Financial activity public opinion comprehensive analysis method based on multi-source data fusion

Abstract

The invention belongs to the technical field of financial monitoring and data analysis, and particularly relates to a financial activity public opinion comprehensive analysis method based on multi-source data fusion, and the method comprises the following steps: 1, a data collection and processing module collects information from news report multi-source data for data preprocessing; 2, a multi-source data fusion module embeds information features from different data sources through a graph convolutional network, and converts the information features into high-dimensional feature vectors; 3, a risk feature extraction module uses a deep neural network to carry out further processing on the fused high-dimensional feature vector, and potential risk features are extracted; 4, a risk assessment and early warning module assesses the risk probability of the financial activity through a classification model according to the extracted risk features, and sets a threshold to trigger early warning; and 5, the data security module adopts homomorphic encryption, differential privacy and distributed storage technologies for the data based on the risk assessment and early warning module to ensure the security of the data and the privacy of the privacy.

Description

Technical Field

[0001] The present invention relates to the technical field of financial monitoring and data analysis, and in particular to a comprehensive analysis method of public opinion on financial activities based on multi-source data fusion. Background Art

[0002] Traditional financial monitoring mainly relies on internal data of financial institutions or a small amount of public data, which makes it difficult to comprehensively and timely identify the potential risks of illegal financial activities. With the rise of diversified data sources such as social media, news reports and government data, how to effectively use these heterogeneous data sources for comprehensive analysis and improve the accuracy and timeliness of monitoring has become an important challenge for current financial supervision. At present, although there are some illegal financial activity monitoring technologies, most of them have the following problems:

[0003] 1. The data source is single and multi-source heterogeneous data cannot be fully integrated;

[0004] 2. Risk assessment and early warning mechanisms are not precise enough, lacking deep learning and real-time feedback;

[0005] 3. Lack of data security and privacy protection may lead to leakage of sensitive data.

[0006] Therefore, there is an urgent need for a new method and technology to improve the efficiency and accuracy of monitoring illegal financial activities and ensure the security of data through the integration and intelligent analysis of multi-dimensional data.

[0007] In response to the above-mentioned technical defects, a comprehensive analysis method solution for public opinion on financial activities based on multi-source data fusion is proposed. Summary of the invention

[0008] To solve the above problems, the present invention provides the following technical solutions:

[0009] The comprehensive analysis system of public opinion on financial activities based on multi-source data fusion includes a data collection and preprocessing module, a multi-source heterogeneous data fusion module, a risk feature extraction module, a risk assessment and early warning mechanism module, and a data security assurance module. The data collection and preprocessing module collects information from multi-source data such as social media, news reports, and financial regulatory data, and cleans and standardizes the data through the module, removes irrelevant information, and converts it into structured data;

[0010] The multi-source heterogeneous data fusion module includes a graph convolutional network, which embeds and fuses features of information from different data sources through the graph convolutional network based on the data processed by the data acquisition and preprocessing module, and converts it into a unified high-dimensional feature vector;

[0011] The risk feature extraction module includes a deep neural network model. Based on the data fused by the multi-source heterogeneous data fusion module, the risk feature extraction module further processes the fused features based on the deep neural network model to extract risk features related to illegal financial activities.

[0012] The risk assessment and early warning mechanism module includes a classification model (multi-layer perceptron), which assesses the risk based on the risk features extracted by the risk feature extraction module, quantifies the risk using the classification model (multi-layer perceptron (MLP)), calculates the probability P of illegal financial activities, and triggers an early warning according to a set threshold;

[0013] The data security assurance module adopts homomorphic encryption, differential privacy and distributed storage technology to ensure the security of data and the confidentiality of privacy.

[0014] Furthermore, the data collection and preprocessing module includes a data collection unit and a preprocessing unit, wherein the data collection unit collects financial public opinion data, social public data, news report data, and social media data from multi-source heterogeneous data sources, and defines the collected data as D;

[0015] The preprocessing unit cleans and standardizes the data, removes irrelevant information, and converts the data into structured data;

[0016] The preprocessing unit includes data cleaning, standardization, format conversion and feature extraction; the specific processing steps of the preprocessing unit are as follows:

[0017] Step 1: Clean the collected data to remove noise, duplicate data and irrelevant data;

[0018] Step 2: For text data, it is necessary to perform word segmentation, remove stop words and stemming for subsequent feature extraction;

[0019] Step 3: For unstructured data (such as text or images), convert it into structured feature vectors through natural language processing (NLP) or computer vision (CV) technology to ensure that the data can be effectively processed by the model;

[0020] Step 4: Convert each piece of financial event or public opinion information into a feature vector as input for subsequent analysis. Use the TF-IDF (term frequency-inverse document frequency) method to convert text into numerical features, or use image processing methods to extract visual features of images.

[0021] Step 5: Define the processed data as D'=Preprocess(D).

[0022] Furthermore, the fusion unit of the multi-source heterogeneous data includes a graph convolutional network, which embeds heterogeneous data from different sources to ensure that they can be effectively compared and fused in the same feature space. The specific analysis formula steps are as follows:

[0023] Z f =GCN(A f ,D f ),Z s =GCN(A s ,D s )

[0024] A f , A s Wherein, D is the adjacency matrix of financial public opinion data and social public data, respectively. f , D s are the characteristic matrices of financial public opinion data and social public data, Z f , Z s Represented as the embedded representation obtained after processing through a graph convolutional network (GCN).

[0025] Furthermore, the multi-source heterogeneous data fusion unit analyzes the data by embedding heterogeneous data from different sources based on a graph convolutional network, and the multi-source heterogeneous data fusion unit obtains a unified multi-dimensional representation through weighted fusion:

[0026] X=αZ f +βZ s

[0027] Where α and β are weight coefficients, and α+β=1.

[0028] Furthermore, the risk feature extraction module includes a deep neural network model, which is based on the data fused by the multi-source heterogeneous data fusion module, and uses a deep neural network (DNN) method to extract the risk feature from X = αZ f +βZ s The risk feature R associated with illegal financial activities is extracted from the multi-dimensional representation, and trusted and secure calculations are performed to ensure the security of sensitive data.

[0029] R = DNN(X)

[0030] Where X represents the feature vector after multi-source data fusion, which contains information from different data sources, DNN(·) represents the deep neural network (DNN) model, and R is the risk feature extracted from the fused feature vector, indicating the risk level of potential illegal financial activities.

[0031] Furthermore, the risk assessment and early warning mechanism module includes a risk assessment unit and an early warning mechanism unit. The risk assessment includes a classification model (multi-layer perceptron), which performs risk assessment based on the risk features extracted by the risk feature extraction module, quantifies the risk using the classification model (multi-layer perceptron (MLP)), calculates the probability P of illegal financial activities, and triggers an early warning according to a set threshold. The calculation formula of the classification model is as follows:

[0032] P=(Risk|R)=σ(W T R+b)

[0033] Where σ is the activation function, W and b are the weight and bias of the model respectively, and P = (Risk|R) represents the predicted probability of the risk of illegal financial activities.

[0034] Specifically, the early warning mechanism unit analyzes the risk assessment data as follows:

[0035]

[0036] Where P = (Risk|R) represents the risk probability value output by the risk assessment model, where θ low ,θ hight Indicates the set high-risk and low-risk thresholds, and Alert indicates whether an alert is triggered.

[0037] Furthermore, the early warning mechanism unit specifically analyzes and further includes the following steps:

[0038] The early warning mechanism unit has its corresponding three warning levels as green, yellow and red colors Monitor, False and True, and works through the warning levels to ensure that the system can take appropriate actions at different risk levels.

[0039] Furthermore, the data security assurance module includes a homomorphic encryption unit, a differential privacy unit and a distributed storage and access control unit: the data security assurance module uses homomorphic encryption, differential privacy and distributed storage technology based on the risk assessment and early warning mechanism module data to ensure the security of data and the confidentiality of privacy;

[0040] The homomorphic encryption unit: allows calculations to be performed directly on encrypted data, avoiding exposure of the original data;

[0041] The differential privacy unit: ensures that the user's private information will not be leaked during the data analysis process by adding noise;

[0042] The distributed storage and access control unit ensures that sensitive data is only accessible to authorized users by distributing data storage and using a role-based access control (RBAC) mechanism.

[0043] Furthermore, the method for comprehensive analysis of public opinion on financial activities based on multi-source data fusion is characterized in that the steps of using the comprehensive analysis system for public opinion on financial activities based on multi-source data fusion described in any one of claims 1 to 9 are as follows:

[0044] Step 1: Use the data collection and processing module to collect information from multiple sources such as social media, news reports, and financial regulatory data, clean and standardize the data, remove irrelevant information, and convert it into structured data;

[0045] Step 2: Use the multi-source data fusion module to embed features from different data sources through the graph convolutional network (GCN) and convert them into high-dimensional feature vectors;

[0046] Step 3: Use the risk feature extraction module to further process the fused high-dimensional feature vector using a deep neural network to extract potential risk features;

[0047] Step 4: Use the risk assessment and early warning module to evaluate the risk probability of financial activities based on the extracted risk features through the classification model, and trigger an early warning based on the set threshold;

[0048] Step 5: Use the data security module to apply homomorphic encryption, differential privacy and distributed storage technology to the data based on the risk assessment and early warning module to ensure the security of the data and the privacy of the privacy.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] 1. In the comprehensive analysis method of public opinion on financial activities based on multi-source data fusion of the present invention, multi-source heterogeneous data fusion technology can more comprehensively and accurately identify potential illegal financial activities.

[0051] 2. In the comprehensive analysis method of public opinion on financial activities based on multi-source data fusion, this technology supports real-time risk assessment and early warning, and can issue warnings at the early stage of illegal financial activities, thereby helping regulatory authorities to take timely measures;

[0052] 3. In the comprehensive analysis method of public opinion on financial activities based on multi-source data fusion, the present invention uses homomorphic encryption and differential privacy technology during data processing to ensure the security of sensitive data and prevent data leakage;

[0053] 4. The comprehensive analysis method of public opinion on financial activities based on multi-source data fusion in the present invention utilizes advanced technologies such as multi-dimensional learning and graph convolutional networks to deeply explore the potential correlations between different data sources and achieve more sophisticated analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to facilitate understanding by those skilled in the art, the present invention is further described below in conjunction with the accompanying drawings;

[0055] Figure 1 It is a schematic diagram of the overall steps of the comprehensive analysis method of public opinion on financial activities based on multi-source data fusion of the present invention. DETAILED DESCRIPTION

[0056] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0057] like Figure 1 As shown in the figure, the comprehensive analysis system of public opinion on financial activities based on multi-source data fusion includes a data collection and preprocessing module, a multi-source heterogeneous data fusion module, a risk feature extraction module, a risk assessment and early warning mechanism module, and a data security assurance module. The data collection and preprocessing module collects information from multi-source data such as social media, news reports, and financial regulatory data, and cleans and standardizes the data through this module, removes irrelevant information, and converts it into structured data.

[0058] The multi-source heterogeneous data fusion module includes a graph convolutional network. Based on the data processed by the data acquisition and preprocessing modules, the graph convolutional network is used to embed and fuse the information from different data sources and convert it into a unified high-dimensional feature vector.

[0059] The risk feature extraction module includes a deep neural network model. Based on the data fused by the multi-source heterogeneous data fusion module, the risk feature extraction module further processes the fused features based on the deep neural network model to extract risk features related to illegal financial activities.

[0060] The risk assessment and early warning mechanism module includes a classification model (multi-layer perceptron), which assesses the risk based on the risk features extracted by the risk feature extraction module, quantifies the risk using the classification model (multi-layer perceptron (MLP)), calculates the probability P of illegal financial activities, and triggers an early warning based on the set threshold;

[0061] The data security assurance module uses homomorphic encryption, differential privacy and distributed storage technology to ensure the security of data and the confidentiality of privacy.

[0062] Specifically, the data collection and preprocessing module includes a data collection unit and a preprocessing unit. The data collection unit collects financial public opinion data, social public data, news report data, and social media data from multi-source heterogeneous data sources, and defines the collected data as D;

[0063] The preprocessing unit includes data cleaning, standardization, format conversion and feature extraction; the specific processing steps of the preprocessing unit are as follows:

[0064] Step 1: Clean the collected data to remove noise, duplicate data and irrelevant data;

[0065] Step 2: For text data, it is necessary to perform word segmentation, remove stop words and stemming for subsequent feature extraction;

[0066] Step 3: For unstructured data (such as text or images), convert it into structured feature vectors through natural language processing (NLP) or computer vision (CV) technology to ensure that the data can be effectively processed by the model;

[0067] Step 4: Convert each piece of financial event or public opinion information into a feature vector as input for subsequent analysis. Use the TF-IDF (term frequency-inverse document frequency) method to convert text into numerical features, or use image processing methods to extract visual features of images.

[0068] Step 5: Define the processed data as D'=Preprocess(D).

[0069] By cleaning the data, the data can be made consistent, interference factors can be removed, the quality and effectiveness of the data can be improved, and subsequent feature extraction and analysis can be facilitated.

[0070] Specifically, the fusion unit of multi-source heterogeneous data includes a graph convolutional network, which embeds heterogeneous data from different sources to ensure that they can be effectively compared and fused in the same feature space. The specific analysis formula steps are as follows:

[0071] Z f =GCN(A f ,D f ),Z s =GCN(A s ,D s )

[0072] A f , A s Wherein, D is the adjacency matrix of financial public opinion data and social public data, respectively. f , D s are the characteristic matrices of financial public opinion data and social public data, Z f , Zs Represented as the embedded representation obtained after processing through a graph convolutional network (GCN).

[0073] A f , A s Especially the adjacency matrix of financial public opinion data and social public data. The adjacency matrix describes the relationship between data sources, for example, the connection or similarity between different data nodes (such as financial events or people in social media);

[0074] D f , D s They are the feature matrices of financial public opinion data and social public data, which contain the attributes or features of each data node (such as text content, risk score, etc.);

[0075] Z f , Z s Represented as an embedded representation obtained after processing through a graph convolutional network (GCN). GCN is a graph-based neural network that can effectively process complex data with graph structures. After processing through GCN, the representation of the data will be updated according to the relationship between data nodes and their characteristics;

[0076] Through graph convolutional networks, multi-source data (such as financial data and social public data) are converted into high-dimensional embedding representations to capture the complex relationships in the data and provide a basis for subsequent multi-source data fusion.

[0077] Specifically, the fusion unit of multi-source heterogeneous data analyzes the data by embedding heterogeneous data from different sources based on the graph convolutional network, and obtains a unified multi-dimensional representation through weighted fusion:

[0078] X=αZ f +βZ s

[0079] Among them, α and β are weight coefficients, which are used to control the importance of financial data and social data in the fusion process, and satisfy α+β=1. Among them, X represents the fused feature vector, which is the final multi-source data fusion result, covering information from different data sources, and Z f , Z s It is represented as a high-dimensional feature vector obtained after graph convolutional network processing; information from different data sources is merged into a unified feature vector X through weighted fusion, which provides basic data for subsequent risk assessment and early warning mechanisms.

[0080] Specifically, the risk feature extraction module includes a deep neural network model, which uses a deep neural network (DNN) method to extract the risk feature from X = αZ based on the data fused by the multi-source heterogeneous data fusion module. f +βZ sThe risk feature R associated with illegal financial activities is extracted from the multi-dimensional representation, and trusted and secure calculations are performed to ensure the security of sensitive data.

[0081] R = DNN(X)

[0082] Where X represents the feature vector after multi-source data fusion, which contains information from different data sources, DNN(·) represents the deep neural network (DNN) model, and R is the risk feature extracted from the fused feature vector, indicating the risk level of potential illegal financial activities.

[0083] Deep neural networks extract risk features related to illegal financial activities by training and reasoning on multi-source fusion data. These features may include abnormal patterns of financial transactions and risk signals in social media.

[0084] Specifically, the risk assessment and early warning mechanism module includes a risk assessment unit and an early warning mechanism unit. The risk assessment includes a classification model (multi-layer perceptron), which assesses the risk based on the risk features extracted by the risk feature extraction module, quantifies the risk using the classification model (multi-layer perceptron (MLP)), calculates the probability P of illegal financial activities, and triggers an early warning based on the set threshold. The calculation formula of the classification model is as follows:

[0085] P=(Risk|R)=σ(W T R+b)

[0086] Where σ is the activation function, which generally uses functions such as Sigmoid or ReLU to ensure that the output value is within the specified range. W and b are the weight and bias of the model, respectively. W contains the parameters learned in the neural network or classification model, and b helps the model to make translation adjustments. P = (Risk|R) represents the predicted probability of the risk of illegal financial activities, which indicates the probability of illegal financial activities occurring under a given feature R. R is the risk feature vector extracted from the fused feature vector, which contains the risk features related to illegal financial activities extracted after multi-source data fusion;

[0087] Formulate a typical classification problem, in which the extracted risk feature R is converted into a probability value P = (Risk|R) = σ(W T R+b) represents the probability of an event occurring.

[0088] Specifically, the early warning mechanism unit analyzes the risk assessment data as follows:

[0089]

[0090] Where P = (Risk|R) represents the risk probability value output by the risk assessment model, where θ low ,θ hight Indicates the high-risk and low-risk thresholds that are set. Alert indicates whether an alert is triggered. Monitor, False, and True indicate continuous monitoring, no alert, and triggering of an alert, respectively.

[0091] Specifically, the early warning mechanism unit specific analysis also includes the following steps:

[0092] The early warning mechanism unit assigns its corresponding three warning levels to the three colors Monitor, False, and True, respectively, in green, yellow, and red. It works through the warning levels to ensure that the system can take appropriate actions at different risk levels;

[0093] When the risk probability is greater than the high risk threshold θ hight , the system triggers the emergency warning Alert = True, and the red light turns on;

[0094] When the risk probability is between low risk and high risk, the system maintains the monitoring state Alert = Monitor and the green light is on;

[0095] When the risk probability is less than the low risk threshold θ low When it is considered that there is no risk, Alert = False, the system does not trigger an early warning, and the yellow light is on;

[0096] The early warning mechanism sets different thresholds to ensure that the system can take appropriate actions at different risk levels.

[0097] Specifically, the data security assurance module includes a homomorphic encryption unit, a differential privacy unit, and a distributed storage and access control unit: the data security assurance module is based on the risk assessment and early warning mechanism module data, and uses homomorphic encryption, differential privacy, and distributed storage technologies to ensure data security and privacy;

[0098] Homomorphic encryption unit: allows calculations to be performed directly on encrypted data, avoiding exposure of the original data;

[0099] Differential privacy unit: By adding noise, it ensures that the user's private information will not be leaked during the data analysis process;

[0100] Distributed storage and access control unit: By distributing data storage and using a role-based access control (RBAC) mechanism, sensitive data is ensured to be accessible only to authorized users.

[0101] Specifically, the method for comprehensive analysis of public opinion on financial activities based on multi-source data fusion adopts the system for comprehensive analysis of public opinion on financial activities based on multi-source data fusion described in any one of claims 1 to 9, and the steps are as follows:

[0102] Step 1: Use the data collection and processing module to collect information from multiple sources such as social media, news reports, and financial regulatory data, clean and standardize the data, remove irrelevant information, and convert it into structured data;

[0103] Step 2: Use the multi-source data fusion module to embed features from different data sources through the graph convolutional network (GCN) and convert them into high-dimensional feature vectors;

[0104] Step 3: Use the risk feature extraction module to further process the fused high-dimensional feature vector using a deep neural network to extract potential risk features;

[0105] Step 4: Use the risk assessment and early warning module to evaluate the risk probability of financial activities based on the extracted risk features through the classification model, and trigger an early warning based on the set threshold;

[0106] Step 5: Use the data security module to apply homomorphic encryption, differential privacy and distributed storage technology to the data based on the risk assessment and early warning module to ensure the security of the data and the privacy of the privacy.

[0107] Working principle: Using the data collection and processing module, information D is collected from multiple sources of data such as social media, news reports, and financial regulatory data. The data is cleaned and standardized, irrelevant information is removed, and converted into structured data D'=Preprocess(D); then the multi-source data fusion module is used to embed features of information from different data sources through a graph convolutional network (GCN) and convert it into a high-dimensional feature vector. The specific analysis formula steps are as follows:

[0108] Z f =GCN(A f ,D f ),Z s =GCN(A s ,D s )

[0109] A f , A s Wherein, D is the adjacency matrix of financial public opinion data and social public data, respectively. f , D s are the characteristic matrices of financial public opinion data and social public data, Z f , Z sIt is represented as an embedded representation obtained after processing by a graph convolutional network (GCN). The fusion unit of multi-source heterogeneous data analyzes the data by embedding heterogeneous data from different sources based on the graph convolutional network. The fusion unit of multi-source heterogeneous data obtains a unified multi-dimensional representation through weighted fusion:

[0110] X=αZ f +βZ s

[0111] Among them, α and β are weight coefficients, and they satisfy α+β=1;

[0112] Then, the risk feature extraction module uses a deep neural network to further process the fused high-dimensional feature vector to extract potential risk features and ensure the security of sensitive data.

[0113] R = DNN(X)

[0114] Then, the risk assessment and early warning module uses the extracted risk features to evaluate the risk probability of financial activities through the classification model, and triggers an early warning based on the set threshold, P = (Risk|R) = σ (W T R+b)

[0115] Where σ is the activation function, W and b are the weight and bias of the model respectively, P = (Risk|R) represents the predicted probability of the risk of illegal financial activities. The early warning mechanism unit is specifically analyzed based on the risk assessment data as follows:

[0116]

[0117] Where P = (Risk|R) represents the risk probability value output by the risk assessment model, where θ low ,θ hight It represents the set high-risk and low-risk thresholds. Alert indicates whether the warning is triggered. The warning mechanism unit corresponds to three warning levels: Monitor, False, and True, which are green, yellow, and red. It works through the warning levels to ensure that the system can take appropriate actions at different risk levels. After that, the data security module uses homomorphic encryption, differential privacy, and distributed storage technology for the data based on the risk assessment and warning module to ensure data security and privacy.

[0118] The preferred embodiments of the present invention disclosed above are only used to help explain the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to only specific implementation methods. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A comprehensive analysis system of public opinion on financial activities based on multi-source data fusion, characterized by: It includes a data collection and preprocessing module, a multi-source heterogeneous data fusion module, a risk feature extraction module, a risk assessment and early warning mechanism module, and a data security assurance module. The data collection and preprocessing module collects information from social media, news reports, and financial regulatory data and uses the module to clean and standardize the data, remove irrelevant information, and convert it into structured data; The multi-source heterogeneous data fusion module includes a graph convolutional network, which embeds and fuses features of information from different data sources through the graph convolutional network based on the data processed by the data acquisition and preprocessing module, and converts it into a unified high-dimensional feature vector; The risk feature extraction module includes a deep neural network model. Based on the data fused by the multi-source heterogeneous data fusion module, the risk feature extraction module further processes the fused features based on the deep neural network model to extract risk features related to illegal financial activities. The risk assessment and early warning mechanism module includes a classification model (multi-layer perceptron), which assesses the risk based on the risk features extracted by the risk feature extraction module, quantifies the risk using the classification model (multi-layer perceptron (MLP)), calculates the probability P of illegal financial activities, and triggers an early warning according to a set threshold; The data security assurance module adopts homomorphic encryption, differential privacy and distributed storage technology to ensure the security of data and the confidentiality of privacy.

2. The comprehensive analysis system of public opinion on financial activities based on multi-source data fusion according to claim 1 is characterized in that: The data collection and preprocessing module includes a data collection unit and a preprocessing unit. The data collection unit collects financial public opinion data, social public data, news report data, and social media data from multi-source heterogeneous data sources, and defines the collected data as D; The preprocessing unit cleans and standardizes the data, removes irrelevant information, and converts the data into structured data; The preprocessing unit includes data cleaning, standardization, format conversion and feature extraction; the specific processing steps of the preprocessing unit are as follows: Step 1: Clean the collected data to remove noise, duplicate data and irrelevant data; Step 2: For text data, it is necessary to perform word segmentation, remove stop words and stemming for subsequent feature extraction; Step 3: For unstructured data (such as text or images), convert it into structured feature vectors through natural language processing (NLP) or computer vision (CV) technology to ensure that the data can be effectively processed by the model; Step 4: Convert each piece of financial event or public opinion information into a feature vector as input for subsequent analysis. Use the TF-IDF (term frequency-inverse document frequency) method to convert text into numerical features, or use image processing methods to extract visual features of images. Step 5: Define the processed data as D'=Preprocess(D).

3. The comprehensive analysis system of public opinion on financial activities based on multi-source data fusion according to claim 2 is characterized in that: The multi-source heterogeneous data fusion unit includes a graph convolutional network, which embeds heterogeneous data from different sources to ensure that they can be effectively compared and fused in the same feature space. The specific analysis formula steps are as follows: Z f =GCN(A f ,D f ),Z s =GCN(A s ,D s ) A f , A s Wherein, D is the adjacency matrix of financial public opinion data and social public data, respectively. f , D s are the characteristic matrices of financial public opinion data and social public data, Z f , Z s Represented as the embedded representation obtained after processing through a graph convolutional network (GCN).

4. The comprehensive analysis system of public opinion on financial activities based on multi-source data fusion according to claim 3 is characterized in that: The multi-source heterogeneous data fusion unit analyzes the data by embedding heterogeneous data from different sources based on a graph convolutional network, and obtains a unified multi-dimensional representation through weighted fusion: X=αZ f +βZ s Where α and β are weight coefficients, and α+β=1.

5. The financial activity public opinion comprehensive analysis system based on multi-source data fusion according to claim 4 is characterized in that: The risk feature extraction module includes a deep neural network model, which is based on the data fused by the multi-source heterogeneous data fusion module, and uses a deep neural network (DNN) method to extract the risk feature from X=αZ f +βZ s The risk feature R associated with illegal financial activities is extracted from the multi-dimensional representation, and trusted and secure calculations are performed to ensure the security of sensitive data. R = DNN(X) Where X represents the feature vector after multi-source data fusion, which contains information from different data sources, DNN(·) represents the deep neural network (DNN) model, and R is the risk feature extracted from the fused feature vector, indicating the risk level of potential illegal financial activities.

6. The financial activity public opinion comprehensive analysis system based on multi-source data fusion according to claim 5 is characterized in that: The risk assessment and early warning mechanism module includes a risk assessment unit and an early warning mechanism unit. The risk assessment includes a classification model (multi-layer perceptron), which performs risk assessment based on the risk features extracted by the risk feature extraction module, quantifies the risk using the classification model (multi-layer perceptron (MLP)), calculates the probability P of illegal financial activities, and triggers an early warning according to a set threshold. The calculation formula of the classification model is as follows: P=(Risk|R)=σ(W T R+b) Where σ is the activation function, W and b are the weight and bias of the model respectively, and P = (Risk|R) represents the predicted probability of the risk of illegal financial activities.

7. The comprehensive analysis system of public opinion on financial activities based on multi-source data fusion according to claim 6 is characterized in that: The early warning mechanism unit specifically analyzes the risk assessment data as follows: Where P = (Risk|R) represents the risk probability value output by the risk assessment model, where θ low ,θ hight Indicates the set high-risk and low-risk thresholds, Alert indicates whether an alert is triggered, Monitor, False, and True indicate continuous monitoring, no alert, and triggering of an alert, respectively.

8. The financial activity public opinion comprehensive analysis system based on multi-source data fusion according to claim 7 is characterized in that: The specific analysis of the early warning mechanism unit also includes the following steps: The early warning mechanism unit has its corresponding three warning levels as green, yellow and red colors Monitor, False and True, and works through the warning levels to ensure that the system can take appropriate actions at different risk levels.

9. The financial activity public opinion comprehensive analysis system based on multi-source data fusion according to claim 8 is characterized in that: The data security assurance module includes a homomorphic encryption unit, a differential privacy unit and a distributed storage and access control unit: the data security assurance module uses homomorphic encryption, differential privacy and distributed storage technology based on the risk assessment and early warning mechanism module data to ensure data security and privacy; The homomorphic encryption unit: allows calculations to be performed directly on encrypted data, avoiding exposure of the original data; The differential privacy unit: ensures that the user's private information will not be leaked during the data analysis process by adding noise; The distributed storage and access control unit ensures that sensitive data is only accessible to authorized users by distributing data storage and using a role-based access control (RBAC) mechanism.

10. The method for comprehensive analysis of public opinion on financial activities based on multi-source data fusion according to claim 9 is characterized in that: The steps of using the financial activity public opinion comprehensive analysis system based on multi-source data fusion as described in any one of claims 1 to 9 are as follows: Step 1: Use the data collection and processing module to collect information from multiple sources such as social media, news reports, and financial regulatory data, clean and standardize the data, remove irrelevant information, and convert it into structured data; Step 2: Use the multi-source data fusion module to embed features from different data sources through the graph convolutional network (GCN) and convert them into high-dimensional feature vectors; Step 3: Use the risk feature extraction module to further process the fused high-dimensional feature vector using a deep neural network to extract potential risk features; Step 4: Use the risk assessment and early warning module to evaluate the risk probability of financial activities based on the extracted risk features through the classification model, and trigger an early warning based on the set threshold; Step 5: Use the data security module to apply homomorphic encryption, differential privacy and distributed storage technology to the data based on the risk assessment and early warning module to ensure the security of the data and the privacy of the privacy.

Images