Financial market trend prediction and optimization transaction system based on machine learning

Abstract

The invention provides a financial market trend prediction and optimization transaction system based on machine learning. Feature extraction is performed on the user behavior data, the financial product data and the market dynamic data through a feature extraction module, static features and time features are obtained, and user historical behavior features and user feedback features are obtained. And combining the static features and the time features through a vector construction module to obtain a time perception vector, and constructing feedback behavior data in the user behavior data into a user feedback vector. And training the constructed TAFM model by using the time perception vector, the user feedback vector and the user transaction result data to obtain a prediction model. And predicting the new data through the prediction model, and recommending the financial product to the user according to a prediction result. The system can provide more accurate and personalized financial product recommendation service for users in a complex and changeable financial market.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a financial market trend prediction and trading optimization system based on machine learning. Background Art

[0002] With the rapid development of the financial industry and the gradual increase in per capita income, consumers' financial needs tend to be diversified, and personal wealth management has shifted from a single savings deposit product to a product configuration such as funds, bonds, wealth management and insurance. At the same time, the number of bank customers is huge and the financial products are rich. The traditional recommendation method that relies on account managers can no longer meet customers' personalized financial service needs. As an important part of bank development, retail business is gradually transforming from high-speed to high-quality development driven by big data. The "customer-centric" development model has been upgraded again. In the new development stage, digital drive and intelligent drive have become new engines for retail growth, and many researchers are also actively exploring the application of artificial intelligence in the financial field. How to provide tens of millions or even hundreds of millions of retail customers with the opportunity to accelerate the deep integration of new generation information technology and manufacturing is the only way to adapt to the trend of the new round of scientific and technological revolution and industrial transformation and promote the high-quality development of the manufacturing industry. Intelligent transformation and digital transformation diagnosis can comprehensively evaluate the current development level of enterprises, clarify the gaps and shortcomings, formulate transformation paths according to the enterprise, guide enterprises to benchmark advanced and break through development bottlenecks. The diagnostic results can provide accurate basis for enterprise policy formulation, optimize resource allocation, and create a good ecology conducive to the digital transformation of the manufacturing industry.

[0003] In recent years, with the development of big data and related technologies, by using data mining technology and intelligent recommendation algorithms, users' financial management preferences and consumption habits can be analyzed from massive user data, and universal data laws can be extracted and personalized data models can be constructed to recommend products that users may like, which provides a solution for personalized financial services for bank customers. However, commercial bank financial products have the characteristics of serious homogeneity and concentrated customer purchase of products. In addition, most traditional recommendation systems use algorithms such as LR, GBDT, and collaborative filtering, which have insufficient expression capabilities and weak ability to capture interactive information between tens of millions of customers and products. The algorithm itself is also difficult to solve the problem of excessive concentration of head effects and lack of generalization, which makes the product recommendation structure tend to be homogeneous when banks recommend products to customers, which is not conducive to the bank's liquidity risk control. Therefore, how to improve the customer product recommendation service experience, improve the trend of customer product purchase concentration, help customers understand more products that better match their financial needs, and how to facilitate customers to trade products have become urgent problems to be solved. Summary of the invention

[0004] The present invention describes a financial market trend prediction and trading optimization system based on machine learning, which can solve the above technical problems.

[0005] According to a first aspect, a financial market trend prediction and trading optimization system based on machine learning comprises:

[0006] A data acquisition module, used to acquire user behavior data, financial product data, market dynamics data, and user transaction result data; wherein the user behavior data, financial product data, and market dynamics data are all recorded with corresponding time information, and the user behavior data includes historical behavior events and user feedback behavior data on financial products;

[0007] The feature extraction module is used to extract features from the three types of data obtained from the data acquisition module, obtain the static features and time features corresponding to the three types of data, and obtain the historical behavior features and user feedback features of the user behavior data; the historical behavior features are used to introduce the time decay mechanism into the TAFM model, and the user feedback features are used to introduce the user feedback mechanism into the TAFM model;

[0008] A vector construction module is used to combine the static features and time features of user behavior data, financial product data, and market dynamic data to obtain the time perception vectors corresponding to the three types of data, and to construct the feedback behavior data in the user behavior data into a user feedback vector;

[0009] A model building module is used to build a TAFM model, wherein the TAFM model includes an input layer, an embedding layer, a time decay layer, and a factor decomposition machine layer; wherein the input layer is used to use the time perception vectors and user feedback vectors corresponding to the three types of data as inputs of the TAFM model; the embedding layer is used to convert the time perception vectors and user feedback vectors corresponding to the three types of data into latent vector representations of fixed lengths; the time decay layer is used to adjust the importance of historical behavior features to update the time perception vectors; the factor decomposition machine layer includes a linear calculation layer, an interactive calculation layer, and an output layer, wherein the linear calculation layer is used to perform linear calculations on the latent vectors corresponding to the time perception vectors and user feedback vectors processed by the time decay layer, the interactive calculation layer is used to perform second-order interactive calculations on the latent vectors corresponding to the time perception vectors and user feedback vectors processed by the time decay layer, and the output layer is used to combine the outputs of the linear calculation layer and the interactive calculation layer to generate a final prediction value;

[0010] The model training module is used to input the time perception vector, user feedback vector and user transaction result data corresponding to the three types of data into the constructed TAFM model, train the TAFM model and obtain a prediction model;

[0011] Model prediction module, which is used to input new user behavior data, financial product data and market dynamic data into the prediction model and output the model prediction results;

[0012] The financial product recommendation module is used to recommend financial products to users based on the prediction results of the model.

[0013] In some embodiments, the method of using the time decay layer to adjust the importance of the historical behavior feature to update the time perception vector specifically includes:

[0014] Based on each historical event, calculate the difference between the current time and the time when the event occurred;

[0015] Apply exponential decay function or linear decay function to calculate the weight of historical behavior characteristics;

[0016] For each historical behavior event, multiply its historical behavior feature vector by the corresponding time decay weight to adjust the historical behavior feature vector;

[0017] The adjusted historical behavior feature vector replaces the original feature vector to update the time-aware vector.

[0018] In some embodiments, a visual report is generated based on the prediction results of the model.

[0019] In some embodiments, the financial product recommendation module also includes a user quick transaction entry, which is used to facilitate users to conduct quick transactions based on pushed financial products.

[0020] In some embodiments, the financial product recommendation module further includes a user feedback entry, which is used to facilitate users to evaluate the recommendation results of the recommended financial products.

[0021] In some embodiments, the financial product recommendation module further includes a user favorite mark button, which is used to facilitate users to collect or mark products of interest.

[0022] The present invention provides a financial market trend prediction and optimization trading system based on machine learning. Through the feature extraction module, this solution can extract rich features from multiple data sources, including static features, time features, historical behavior features and user feedback features, to provide more information support for the model. Through the time-aware vector, the model can capture the trend of user behavior and market dynamics changing over time, reflecting the changes in user interests and market environment. User feedback features directly reflect the user's satisfaction and preference for financial products. By analyzing these feedbacks, the model can more accurately understand the user's actual needs and preferences. The financial product recommendation system constructed by the above method has comprehensive data coverage, powerful feature extraction capabilities, time-aware mechanism, user feedback mechanism, efficient model construction and training, accurate prediction and recommendation, optimized user experience and strict security and compliance guarantees. These advantages enable the system to provide users with more accurate, personalized and secure financial product recommendation services in complex and changing financial markets. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0024] Figure 1 A schematic diagram of a process framework of a financial market trend prediction and optimized trading system based on machine learning provided in an embodiment of this specification is shown. DETAILED DESCRIPTION

[0025] The solution provided in this specification is described below in conjunction with the accompanying drawings.

[0026] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0027] In the description of the embodiments of the present application, words such as "exemplary", "for example" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary", "for example" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary", "for example" or "for example" is intended to present related concepts in a concrete way.

[0028] In the description of the embodiments of the present application, the term "and / or" is merely a description of the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B may represent: A exists alone, B exists alone, and A and B exist at the same time. In addition, unless otherwise specified, the term "plurality" means two or more.

[0029] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. The terms "include", "comprises", "has" and their variations all mean "including but not limited to", unless otherwise specifically emphasized.

[0030] Figure 1 A schematic diagram of a financial market trend prediction and trading optimization system process framework based on machine learning provided in an embodiment of this specification is shown, such as Figure 1 As shown, the system includes:

[0031] Data acquisition module S101, data acquisition module, is used to acquire user behavior data, financial product data, market dynamics data and user transaction result data. Specifically, the user behavior data, financial product data and market dynamics data are all recorded with corresponding time information, and the user behavior data includes historical behavior events and user feedback behavior data on financial products.

[0032] Specifically, user behavior data can be obtained from the company's internal user behavior record system, such as user click, purchase, favorite and other behavior logs. Its user behavior data includes but is not limited to basic user information, interaction with financial products, transaction behavior, social behavior, feedback on financial products, geographic location information, and time information. Among them, basic user information includes user ID, registration channel, registration time, etc., interaction with financial products includes page access records, click records, search records, collection records, sharing records, etc., transaction behavior includes purchase records, redemption records, transfer records, investment records, etc., social behavior includes attention records, like records, comment interaction, etc., and feedback on financial products includes rating records, comment records, complaint records, customer service records, etc. Specifically, user historical behavior events refer to specific behavior events described in the user's interaction with financial products, transaction behavior, social behavior, and feedback on financial products. For example, in the interaction with financial products, the user clicks on a certain financial product. The specific information of this event can be obtained from the click record. Time information includes the timestamp of the user's historical behavior event, the start time and end time of the user in a session, and the user's active time period on the platform.

[0033] Financial product data can be obtained from the record system of financial products within the financial institution, and can be obtained through database query, API interface, etc. Specifically, financial product data includes but is not limited to basic information of financial products, product details, price information, market performance, user feedback results data on financial products, technical indicators, and transaction records. Among them, the basic information of financial products at least includes product ID, product name, product type, issuing institution, and issuance time. Product details at least include product description, risk level, expected return, and investment period. Price information at least includes current price, historical price, net value, transaction fee, management fee, and custody fee. Market performance at least includes historical performance, ranking in the same category, and credit rating. User feedback results data on financial products at least includes user ratings of financial products, user comments on financial products, and user feedback and opinions on financial products. Technical indicators at least include technical analysis indicators of financial products, such as MACD, RSI, etc., and financial indicators of financial products, such as price-earnings ratio, price-to-book ratio, etc. Transaction records at least include transactions of financial products, transaction prices of financial products, and buy and sell order information of financial products. Specifically, financial product data records have corresponding time information. The time information includes the timestamp corresponding to the financial product data record.

[0034] Market dynamics data refers to information that reflects changes and trends in the financial markets, which can be obtained from financial news websites, international organizations, statistical departments and other channels. Specifically, market dynamics data includes but is not limited to macroeconomic indicators, market indices, industry trends, news reports, and policy changes.

[0035] User transaction result data refers to the result records of users' transaction activities on the financial platform, which are used as labels for model training. In other words, user transaction result data is used as output data for model training.

[0036] The feature extraction module S102 is used to extract features from the three types of data obtained from the data acquisition module, obtain static features and time features corresponding to the three types of data, and obtain historical behavior features and user feedback features of the user behavior data. Specifically, the historical behavior features are used to introduce a time decay mechanism into the AFM model, and the user feedback features are used to introduce a user feedback mechanism into the AFM model.

[0037] Before extracting features from data, you can preprocess the data. First, clean the data to remove duplicates, anomalies, and missing values, then format the cleaned data to unify the data format and ensure data consistency. After preprocessing the data, extract features from it.

[0038] Specifically, for user behavior data, first, static features are extracted based on the user behavior data. Static features in user behavior data refer to those features that are relatively stable and do not change frequently over a period of time, such as basic information of users, geographic location information, etc. Specifically, they include the user's age, gender, occupation, education level, income level, time when the user registered an account, registration channel, geographic location of the user, etc. Secondly, time features are extracted based on time-related features and time information in user behavior data, that is, time features are extracted based on the interaction behavior with financial products, transaction behavior, social behavior, and feedback behavior on financial products in user behavior data and their corresponding time information, such as the timestamp of the user's behavior, the time interval between two user behaviors, the user's active time period on the platform, the time of the user's most recent behavior, etc.

[0039] In addition, for financial product data, first of all, static features are extracted from the financial product data. Static features in financial product data refer to those features that are relatively stable and do not change frequently over a period of time, such as basic information, product details, price information, etc. of financial products. Specifically, they include product ID, product name, product type, issuing institution, issuance date, expiration date, product description, risk level, minimum investment amount, investment period, initial price, management fee, custody fee, and transaction fee. Secondly, time features are extracted based on time-related features and time information in financial product data, that is, time features can be extracted based on historical price information, market performance, technical indicators, transaction records and their corresponding time information in financial product data. For example, historical price records of financial products presented in the form of time series, technical analysis indicators of financial products, trading volume of financial products, transaction price of financial products, performance of financial products in the past period of time (such as yield curve, volatility, etc.), etc.

[0040] For market dynamic data, firstly, static features are extracted based on market dynamic data, such as GDP, CPI, PPI, unemployment rate, foreign exchange reserves and other static features extracted from macroeconomic indicators; static features such as the definition and calculation method of market index, base period of market index, weight composition of market index and other static features extracted from market index; static features such as industry classification standard, industry market size, and major participants of industry extracted from industry dynamics; static features such as policy type, policy objectives, and policy formulation agency extracted from policy changes. Secondly, time features are extracted based on market dynamic data combined with time information, such as GDP growth rate, CPI growth rate, PPI growth rate, unemployment rate change, and foreign exchange reserve change recorded quarterly or annually extracted from macroeconomic indicators. Time features such as the value of market index recorded in time units such as day, week, and month, the rise and fall of market index, and trading volume of market index are extracted from market index. Time features such as the release time and content of industry reports recorded monthly or quarterly, market performance index of specific industries, and quarterly and annual financial reports of listed companies are extracted from industry dynamics. The release time and content of financial news recorded daily, and the impact of natural disasters and political events on the market recorded by the time of event occurrence are extracted from news reports. Based on policy changes, time characteristics such as the release time and content of the policy recorded daily, and the effect evaluation after policy implementation are extracted.

[0041] Regarding the extraction of historical behavior features of user behavior data, specifically, historical behavior features reflect the various behavior records of users related to financial products in the past. Based on the interaction behavior with financial products, historical behavior features such as the total number of times users view a certain financial product, the time users stay on each financial product detail page, the timestamp of each user browsing, the total number of times users click on a certain financial product-related link, the total number of times users add a certain financial product to favorites, the total number of times users share a certain financial product to social media or other platforms, and the channels through which users share can be extracted. Based on transaction behavior, historical behavior features such as the total number of times users purchase a certain financial product, the timestamp of each purchase by users, the amount of each purchase by users, the quantity of each purchase by users, the total number of times users redeem a certain financial product, the amount of each redemption by users, and the total number of times users transfer a certain financial product can be extracted. Based on social behavior, historical behavior features such as the total number of times users follow a certain financial product, the total number of times users unfollow a certain financial product, the total number of times users recommend a certain financial product, and the total number of times users accept financial products recommended by others can be extracted.

[0042] For the extraction of user feedback features from user behavior data, specifically, user feedback features reflect the user's satisfaction and opinions on financial products. The total number of times a user has rated a financial product and its specific rating can be extracted based on the rating record, the total number of times a user has commented on a financial product and its specific content can be extracted based on the comment record, the total number of times a user has complained about a financial product and its content can be extracted based on the complaint record, and the total number of times a user has participated in a satisfaction survey, the user's answers and ratings for each survey can be extracted based on the satisfaction survey.

[0043] In summary, static features can fully describe the attributes of users (such as age, gender, occupation, etc.) and products (such as product type, yield, risk level, etc.), providing rich background information for the model. Temporal features can capture the trend of user behavior and market dynamics over time, reflecting changes in user interests and market environment. The model can more accurately reflect the impact of recent behavior on user interests, thereby improving the timeliness and accuracy of predictions. User feedback directly reflects the user's satisfaction and preference for financial products. By analyzing these feedbacks, the model can more accurately understand the user's actual needs and preferences. User feedback can reflect the user's latest opinions in real time, helping the model to dynamically adjust the recommendation strategy to ensure that the recommendation results always meet the user's current needs.

[0044] The vector construction module S103 is used to combine the static features and time features of the user behavior data, financial product data and market dynamic data to obtain the time perception vectors corresponding to the three types of data respectively, and to construct the feedback behavior data in the user behavior data into a user feedback vector.

[0045] All static features corresponding to user behavior data, financial product data, and market dynamics data are extracted to form a structured list or dictionary. Similarly, all time features are extracted to form a structured list or dictionary. The static features and time features are then combined to obtain the time-aware vectors corresponding to the three types of data.

[0046] For the construction of the user feedback vector, the user feedback vector can be constructed based on the user feedback features extracted in the feature extraction module S102. Before constructing the vector, its user feedback features need to be processed. Specifically, for its rating features, the average value, standard deviation, and the time of the most recent rating of all ratings can be calculated. For the comment features, the total number of user comments, the time of the user's most recent comment, sentiment analysis of the comment content, and the extraction of positive and negative emotion scores can be calculated. For the complaint features, the total number of user complaints, the time of the user's most recent complaint, sentiment analysis of the complaint content, and the extraction of negative emotion scores can be calculated. For the satisfaction survey features, the total number of users participating in the satisfaction survey, the time of the user's most recent participation in the survey, the user's answers and scores in the survey, etc. can be calculated. The extracted features are combined into a user feedback vector, exemplarily:

[0047] User ID: 12345

[0048] Product ID: 1001

[0049] Average rating: 4.5

[0050] Rating variance: 0.29

[0051] Last rating time: 2024-10-03 10:10:00

[0052] Number of comments: 2

[0053] Last comment time: 2024-10-02 10:20:00

[0054] Comment sentiment score: 0.85 (positive sentiment score)

[0055] Number of complaints: 1

[0056] The latest complaint time: 2024-10-01 10:25:00

[0057] Complaint sentiment score: -0.75 (negative sentiment score)

[0058] Number of surveys: 1

[0059] The latest survey time: 2024-10-01 10:30:00

[0060] Survey results: Very satisfied

[0061] The model building module S104 is used to build a TAFM model, wherein the TAFM model includes an input layer, an embedding layer, a time decay layer, and a factor decomposition layer.

[0062] Specifically, the input layer is used to use the time perception vectors and user feedback vectors corresponding to the three types of data as inputs of the TAFM model. The input layer may have two input interfaces, one of which is used to input the time perception vectors corresponding to the three types of data, and the other is used to input the user feedback vector.

[0063] The embedding layer is used to convert the time perception vector and user feedback vector corresponding to the three types of data into a fixed-length latent vector representation. Specifically, an embedding matrix can be created for each feature dimension, and the size of the matrix can be feature dimension × latent vector dimension. Then, according to the eigenvalue in the input vector, the corresponding latent vector is searched in the embedding matrix, and all the latent vectors found are concatenated into a fixed-length vector.

[0064] The time decay layer is used to adjust the importance of historical behavior features to update the time-aware vector.

[0065] In one embodiment, the method of using the time decay layer to adjust the importance of the historical behavior feature to update the time perception vector specifically includes:

[0066] Based on each historical event, calculate the difference between the current time and the time when the event occurred. Specifically, first determine the current time when the model is running, extract the occurrence time of each historical event from the time-aware vector, and calculate the difference between the current time and the occurrence time of each historical event. Apply an exponential decay function or a linear decay function to calculate the weight of the historical behavior characteristics. For each historical behavior event, multiply its historical behavior feature vector by the corresponding time decay weight to adjust the historical behavior feature vector. Specifically, extract the feature vector of each historical behavior event from the time-aware vector, and multiply each historical behavior feature vector by its corresponding time decay weight. Replace the original feature vector with the adjusted historical behavior feature vector to update the time-aware vector.

[0067] The factor decomposition machine layer includes a linear calculation layer, an interactive calculation layer, and an output layer. Specifically, the linear calculation layer is used to perform linear calculations on the latent vectors corresponding to the time perception vector processed by the time decay layer and the user feedback vector.

[0068] The latent vector of each feature is extracted from the time perception vector and user feedback vector processed by the time decay layer, and the weighted sum of each latent vector is performed to obtain the linear term. The formula can be expressed as:

[0069]

[0070] Among them, w i is feature x i The weight of , n is the total number of features.

[0071] The interactive calculation layer is used to perform second-order interactive calculations on the latent vectors corresponding to the time perception vector and the user feedback vector processed by the time decay layer. First, the latent vector of each feature is extracted from the time perception vector and the user feedback vector processed by the time decay layer. Then, the inner product operation is performed on each pair of latent vectors to obtain the interaction term. The formula can be expressed as:

[0072]

[0073] Among them, v i and v j They are the features x i x j The hidden vector of <v i , v j > represents the inner product of two vectors.

[0074] The output layer is used to combine the outputs of the linear calculation layer and the interactive calculation layer to generate the final prediction value. The formula can be expressed as: prediction value = linear term + interactive term + b, where b is the bias term.

[0075] The time perception vector processed by the time decay layer and the latent vector corresponding to the user feedback vector can be linearly calculated and second-order interactively calculated to finally generate the prediction value of the model. The application of the factorization layer in this solution can capture the linear and nonlinear relationship between the characteristics of the time perception vector and the user feedback vector, and improve the prediction ability of the model.

[0076] The model training module S105 is used to input the time perception vector, user feedback vector and user transaction result data corresponding to the three types of data into the constructed TAFM model, train the TAFM model, and obtain a prediction model.

[0077] The time perception vector, user feedback vector and user transaction results corresponding to the three types of data are used as data sets, and the data sets are divided into training set, validation set and test set. The training set is used to train the model, that is, to adjust the model parameters through the optimization algorithm so that the model performs well on the training data. It is used to evaluate the performance of the model during the training process, and to perform hyperparameter tuning and model selection. The test set is used to finally evaluate the generalization ability of the model, that is, the performance of the model on new data that has not been seen. During the training process, you can choose a suitable loss function, such as mean square error (MSE) for regression tasks, cross-entropy loss (Cross-Entropy Loss) for classification tasks, and choose a suitable optimizer, such as Adam, RMSprop, etc. To prevent overfitting, you can add L1 or L2 regularization terms and set early stopping conditions. When the performance on the validation set no longer improves, the training is terminated in advance, and finally a trained prediction model is obtained.

[0078] The model prediction module S106 is used to input new user behavior data, financial product data and market dynamic data into the prediction model and output the model prediction results.

[0079] It is easy to understand that before inputting new user behavior data, financial product data, and market dynamic data into the constructed TAFM model, it is necessary to perform feature extraction and vector construction on the three types of data respectively through the feature extraction module S102 and the vector construction module S103 to obtain the time perception vector and user feedback vector corresponding to the three new data respectively. Then the time perception vector and the user feedback vector are input into the prediction model, and the model prediction result is output. The prediction result can be the possibility of the user to trade a certain financial product in a certain time period in the future, the amount of the user's future transaction is predicted, the type of the user's future transaction (such as purchase, redemption, transfer, etc.), or it can be a recommendation of the financial product that is most likely to arouse the user's interest, or a recommendation score is assigned to each recommended product to indicate the user's interest in the product.

[0080] In one embodiment, a visualization report can be generated based on the prediction results of the model. The prediction results can be converted into a format suitable for visualization, such as DataFrame. Select common visualization tools such as Python: Matplotlib, Seaborn, Plotly, Bokeh, Web: D3.js, ECharts, business intelligence tools: Tableau, Power BI. Select the appropriate chart type, such as:

[0081] Bar chart: Displays the scores of recommended products.

[0082] Pie chart: shows the distribution of user behavior.

[0083] Line chart: Displays the predicted trend of market indices.

[0084] Scatter plot: shows the relationship between user transaction probability and transaction amount.

[0085] Dashboards: Display key metrics such as credit score and fraud risk.

[0086] The financial product recommendation module S107 is used to recommend financial products to users based on the prediction results of the model.

[0087] Generate a personalized list of financial product recommendations based on the prediction results. The recommended products can be sorted according to the recommendation scores to ensure that the products in the recommendation list are displayed in priority.

[0088] In one embodiment, the financial product recommendation module also includes a user quick transaction entry, which is used to facilitate users to conduct quick transactions based on the pushed financial products.

[0089] The TAFM model can be used to generate a personalized list of financial product recommendations based on the user's transaction history, behavioral data, market dynamics and other information. The recommended products are presented to the user in a list format, and each product should include information such as name, introduction, and recommendation score. A quick transaction button is provided for each recommended financial product, and the user can directly enter the transaction process after clicking it. Simplify the transaction process, reduce the user's operation steps, and ensure that the user can complete the transaction quickly.

[0090] In one embodiment, the financial product recommendation module also includes a user feedback portal, which is used to facilitate users to evaluate the recommended financial products based on the recommended results. A feedback button can be provided for each recommended financial product, and the user can click the button to enter the feedback page. The user can evaluate the recommended financial product on the feedback page, including rating and text comments.

[0091] In one embodiment, the financial product recommendation module further includes a user favorite mark button, which is used to facilitate users to collect or mark products of interest.

[0092] A favorites mark button can be provided for each recommended financial product, and users can click the button to add the product of interest to their favorites. A favorites page is provided to display all the products collected by the user, making it easier for users to view and manage. For example, a button displays as "Favorites", and after the user clicks it, it changes to "Favorited". A button displays as "Favorited", and after the user clicks it again to cancel the favorite, the button returns to "Favorites". Information about user favorited products is stored in the database to ensure that users can see their favorite records on different devices and sessions. Users can also sort and filter favorite products based on different criteria (such as recommendation score, collection time, etc.).

[0093] Those skilled in the art should be aware that in one or more of the above examples, the functions described in the present invention can be implemented by hardware, software, firmware or any combination thereof. When implemented by software, these functions can be stored in a computer-readable medium or transmitted as one or more instructions or codes on a computer-readable medium.

[0094] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications made on the basis of the technical solution of the present invention are subject to the provisions of the present invention.

Claims

1. A financial market trend prediction and optimization trading system based on machine learning, characterized in that: The system includes: a data acquisition module, which is used to acquire user behavior data, financial product data, market dynamics data and user transaction result data; wherein the user behavior data, financial product data and market dynamics data are all recorded with corresponding time information, and the user behavior data includes historical behavior events and user feedback behavior data on financial products; a feature extraction module, which is used to extract features of the three types of data obtained from the data acquisition module, obtain static features and time features corresponding to the three types of data, and obtain historical behavior features and user feedback features of the user behavior data; the historical behavior features are used to introduce a time decay mechanism into the TAFM model, and the user feedback features are used to introduce a user feedback mechanism into the TAFM model; a vector construction module, which is used to combine the static features and time features of the user behavior data, financial product data and market dynamics data to obtain the time perception vectors corresponding to the three types of data, and to construct the feedback behavior data in the user behavior data into a user feedback vector; a model construction module, which is used to construct a TAFM model, and the TAFM model includes an input layer, an embedding layer, a time decay layer and a factor decomposition layer; wherein the input layer is used to extract the three types of data The corresponding time perception vectors and user feedback vectors are used as inputs of the TAFM model; the embedding layer is used to convert the time perception vectors and user feedback vectors corresponding to the three types of data into latent vector representations of fixed length; the time decay layer is used to adjust the importance of historical behavior features to update the time perception vector; the factor decomposition machine layer includes a linear calculation layer, an interactive calculation layer and an output layer, wherein the linear calculation layer is used to perform linear calculations on the latent vectors corresponding to the time perception vectors and user feedback vectors processed by the time decay layer, the interactive calculation layer is used to perform second-order interactive calculations on the latent vectors corresponding to the time perception vectors and user feedback vectors processed by the time decay layer, and the output layer is used to combine the outputs of the linear calculation layer and the interactive calculation layer to generate the final prediction value; the model training module is used to input the time perception vectors, user feedback vectors and user transaction result data corresponding to the three types of data into the constructed TAFM model, train the TAFM model to obtain a prediction model; the model prediction module is used to input new user behavior data, financial product data and market dynamics data into the prediction model, and output the model prediction results; the financial product recommendation module is used to recommend financial products to users based on the prediction results of the model.

2. A financial market trend prediction and trading optimization system based on machine learning according to claim 1, characterized in that: The method of using a time decay layer to adjust the importance of historical behavior features to update a time perception vector specifically includes: based on each historical event, calculating the difference between the current time and the time when the event occurred; applying an exponential decay function or a linear decay function to calculate the weight of the historical behavior features; for each historical behavior event, multiplying its historical behavior feature vector by the corresponding time decay weight to adjust the historical behavior feature vector; and replacing the original feature vector with the adjusted historical behavior feature vector to update the time perception vector.

3. The financial market trend prediction and trading optimization system based on machine learning according to claim 1, characterized in that: Generate a visual report based on the model's prediction results.

4. The financial market trend prediction and trading optimization system based on machine learning according to claim 1, characterized in that: The financial product recommendation module also includes a user quick transaction entry, which is used to facilitate users to conduct quick transactions based on pushed financial products.

5. The financial market trend prediction and trading optimization system based on machine learning according to claim 1, characterized in that: The financial product recommendation module also includes a user feedback entry, which is used to facilitate users to evaluate the recommended financial products based on the recommendation results.

6. A financial market trend prediction and trading optimization system based on machine learning according to claim 1, characterized in that: The financial product recommendation module also includes a user favorite mark button, which is used to facilitate users to collect or mark products of interest.

Images