaluation result and the decision path report. According to the method, the efficiency and accuracy of financial analysis report generation can be improved.

Description

Technical Field

[0001] The present application relates to the field of big data, and in particular to a method and related device for generating a financial analysis report. Background Art

[0002] Financial analysis reports reflect a company's financial status and operating results over a specific period of time and serve as a crucial support document for banks' credit assessments. Data sources for financial analysis reports include multimodal data such as tables, text, images, audio, and video.

[0003] Currently, generating financial analysis reports requires processing data from different modalities separately to generate multiple analysis reports, which are then combined. This method of generating financial analysis reports is inefficient and inaccurate. Summary of the Invention

[0004] The present invention provides a method and apparatus for generating a financial analysis report to improve the efficiency and accuracy of the financial analysis report.

[0005] In a first aspect, an embodiment of the present application provides a method for generating a financial analysis report, comprising:

[0006] Obtain multimodal financial data of the target object;

[0007] Extracting features from structured data and unstructured data in the financial data to obtain a feature vector set; the feature vector set includes numerical features, text features, and graph features;

[0008] fusing the numerical features, the text features, and the graph features, and determining a financial evaluation result based on the fused context vector;

[0009] Obtaining the contribution of the numerical feature, the text feature, and the graph feature to the financial evaluation result, and generating a decision path report based on the contribution;

[0010] The financial analysis report is generated based on the financial evaluation results and the decision path report.

[0011] In some embodiments, extracting features from the structured data and unstructured data in the financial data to obtain a set of feature vectors includes:

[0012] Performing index extraction and time series feature extraction on the structured data to obtain the numerical features;

[0013] Extracting semantic vectors from the unstructured data, and calculating key information density of key data in the unstructured data to obtain the text features;

[0014] Entity data corresponding to the target object is extracted from the structured data and the unstructured data, a graph network of the target object is constructed based on the entity data, and path analysis is performed on the graph network to obtain the graph features.

[0015] In some embodiments, the fusing of the numerical features, the text features, and the graph features includes:

[0016] Processing the graph features to obtain continuous numerical graph vectors;

[0017] Processing the text features to obtain a text semantic representation vector;

[0018] The digitized graph vector, the text semantic representation vector and the numerical feature are fused.

[0019] In some embodiments, fusing the digitized graph vector, the text semantic representation vector, and the numerical feature includes:

[0020] Using the text semantic representation vector, the numerical feature and the numerical graph vector as the key and value of a predefined vector;

[0021] Obtaining the similarity between the predefined vector and each of the keys to obtain an attention score;

[0022] Normalizing the attention scores to obtain the weights corresponding to the keys;

[0023] The vectors corresponding to the values ​​are weightedly summed according to the weights to obtain the context vector.

[0024] In some embodiments, determining a financial evaluation result based on the fused context vector includes:

[0025] Performing weighted summation and nonlinear activation processing on the context vector to obtain a hidden feature vector result;

[0026] The hidden feature vector results are weighted summed and linearly activated to obtain the financial evaluation result.

[0027] In some embodiments, obtaining the contribution of the numerical feature, the text feature, and the graphical feature to the financial evaluation result includes:

[0028] respectively obtaining the contribution of each sub-feature of the numerical feature, the text feature, and the graph feature to the financial evaluation result;

[0029] The contribution degrees corresponding to each sub-feature of the numerical feature, the text feature, and the graph feature are aggregated to obtain the contribution degrees corresponding to each of the numerical feature, the text feature, and the graph feature.

[0030] In some embodiments, generating a decision path report based on the contribution includes:

[0031] Constructing three modal total nodes according to the respective contributions of the numerical feature, the text feature, and the graph feature, and constructing subnodes for the three modal total nodes according to the respective corresponding subfeatures and their corresponding contributions; the greater the contribution, the larger the total node;

[0032] The sub-nodes having the association relationship are connected to generate the decision path report.

[0033] In some embodiments, generating the financial analysis report based on the financial evaluation results and the decision path report includes:

[0034] Get the placeholder information in the financial analysis report template;

[0035] combining the placeholder information, the financial assessment results, the decision path report, and target data extracted from the structured data and the unstructured data;

[0036] The combined information is subjected to natural language processing to fill in the financial analysis report template and generate the financial analysis report.

[0037] In a second aspect, an embodiment of the present application provides a financial analysis report generating device, comprising:

[0038] An acquisition module, used to obtain multimodal financial data of the target object;

[0039] An extraction module, configured to extract features from the structured data and unstructured data in the financial data to obtain a feature vector set; the feature vector set includes numerical features, text features, and graph features;

[0040] A fusion module, configured to fuse the numerical features, the text features, and the graph features, and determine a financial evaluation result based on the fused context vector;

[0041] a processing module, configured to obtain the contribution of the numerical feature, the text feature, and the graph feature to the financial evaluation result, and generate a decision path report based on the contribution;

[0042] A generating module is used to generate the financial analysis report based on the financial evaluation result and the decision path report.

[0043] In a third aspect, the present application provides an electronic device, comprising: a processor, a transceiver, and a memory; the processor is communicatively connected to the transceiver and the memory respectively;

[0044] The memory is used to store computer programs; the transceiver is used to communicate and interact with external devices; and the processor is used to execute computer instructions stored in the memory to implement any method in the first aspect.

[0045] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, and the computer program is executed by a processor to implement any one of the methods in the first aspect.

[0046] In a fifth aspect, the present application provides a computer program product, comprising a computer program, which implements any one of the methods in the first aspect when executed by a processor.

[0047] The financial analysis report generation method and related devices provided in the embodiments of the present application can efficiently process multi-source heterogeneous data, deeply mine the multi-dimensional signals of the target object, and provide high-precision evaluation results, thereby improving the efficiency and accuracy of financial analysis report generation. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 A schematic diagram of a method for generating a financial analysis report provided in an embodiment of the present application Figure 1 ;

[0049] Figure 2 A schematic diagram of a method for generating a financial analysis report provided in an embodiment of the present application Figure 2 ;

[0050] Figure 3 A schematic diagram of the structure of a financial analysis report generating device provided in an embodiment of the present application;

[0051] Figure 4 A schematic diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0052] To make the purpose, technical solutions, and advantages of the embodiments of this application more clear, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0053] In the embodiments of the present application, words such as "first" and "second" are used to distinguish between identical or similar items with substantially the same functions and effects, and do not limit their order. Those skilled in the art will understand that words such as "first" and "second" do not limit the quantity or execution order, and words such as "first" and "second" do not necessarily mean different.

[0054] It should be noted that in the embodiments of this application, words such as "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described in this application as "exemplary" or "for example" should not be interpreted as being preferred or advantageous over other embodiments or designs. Rather, the use of words such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0055] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data comply with the relevant laws, regulations and standards of the relevant countries and regions, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0056] In addition, this application involves conducting big data analysis of user information (including but not limited to personal biometrics, identity data, consumption data, asset data, electronic terminal operation data, etc.), and using artificial intelligence technology to make automated decisions, and making technical solutions that have a significant impact on personal rights and interests based on the results of automated decisions. The application provides users with corresponding operation entrances for them to choose to agree or reject the results of automated decisions; if the user chooses to reject, the expert decision-making process will be entered.

[0057] It should be noted that the financial analysis report generation method and related devices provided in this application can be used in the field of big data, and can also be used in any field other than big data. This application does not limit the application field of the financial analysis report generation method and related devices.

[0058] As mentioned above, the current method of generating financial reports is inefficient and inaccurate. Therefore, the embodiments of the present application provide a financial report generation method and related devices to improve the efficiency and accuracy of financial report generation.

[0059] Figure 1The flowchart of a financial report generation method provided by the embodiment of the present application is as follows. The execution subject of the embodiment of the present application can be a computing platform, server, etc. with specific data analysis and processing capabilities. Taking the computing platform as an example, Figure 1 As shown, the method includes:

[0060] S101. Acquire multimodal financial data of a target object.

[0061] In some embodiments, the target object may be an object that requires financial analysis (ie, a financial report for the target object needs to be generated), for example, a certain enterprise.

[0062] In some embodiments, the multimodal financial data may include financial data in the form of pictures, tables, text, video, audio, and other modalities.

[0063] The computing platform can obtain the multimodal financial data from the outside through a data interface.

[0064] S102 : Extract features from the structured data and unstructured data in the financial data to obtain a feature vector set; the feature vector set includes numerical features, text features, and graph features.

[0065] In some embodiments, data extraction may be performed directly on the tables in the multimodal financial data (for example, the tables may be processed using a document analysis model (such as PP-StructureV2)) to obtain structured data.

[0066] Unstructured financial data in the multimodal financial data, such as images, videos, and audio, can be analyzed and processed to convert it into unstructured data. For example, optical character recognition (OCR) can be used to identify text within images and obtain the corresponding text. For video and audio data, speech recognition and computer vision technologies can be used to convert them into text and obtain the corresponding text.

[0067] In some embodiments, after obtaining the structured data and unstructured data, feature extraction may be performed on the structured data and the unstructured data respectively to obtain a feature vector set.

[0068] Exemplarily, the structured data is subjected to indicator extraction and time series feature extraction to obtain the numerical features; the unstructured data is subjected to semantic vector extraction, and the target data in the unstructured data is subjected to key information density calculation to obtain the text features; the entity data corresponding to the target object is extracted from the structured data and the unstructured data, a graph network of the target object is constructed based on the entity data, and the graph network is subjected to path analysis to obtain the graph features.

[0069] In some embodiments, the constructed financial calculation engine can be used to extract key indicators in the table, and corresponding time series features can be extracted based on the time series data in the table.

[0070] Key indicators include liquidity indicators (current ratio, quick ratio), debt-paying ability indicators (asset-liability ratio, interest coverage ratio), profitability indicators (sales profit margin, return on equity (ROE), and operating ability indicators (accounts receivable turnover, inventory turnover). Time series features may include the company's operating income growth rate over the past six months, a year, or other period, as well as the standard deviation of each growth rate.

[0071] In some embodiments, a financial text sentiment analysis model (such as FinBERT) can be used to extract semantic vectors from each of the above texts. For example, FinBERT is used to convert the text into a high-dimensional vector of a fixed length (usually 768 dimensions).

[0072] Key data can be important objects or target scenarios mentioned in a text, such as equipment, factories, cross-border transactions, and cryptocurrencies. The number of times the target data appears in a text can be counted and divided by the total length of the text to obtain the key information density of the key data. This key information density reflects the prominence of the key data in the text.

[0073] In some embodiments, the entity data may refer to an object associated with the target object. For example, if the target object is an enterprise, the corresponding entity data may refer to all enterprises associated with the enterprise.

[0074] Entity data corresponding to the target object can be extracted from the structured data and unstructured data, and a graph network of the target object can be constructed with the entity data as nodes and the relationships between the nodes (such as supply chain relationships, investment relationships, guarantee relationships, etc.) as edges.

[0075] After constructing the graph network, we can use analysis methods such as large-scale interconnected subgraphs and graph indicators to perform path analysis on the graph network to obtain a graph embedding vector (i.e., graph feature) that represents the structural characteristics of the relational network where the target object is located.

[0076] S103: Fusing the numerical features, the text features, and the graph features, and determining a financial evaluation result based on the fused context vector.

[0077] In some embodiments, a multimodal fusion network may be used to fuse the numerical features, the text features, and the graph features to obtain a context vector. A pre-trained machine learning model is then used to jointly learn and predict the context vector to obtain a financial assessment result for the target object.

[0078] S104: Obtain the contribution of the numerical feature, the text feature, and the graph feature to the financial evaluation result, and generate a decision path report according to the contribution.

[0079] In some embodiments, the decision path report is used to characterize the key basis for the machine learning model to generate the financial assessment results. For example, it can identify which types of features (e.g., current ratio, negative public opinion keywords, stability of key suppliers) have the greatest impact on the model's predictions; which key factors (e.g., low quick ratio, vague collateral description, recent negative news about a related party) significantly lower the credit score; and which factors (e.g., stable revenue growth over the past three years, a clear debt repayment plan as shown in the annual report, and strong core customer base) support a strong rating.

[0080] In some embodiments, the contribution of each feature to the financial evaluation result can be calculated based on game theory; and the decision path report can be generated using a visualization tool.

[0081] S105: Generate the financial analysis report based on the financial evaluation results and the decision path report.

[0082] In some embodiments, the financial evaluation results and relevant information in the decision path report may be filled into a predefined financial analysis report template to output the financial analysis report.

[0083] The financial analysis report generation method provided in the embodiment of the present application obtains multimodal financial data of the target object; extracts features of the structured data and unstructured data in the financial data to obtain a feature vector set; the feature vector set includes numerical features, text features and graph features; fuses the numerical features, the text features and the graph features, and determines the financial evaluation result based on the fused context vector; obtains the contribution of the numerical features, the text features and the graph features to the financial evaluation result, and generates a decision path report based on the contribution; generates the financial analysis report based on the financial evaluation result and the decision path report. The above method can efficiently process multi-source heterogeneous data, deeply mine the multi-dimensional signals of the target object, and provide high-precision evaluation results, thereby improving the efficiency and accuracy of financial analysis report generation.

[0084] Based on the above embodiments, Figure 2 The financial analysis report generation method provided in the embodiment of the present application is further explained.

[0085] Figure 2 A flowchart of a method for generating a financial analysis report provided in an embodiment of the present application is shown in FIG. Figure 2 As shown, including:

[0086] S201: Acquire multimodal financial data of a target object.

[0087] S202: Extract features from the structured data and unstructured data in the financial data to obtain a set of feature vectors.

[0088] The specific implementation of S201-S202 in this application is the same as Figure 1 The specific implementation methods shown in S101-S102 in the illustrated embodiment are similar and will not be repeated here.

[0089] S203: Process the graph features to obtain continuous numerical graph vectors, and process the text features to obtain text semantic representation vectors.

[0090] In some embodiments, a graph neural network (GNN) can be used to process the graph features to obtain continuous numerical graph vectors. The core idea of ​​the graph neural network is message passing. After the graph features are processed by GNN, each node will aggregate the information of its neighboring nodes (directly connected nodes) and combine its own information to update its own state (i.e., generate a new vector representation). This process is iterated (usually several layers) so that the final representation of each node contains not only its own information, but also the information of its multi-hop neighbors (indirectly related parties) and the information of the entire local network structure. In the continuous numerical graph vector, each node of the graph network included in the graph feature will be encoded into a fixed-length, low-dimensional, dense real number vector (for example, 64 dimensions, 128 dimensions), thereby obtaining a continuous numerical graph vector.

[0091] In some embodiments, a deep learning network (such as a Transformer) can be used to process the text features to generate a text semantic representation vector. The core of the Transformer is the self-attention mechanism, which allows each element in a sequence to dynamically aggregate information based on the importance (weight) of all other elements. Since the text features are already semantic vectors output by FinBERT, the Transformer can further explore and refine more complex semantic associations and dependencies within the text. The Transformer (typically using an encoder structure similar to BERT) performs deep nonlinear transformations and refinements on the input semantic vector. The final output is still a fixed-length, high-dimensional dense real number vector (for example, maintaining 768 dimensions or through pooling / compression to other dimensions).

[0092] S204: fusing the digitized graph vector, the text semantic representation vector, and the numerical feature.

[0093] In some embodiments, the digitized graph vector, the text semantic representation vector, and the numerical feature may be fused in a weighted fusion manner.

[0094] Exemplarily, the text semantic representation vector, the numerical features and the digitized graph vector can be used as the keys and values ​​of a predefined vector; the similarity between the predefined vector and each of the keys is obtained to obtain an attention score; each of the attention scores is normalized to obtain a weight corresponding to each key; and the vectors corresponding to each of the values ​​are weightedly summed according to the weight to obtain the context vector.

[0095] For example, if the predefined vector includes three keys and three values, a dot product or a small neural network can be used to obtain the similarity between the predefined vector and each key to obtain an attention score. The attention score is then softmax-normalized to obtain a weight corresponding to each key. This weight is then used to perform a weighted summation with the corresponding value to obtain the context vector.

[0096] Optionally, the fusion of the digitized graph vector, the text semantic representation vector, and the numerical feature may be performed by bilinear fusion, gating mechanism, multimodal transformer, etc. This embodiment of the present application does not limit this.

[0097] S205 : Determine a financial evaluation result based on the fused context vector.

[0098] Exemplarily, the context vector is subjected to weighted summation and nonlinear activation processing to obtain a hidden feature vector result; the hidden feature vector result is subjected to weighted summation and linear activation processing to obtain the financial evaluation result.

[0099] For example, the above processing process can be implemented using a fully connected neural network, which includes at least one hidden layer and an output layer, the hidden layer includes at least one neuron and a nonlinear activation function; the output layer includes one neuron and a linear activation function.

[0100] The neurons in the hidden layer perform weighted summation on the context vector, and a nonlinear activation function (such as ReLU, GELU) extracts and combines the results of the weighted summation to obtain a hidden feature vector result.

[0101] The neurons in the output layer perform weighted summation on the hidden feature vector results, and a linear activation function (such as Sigmoid) extracts and combines the weighted summation results to output the financial evaluation results.

[0102] In some embodiments, if multiple hidden layers are included, the output of the previous hidden layer becomes the input of the next hidden layer.

[0103] Optionally, the financial assessment result may include a risk level and a credit score, and the output result of the output layer may be only a credit score, and the credit score is mapped to the corresponding risk level through a preset mapping relationship.

[0104] Optionally, the financial assessment results can include both the risk level and the credit score. The output layer can directly output both the credit score and the risk level. For example, the output layer can have two branches: one branch (1 neuron + Sigmoid) outputs the credit score, and the other branch (1 neuron + Softmax) outputs the risk level.

[0105] S206: Obtain the contribution of the numerical feature, the text feature, and the graph feature to the financial evaluation result.

[0106] In some embodiments, the contribution of the numerical feature, the text feature, and the graph feature to the financial evaluation result may be calculated based on game theory (Shapley value).

[0107] Exemplarily, the contribution of each sub-feature of the numerical feature, the text feature and the graphical feature to the financial evaluation result is obtained respectively; the contribution corresponding to each sub-feature of the numerical feature, the text feature and the graphical feature is aggregated respectively to obtain the contribution corresponding to each of the numerical feature, the text feature and the graphical feature.

[0108] For example, for numerical features, the SHAP value of each sub-feature is calculated for the structured financial indicators (current ratio, debt-to-asset ratio, etc.) and time series statistics (standard deviation of revenue growth rate, etc.), and the contribution of the numerical feature to the financial evaluation result is determined based on TreeSHAP or KernelSHAP.

[0109] For text features, a hierarchical attribution method can be used to calculate contributions. For example, for the semantic vector (768 dimensions) output by FinBERT, the integral gradient algorithm is used to calculate the contribution value of each dimension. The top-K words with the highest FinBERT attention weights are extracted, and the dimension contribution values ​​corresponding to each word are aggregated to obtain the corresponding contribution of the semantic vector. For key data (keyword) density, SHAP calculations can be performed as independent features.

[0110] For graph features, the GNN-SHAP coupled algorithm can be used to calculate contribution. For example, the input data includes the target object GNN embedding vector A1, the top-three related party embedding vectors A2, A3, and A4 associated with the target object, and the size of the largest connected subgraph. The node-level contribution and the target object's own contribution are calculated separately, and the sum of these two contributions is used to obtain the corresponding contribution of the graph feature.

[0111] S207: Generate a decision path report based on the contribution.

[0112] In some embodiments, the decision path report can convert the calculated SHAP values ​​into intuitive and actionable visual charts and textual analysis to explain why the model made a specific assessment.

[0113] Exemplarily, three modal total nodes are constructed according to the corresponding contribution degrees of the numerical features, the text features and the graph features, and sub-nodes are constructed for the three modal total nodes according to the corresponding sub-features and their corresponding contribution degrees; the greater the contribution degree, the larger the total node; the sub-nodes with associated relationships are connected to generate the decision path report.

[0114] For example, create three modal master nodes. For example, numerical features (blue), text features (orange), and graph features (red). Add key sub-feature nodes for each modality. For example, for numerical features, add the financial indicator with the highest contribution (such as the quick ratio); for text features, add key risk terms (such as inventory backlog); and for graph features, add important related parties (such as Guarantor X). Arrange the three modal master nodes horizontally, and arrange their sub-feature nodes vertically below each modal node. Scale the node size by the absolute value of the contribution, and use different lines to connect modalities to sub-features, as well as sub-features across modalities. Display the total contribution value on the modal node, the feature value and / or contribution value on the sub-feature node, and annotate the association relationship on the connecting line to obtain the decision path report.

[0115] S208: Generate the financial analysis report based on the financial evaluation results and the decision path report.

[0116] In some embodiments, the financial evaluation results and the decision path report may be filled into a preset financial analysis report template to generate the financial analysis report.

[0117] Exemplarily, placeholder information in a financial analysis report template is obtained; the placeholder information shown, the financial assessment results, the decision path report, and the target data extracted from the structured data and the unstructured data are combined; natural language processing is performed on the combined information to fill in the financial analysis report template and generate the financial analysis report.

[0118] For example, the target data extracted from the structured data and unstructured data includes calculated core financial ratio values, key information summaries extracted from text, key conclusions from graph analysis, etc. The placeholder information in the financial analysis report template can indicate the location of the content that needs to be filled in the financial analysis report template.

[0119] The placeholder information in the template, the financial assessment results, the decision path report, the target data, and the report filling instructions are input into the large language model (LLM). The large language model can accurately fill in the placeholder information in the template and generate professional analysis text and corresponding suggestions based on the financial assessment results, the decision path report and the target data to obtain the financial analysis report.

[0120] In summary, the financial analysis report generation method provided in the embodiment of the present application, by making full use of technologies such as OCR, NLP (BERT / FinBERT), graph analysis, multimodal fusion AI model (GNN+Transformer), explainable AI (SHAP) and large language model, efficiently processes multi-source heterogeneous data, deeply mines multi-dimensional signals of corporate credit risk, provides high-precision evaluation results, and greatly improves the efficiency and accuracy of financial analysis report generation through transparent interpretation and professional report generation.

[0121] Based on the above embodiments, the embodiments of the present application also provide a financial analysis report generating device.

[0122] Figure 3 A schematic diagram of the structure of a financial analysis report generating device 30 provided in an embodiment of the present application is shown as follows: Figure 3 As shown, including:

[0123] The acquisition module 301 is used to acquire multimodal financial data of a target object.

[0124] The extraction module 302 is used to extract features from the structured data and unstructured data in the financial data to obtain a feature vector set; the feature vector set includes numerical features, text features, and graph features.

[0125] The fusion module 303 is configured to fuse the numerical features, the text features, and the graph features, and determine a financial evaluation result based on the fused context vector.

[0126] The processing module 304 is configured to obtain the contribution of the numerical feature, the text feature, and the graph feature to the financial evaluation result, and generate a decision path report according to the contribution.

[0127] The generating module 305 is configured to generate the financial analysis report according to the financial evaluation result and the decision path report.

[0128] In some embodiments, the extraction module 302 is used to perform indicator extraction and time series feature extraction on the structured data to obtain the numerical features; perform semantic vector extraction on the unstructured data, and perform key information density calculation on the key data in the unstructured data to obtain the text features; extract entity data corresponding to the target object from the structured data and the unstructured data, construct a graph network of the target object based on the entity data, and perform path analysis on the graph network to obtain the graph features.

[0129] In some embodiments, the fusion module 303 is used to process the graph features to obtain continuous numerical graph vectors; process the text features to obtain text semantic representation vectors; and fuse the numerical graph vectors, the text semantic representation vectors and the numerical features.

[0130] In some embodiments, the fusion module 303 is used to use the text semantic representation vector, the numerical features and the digitized graph vector as the keys and values ​​of a predefined vector; obtain the similarity between the predefined vector and each of the keys to obtain an attention score; normalize each of the attention scores to obtain a weight corresponding to each key; and perform weighted summation on the vectors corresponding to each of the values ​​according to the weight to obtain the context vector.

[0131] In some embodiments, the processing module 304 is configured to perform weighted summation and nonlinear activation processing on the context vector to obtain a hidden feature vector result; and perform weighted summation and linear activation processing on the hidden feature vector result to obtain the financial evaluation result.

[0132] In some embodiments, the processing module 304 is used to respectively obtain the contribution of each sub-feature of the numerical feature, the text feature and the graphical feature to the financial evaluation result; and aggregate the contribution corresponding to each sub-feature of the numerical feature, the text feature and the graphical feature to obtain the contribution corresponding to each of the numerical feature, the text feature and the graphical feature.

[0133] In some embodiments, the processing module 304 is used to construct three modal total nodes based on the corresponding contribution degrees of the numerical features, the text features and the graph features, and to construct sub-nodes for the three modal total nodes based on the corresponding sub-features and their corresponding contribution degrees; the greater the contribution degree, the larger the total node; the sub-nodes with associated relationships are connected to generate the decision path report.

[0134] In some embodiments, the generation module 305 is used to obtain placeholder information in the financial analysis report template; combine the shown placeholder information, the financial assessment results, the decision path report, and the target data extracted from the structured data and the unstructured data; perform natural language processing on the combined information to fill in the financial analysis report template and generate the financial analysis report.

[0135] The financial analysis report generation device provided in the embodiment of the present application can execute the financial analysis report generation method shown in any of the above embodiments. Its principles and technical effects are similar and will not be repeated here.

[0136] An embodiment of the present application also provides an electronic device.

[0137] Figure 4 This is a structural diagram of an electronic device 40 provided in an embodiment of the present application, as shown in FIG. Figure 4 As shown, the electronic device may include: a transceiver 401 , a processor 402 , and a memory 403 .

[0138] The processor 402 executes the computer-executable instructions stored in the memory, so that the processor 402 implements the solution in the above embodiment. The processor 402 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0139] The memory 403 is connected to the processor 402 via a system bus and implements communication therebetween. The memory 403 is used to store computer program instructions.

[0140] The transceiver 401 can receive and transmit data and instructions.

[0141] Optionally, electronic device 40 may further include a communication interface 404 to enable communication interaction with external or internal devices via communication interface 403. For example, the external device may be a client (e.g., a mobile phone or tablet). In a specific implementation, if communication interface 404, memory 403, and processor 402 are implemented independently, communication interface 404, memory 403, and processor 402 may be interconnected via a bus to enable communication between them.

[0142] The system bus can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, among others. System buses can be divided into address buses, data buses, and control buses. For ease of illustration, the diagram uses only a single thick line, but this does not imply a single bus or type of bus. Transceivers enable communication between the database access device and other computers (such as clients, read-write libraries, and read-only libraries). Memory may include random access memory (RAM) and non-volatile memory.

[0143] Optionally, in a specific implementation, if the communication interface 404, the memory 403 and the processor 402 are integrated on a chip, the communication interface 404, the memory 403 and the processor 402 can complete communication through an internal interface.

[0144] An embodiment of the present application also provides a chip for executing instructions, which is used to execute the technical solutions in the above embodiments.

[0145] In an embodiment of the present application, a computer-readable storage medium is further provided, on which a computer program is stored. When the computer program is executed by a processor, the technical solution of the above embodiment is implemented. The implementation principle and technical effect are similar and will not be repeated here.

[0146] In one possible implementation, a computer-readable medium may include random access memory (RAM), read-only memory (ROM), compact disc read-only memory (CD-ROM) or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other medium designed to carry or store desired program code in the form of instructions or data structures and accessible by a computer. Furthermore, any connection is appropriately termed a computer-readable medium. For example, if software is transmitted from a website, server, or other remote source using a coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technologies such as infrared, radio, and microwave, then the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwave are included in the definition of medium. Disk and disc, as used herein, include optical disc, laser disc, optical disc, digital versatile disc (DVD), floppy disk, and Blu-ray disc, where disks typically reproduce data magnetically, while discs reproduce data optically using lasers. Combinations of the above are also intended to be included within the scope of computer-readable media.

[0147] A computer program product is also provided in an embodiment of the present application, including a computer program. When the computer program is executed by a processor, the technical solution of the above embodiment is implemented. The implementation principle and technical effect are similar and will not be repeated here.

[0148] In the specific implementation of the above-mentioned terminal device or server, it should be understood that the processor can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in the embodiments of the present application can be directly implemented as a hardware processor, or can be implemented by a combination of hardware and software modules in the processor.

[0149] Those skilled in the art will appreciate that all or part of the steps of any of the above method embodiments may be accomplished by hardware associated with program instructions. The aforementioned program may be stored in a computer-readable storage medium, and when the program is executed, all or part of the steps of the above method embodiments are executed.

[0150] If the technical solution of this application is implemented in the form of software and sold or used as a product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the technical solution of this application can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes a computer program or several instructions. This computer software product enables a computer device (which can be a personal computer, server, network device, or similar electronic device) to perform all or part of the steps of the method described in the embodiments of this application.

[0151] It should be noted that for the aforementioned method embodiments, for the sake of simplicity, they are all expressed as a series of action combinations, but those skilled in the art should be aware that this application is not limited by the order of the actions described, because according to this application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in this specification are all optional embodiments, and the actions and modules involved are not necessarily required by this application.

[0152] It should be further noted that, although the various steps in the flowchart are shown in sequence as indicated by the arrows, these steps are not necessarily performed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps may be performed in other orders. Moreover, at least a portion of the steps in the flowchart may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily performed at the same time, but may be performed at different times. The execution order of these sub-steps or stages is not necessarily to be performed in sequence, but may be performed in turn or alternately with other steps or at least a portion of the sub-steps or stages of other steps.

[0153] It should be understood that the above-described device embodiments are merely illustrative, and the device of the present application may also be implemented in other ways. For example, the division of units / modules in the above-described embodiments is merely a logical functional division, and actual implementations may employ other division methods. For example, multiple units, modules, or components may be combined or integrated into another system, or some features may be omitted or not implemented.

[0154] In addition, unless otherwise specified, the functional units / modules in the various embodiments of the present application may be integrated into a single unit / module, each unit / module may exist physically separately, or two or more units / modules may be integrated together. The aforementioned integrated units / modules may be implemented in the form of hardware or software program modules.

[0155] If an integrated unit / module is implemented in hardware, the hardware may be digital circuits, analog circuits, etc. The physical implementation of the hardware structure includes, but is not limited to, transistors, memristors, etc. Unless otherwise specified, the processor may be any appropriate hardware processor, such as a CPU, GPU, FPGA, DSP, and ASIC. Unless otherwise specified, the storage unit may be any appropriate magnetic storage medium or magneto-optical storage medium, such as resistive random access memory (RRAM), dynamic random access memory (DRAM), static random access memory (SRAM), enhanced dynamic random access memory (EDRAM), high-bandwidth memory (HBM), hybrid memory cube (HMC), etc.

[0156] If the integrated unit / module is implemented in the form of a software program module and sold or used as an independent product, it can be stored in a computer-readable memory. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a memory and includes a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the various embodiments of the present application. The aforementioned memory includes: U disk, read-only memory (ROM), random access memory (RAM), mobile hard disk, magnetic disk, or optical disk, etc., various media that can store program code.

[0157] In the above embodiments, the description of each embodiment has its own emphasis. For parts not described in detail in a particular embodiment, please refer to the relevant description of other embodiments. The technical features of the above embodiments can be combined in any way. To keep the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0158] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some or all of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for generating a financial analysis report, characterized in that: include: Obtain multimodal financial data of the target object; Extracting features from structured data and unstructured data in the financial data to obtain a set of feature vectors; The feature vector set includes numerical features, text features and graph features; fusing the numerical features, the text features, and the graph features, and determining a financial evaluation result based on the fused context vector; Obtaining the contribution of the numerical feature, the text feature, and the graph feature to the financial evaluation result, and generating a decision path report based on the contribution; The financial analysis report is generated based on the financial evaluation results and the decision path report.

2. The method according to claim 1, characterized in that The feature extraction of the structured data and the unstructured data in the financial data to obtain a feature vector set includes: Performing index extraction and time series feature extraction on the structured data to obtain the numerical features; Extracting semantic vectors from the unstructured data, and calculating key information density of key data in the unstructured data to obtain the text features; Entity data corresponding to the target object is extracted from the structured data and the unstructured data, a graph network of the target object is constructed based on the entity data, and path analysis is performed on the graph network to obtain the graph features.

3. The method according to claim 1, characterized in that The fusing of the numerical features, the text features, and the graph features includes: Processing the graph features to obtain continuous numerical graph vectors; Processing the text features to obtain a text semantic representation vector; The digitized graph vector, the text semantic representation vector and the numerical feature are fused.

4. The method according to claim 3, characterized in that The fusing of the digitized graph vector, the text semantic representation vector, and the numerical feature includes: Using the text semantic representation vector, the numerical feature and the numerical graph vector as the key and value of a predefined vector; Obtaining the similarity between the predefined vector and each of the keys to obtain an attention score; Normalizing the attention scores to obtain the weights corresponding to the keys; The vectors corresponding to the values ​​are weightedly summed according to the weights to obtain the context vector.

5. The method according to claim 4, characterized in that Determining a financial evaluation result based on the fused context vector includes: Performing weighted summation and nonlinear activation processing on the context vector to obtain a hidden feature vector result; The hidden feature vector results are weighted summed and linearly activated to obtain the financial evaluation result.

6. The method according to any one of claims 1 to 5, characterized in that The obtaining of the contribution of the numerical feature, the text feature, and the graph feature to the financial evaluation result includes: respectively obtaining the contribution of each sub-feature of the numerical feature, the text feature, and the graph feature to the financial evaluation result; The contribution degrees corresponding to each sub-feature of the numerical feature, the text feature, and the graph feature are aggregated to obtain the contribution degrees corresponding to each of the numerical feature, the text feature, and the graph feature.

7. The method according to claim 6, characterized in that Generating a decision path report according to the contribution degree includes: Constructing three modal total nodes according to the respective contributions of the numerical feature, the text feature, and the graph feature, and constructing subnodes for the three modal total nodes according to the respective corresponding subfeatures and their corresponding contributions; the greater the contribution, the larger the total node; The sub-nodes having the association relationship are connected to generate the decision path report.

8. The method according to any one of claims 1 to 5, characterized in that Generating the financial analysis report based on the financial evaluation results and the decision path report includes: Get the placeholder information in the financial analysis report template; combining the placeholder information, the financial assessment results, the decision path report, and target data extracted from the structured data and the unstructured data; The combined information is subjected to natural language processing to fill in the financial analysis report template and generate the financial analysis report.

9. A financial analysis report generating device, characterized in that: include: An acquisition module, used to obtain multimodal financial data of the target object; An extraction module, configured to extract features from structured data and unstructured data in the financial data to obtain a set of feature vectors; The feature vector set includes numerical features, text features and graph features; A fusion module, configured to fuse the numerical features, the text features, and the graph features, and determine a financial evaluation result based on the fused context vector; a processing module, configured to obtain the contribution of the numerical feature, the text feature, and the graph feature to the financial evaluation result, and generate a decision path report based on the contribution; A generating module is used to generate the financial analysis report based on the financial evaluation result and the decision path report.

10. An electronic device, characterized in that: include: A processor, a transceiver, and a memory; the processor is communicatively connected to the transceiver and the memory respectively; The memory is used to store computer programs; The transceiver is used to communicate and interact with external devices; The processor is configured to execute the computer program to implement the method according to any one of claims 1 to 9.

11. A computer-readable storage medium, characterized in that A computer program is stored thereon, and the computer program is executed by a processor to implement the method according to any one of claims 1 to 9.

12. A computer program product, characterized in that The computer program comprises a computer program, which implements the method according to any one of claims 1 to 9 when executed by a controller.

Images