Data fraud detection method and device, computer equipment and storage medium

Abstract

The embodiment of the invention belongs to the field of data processing, and relates to a data fraud detection method and device, computer equipment and a storage medium, and the method comprises the following steps: obtaining financial information data, analyzing the financial information data, and obtaining financial time series data and financial text information data; performing feature extraction on the financial time series data and the financial text information data to obtain financial time features and financial emotion features; performing feature representation transformation on the financial time features and the financial emotion features based on a graph attention network to obtain fused feature representation; the financial time features and the financial emotion features are merged into fused feature representation, and financial information features are obtained; and performing normalization processing on the financial information features to obtain financial effective features, and inputting the financial effective features into a pre-trained fraud probability prediction model to obtain a fraud probability result. According to the method, the fraud probability result of the financial information data can be accurately obtained.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, specifically to the field of financial technology, and in particular to a data fraud detection method, device, computer equipment and storage medium. Background Art

[0002] With the rapid increase of data in the financial field, the judgment of data authenticity and validity has become increasingly complex. The strategies of fraudsters are constantly upgraded and changed with the flood of data. They use advanced technology and complex means to tamper with data and commit fraud, which makes the analysis of financial data face huge risks and challenges. Traditional fraud detection methods, such as relying on professionals to conduct detailed analysis of data, are not only time-consuming and labor-intensive, occupying a large amount of human resources, but also difficult to guarantee efficiency and difficult to meet the growing demand for financial data processing.

[0003] In order to meet this challenge, the industry has begun to try to use deep learning models for fraud detection. Deep learning models have powerful data processing capabilities and can automatically complete data fraud detection tasks to a certain extent, thereby improving detection efficiency. However, existing deep learning models still have many problems in fraud detection. First, these models can often only output a simple data fraud detection result, and cannot provide detailed fraud behavior analysis and evidence support, which makes the accuracy and credibility of the results questionable. Secondly, the application of these models in fraud detection is often one-time, and the detection results cannot be used for subsequent fraud analysis, so it is impossible to form an effective feedback mechanism to continuously optimize and train the model. Summary of the invention

[0004] The purpose of the embodiments of the present application is to provide a data fraud detection method, apparatus, computer equipment and storage medium to solve the problem of being unable to generate continuous and content-consistent output video data.

[0005] In order to solve the above technical problems, the present application embodiment provides a data fraud detection method, which adopts the following technical solution:

[0006] Acquire financial information data, and parse the financial information data to obtain financial time series data and financial text information data;

[0007] Performing feature extraction on the financial time series data and the financial text information data to obtain financial time features and financial sentiment features;

[0008] Based on the graph attention network, the financial time feature and the financial sentiment feature are transformed into feature representations to obtain a fused feature representation;

[0009] Merging the financial time feature and the financial sentiment feature into the fused feature representation to obtain a financial information feature;

[0010] The financial information features are normalized to obtain financially effective features, and the financially effective features are input into a pre-trained fraud probability prediction model to obtain a fraud probability result.

[0011] Furthermore, the steps of acquiring financial information data, parsing the financial information data, and obtaining financial time series data and financial text information data specifically include:

[0012] Acquire a data extraction identifier, and extract initial financial data from a database according to the data extraction identifier;

[0013] Preprocessing the initial financial data to obtain the financial information data;

[0014] The financial information data is classified according to data types to obtain the financial time series data and the financial text information data.

[0015] Furthermore, the step of extracting features from the financial time series data and the financial text information data to obtain financial time features and financial sentiment features specifically includes:

[0016] Inputting the financial time series data into a preset memory network to extract time features, thereby obtaining the financial time features;

[0017] The financial text information data is input into a pre-trained language model for sentiment quantification to obtain the financial sentiment features.

[0018] Furthermore, the financial text information data includes financial news text data and financial media text data, and the step of inputting the financial text information data into a pre-trained language model for sentiment quantification to obtain the financial sentiment feature specifically includes:

[0019] Preprocessing the financial news text data and the financial media text data to obtain standard financial news text data and standard financial media text data;

[0020] Inputting the standard financial news text data and the standard financial media text data into the pre-trained language model to perform sentiment analysis on a semantic dimension basis to obtain news text sentiment information and media text sentiment information;

[0021] Extracting fine-grained sentiment features from the news text sentiment information and the media text sentiment information to obtain sentiment intensity and sentiment polarity;

[0022] A preset emotion quantification index is obtained, and the news text emotion information and the media text emotion information are quantified according to the preset emotion quantification index, the emotion intensity, and the emotion polarity to obtain the financial emotion feature.

[0023] Furthermore, the step of performing feature representation transformation on the financial time feature and the financial sentiment feature based on the graph attention network to obtain a fused feature representation specifically includes:

[0024] Perform feature alignment according to the financial time feature and the financial sentiment feature to obtain an effective time feature and an effective sentiment feature;

[0025] Define network nodes according to the effective time features and the effective emotion features to obtain network feature nodes;

[0026] Calculating the cosine similarity of the network feature nodes, and constructing network edges according to the calculated cosine similarity;

[0027] Calculate the attention weights between network feature nodes based on multi-head attention;

[0028] The network feature nodes are weightedly summed according to the attention weights and the network edges to obtain the fused feature representation.

[0029] Furthermore, the step of merging the financial time feature and the financial sentiment feature into the fused feature representation to obtain the financial information feature specifically includes:

[0030] Standardizing the financial time feature and the financial sentiment feature to obtain a standard financial time feature and a standard financial sentiment feature;

[0031] Performing feature selection on the standard financial time feature and the standard financial sentiment feature to obtain effective financial time feature and effective financial sentiment feature;

[0032] Performing weighted fusion on the effective financial time feature and the effective financial sentiment feature to obtain a fusion feature vector;

[0033] The fused feature vector and the fused feature representation are concatenated to obtain the financial information feature.

[0034] Furthermore, before the step of normalizing the financial information features to obtain financial effective features and inputting the financial effective features into a pre-trained fraud probability prediction model to obtain the fraud probability, the following steps are also included:

[0035] Obtain fraudulent transaction sample data and normal transaction sample data;

[0036] Preprocessing the fraudulent transaction sample data and the normal transaction sample data to obtain standard fraudulent transaction sample data and standard normal transaction sample data;

[0037] The preset decision tree model is trained and optimized according to the standard fraudulent transaction sample data and the standard normal transaction sample data to obtain the fraud probability prediction model.

[0038] In order to solve the above technical problems, the embodiment of the present application further provides a data fraud detection device, which adopts the following technical solution:

[0039] A data acquisition module, used to acquire financial information data, parse the financial information data, and obtain financial time series data and financial text information data;

[0040] A feature extraction module, used to extract features from the financial time series data and the financial text information data to obtain financial time features and financial sentiment features;

[0041] A feature representation module, used for performing feature representation transformation on the financial time feature and the financial sentiment feature based on a graph attention network to obtain a fused feature representation;

[0042] A feature fusion module, used for merging the financial time feature and the financial sentiment feature into the fused feature representation to obtain a financial information feature;

[0043] The model prediction module is used to normalize the financial information features to obtain financial effective features, and input the financial effective features into a pre-trained fraud probability prediction model to obtain a fraud probability result.

[0044] In order to solve the above technical problems, the embodiment of the present application further provides a computer device, which adopts the following technical solution:

[0045] A computer device comprises a memory and a processor, wherein the memory stores computer-readable instructions, and the processor implements the steps of any one of the above data fraud detection methods when executing the computer-readable instructions.

[0046] In order to solve the above technical problems, the embodiment of the present application further provides a computer-readable storage medium, which adopts the following technical solution:

[0047] A computer-readable storage medium having computer-readable instructions stored thereon, wherein the computer-readable instructions, when executed by a processor, implement the steps of any of the above data fraud detection methods.

[0048] Compared with the prior art, the embodiments of the present application have the following beneficial effects: the embodiments obtain financial information data, parse the financial information data, and obtain financial time series data and financial text information data; extract features from the financial time series data and the financial text information data to obtain financial time features and financial sentiment features; perform feature representation transformation on the financial time features and the financial sentiment features based on the graph attention network to obtain fused feature representation; merge the financial time features and the financial sentiment features into the fused feature representation to obtain financial information features; normalize the financial information features to obtain financial effective features, and input the financial effective features into the pre-trained fraud probability prediction model to obtain fraud probability results. This effectively achieves accurate acquisition of fraud probability results of financial information data, so as to identify whether the financial information data is fraud information and improve the reliability of information acquisition. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the scheme in the present application, a brief introduction is given below to the drawings required for use in the description of the embodiments of the present application. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0050] Figure 1 is an exemplary system architecture diagram to which the present application may be applied;

[0051] Figure 2 A flowchart of an embodiment of a data fraud detection method according to the present application;

[0052] Figure 3 yes Figure 2 A flowchart of a specific implementation of step S10;

[0053] Figure 4 yes Figure 2 A flowchart of a specific implementation of step S20;

[0054] Figure 5 yes Figure 4 A flowchart of a specific implementation of step S203;

[0055] Figure 6 yes Figure 2 A flowchart of a specific implementation of step S30;

[0056] Figure 7 yes Figure 2 A flowchart of a specific implementation of step S40;

[0057] Figure 8is a structural schematic diagram of an embodiment of a data fraud detection device according to the present application;

[0058] Fig. 9 It is a structural diagram of an embodiment of a computer device according to the present application. DETAILED DESCRIPTION

[0059] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of the present application; the terms used in the specification of the application herein are only for the purpose of describing specific embodiments and are not intended to limit the present application; the terms "including" and "having" and any variations thereof in the specification and claims of the present application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of the present application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0060] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor does it refer to non-related or alternative embodiments that are mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0061] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings.

[0062] like Figure 1 As shown, the system architecture 100 may include a terminal device 101, a network 102 and a server 103. The terminal device 101 may be a laptop 1011, a tablet computer 1012 or a mobile phone 1013. The network 102 is used to provide a medium for a communication link between the terminal device 101 and the server 103. The network 102 may include various connection types, such as wired, wireless communication links or optical fiber cables.

[0063] The user can use the terminal device 101 to interact with the server 103 through the network 102 to receive or send messages, etc. Various communication client applications can be installed on the terminal device 101, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.

[0064] The terminal device 101 can be any electronic device with a display screen and supporting web browsing. In addition to a laptop computer 1011, a tablet computer 1012 or a mobile phone 1013, the terminal device 101 can also be an e-book reader, an MP3 player (Moving Picture Experts Group Audio Layer III), an MP4 (Moving Picture Experts Group Audio Layer IV), a laptop computer, a desktop computer, etc.

[0065] The server 103 may be a server that provides various services, such as a background server that provides support for a web page displayed on the terminal device 101 .

[0066] It should be noted that the data fraud detection method provided in the embodiment of the present application is generally executed by a server / terminal device, and accordingly, the data fraud detection device is generally arranged in the server / terminal device.

[0067] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0068] Continue to refer Figure 2 , shows a flow chart of an embodiment of a method for data fraud detection according to the present application. The data fraud detection method comprises the following steps:

[0069] Step S10, acquiring financial information data, parsing the financial information data, and obtaining financial time series data and financial text information data;

[0070] In this embodiment, the financial information data includes financial time series data and financial text information data, wherein the financial time series data includes historical transaction data and related financial indicators. The historical transaction data refers to the historical transaction records of financial products such as stocks, bonds, futures, and foreign exchange, including transaction prices, transaction volumes, opening prices, closing prices, highest prices, lowest prices, etc. The related financial indicators are calculated based on the historical transaction data and are used to reflect the overall situation of the financial market or the performance of specific financial products. Common financial indicators include price-earnings ratio, price-to-book ratio, dividend yield, moving average, relative strength index (RSI), etc.; the financial text information data includes financial news text data and financial media text data. The financial news text data refers to news reports on financial markets, financial products, financial institutions, etc., and the news reports include the latest market trends, policy changes, company performance, etc. The financial media text data refers to text information such as comments, analysis, and forecasts on the financial market published by financial media (such as financial websites, blogs, social media, etc.), and the text information includes information such as expert opinions, market sentiment, and investor expectations.

[0071] Step S20, extracting features from the financial time series data and the financial text information data to obtain financial time features and financial sentiment features;

[0072] In this embodiment, the feature extraction of financial time series data is realized by a preset memory network, which is a LSTM (Long Short-Term Memory Network). LSTM effectively captures the long-term dependencies in the financial time series data through a gating mechanism (input gate, forget gate, and output gate), and processes the high nonlinearity and volatility in the financial time series data through a nonlinear activation function and a gating mechanism, thereby realizing accurate and effective feature extraction to obtain financial time features. The feature extraction of financial text information data is performed by language processing through a pre-trained language model to identify the emotional vocabulary in the financial text information data, and then the emotional vocabulary is quantified to obtain the financial emotional features of the emotional representation of the financial text information data. Among them, the pre-trained language model adopts the BERT model, and the BERT model adopts a bidirectional Transformer structure, which can simultaneously consider the contextual information of each word in the financial text information data, so that the contextual meaning of each word can be more accurately understood to improve the accuracy and reliability of feature extraction.

[0073] Step S30, performing feature representation transformation on the financial time feature and the financial sentiment feature based on a graph attention network to obtain a fused feature representation;

[0074] In this embodiment, a graph attention network (GAT) is a graph neural network (GNN) model, which defines the network nodes of the graph attention network according to financial time features and financial sentiment features, and then introduces the calculation of the attention score weights of the current network node and each adjacent network node to determine the relative importance between them. The score is usually calculated by a learnable function (such as a dot product, bilinear or additive function), which takes the features of the current network node and the adjacent network nodes as input to calculate the attention weight. After obtaining the attention weight, the graph attention network uses these attention weights to perform weighted summation on the features of the adjacent network nodes, thereby generating a representation of the current network node and obtaining a fused feature representation.

[0075] Step S40, merging the financial time feature and the financial sentiment feature into the fusion feature representation to obtain a financial information feature;

[0076] In this embodiment, the financial information feature is the feature information obtained by combining the financial time feature, the financial sentiment feature, and the fusion feature representation. The combination of the financial time feature and the financial sentiment feature is achieved through feature selection and weighted fusion. Feature selection is the process of selecting features with significant influence from the financial time feature and the financial sentiment feature. The key information related to the financial time feature and the financial sentiment feature is screened out through feature selection. In this embodiment, feature selection can be achieved through the correlation coefficient method. Weighted fusion is the process of weighting multiple features according to their importance to obtain a more comprehensive feature representation. Through weighted fusion, we can combine the financial time feature and the financial sentiment feature to form a more comprehensive feature set. In this embodiment, weighted fusion can be achieved by using the decision tree method.

[0077] Step S50, normalizing the financial information features to obtain financially effective features, and inputting the financially effective features into a pre-trained fraud probability prediction model to obtain a fraud probability result.

[0078] In this embodiment, normalization is the process of converting features of different dimensions to the same dimension so that the model can compare these features more fairly. In this embodiment, the normalization method can use minimum-maximum normalization (scaling the data to between 0 and 1) to normalize the financial information features to obtain financial effective features with unified representation. The pre-trained fraud probability prediction model uses the LightGBM model. LightGBM (Light GradientBoosting Machine) is a gradient boosting framework based on a decision tree. By analyzing the importance of each feature in the LightGBM model, it is understood which features have the greatest impact on fraud detection, and then the corresponding fraud probability results are output according to the impact.

[0079] This embodiment acquires financial information data, parses the financial information data, obtains financial time series data and financial text information data; extracts features from the financial time series data and the financial text information data to obtain financial time features and financial sentiment features; transforms the feature representation of the financial time features and the financial sentiment features based on the graph attention network to obtain a fused feature representation; merges the financial time features and the financial sentiment features into the fused feature representation to obtain financial information features; normalizes the financial information features to obtain financial effective features, and inputs the financial effective features into a pre-trained fraud probability prediction model to obtain a fraud probability result. This effectively achieves accurate acquisition of the fraud probability results of the financial information data, so as to identify whether the financial information data is fraud information and improve the reliability of information acquisition.

[0080] refer to Figure 3 In some optional implementations of this embodiment, step S10 includes the following steps:

[0081] Step S101, obtaining a data extraction identifier, and extracting initial financial data from a database according to the data extraction identifier;

[0082] In this embodiment, the data extraction identifier is a unique information identifier corresponding to the initial financial data. The data extraction identifier is used to perform traversal matching in the database, and then the matched data is extracted to obtain the initial financial data.

[0083] Step S102, preprocessing the initial financial data to obtain the financial information data;

[0084] In this embodiment, preprocessing includes cleaning, conversion and formatting. Cleaning data includes removing duplicates, processing missing values, correcting errors, etc. Converting data includes converting the initial financial data into a format suitable for analysis, such as converting a date string into a date object. Formatting data refers to standardizing the initial financial data to ensure data consistency and readability.

[0085] Step S103, classifying the financial information data according to data types to obtain the financial time series data and the financial text information data.

[0086] In this embodiment, the corresponding data type is determined according to the structure and content of the financial information data, and the financial information data is classified according to the data type, wherein the financial time series data is financial numerical data arranged in chronological order, and the financial text information data is text data related to finance.

[0087] This embodiment obtains a data extraction identifier and extracts initial financial data from a database according to the data extraction identifier; preprocesses the initial financial data to obtain the financial information data; and classifies the financial information data according to the data type, thereby effectively obtaining standard and valid financial time series data and financial text information data to facilitate subsequent feature extraction processing.

[0088] refer to Figure 4 In some optional implementations of this embodiment, step S20 includes the following steps:

[0089] Step S201, obtaining a network and model extraction identifier, and extracting a preset memory network and a pre-trained language model from a database according to the network and model extraction identifier;

[0090] In this embodiment, the network and model extraction identifier is identification information corresponding to the preset memory network and the pre-trained language model. A matching query is performed in the database according to the network and model extraction identifier to extract the corresponding preset memory network and pre-trained language model.

[0091] Step S202, inputting the financial time series data into the preset memory network to extract time features, thereby obtaining the financial time features;

[0092] In this embodiment, the preset memory network adopts a long short-term memory network. By inputting the financial time series data into the long short-term memory network, the LSTM network learns and extracts the features in the time series data through its internal mechanism (especially its gating structure and cell state) to obtain financial time features. Financial time features include trend (the overall upward or downward trend of financial time series data in a certain period of time), seasonality (a pattern of financial time series data that repeats in a specific time period (such as a quarter of a year)), periodicity (a pattern of financial time series data that repeats in a longer time period (such as several years or decades)), volatility (dramatic fluctuations in financial time series data in a short period of time), momentum (a trend of financial time series data that continues to rise or fall in a short period of time), the relationship between trading volume and price, mean reversion (financial time series data tends to return to its long-term average after deviating from its long-term average), etc.

[0093] Step S203: input the financial text information data into the pre-trained language model to perform sentiment quantification to obtain the financial sentiment features.

[0094] In this embodiment, by performing sentiment analysis and fine-grained sentiment feature extraction on the financial text information data in each semantic dimension, information capable of sentiment quantification is obtained, and then sentiment quantification is performed on the obtained information according to preset sentiment quantification indicators, thereby obtaining financial sentiment features. Financial sentiment features include sentiment polarity features, sentiment intensity features, sentiment theme features, sentiment tendency features, sentiment change trend features, etc.

[0095] This embodiment obtains the network and model extraction identifier, extracts the preset memory network and the pre-trained language model from the database according to the network and model extraction identifier; inputs the financial time series data into the preset memory network to extract the time feature and obtain the financial time feature; inputs the financial text information data into the pre-trained language model to quantify the emotion and obtain the financial emotion feature. Thus, effective time feature extraction and emotion feature extraction are respectively realized from the financial time series data and the financial text information data, so as to facilitate the subsequent acquisition of the fusion feature representation.

[0096] refer to Figure 5 In some optional implementations of this embodiment, step S203 includes the following steps:

[0097] Step S2031, preprocessing the financial news text data and the financial media text data to obtain standard financial news text data and standard financial media text data;

[0098] In this embodiment, the preprocessing of financial news text data and financial media text data includes data cleaning, text segmentation and stop word removal, stem extraction and word form restoration, data standardization and normalization, etc., among which data cleaning includes removing irrelevant information, processing duplicate data, and processing missing values.

[0099] Step S2032, inputting the standard financial news text data and the standard financial media text data into the pre-trained language model to perform sentiment analysis on a semantic dimension basis, to obtain news text sentiment information and media text sentiment information;

[0100] In this embodiment, the pre-trained language model can adopt the BERT model, and the sentiment analysis by semantic dimension refers to performing in-depth semantic analysis of standard financial news text data and standard financial media text data through the pre-trained language model, and extracting sentiment features from multiple semantic dimensions (such as vocabulary, phrases, sentences, paragraphs, etc.) to obtain news text sentiment information and media text sentiment information.

[0101] Step S2033, extracting fine-grained emotional features from the news text emotional information and the media text emotional information to obtain emotional intensity and emotional polarity;

[0102] In this embodiment, fine-grained sentiment feature extraction refers to extracting smaller and more specific sentiment features from news text sentiment information and media text sentiment information. These features can more accurately reflect the sentiment tendencies and nuances in the text. Fine-grained sentiment feature extraction can be performed based on a preset sentiment dictionary, which contains a series of sentiment words and corresponding sentiment polarities (positive, negative or neutral).

[0103] Step S2034, obtaining a preset emotion quantification index, and quantifying the news text emotion information and the media text emotion information according to the preset emotion quantification index, the emotion intensity, and the emotion polarity to obtain the financial emotion feature.

[0104] In this embodiment, the preset sentiment quantification index is preset information for evaluating text sentiment, which is used to define the scope and accuracy of sentiment analysis. The preset sentiment quantification index includes sentiment polarity: positive, negative, neutral, sentiment intensity: weak, medium, strong, and sentiment dimension: such as trust, fear, optimism, pessimism, etc. Among them, each preset sentiment quantification index corresponds to a sentiment feature value, which can be quantified by mapping the sentiment intensity and sentiment polarity to the corresponding sentiment value. Alternatively, a sentiment analysis model can be constructed by collecting a large amount of financial news and media text data with marked sentiment polarity and intensity, and then extracting sentiment-related features from the text, such as bag-of-words features, TF-IDF features, word embedding features, etc., and using the marked data to train a sentiment classification model so that the model can predict the sentiment polarity and sentiment intensity of text data, and then evaluate the performance of the model through methods such as cross-validation, and perform necessary optimization based on the evaluation results to obtain a sentiment analysis model that can effectively quantify the sentiment polarity and sentiment intensity of text data.

[0105] This embodiment obtains standard financial news text data and standard financial media text data by preprocessing the financial news text data and the financial media text data; inputs the standard financial news text data and the standard financial media text data into the pre-trained language model for sentiment analysis of each semantic dimension to obtain news text sentiment information and media text sentiment information; performs fine-grained sentiment feature extraction on the news text sentiment information and the media text sentiment information to obtain sentiment intensity and sentiment polarity; obtains a preset sentiment quantification index, and quantifies the news text sentiment information and the media text sentiment information according to the preset sentiment quantification index, the sentiment intensity, and the sentiment polarity, thereby effectively obtaining financial sentiment features that accurately represent the sentiment information of the financial news text data and the financial media text data, so as to facilitate the subsequent acquisition of fusion feature representation.

[0106] Continue to refer Figure 6 In some optional implementations of this embodiment, step S30 includes the following steps:

[0107] Step S301, performing feature alignment according to the financial time feature and the financial sentiment feature to obtain an effective time feature and an effective sentiment feature;

[0108] In this embodiment, the feature alignment of the financial time feature and the financial emotion feature is achieved by associating the emotion feature value corresponding to the financial emotion feature with the time point or time period corresponding to the financial time feature.

[0109] Step S302, defining network nodes according to the effective time feature and the effective emotion feature to obtain network feature nodes;

[0110] In this embodiment, the network feature node is the basic component unit in the graph attention network, which represents an entity or individual in the network. These nodes are interconnected through edges and together form a complex network structure. By integrating the effective time feature and the effective emotion feature into the network feature node to complete the network node definition, a network feature node including the effective time feature and the effective emotion feature is obtained.

[0111] Step S303, calculating the cosine similarity of the network feature nodes, and constructing network edges according to the calculated cosine similarity;

[0112] In this embodiment, cosine similarity refers to the cosine value of the angle between the feature vector corresponding to the network feature node and the feature vector corresponding to other network feature nodes in the feature dimension space. The cosine value is used to measure the directional similarity of financial time features and financial sentiment features, where the feature vector corresponding to the network feature node is the feature vector corresponding to the effective time feature and the effective sentiment feature. The value range of cosine similarity is [-1,1]. The larger the value, the more similar the two vectors are, and the smaller the value, the less similar the two vectors are. The calculation formula of cosine similarity is: cosθ=(A·B) / (‖A‖×‖B‖), where A and B are the feature vector of the current network node and the feature vector of another network node, respectively, A·B represents the dot product of A and B (i.e., the sum of the corresponding elements after multiplication), and ‖A‖ and ‖B‖ represent the modulus of A and B respectively (i.e., the length of the vector, which can be obtained by taking the square root of the sum of the squares of each element). After the cosine similarity is calculated, the cosine similarity is compared with a preset threshold to determine which nodes need to build network edges (i.e., establish connections between network feature nodes). The preset threshold can be set and adjusted according to actual conditions. By building network edges, the network feature nodes are effectively connected to form a financial network based on cosine similarity.

[0113] Step S304, calculating the attention weights between network feature nodes based on multi-head attention;

[0114] In this embodiment, the step of calculating the attention weight based on multi-head attention includes node feature representation: using a trainable embedding matrix (or called weight matrix) to map the network feature node features to a new space to obtain a new node feature representation; calculating the attention head: for each attention head, using two trainable weight matrices (query matrix Q, key matrix K and value matrix V) to calculate the attention score between the nodes, wherein the attention score is usually obtained by calculating the dot product between the query vector and the key vector, and applying the softmax function to normalize, and the attention score is the attention weight between the network feature nodes.

[0115] Step S305, performing weighted summation on the network feature nodes according to the attention weights and the network edges to obtain the fused feature representation.

[0116] In this embodiment, the value vectors of the network feature nodes are weighted and summed using the attention weights to obtain the output of each attention head, and then the outputs of each attention head are concatenated and merged using an additional linear layer (or output layer) to obtain the final fused feature representation, where each fused feature representation corresponds to a single network feature node.

[0117] This embodiment obtains effective time features and effective emotion features by aligning features according to the financial time features and the financial emotion features; defines network nodes according to the effective time features and the effective emotion features to obtain network feature nodes; calculates cosine similarity for the network feature nodes, and constructs network edges according to the calculated cosine similarity; calculates attention weights between network feature nodes based on multi-head attention; and performs weighted summation of the network feature nodes according to the attention weights and the network edges, thereby effectively obtaining a fusion feature representation that effectively integrates the financial time features and the financial emotion features, so as to facilitate the subsequent acquisition of financial information features.

[0118] Continue to refer Figure 7 In some optional implementations of this embodiment, step S40 includes the following steps:

[0119] Step S401, performing standardization processing on the financial time feature and the financial sentiment feature to obtain a standard financial time feature and a standard financial sentiment feature;

[0120] In this embodiment, the standardized processing of financial time features includes timestamp conversion, time period processing, time scale transformation, etc., and the standardized processing of financial sentiment features includes sentiment polarity classification, feature scaling, etc.

[0121] Step S402, performing feature selection on the standard financial time feature and the standard financial sentiment feature to obtain effective financial time feature and effective financial sentiment feature;

[0122] In this embodiment, feature selection can be achieved by the correlation coefficient method. The correlation coefficient is a statistic used to measure the linear correlation between the standard financial time feature and the standard financial sentiment feature. In this embodiment, the standard financial time feature and the standard financial sentiment feature correspond to the target financial time feature and the target financial sentiment feature respectively. The target financial time feature and the target financial sentiment feature are the feature information most relevant to the probability of fraud, which can be obtained by analyzing the historical financial information data. The correlation coefficient method uses the Pearson correlation coefficient, and the Pearson correlation coefficient is used to calculate the correlation coefficient value between the standard financial time feature and the target financial time feature, as well as the correlation coefficient value between the standard financial sentiment feature and the target financial sentiment feature, and then the effective financial time feature and the effective financial sentiment feature that meet the correlation coefficient threshold are screened out according to the correlation coefficient value. Among them, the Pearson correlation coefficient is obtained by calculating the mean of the product of the difference between the standard financial time feature and the target financial time feature and their respective means, and then dividing it by the product of the standard deviation of the standard financial time feature and the target financial time feature.

[0123] Step S403, performing weighted fusion on the effective financial time feature and the effective financial sentiment feature to obtain a fusion feature vector;

[0124] In this embodiment, weighted fusion can be achieved by using a decision tree method. The weights of the features are extracted by calculating the contribution of the effective financial time features and the effective financial sentiment features in the decision tree (such as the Gini index, information gain, etc.). The effective financial time features and the effective financial sentiment features are weightedly fused based on the extracted feature weights. The weighted fusion can be achieved by calculating the weighted value of each feature, that is, multiplying the feature value by its weight. Finally, the weighted feature values ​​are combined into a fused feature vector, thereby facilitating subsequent fraud prediction.

[0125] Step S404, performing feature concatenation on the fused feature vector and the fused feature representation to obtain the financial information feature.

[0126] In this embodiment, the fused feature vector and the fused feature representation are concatenated end to end to obtain the financial information feature including the fused feature vector and the fused feature representation.

[0127] This embodiment obtains standard financial time features and standard financial sentiment features by standardizing the financial time features and the financial sentiment features; performs feature selection on the standard financial time features and the standard financial sentiment features to obtain effective financial time features and effective financial sentiment features; performs weighted fusion on the effective financial time features and the effective financial sentiment features to obtain a fused feature vector; performs feature splicing on the fused feature vector and the fused feature representation, thereby effectively obtaining financial information features that combine financial time features, financial sentiment features, and fused feature representations, so as to obtain financial information features that accurately represent financial information data, and facilitate subsequent fraud prediction processing of information.

[0128] In some optional implementations of this embodiment, before step S50, the following steps are further included:

[0129] Obtain fraudulent transaction sample data and normal transaction sample data;

[0130] In this embodiment, transaction data can be collected from financial institutions, payment platforms or other relevant channels, and then samples of fraudulent transactions and normal transactions can be screened based on known labels (fraud or non-fraud), thereby obtaining fraudulent transaction sample data and normal transaction sample data.

[0131] Preprocessing the fraudulent transaction sample data and the normal transaction sample data to obtain standard fraudulent transaction sample data and standard normal transaction sample data;

[0132] In this embodiment, the preprocessing of fraudulent transaction sample data and normal transaction sample data includes: data cleaning, feature selection, feature encoding, data standardization, and data segmentation (dividing fraudulent transaction sample data and normal transaction sample data into training set, verification set, and test set).

[0133] The preset decision tree model is trained and optimized according to the standard fraudulent transaction sample data and the standard normal transaction sample data to obtain the fraud probability prediction model.

[0134] In this embodiment, the preset decision tree model adopts the LightGBM model, and the LightGBM model is trained by the training set obtained by segmenting the standard fraudulent transaction sample data and the standard normal transaction sample data, and then the performance of the trained LightGBM model is evaluated by the validation set (such as accuracy, recall rate, F1 score, etc.), so as to optimize the model according to the evaluated performance, and finally the optimized LightGBM model is finally evaluated by the test set to ensure the prediction ability of the model. The LightGBM model that meets the model prediction ability requirements is used as the fraud probability prediction model.

[0135] This embodiment obtains fraudulent transaction sample data and normal transaction sample data; pre-processes the fraudulent transaction sample data and the normal transaction sample data to obtain standard fraudulent transaction sample data and standard normal transaction sample data; trains and optimizes a preset decision tree model based on the standard fraudulent transaction sample data and the standard normal transaction sample data, thereby effectively obtaining a fraud probability prediction model that can accurately predict the probability that information is fraudulent information, so as to facilitate the subsequent acquisition of fraud probability results of financial information data.

[0136] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through computer-readable instructions, and the computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the aforementioned storage medium can be a non-volatile storage medium such as a disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0137] It should be understood that, although the steps in the flowchart of the accompanying drawings are displayed in sequence as indicated by the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least a part of the steps in the flowchart of the accompanying drawings may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

[0138] Further references Figure 8 , as a response to the above Figure 1 The present application provides an embodiment of a data fraud detection device, and the device embodiment is similar to Figure 1 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.

[0139] like Figure 8 As shown, the data fraud detection device 600 described in this embodiment includes: a data acquisition module 601, a feature extraction module 602, a feature representation module 603, a feature fusion module 604, and a model prediction module 605. Among them:

[0140] The data acquisition module 601 is used to acquire financial information data, parse the financial information data, and obtain financial time series data and financial text information data;

[0141] A feature extraction module 602 is used to extract features from the financial time series data and the financial text information data to obtain financial time features and financial sentiment features;

[0142] A feature representation module 603 is used to perform feature representation transformation on the financial time feature and the financial sentiment feature based on a graph attention network to obtain a fused feature representation;

[0143] A feature fusion module 604 is used to merge the financial time feature and the financial sentiment feature into the fusion feature representation to obtain a financial information feature;

[0144] The model prediction module 605 is used to normalize the financial information features to obtain financial effective features, and input the financial effective features into the pre-trained fraud probability prediction model to obtain a fraud probability result.

[0145] By adopting the above-mentioned data fraud detection device, this embodiment can obtain financial information data, parse the financial information data to obtain financial time series data and financial text information data; extract features from the financial time series data and the financial text information data to obtain financial time features and financial sentiment features; transform the feature representation of the financial time features and the financial sentiment features based on the graph attention network to obtain a fused feature representation; merge the financial time features and the financial sentiment features into the fused feature representation to obtain financial information features; normalize the financial information features to obtain financial effective features, and input the financial effective features into the pre-trained fraud probability prediction model to obtain fraud probability results. This effectively achieves accurate acquisition of fraud probability results of financial information data, so as to identify whether the financial information data is fraud information and improve the reliability of information acquisition.

[0146] To solve the above technical problems, the present application also provides a computer device. Fig. 9 , Fig. 9 This is a basic structural block diagram of the computer device in this embodiment.

[0147] The computer device 7 includes a memory 71, a processor 72, and a network interface 73 that are interconnected through a system bus. It should be noted that the figure only shows a computer device 7 with components 71-73, but it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented instead. Among them, those skilled in the art can understand that the computer device here is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application specific integrated circuits (Application Specific Integrated Circuit, ASIC), programmable gate arrays (Field-Programmable Gate Array, FPGA), digital processors (Digital Signal Processor, DSP), embedded devices, etc.

[0148] The computer device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The computer device may interact with a user through a keyboard, a mouse, a remote controller, a touch pad, or a voice control device.

[0149] The memory 71 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, disk, optical disk, etc. In some embodiments, the memory 71 can be an internal storage unit of the computer device 7, such as a hard disk or memory of the computer device 7. In other embodiments, the memory 71 can also be an external storage device of the computer device 7, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (FlashCard), etc. equipped on the computer device 7. Of course, the memory 71 can also include both the internal storage unit of the computer device 7 and its external storage device. In this embodiment, the memory 71 is generally used to store the operating system and various application software installed on the computer device 7, such as computer-readable instructions of a data fraud detection method, etc. In addition, the memory 71 can also be used to temporarily store various types of data that have been output or are to be output.

[0150] The processor 72 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chips in some embodiments. The processor 72 is generally used to control the overall operation of the computer device 7. In this embodiment, the processor 72 is used to run computer-readable instructions stored in the memory 71 or process data, such as computer-readable instructions for running the data fraud detection method.

[0151] The network interface 73 may include a wireless network interface or a wired network interface. The network interface 73 is generally used to establish a communication connection between the computer device 7 and other electronic devices.

[0152] By using the above-mentioned computer device, this embodiment can obtain financial information data, parse the financial information data to obtain financial time series data and financial text information data; extract features from the financial time series data and the financial text information data to obtain financial time features and financial sentiment features; perform feature representation transformation on the financial time features and the financial sentiment features based on the graph attention network to obtain fused feature representation; merge the financial time features and the financial sentiment features into the fused feature representation to obtain financial information features; normalize the financial information features to obtain financial effective features, and input the financial effective features into the pre-trained fraud probability prediction model to obtain fraud probability results. This effectively achieves accurate acquisition of fraud probability results of financial information data, so as to identify whether the financial information data is fraud information and improve the reliability of information acquisition.

[0153] The present application also provides another embodiment, namely, providing a computer-readable storage medium, wherein the computer-readable storage medium stores computer-readable instructions, and the computer-readable instructions can be executed by at least one processor to enable the at least one processor to perform the steps of the data fraud detection method as described above.

[0154] By adopting the above-mentioned computer-readable storage medium, this embodiment can obtain financial information data, parse the financial information data to obtain financial time series data and financial text information data; extract features from the financial time series data and the financial text information data to obtain financial time features and financial sentiment features; perform feature representation transformation on the financial time features and the financial sentiment features based on the graph attention network to obtain fused feature representation; merge the financial time features and the financial sentiment features into the fused feature representation to obtain financial information features; normalize the financial information features to obtain financial effective features, and input the financial effective features into the pre-trained fraud probability prediction model to obtain fraud probability results. This effectively achieves accurate acquisition of fraud probability results of financial information data, so as to identify whether the financial information data is fraud information and improve the reliability of information acquisition.

[0155] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0156] Obviously, the embodiments described above are only some embodiments of the present application, rather than all embodiments. The preferred embodiments of the present application are given in the accompanying drawings, but they do not limit the patent scope of the present application. The present application can be implemented in many different forms. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present application more thorough and comprehensive. Although the present application is described in detail with reference to the aforementioned embodiments, for those skilled in the art, it is still possible to modify the technical solutions recorded in the aforementioned specific implementation methods, or to perform equivalent replacement of some of the technical features therein. Any equivalent structure made using the contents of the specification and drawings of this application, directly or indirectly used in other related technical fields, is similarly within the scope of patent protection of this application.

[0157] The non-Company software tools or components appearing in the embodiments of this application are merely examples and do not represent actual use.

Claims

1. A data fraud detection method, characterized in that: The steps include: Acquire financial information data, and parse the financial information data to obtain financial time series data and financial text information data; Performing feature extraction on the financial time series data and the financial text information data to obtain financial time features and financial sentiment features; Based on the graph attention network, the financial time feature and the financial sentiment feature are transformed into feature representations to obtain a fused feature representation; Merging the financial time feature and the financial sentiment feature into the fused feature representation to obtain a financial information feature; The financial information features are normalized to obtain financially effective features, and the financially effective features are input into a pre-trained fraud probability prediction model to obtain a fraud probability result.

2. The data fraud detection method according to claim 1, characterized in that: The steps of acquiring financial information data, parsing the financial information data, and obtaining financial time series data and financial text information data specifically include: Acquire a data extraction identifier, and extract initial financial data from a database according to the data extraction identifier; Preprocessing the initial financial data to obtain the financial information data; The financial information data is classified according to data types to obtain the financial time series data and the financial text information data.

3. The data fraud detection method according to claim 1, characterized in that: The step of extracting features from the financial time series data and the financial text information data to obtain financial time features and financial sentiment features specifically includes: Obtaining a network and model extraction identifier, and extracting a preset memory network and a pre-trained language model from a database according to the network and model extraction identifier; Inputting the financial time series data into the preset memory network to extract time features, thereby obtaining the financial time features; The financial text information data is input into the pre-trained language model for sentiment quantification to obtain the financial sentiment features.

4. The data fraud detection method according to claim 3, characterized in that: The financial text information data includes financial news text data and financial media text data. The step of inputting the financial text information data into a pre-trained language model for sentiment quantification to obtain the financial sentiment feature specifically includes: Preprocessing the financial news text data and the financial media text data to obtain standard financial news text data and standard financial media text data; Inputting the standard financial news text data and the standard financial media text data into the pre-trained language model to perform sentiment analysis on a semantic dimension basis to obtain news text sentiment information and media text sentiment information; Extracting fine-grained sentiment features from the news text sentiment information and the media text sentiment information to obtain sentiment intensity and sentiment polarity; A preset emotion quantification index is obtained, and the news text emotion information and the media text emotion information are quantified according to the preset emotion quantification index, the emotion intensity, and the emotion polarity to obtain the financial emotion feature.

5. The data fraud detection method according to claim 1, characterized in that: The step of performing feature representation transformation on the financial time feature and the financial sentiment feature based on the graph attention network to obtain a fused feature representation specifically includes: Perform feature alignment according to the financial time feature and the financial sentiment feature to obtain an effective time feature and an effective sentiment feature; Define network nodes according to the effective time features and the effective emotion features to obtain network feature nodes; Calculating the cosine similarity of the network feature nodes, and constructing network edges according to the calculated cosine similarity; Calculate the attention weights between network feature nodes based on multi-head attention; The network feature nodes are weightedly summed according to the attention weights and the network edges to obtain the fused feature representation.

6. The data fraud detection method according to claim 1, characterized in that: The step of merging the financial time feature and the financial sentiment feature into the fusion feature representation to obtain the financial information feature specifically includes: Standardizing the financial time feature and the financial sentiment feature to obtain a standard financial time feature and a standard financial sentiment feature; Performing feature selection on the standard financial time feature and the standard financial sentiment feature to obtain effective financial time feature and effective financial sentiment feature; Performing weighted fusion on the effective financial time feature and the effective financial sentiment feature to obtain a fusion feature vector; The fused feature vector and the fused feature representation are concatenated to obtain the financial information feature.

7. The data fraud detection method according to claim 1, characterized in that: Before the step of normalizing the financial information features to obtain financial effective features and inputting the financial effective features into a pre-trained fraud probability prediction model to obtain the fraud probability, the following steps are also included: Obtain fraudulent transaction sample data and normal transaction sample data; Preprocessing the fraudulent transaction sample data and the normal transaction sample data to obtain standard fraudulent transaction sample data and standard normal transaction sample data; The preset decision tree model is trained and optimized according to the standard fraudulent transaction sample data and the standard normal transaction sample data to obtain the fraud probability prediction model.

8. A data fraud detection device, characterized in that: include: A data acquisition module, used to acquire financial information data, parse the financial information data, and obtain financial time series data and financial text information data; A feature extraction module, used to extract features from the financial time series data and the financial text information data to obtain financial time features and financial sentiment features; A feature representation module, used for performing feature representation transformation on the financial time feature and the financial sentiment feature based on a graph attention network to obtain a fused feature representation; A feature fusion module, used for merging the financial time feature and the financial sentiment feature into the fused feature representation to obtain a financial information feature; The model prediction module is used to normalize the financial information features to obtain financial effective features, and input the financial effective features into a pre-trained fraud probability prediction model to obtain a fraud probability result.

9. A computer device, characterized in that: The method comprises a memory and a processor, wherein the memory stores computer-readable instructions, and the processor implements the steps of the data fraud detection method according to any one of claims 1 to 7 when executing the computer-readable instructions.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-readable instructions, and when the computer-readable instructions are executed by a processor, the steps of the data fraud detection method according to any one of claims 1 to 7 are implemented.

Images