Financial product recommendation method and device, electronic equipment and storage medium

Abstract

The invention discloses a financial product recommendation method and device, electronic equipment and a storage medium, and relates to the field of financial science and technology or other related technical fields, and the method comprises the steps: obtaining multi-source data and user portrait data of a financial product; aggregating the multi-source data, and constructing a feature vector of the financial product; constructing a feature space based on the feature vectors of all the financial products, calculating a local outlier factor value of each financial product in the feature space, screening the financial products based on the local outlier factor values, and constructing an abnormal product candidate pool; and performing attribution analysis on the abnormal financial products in the abnormal product candidate pool, determining abnormal attributes of the abnormal financial products, and generating a matching result for the user in combination with the user portrait data and the abnormal attributes. According to the method and the device, the technical problems of relatively low accuracy and relatively weak risk identification capability of a financial product recommendation mode based on collaborative filtering and content recommendation in related technologies are solved.

Description

Technical Field

[0001] This invention relates to the field of financial technology or other related technical fields. Specifically, it relates to a method, apparatus, electronic device, and storage medium for recommending financial products. Background Technology

[0002] With the rapid development of internet finance, the variety and quantity of financial products in the financial market have shown unprecedented richness and growth. From stocks, bonds, and mutual funds to various wealth management products and derivatives, users face a dazzling array of choices. Especially in the field of standardized financial products such as mutual funds, the explosive growth in the number has created serious selection barriers for ordinary investors when screening suitable investment targets, making it difficult to effectively identify and select products that match their personal preferences and investment goals.

[0003] Among related technologies, financial product recommendations mostly rely on collaborative filtering, content recommendation, or rule-based recommendation techniques. Collaborative filtering (including user-based and item-based filtering) recommends similar products by analyzing user behavior or relationships between products. However, this technique is ineffective when dealing with new users (cold start problem) or new products, and is prone to falling into the popularity trap, resulting in overly similar recommended products and making it difficult to uncover potential but less well-known dark horse products. Content recommendation technology focuses on using product attribute information (such as historical performance, risk management characteristics, etc.) to match user needs, but its effectiveness is limited by fixed rules and weight settings, lacking the ability to respond to market changes in real time and capture complex relationships between products. Current financial product recommendation methods have low accuracy and fail to meet user needs.

[0004] There is currently no effective solution to the above problems. Summary of the Invention

[0005] This invention provides a method, apparatus, electronic device, and storage medium for recommending financial products, to at least solve the technical problems of low accuracy and weak risk identification ability in related technologies that rely on collaborative filtering and content recommendation for recommending financial products.

[0006] According to one aspect of the present invention, a method for recommending financial products is provided, comprising: acquiring multi-source data of financial products and user profile data; aggregating the multi-source data to construct feature vectors of the financial products; constructing a feature space based on the feature vectors of all the financial products, calculating a local outlier factor value for each financial product in the feature space, and screening the financial products based on the local outlier factor value to construct an abnormal product candidate pool, wherein the local outlier factor value is used to characterize the local density deviation between a product node and its neighboring nodes in the feature space, and the product node is the representation of the financial product in the feature space; performing attribution analysis on the abnormal financial products in the abnormal product candidate pool to determine the abnormal attributes of the abnormal financial products, and generating matching results for users by combining the user profile data and the abnormal attributes, wherein the matching results include at least one of the following: recommending investment financial products and risky financial products.

[0007] Furthermore, the steps for obtaining multi-source data of financial products include: collecting raw data of the financial products from multi-source heterogeneous data sources; cleaning and integrating the raw data to obtain the multi-source data.

[0008] Furthermore, the step of constructing a feature space based on the feature vectors of all the financial products includes: receiving the feature vectors of all the financial products, aggregating the feature vectors of all the financial products to construct an initial feature space; and standardizing the initial feature space to obtain the feature space.

[0009] Further, the step of calculating the local outlier factor value of each financial product in the feature space includes: calculating the distance value between any two product nodes in the feature space, and defining the neighborhood of the product node based on the distance value between the product nodes; calculating the neighborhood reachability distance between the product node and the neighboring nodes in the neighborhood, and calculating the local reachability density of the product node based on all neighborhood reachability distances of the product node; calculating the average local reachability density of all neighboring nodes in the neighborhood of the product node to obtain the neighborhood average local reachability density of the product node; and calculating the ratio between the neighborhood average local reachability density of the product node and the local reachability density of the product node to obtain the local outlier factor value of the financial product.

[0010] Further, the step of screening the financial products based on the local outlier factor values ​​and constructing an abnormal product candidate pool includes: sorting the local outlier factor values ​​to obtain a sorting result; determining a local outlier factor threshold based on the sorting result, screening out all financial products that are greater than the local outlier factor threshold, and marking the screened financial products as the abnormal financial products; and constructing the abnormal product candidate pool based on the abnormal financial products.

[0011] Further, the step of performing attribution analysis on the abnormal financial products in the abnormal product candidate pool to determine the abnormal attributes of the abnormal financial products includes: inputting multi-source data of the abnormal financial products into a classification model, wherein the classification model is a pre-built machine learning model for identifying the abnormal attributes of the abnormal financial products; performing feature extraction and attribution analysis on the abnormal financial products through the classification model, and outputting the abnormal attributes of the abnormal financial products.

[0012] Furthermore, the step of generating matching results for users by combining the user profile data and the abnormal attributes includes: calculating a risk preference value for users based on the user profile data; classifying the abnormal financial products into positive and negative abnormal financial products based on the abnormal attributes, and calculating the risk value of the positive abnormal financial products; matching positive abnormal financial products to users by combining the user's risk preference value and the risk value of the positive abnormal financial products, and pushing the successfully matched positive abnormal financial products as recommended investment financial products to the user's end; and pushing all negative abnormal financial products as risky financial products to the user's end.

[0013] According to another aspect of the present invention, a financial product recommendation apparatus is also provided, comprising: an acquisition unit for acquiring multi-source data of financial products and user profile data; a construction unit for aggregating the multi-source data to construct feature vectors of the financial products; a calculation unit for constructing a feature space based on the feature vectors of all the financial products, calculating a local outlier factor value for each financial product in the feature space, and filtering the financial products based on the local outlier factor value to construct an abnormal product candidate pool, wherein the local outlier factor value is used to characterize the local density deviation between a product node and its neighboring nodes in the feature space, and the product node is the representation of the financial product in the feature space; and an analysis unit for performing attribution analysis on the abnormal financial products in the abnormal product candidate pool, determining the abnormal attributes of the abnormal financial products, and generating matching results for users by combining the user profile data and the abnormal attributes, wherein the matching results include at least one of the following: recommending investment financial products and risky financial products.

[0014] Furthermore, the acquisition unit includes: a first acquisition module, used to acquire raw data of the financial product from multiple heterogeneous data sources; and a first processing module, used to perform data cleaning and data integration on the raw data to obtain the multi-source data.

[0015] Furthermore, the construction unit includes: a first aggregation module, used to receive feature vectors of all the financial products and aggregate the feature vectors of all the financial products to construct an initial feature space; and a second processing module, used to perform standardization processing on the initial feature space to obtain the feature space.

[0016] Further, the calculation unit includes: a first calculation module, used to calculate the distance value between any two product nodes in the feature space, and define the neighborhood of the product node based on the distance value between the product nodes; a second calculation module, used to calculate the neighborhood reachability distance between the product node and the neighboring nodes in the neighborhood, and calculate the local reachability density of the product node based on all neighborhood reachability distances of the product node; a third calculation module, used to calculate the average local reachability density of all neighboring nodes in the neighborhood of the product node, to obtain the neighborhood average local reachability density of the product node; and a fourth calculation module, used to calculate the ratio between the neighborhood average local reachability density of the product node and the local reachability density of the product node, to obtain the local outlier factor value of the financial product.

[0017] Furthermore, the calculation unit further includes: a first sorting module, used to sort the local outlier factor values ​​to obtain a sorting result; a first filtering module, used to determine a local outlier factor threshold based on the sorting result, filter out all financial products that are greater than the local outlier factor threshold, and mark the filtered financial products as the abnormal financial products; and a first construction module, used to construct the abnormal product candidate pool based on the abnormal financial products.

[0018] Furthermore, the analysis unit includes: a first input module, used to input multi-source data of the abnormal financial product into a classification model, wherein the classification model is a pre-built machine learning model for identifying the abnormal attributes of the abnormal financial product; and a first analysis module, used to perform feature extraction and attribution analysis on the abnormal financial product through the classification model, and output the abnormal attributes of the abnormal financial product.

[0019] Furthermore, the analysis unit further includes: a fifth calculation module, used to calculate a risk preference value for the user based on the user profile data; a first classification module, used to classify the abnormal financial products into positive and negative abnormal financial products based on the abnormal attributes, and calculate the risk value of the positive abnormal financial products; a first matching module, used to match positive abnormal financial products to the user by combining the user's risk preference value and the risk value of the positive abnormal financial products, and push the successfully matched positive abnormal financial products as recommended investment financial products to the user terminal; and a first push module, used to push all negative abnormal financial products as risky financial products to the user terminal. According to another aspect of the present invention, a computer-readable storage medium is also provided, the computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to execute the method for recommending any of the above-mentioned financial products.

[0020] According to another aspect of the present invention, an electronic device is also provided, including one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the method for recommending any of the above-mentioned financial products.

[0021] In this application, the following steps are performed: acquiring multi-source data of financial products and user profile data, aggregating the multi-source data to construct feature vectors for financial products, then constructing a feature space based on the feature vectors of all financial products, calculating the local outlier factor value for each financial product in the feature space, and filtering financial products based on the local outlier factor value to construct an abnormal product candidate pool. The local outlier factor value is used to characterize the local density deviation between product nodes and their neighboring nodes in the feature space. Product nodes represent the representation of financial products in the feature space. Finally, attribution analysis is performed on the abnormal financial products in the abnormal product candidate pool to determine the abnormal attributes of the abnormal financial products. Matching results are generated for users by combining user profile data with the abnormal attributes. The matching results include at least one of the following: recommended investment financial products, risky financial products.

[0022] In this application, for financial products, abnormal financial products are screened by calculating local outlier factor values, and an abnormal product candidate pool is constructed. This enables the identification of financial products with unique advantages or risk characteristics based solely on their intrinsic features, even without behavioral data, effectively solving the cold start problem of financial product recommendation. Subsequently, attribution analysis is performed on the financial products in the candidate pool, classifying abnormal financial products into positive and negative anomalies, and accurately matching them with users. Positive abnormal financial products with investment potential that meet the user's investment risk and investment needs are recommended, while negative abnormal financial products are given early warnings. This improves risk identification capabilities, avoids potential investment risks, and achieves accurate recommendations. Thus, it solves the technical problems of low accuracy and weak risk identification capabilities in financial product recommendation methods based on collaborative filtering and content recommendation in related technologies. Attached Figure Description

[0023] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0024] Figure 1 A hardware block diagram of a computer terminal (or mobile device) for implementing a recommendation method for financial products is shown.

[0025] Figure 2 This is a flowchart of an optional financial product recommendation method according to an embodiment of the present invention;

[0026] Figure 3 This is a schematic diagram of an optional screening process based on local outlier factors according to an embodiment of the present invention;

[0027] Figure 4 This is a schematic diagram of an optional attribution analysis process according to an embodiment of the present invention;

[0028] Figure 5 This is a schematic diagram of an optional financial product recommendation device according to an embodiment of the present invention;

[0029] Figure 6 This is a hardware structure block diagram of an electronic device (or mobile device) for performing an optional method of recommending financial products according to an embodiment of the present invention. Detailed Implementation

[0030] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0031] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0032] To facilitate understanding of the present invention by those skilled in the art, some terms or nouns involved in the various embodiments of the present invention are explained below:

[0033] Annualized rate of return is a method of measuring the average return of an investment over a certain period of time, converting the rate of return for different time periods into an annual percentage.

[0034] Volatility is an important indicator in finance for measuring the magnitude of changes in asset prices or portfolio value.

[0035] The Sharpe ratio is an indicator used in the financial investment field to measure investment returns and the risks undertaken.

[0036] The Sortino ratio is a metric for measuring investment performance, primarily used to assess the excess return of a portfolio per unit of downside risk.

[0037] The Calmar ratio is a risk-adjusted return metric used to measure the performance of an investment strategy or fund.

[0038] Turnover rate, also known as turnover ratio, is an indicator that measures the trading activity of the market or a particular stock, fund, or other financial asset within a certain period of time.

[0039] It should be noted that the financial product recommendation method and apparatus in this application can be used in the field of fintech for reverse recommendation of financial products based on local outlier analysis and attribution analysis, and can also be used in any field other than fintech for reverse recommendation of financial products based on local outlier analysis and attribution analysis. This application does not limit the application field of the financial product recommendation method and apparatus.

[0040] It should be noted that the information collected in this application (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for display, data used for analysis, etc.) are information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of this data all comply with relevant laws, regulations, and standards, necessary confidentiality measures have been taken, and they do not violate public order and good morals. Corresponding access points are provided for users to choose to authorize or refuse. For example, interfaces are set up between this system and relevant users or organizations, providing users with corresponding access points to choose to agree to or refuse automated decision-making results; if the user chooses to refuse, the process proceeds to the expert decision-making stage.

[0041] The following embodiments of the present invention can be applied to various financial product recommendation systems / applications / devices. The present invention proposes a financial product recommendation method based on the calculation of local outlier factor values. This method identifies potentially anomalous financial products by calculating local outlier factor values, performs anomaly attribution analysis, filters out financial products with investment potential for recommendation to users, and provides early warnings for financial products with investment risks. This addresses problems in existing technologies such as cold start, homogeneous recommendations, and weak dynamic risk identification capabilities.

[0042] The present invention will now be described in detail with reference to various embodiments.

[0043] Example 1

[0044] According to an embodiment of the present invention, an embodiment of a method for recommending financial products is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0045] The method embodiment provided in Embodiment 1 of this application can be executed on a mobile terminal, computer terminal, or similar computing device. Figure 1 A hardware block diagram of a computer terminal (or mobile device) for implementing a recommendation method for financial products is shown. Figure 1As shown, the computer terminal 10 (or mobile device) may include one or more processors 102 (shown as 102a, 102b, ..., 102n in the figure) 102 (processor 102 may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of a BUS bus), a network interface, a power supply, and / or a camera. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, computer terminal 10 may also include... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0046] It should be noted that the aforementioned one or more processors 102 and / or other data processing circuits are generally referred to herein as "data processing circuits". These data processing circuits may be embodied, in whole or in part, in software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuits may be a single, independent processing module, or may be integrated, in whole or in part, into any other element within the computer terminal 10 (or mobile device). As involved in the embodiments of this application, the data processing circuits serve as a processor control mechanism (e.g., selection of a variable resistor termination path connected to an interface).

[0047] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the financial product recommendation method in this embodiment. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, thereby realizing the aforementioned financial product recommendation method. The memory 104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the computer terminal 10 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0048] The transmission device 106 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of the computer terminal 10. In one example, the transmission device 106 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 106 may be a Radio Frequency (RF) module, used for wireless communication with the Internet.

[0049] The display can be, for example, a touchscreen liquid crystal display (LCD) that allows the user to interact with the user interface of the computer terminal 10 (or mobile device).

[0050] Under the aforementioned operating environment, this application provides the following: Figure 2 The method for recommending financial products shown is implemented by a financial product recommendation system.

[0051] Figure 2 This is a flowchart of an optional financial product recommendation method according to an embodiment of the present invention, such as... Figure 2 As shown, the method includes the following steps:

[0052] The implementation subject of this invention is a financial product recommendation system. This system uses the calculation of local outlier factor values ​​and attribution analysis to identify financial products with investment potential and risks. Especially in the scenario of recommending fund products, traditional recommendation methods mostly look for good funds from a positive perspective. However, financial markets involve both risk and return, and identifying and avoiding potentially bad funds is equally crucial. Existing technologies lack an effective mechanism to automatically identify funds with abnormal performance, potentially undervalued or overvalued by the market, based on the data itself. This invention, however, recommends financial products with investment potential to users from a reverse perspective, while avoiding abnormal financial products with significant investment risks, achieving accurate recommendation and early warning of financial products.

[0053] Step S201: Obtain multi-source data of financial products and user profile data.

[0054] In step S201 above, a solid data foundation is provided for financial product recommendations by collecting multi-dimensional data. First, the system connects to multiple data sources, including but not limited to publicly released financial reports of financial products, market transaction data, and user transaction records from financial institutions. For financial products, the system needs to collect a series of key attributes, such as product code, name, type, establishment date, management team information, historical performance data, volatility, maximum drawdown, and return ratio indicators. User profile data covers personalized information such as the user's risk preference, investment horizon preference, and past investment behavior and preferences.

[0055] Through the steps outlined above, a comprehensive dataset is constructed, providing rich raw information for subsequent feature engineering and anomaly detection. Multi-source data from financial products ensures the comprehensiveness and depth of the analysis, while user profile data helps the system understand each user's specific needs and preferences, laying the foundation for personalized recommendations.

[0056] Furthermore, the steps for obtaining multi-source data of financial products include: collecting raw data of financial products from multi-source heterogeneous data sources; cleaning and integrating the raw data to obtain multi-source data.

[0057] Specifically, when acquiring multi-source data on financial products, it is necessary to collect raw data from various heterogeneous data sources. These heterogeneous data sources refer to different data providers or interfaces, including but not limited to financial information service providers, official websites of financial product companies, data released by financial regulatory agencies, and third-party research reports. These data sources may use different data formats, storage methods, and update frequencies. During the collection process, it is essential to ensure broad coverage and reliable data sources to obtain comprehensive and accurate information on financial products. The collected raw data mainly includes basic information about the financial product, such as product code, name, type, establishment date, and product management personnel; historical financial data, such as daily / weekly / monthly / yearly net asset value, yield, and volatility; holdings information, such as the industry distribution of stocks and bonds held and the top ten holdings; and market data, such as the risk-free interest rate and benchmark index returns.

[0058] The collected raw data undergoes data cleaning and integration to obtain multi-source data. Data cleaning primarily removes missing values, outliers, or duplicate information to ensure the accuracy of subsequent analysis. For example, missing values ​​in historical return data can be filled by interpolation or using the median of similar financial products; outliers can be reasonably removed or corrected based on statistical principles such as the 3σ principle or interquartile range. Data integration, building upon cleaning, unifies information from different data sources under a common framework, constructing a complete and consistent view of financial product data. This step is particularly important for handling heterogeneous data, as the raw data may have inconsistencies in naming, format, and coverage. The goal of integration is to eliminate these differences, forming a high-quality dataset that can be used for subsequent feature engineering and algorithmic analysis. Data integration can specifically include: data alignment, ensuring that the time series data of all financial products are synchronized, which usually requires adjusting the time interval of the data, such as unifying it to the daily level, to support time series analysis across products; feature construction, based on the original data, constructing a series of quantitative indicators that reflect the characteristics and performance of financial products, such as annualized return, volatility, Sharpe ratio, etc.; and data standardization, so that all features have the same dimensions and scale, thereby avoiding the problem of certain features dominating the analysis results.

[0059] Step S202: Aggregate multi-source data to construct feature vectors for financial products.

[0060] In step S202 above, multi-source data is aggregated to construct a multi-dimensional feature space that comprehensively reflects the characteristics of financial products. Specifically, features are extracted from the multi-source data, such as yield, volatility, drawdown indicators, liquidity indicators, and style indicators. The extracted features are then transformed into high-dimensional feature vectors. By constructing these feature vectors, all financial products can be compared and analyzed within a unified framework, identifying those products that stand out across various dimensions.

[0061] Step S203: Construct a feature space based on the feature vectors of all financial products, calculate the local outlier factor value of each financial product in the feature space, and screen financial products based on the local outlier factor value to construct an abnormal product candidate pool.

[0062] In step S203 above, the system first receives and aggregates the feature vectors of all financial products to construct an initial feature space. This space consists of multi-dimensional feature vectors, each dimension representing a specific analytical indicator, such as risk-return characteristics, dynamic behavior characteristics, and style quality characteristics, thus forming a multi-dimensional coordinate system that comprehensively reflects the characteristics and market position of financial products. Subsequently, the system standardizes this initial feature space to ensure that all feature vectors are compared on the same scale, avoiding result bias caused by the influence of dimensions. This processing is accomplished through Z-Score standardization, which calculates the deviation of each feature vector from the overall mean and divides it by the standard deviation to obtain a standardized feature space in which financial products exist as product nodes.

[0063] In the standardized feature space, the system calculates the Local Outlier Factor (LOF) value for each product node. The LOF value characterizes the local density deviation between a product node and its neighboring nodes in the feature space, measuring the degree of deviation of a point (product node) from its neighborhood density. By comparing the local density deviation between a product node and its nearest k neighbors, outliers are identified. Outliers may indicate financial products with abnormal characteristics. If a product node exhibits higher sparsity (lower local density) relative to its neighbors, its LOF value will be greater than 1, indicating that it is a local outlier, potentially representing a special market opportunity or potential risk, such as a high-risk, high-return financial product, or a high-risk financial product requiring caution.

[0064] Furthermore, the steps for constructing a feature space based on the feature vectors of all financial products include: receiving the feature vectors of all financial products, aggregating the feature vectors of all financial products to construct an initial feature space; and standardizing the initial feature space to obtain the feature space.

[0065] Specifically, the feature space refers to a multi-dimensional space composed of the feature vectors of all observed samples. In this space, each dimension represents a feature, and each sample (such as a financial product) is represented as a point or vector in this multi-dimensional space, with its coordinates determined by the measurements of the sample on each feature. When constructing the feature space, we receive feature vectors from all financial products analyzed across multiple dimensions. These feature vectors encompass various factors such as return and risk, dynamic behavior, and style quality. Subsequently, these feature vectors are aggregated to construct an initial feature space. The aggregation process is essentially an integration of all financial product data; each set of feature vectors is like a point in the space, collectively depicting a multi-layered, detailed map. This map not only encompasses a broad panorama of the market but is also detailed enough to capture subtle differences between individual products. Then, this initial feature space is standardized. The purpose of standardization is to ensure that the data of each feature vector across different dimensions are treated fairly, eliminating misleading results caused by differences in units of measurement. Z-Score standardization can be used to transform the data of each dimension into a form with the same mean and standard deviation, ensuring that data from any dimension does not unduly dominate the analysis results.

[0066] Furthermore, the steps for calculating the local outlier factor value of each financial product in the feature space include: calculating the distance between any two product nodes in the feature space, and defining the neighborhood of a product node based on the distance between product nodes; calculating the neighborhood reachability distance between a product node and its neighboring nodes within the neighborhood, and calculating the local reachability density of the product node based on all neighborhood reachability distances of the product node; calculating the average local reachability density of all neighboring nodes within the neighborhood of the product node, thus obtaining the average local reachability density of the product node's neighborhood; and calculating the ratio between the average local reachability density of the product node's neighborhood and the local reachability density of the product node, thus obtaining the local outlier factor value of the financial product.

[0067] Specifically, when calculating the local outlier factor, the distance between any two product nodes in the feature space is first calculated. Euclidean distance can be used as a metric to quantify the relative position of product nodes in the feature space, reflecting the differences between financial products in characteristics such as annualized return and volatility. Based on these distance values, the neighborhood of each product node is defined, that is, finding a certain number (e.g., k) of other nodes closest to that node, forming the neighborhood set of that node. The definition of the neighborhood is a prerequisite for calculating the local outlier factor, revealing the direct environment of the product node and used to assess its density relative to neighboring products.

[0068] Subsequently, the system calculates the neighborhood reachability distance between the product node and its neighboring nodes. This reachability distance considers not only the direct distance between two nodes but also incorporates the density information of neighboring nodes, ensuring sensitive detection of anomalies in sparse regions. Based on the neighborhood reachability distance, the system further calculates the local reachability density of the product node, which is the reciprocal of the average reachability distance of neighboring nodes to the product node, reflecting the density of the area where the product node is located. Then, the system calculates the average local reachability density of all neighboring nodes within the product node's neighborhood. Finally, by calculating the local reachability density of the product node and the average local reachability density of its neighborhood, the system obtains the local outlier factor value of the product node. Product nodes with a local outlier factor value greater than 1 are considered low-density local outliers, thus filtering out abnormal financial products.

[0069] The calculation of local outlier factor values ​​can identify product nodes in the feature space that exhibit unusual characteristics relative to their neighborhood, thereby screening out abnormal financial products from a massive amount of financial products. This process not only helps to discover high-quality funds that are undervalued by the market, but also to discover unstable investments that may have high risks, providing data support for subsequent risk management and personalized recommendations.

[0070] Furthermore, the steps of screening financial products based on local outlier factor values ​​and constructing an abnormal product candidate pool include: sorting the local outlier factor values ​​to obtain the sorting results; determining the local outlier factor threshold based on the sorting results, screening out all financial products that are greater than the local outlier factor threshold, and marking the screened financial products as abnormal financial products; and constructing an abnormal product candidate pool based on the abnormal financial products.

[0071] Specifically, the system first sorts all financial products by their local outlier values ​​in descending order, generating a ranking result. This ranking reflects the importance of the anomaly; a higher value indicates that the financial product deviates more from its neighborhood's normal performance, and its anomalous characteristics are more pronounced. Based on the ranking result, the system needs to determine an appropriate local outlier threshold. This threshold is typically chosen based on specific business rules or through historical data analysis, aiming to balance the quantity and quality of the selected anomalous financial products. After determining the local outlier threshold, the system filters out all financial products with local outlier values ​​greater than the threshold from the ranking result, marks these selected financial products as anomalous financial products, and uses this to construct an anomalous product candidate pool.

[0072] By ranking local outlier factor values ​​and setting thresholds, financial products exhibiting significant anomalous behavior in the feature space can be accurately identified, constructing a high-quality candidate pool of anomalous products. This candidate pool provides a solid data foundation for subsequent anomaly attribution analysis and personalized recommendations.

[0073] Step S204: Perform attribution analysis on abnormal financial products in the abnormal product candidate pool, determine the abnormal attributes of abnormal financial products, and generate matching results for users by combining user profile data with abnormal attributes.

[0074] The matching results include at least one of the following: recommended investment financial products, risky financial products.

[0075] In step S204 above, the attribution analysis examines the specific characteristics of each anomalous product in dimensions such as return risk, dynamic behavior, style and quality based on the feature vector revealed by the LOF algorithm, thereby determining its anomalous attributes. Anomalous attributes include positive anomalies and negative anomalies. Positive anomalies refer to financial products that are undervalued by the market, while negative anomalies refer to financial products that have high risks.

[0076] Subsequently, personalized matching is further performed by combining user profile data. User profile data includes key information such as risk tolerance, investment horizon preferences, and historical investment behavior, generating matching results for users, including recommended investment financial products and risk warnings for financial products. By comparing user profiles with the abnormal attributes of unusual financial products, the system can infer which unusual products best match the user's investment goals and risk preferences. For example, for users seeking high returns and with a high risk tolerance, the system will recommend financial products marked as having potentially high-yield abnormal attributes; for conservative investors who prioritize capital safety, the system tends to recommend stable, high-performing abnormal products, i.e., financial products with low volatility and low drawdowns; finally, for all users, the system will also push risk warnings to financial products marked as negatively abnormal, helping users fully understand market risks.

[0077] Furthermore, the steps for performing attribution analysis on abnormal financial products in the abnormal product candidate pool to determine the abnormal attributes of abnormal financial products include: inputting multi-source data of abnormal financial products into a classification model, wherein the classification model is a pre-built machine learning model for identifying the abnormal attributes of abnormal financial products; performing feature extraction and attribution analysis on abnormal financial products through the classification model, and outputting the abnormal attributes of abnormal financial products.

[0078] Specifically, after identifying anomalous financial products, in-depth attribution analysis is needed on these products within the candidate pool. A pre-trained classification model is used to extract and analyze features from the multi-source data of these anomalous products to accurately identify their anomalous attributes. First, the original multi-source data of the anomalous financial products, including historical net asset value, return, volatility, maximum drawdown, Sharpe ratio, and product management personnel information, is input into a pre-built classification model. This model, based on machine learning techniques such as support vector machines, random forests, or deep neural networks, aims to identify patterns and relationships in the data, thereby performing feature-level analysis of the anomalous products.

[0079] The classification model automatically extracts features closely related to anomalous attributes from the input data through multi-layered computation. This process not only considers intuitively visible risk-return characteristics but also delves into complex factors such as dynamic market behavior and style changes. Finally, the model performs attribution analysis on anomalous financial products based on the extracted features, aiming to analyze the specific reasons or mechanisms behind abnormal performance, such as high-yield volatility, style switching, and market mismatch. After analysis and processing by the model, the anomalous attributes of each anomalous financial product are finally output.

[0080] Furthermore, the steps for generating matching results for users by combining user profile data and abnormal attributes include: calculating the user's risk preference value based on the user profile data; classifying abnormal financial products into positive and negative abnormal financial products based on abnormal attributes, and calculating the risk value of positive abnormal financial products; matching the user with positive abnormal financial products based on the user's risk preference value and the risk value of positive abnormal financial products, and pushing successfully matched positive abnormal financial products as recommended investment financial products to the user; and pushing all negative abnormal financial products as risky financial products to the user.

[0081] Specifically, to personalize the matching of anomalous financial products for users, the system first calculates a risk preference value based on the user's profile data, including risk tolerance and investment horizon preferences. The risk preference value reflects the user's attitude towards risk in investment decisions and is a core indicator of user inclination during the matching process. Subsequently, the system calculates the risk value of positively anomalous financial products in the anomalous product candidate pool based on their anomalous characteristics, such as return volatility, liquidity risk, and style stability. Positively anomalous financial products refer to those that exhibit anomalousness in the feature space due to their unique advantages (such as high returns and low risk). The calculation of the risk value aims to quantify the potential risks of these positively anomalous products, ensuring that the recommended financial products are not only attractive but also consistent with the user's individual risk tolerance. Combining the user's profile data (mainly the risk preference value) with the risk values ​​of positively anomalous financial products, the system can select financial products that are both suitable for the user's risk tolerance and attractive, forming a list of recommended investment financial products. This process may involve comparing the relationship between the user's risk preference value and the risk value of the financial products to ensure that the recommended financial products achieve a satisfactory balance between risk and return for the user.

[0082] Finally, for all financial products marked as negatively abnormal, alert messages will be generated as if they were risky financial products and uniformly pushed to users. This reminds users to be aware of potential investment risks and not to neglect risk management even when pursuing high returns.

[0083] Through the above steps, multi-source data of financial products and user profile data are obtained, and the multi-source data is aggregated to construct feature vectors for financial products. Then, a feature space is constructed based on the feature vectors of all financial products. In the feature space, the local outlier factor value of each financial product is calculated, and financial products are screened based on the local outlier factor value to construct an abnormal product candidate pool. The local outlier factor value is used to characterize the local density deviation between product nodes and their neighboring nodes in the feature space. The product node is the representation of the financial product in the feature space. Finally, attribution analysis is performed on the abnormal financial products in the abnormal product candidate pool to determine the abnormal attributes of the abnormal financial products. The user profile data and abnormal attributes are combined to generate matching results for users. The matching results include at least one of the following: recommended investment financial products and risky financial products.

[0084] In this embodiment, for financial products, abnormal financial products are screened by calculating local outlier factor values, and an abnormal product candidate pool is constructed. This enables the identification of financial products with unique advantages or risk characteristics based solely on their intrinsic features, even without behavioral data, effectively solving the cold start problem of financial product recommendation. Subsequently, attribution analysis is performed on the financial products in the candidate pool, classifying abnormal financial products into positive and negative anomalies, and accurately matching them with users. Positive abnormal financial products with investment potential that meet the user's investment risk and investment needs are recommended, while negative abnormal financial products are given early warnings. This improves risk identification capabilities, avoids potential investment risks, and achieves accurate recommendations. Thus, it solves the technical problems of low accuracy and weak risk identification capabilities in financial product recommendation methods based on collaborative filtering and content recommendation in related technologies.

[0085] The following describes in detail another optional implementation method.

[0086] The financial product recommendation method of this invention is applied to a financial product recommendation system. It filters potentially abnormal financial products by calculating local outlier factor values ​​and performs attribution analysis on them. Then, it matches the corresponding abnormal financial products to the user and recommends or alerts the user, thereby solving the problems of cold start, homogeneous recommendation and weak dynamic risk identification ability in the prior art.

[0087] The recommendation system for this financial product includes:

[0088] The data acquisition module is used to collect raw data and user profile data of financial products (such as funds) from multiple data sources. The raw data includes, but is not limited to: historical net asset value data, returns (daily, weekly, monthly, yearly), volatility, maximum drawdown, Sharpe ratio, Sortino ratio, management information, size data, and industry distribution of holdings. The user profile data includes the user's risk assessment results, investment horizon, and historical investment behavior preferences.

[0089] For example, the raw data for a fund may include: basic fund information, including but not limited to fund code, name, type, establishment date, fund manager, fund size, etc.; fund net asset value data, including but not limited to daily / weekly adjusted net asset value per unit; fund periodic reports, including but not limited to industry distribution of stocks / bonds held, top ten holdings, etc.; and market data, including but not limited to risk-free interest rate, market benchmark index, and yield.

[0090] The preprocessing module cleans and organizes the collected raw data. Specifically, this includes: handling missing values ​​(for numerical features, filling with the median of the same category, such as equity funds); handling outliers (truncating using the 3σ rule or interquartile range); and data alignment to ensure that all fund data are within the same time interval.

[0091] The multi-dimensional feature engineering module, connected to the data acquisition and preprocessing modules, is used to extract features and construct feature spaces from the preprocessed multi-dimensional data, generating a comprehensive feature vector for each financial product. For example, a fund's feature vector may include: risk-return characteristics such as annualized return, volatility, and Sharpe ratio; dynamic behavior characteristics such as the recent net asset value's position relative to the moving average and momentum factors; and style characteristics such as large-cap / small-cap and growth / value style scores based on portfolio estimates. Taking funds as an example, this module constructs a high-dimensional feature vector for each fund, with features mainly divided into three categories:

[0092] Risk-return characteristics:

[0093] Annualized return (f_ret): f_ret=(NAV_t / NAV_{tn})^(252 / n)–1, where NAV_t is the net asset value at the end of the period, NAV_{tn} is the net asset value at the beginning of the period, and n is the number of trading days in the calculation period.

[0094] Annualized volatility (f_vol): f_vol = Std(R_i) sqrt(252), where Std(R_i) is the standard deviation of the daily return series R_i of the fund.

[0095] Maximum drawdown (f_mdd): f_mdd=max_{t∈(0,T)}(P_t-max_{i∈(0,t)}P_i) / (max_{i∈(0,t)}P_i), where P_t is the net value at time t.

[0096] Sharpe ratio (f_sharpe): f_sharpe=(f_ret-R_f) / f_vol, where R_f is the annualized risk-free rate.

[0097] Sortino ratio (f_sortino): f_sortino=(f_ret-R_f) / DownsideStd, where DownsideStd is the downside risk, which is calculated only over negative returns and annualized.

[0098] Calmar ratio (f_calmar): f_calmar = f_ret / f_mdd.

[0099] Beta coefficient (f_beta): R_fund-R_f=α+β (R_mkt-R_f)+ε, where β is f_beta.

[0100] Alpha coefficient (f_alpha): Same as above, α is f_alpha, representing excess return.

[0101] Dynamic behavioral characteristics:

[0102] Momentum factor (f_momentum): f_momentum=(NAV_t / NAV_{t-21})-1, which is the return rate of the most recent month.

[0103] Volatility change rate (f_vol_chg): f_vol_chg=(Vol_Recent-Vol_LongTerm) / Vol_LongTerm, compares recent (e.g., 3 months) volatility with long-term (e.g., 1 year) volatility.

[0104] Relative Strength Index (f_rsi): Calculated using the standard 14-day RSI formula.

[0105] Net asset value deviation (f_deviation): f_deviation=(NAV_t-MA_t) / MA_t, where MA_t is the 60-day moving average at time t.

[0106] Turnover rate (f_turnover): The annualized turnover rate obtained from the fund's periodic reports.

[0107] Style and quality characteristics:

[0108] Size style (f_size): The median of the market value of the stocks held, normalized to [-1,1], where -1 represents small-cap stocks and 1 represents large-cap stocks.

[0109] Valuation style (f_value): Calculated based on the median price-to-earnings ratio and price-to-book ratio of the stocks held, normalized to [-1,1], where -1 represents growth and 1 represents value.

[0110] Industry concentration (f_industry_con): f_industry_con=Σ(Weight_i)^2, where Weight_i is the weight of the fund in the i-th industry, i.e., the Herfindahl index.

[0111] Stock concentration (f_stock_con): The sum of the weights of the top ten holdings.

[0112] Fund Manager Score (f_manager_score): A comprehensive score based on the fund manager's years of experience and the historical performance of the products they manage.

[0113] Ultimately, each fund i can be represented as an 18-dimensional feature vector: F_i=[f_ret,f_vol,...,f_manager_score].

[0114] Local Outlier Calculation Engine: Connected to the multi-dimensional feature engineering module, this engine is configured to: receive feature vectors from all financial products, forming a high-dimensional feature space; standardize the feature space to eliminate the influence of dimensions; and calculate the LOF (Local Outlier Factor) value for each financial product. The LOF value is calculated by quantifying the local density deviation between a data point and its neighbors in the feature space to identify the degree of anomaly at that point. Points with LOF values ​​significantly greater than 1 are considered local outliers.

[0115] Furthermore, financial products with high LOF values ​​are marked as anomalous financial products. This anomalousness has two sides: positive anomalous financial products may be products with excellent fundamentals but whose prices are temporarily undervalued due to market style or short-term events, and are products with potential; negative anomalous financial products may be products that face extremely high risk of correction due to betting on a single industry or extreme style.

[0116] Figure 3 This is a schematic diagram of an optional screening process based on local outlier factors according to an embodiment of the present invention, such as... Figure 3 As shown, the screening process based on local outliers includes:

[0117] Step 1, Begin;

[0118] Step 2: Construct standardized feature vectors;

[0119] Step 3: Set the neighborhood parameter K;

[0120] Step 4: Calculate the distance matrix;

[0121] For any two points (products p and o) in the standardized feature space, the Euclidean distance (or Mahalanobis distance or other distance calculation methods: d(p,o)=sqrt(Σ(p_j-o_j)^2), where j ranges from 1 to 18) is used.

[0122] Step 5, for each financial product P;

[0123] Step 6: Calculate the K-distance;

[0124] k-distance represents the k-th distance of point p in a spatial feature. k-distance(p) is the distance from point p to its k-th nearest neighbor node.

[0125] Step 7: Determine the K-distance neighborhood;

[0126] The k-distance neighborhood N_k(p) of point p contains all neighboring points whose distance from point p does not exceed k-distance(p), that is: N_k(p)={q|d(p,q)≤k-distance(p)}.

[0127] Step 8: Calculate the locally reachable density;

[0128] The reachable distance of point p with respect to point o is defined as: reach-dist_k(p,o)=max{k-distance(o),d(p,o). This definition guarantees that when point p is in a dense region of point o, reach-dist will not be dominated by the minimum value of d(p,o), but will be smoothed by the local density k-distance(o) of o.

[0129] The local reachability density of point p is the reciprocal of the average reachability distance from all points in its k-distance neighborhood to p: lrd_k(p)=1 / ((Σ_{o∈N_k(p)}reach-dist_k(p,o)) / |N_k(p)|), where |N_k(p)| is the number of points in the neighborhood. lrd_k(p) intuitively reflects the density of the space in which point p is located. The smaller the lrd value, the sparser the area around the point.

[0130] Step 9: Calculate the local outlier factor value;

[0131] The LOF value of a point p is the ratio of the average local reachability density of all points in its neighborhood to the local reachability density of p itself: LOF_k(p)=(Σ_{o∈N_k(p)}lrd_k(o)) / |N_k(p)|) / lrd_k(p).

[0132] LOF_k(p)≈1 indicates that the density of point p is comparable to that of its neighboring points, and p is a normal point.

[0133] LOF_k(p)<1 indicates that point p is in a region that is denser than its neighbors, which is usually a normal point.

[0134] LOF_k(p)>1 indicates that the local reachability density of point p is lower than the average local reachability density of its neighboring points. That is, p is in a relatively sparse region relative to its neighbors and is a local outlier. The larger the value, the higher the degree of anomaly.

[0135] Step 10: Determine if all products have been calculated. If yes, proceed to Step 11; otherwise, return to Step 5.

[0136] Step 11: Output the product list;

[0137] All financial products are sorted according to their LOF values, and products with LOF values ​​greater than a preset threshold (e.g., 1.5 or 2.0) are selected to form an abnormal product candidate pool.

[0138] Step 12, End.

[0139] After calculating the local outlier factor value, attribution analysis was performed on the products in the outlier candidate pool. Figure 4 This is a schematic diagram of an optional attribution analysis process according to an embodiment of the present invention, such as... Figure 4 As shown, the attribution analysis process includes:

[0140] Step 1, Begin;

[0141] Step 2: Obtain the candidate pool of abnormal products;

[0142] Set a LOF threshold θ (e.g., 1.5, 2.0, which can be determined through historical backtesting) and filter out all financial products p that satisfy LOF_k(p)>θ to form a candidate pool C.

[0143] Step 3: Perform attribution analysis on each abnormal financial product;

[0144] For each anomalous financial product in candidate pool C, perform attribution analysis to determine whether it is a positive or negative anomalous product.

[0145] Calculate the Z-Score of the abnormal financial product on the multidimensional features, i.e., Z_j=(f_j-μ_j) / σ_j.

[0146] Rule engine example: If the LOF value is high AND (f_ret_Z>1.5 AND f_sharpe_Z>1 AND f_mdd_Z<-1), then it is marked as a high-return, low-drawdown positive anomaly. If the LOF value is high AND (f_vol_Z>2 AND f_mdd_Z>2 AND f_industry_con_Z>2), then it is marked as a high-risk, concentrated negative anomaly. If the LOF value is high AND (f_vol_Z<-1.5 AND f_mdd_Z<-1.5 AND f_ret_Z>0.5), then it is marked as a robust positive anomaly.

[0147] Step 4: Based on the analysis results, filter out products with positive and negative anomalies;

[0148] Step 5: Construct a positive anomaly pool and a negative anomaly pool;

[0149] Step 6: Generate alarm information for negatively abnormal financial products in the negative anomaly pool;

[0150] Step 7: Obtain user risk profiles;

[0151] Define a user risk preference vector U=[risk_tolerance,investment_horizon,...] and quantify it into a risk score S_user.

[0152] Define a matching function Match(S_fund, S_user), where S_fund is the label and risk score assigned to the abnormal financial product after attribution analysis.

[0153] Step 8: Identify the user's risk preference. If the user's risk preference is aggressive, proceed to step 9. If the user's risk preference is conservative, proceed to step 10.

[0154] Step 9: Select high-yield, high-volatility products from the positive anomaly pool and generate an aggressive recommendation list;

[0155] For aggressive users (S_user greater than the threshold): Recommendation={p|p∈C,p.type="positive anomaly",p.risk_level="high"}, based on LOF(p) The f_ret(p) function is used to sort the data and generate an aggressive recommendation list.

[0156] Step 10: Select low volatility and stable return products from the positive anomaly pool and generate a list of stable recommended products;

[0157] For robust users (S_user less than the threshold): Recommendation={p|p∈C,p.type="positive anomaly",p.risk_level="low"}, and sorted by LOF(p). f_sortino(p) sorts the data to generate a robust recommendation list.

[0158] Step 11: Integrate and output;

[0159] Step 12, End.

[0160] This invention proposes a financial product recommendation method based on the calculation of local outlier factor values. By calculating local outlier factor values, it identifies potentially abnormal financial products, performs anomaly attribution analysis, selects financial products with investment potential for recommendation to users, and issues warnings for financial products with investment risks. This addresses the problems of cold start, homogeneous recommendations, and weak dynamic risk identification capabilities in existing technologies.

[0161] The following is a detailed description with reference to another embodiment.

[0162] Example 2

[0163] The financial product recommendation device provided in this embodiment includes multiple implementation units, each of which corresponds to a specific implementation step in Embodiment 1 above. The specific implementation method and beneficial effects can be referred to the aforementioned method embodiment, and will not be repeated here.

[0164] Figure 5 This is a schematic diagram of an optional financial product recommendation device according to an embodiment of the present invention, such as... Figure 5 As shown, the device for recommending financial products may include: an acquisition unit 51, a construction unit 52, a calculation unit 53, and an analysis unit 54, wherein,

[0165] Acquisition unit 51 is used to acquire multi-source data on financial products and user profile data;

[0166] Construction unit 52 is used to aggregate multi-source data and construct feature vectors for financial products;

[0167] The computing unit 53 is used to construct a feature space based on the feature vectors of all financial products, calculate the local outlier factor value of each financial product in the feature space, and screen financial products based on the local outlier factor value to construct an abnormal product candidate pool. The local outlier factor value is used to characterize the local density deviation between the product node and its neighboring nodes in the feature space. The product node is the representation of the financial product in the feature space.

[0168] Analysis unit 54 is used to perform attribution analysis on abnormal financial products in the abnormal product candidate pool, determine the abnormal attributes of abnormal financial products, and generate matching results for users by combining user profile data and abnormal attributes. The matching results include at least one of the following: recommended investment financial products and risky financial products.

[0169] The aforementioned financial product recommendation device acquires multi-source data and user profile data of financial products through acquisition unit 51; aggregates the multi-source data through construction unit 52 to construct feature vectors of financial products; constructs a feature space based on the feature vectors of all financial products through calculation unit 53, calculates the local outlier factor value of each financial product in the feature space, and filters financial products based on the local outlier factor value to construct an abnormal product candidate pool. The local outlier factor value is used to characterize the local density deviation between product nodes and their neighboring nodes in the feature space, and the product node is the representation of the financial product in the feature space; and performs attribution analysis on the abnormal financial products in the abnormal product candidate pool through analysis unit 54 to determine the abnormal attributes of the abnormal financial products, and generates matching results for users by combining user profile data and abnormal attributes. The matching results include at least one of the following: recommended investment financial products and risky financial products.

[0170] In this embodiment, for financial products, abnormal financial products are screened by calculating local outlier factor values, and an abnormal product candidate pool is constructed. This enables the identification of financial products with unique advantages or risk characteristics based solely on their intrinsic features, even without behavioral data, effectively solving the cold start problem of financial product recommendation. Subsequently, attribution analysis is performed on the financial products in the candidate pool, classifying abnormal financial products into positive and negative anomalies, and accurately matching them with users. Positive abnormal financial products with investment potential that meet the user's investment risk and investment needs are recommended, while negative abnormal financial products are given early warnings. This improves risk identification capabilities, avoids potential investment risks, and achieves accurate recommendations. Thus, it solves the technical problems of low accuracy and weak risk identification capabilities in financial product recommendation methods based on collaborative filtering and content recommendation in related technologies.

[0171] Furthermore, the acquisition unit includes: a first acquisition module, used to acquire raw data of financial products from multiple heterogeneous data sources; and a first processing module, used to clean and integrate the raw data to obtain multi-source data.

[0172] Furthermore, the construction unit includes: a first aggregation module, used to receive the feature vectors of all financial products and aggregate the feature vectors of all financial products to construct an initial feature space; and a second processing module, used to standardize the initial feature space to obtain the feature space.

[0173] Furthermore, the calculation unit includes: a first calculation module, used to calculate the distance between any two product nodes in the feature space, and define the neighborhood of the product node based on the distance between the product nodes; a second calculation module, used to calculate the neighborhood reachability distance between the product node and the neighboring nodes in the neighborhood, and calculate the local reachability density of the product node based on all neighborhood reachability distances of the product node; a third calculation module, used to calculate the average local reachability density of all neighboring nodes in the neighborhood of the product node, to obtain the neighborhood average local reachability density of the product node; and a fourth calculation module, used to calculate the ratio between the neighborhood average local reachability density of the product node and the local reachability density of the product node, to obtain the local outlier factor value of the financial product.

[0174] Furthermore, the calculation unit also includes: a first sorting module, used to sort the local outlier factor values ​​to obtain the sorting result; a first filtering module, used to determine the local outlier factor threshold based on the sorting result, filter out all financial products that are greater than the local outlier factor threshold, and mark the filtered financial products as abnormal financial products; and a first construction module, used to construct an abnormal product candidate pool based on the abnormal financial products.

[0175] Furthermore, the analysis unit includes: a first input module, used to input multi-source data of abnormal financial products into a classification model, wherein the classification model is a pre-built machine learning model for identifying the abnormal attributes of abnormal financial products; and a first analysis module, used to perform feature extraction and attribution analysis on abnormal financial products through the classification model, and output the abnormal attributes of abnormal financial products.

[0176] Furthermore, the analysis unit also includes: a fifth calculation module, used to calculate the user's risk preference value based on user profile data; a first classification module, used to classify abnormal financial products into positive and negative abnormal financial products based on abnormal attributes, and calculate the risk value of positive abnormal financial products; a first matching module, used to match positive abnormal financial products to the user by combining the user's risk preference value and the risk value of positive abnormal financial products, and push the successfully matched positive abnormal financial products as recommended investment financial products to the user's end; and a first push module, used to push all negative abnormal financial products as risky financial products to the user's end.

[0177] It should be noted that the acquisition unit 51, construction unit 52, calculation unit 53, and analysis unit 54 mentioned above correspond to steps S201 to S204 in Embodiment 1. The instances and application scenarios implemented by the above units and corresponding steps are the same, but are not limited to the content disclosed in Embodiment 1. It should be noted that the above modules or units can be hardware or software components stored in memory (e.g., memory 104) and processed by one or more processors (e.g., processors 102a, 102b, ..., 102n). The above modules or units can also be part of a device and run in the computer terminal 10 provided in Embodiment 1.

[0178] The invention will now be described in conjunction with another alternative embodiment.

[0179] Example 3

[0180] The present invention can also provide an electronic device. Figure 6 This is a hardware structure block diagram of an electronic device (or mobile device) for performing an optional method of recommending financial products according to an embodiment of the present invention, such as... Figure 6 As shown, the electronic device may include: one or more ( Figure 6 (Only one is shown) Processor 602, memory 604, memory controller, and peripheral interface, wherein the peripheral interface is connected to the radio frequency module, audio module and display.

[0181] The memory can be used to store software programs and modules, such as the program instructions / modules corresponding to the methods and apparatus in the embodiments of this application. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, thereby implementing the above-described methods. The memory may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include memory remotely located relative to the processor, and these remote memories can be connected to the terminal via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0182] The processor can access information and applications stored in memory via a transmission device to perform the following steps: acquiring multi-source data and user profile data of financial products; aggregating the multi-source data to construct feature vectors for financial products; constructing a feature space based on the feature vectors of all financial products, calculating the local outlier factor value for each financial product in the feature space, and filtering financial products based on the local outlier factor value to construct an abnormal product candidate pool, wherein the local outlier factor value is used to characterize the local density deviation between product nodes and their neighboring nodes in the feature space, and the product node is the representation of the financial product in the feature space; performing attribution analysis on the abnormal financial products in the abnormal product candidate pool to determine the abnormal attributes of the abnormal financial products, and generating matching results for users by combining user profile data and abnormal attributes, wherein the matching results include at least one of the following: recommended investment financial products, risky financial products.

[0183] The processor can access information and applications stored in memory via a transmission device to perform the following steps: collect raw data of financial products from multiple heterogeneous data sources; perform data cleaning and data integration on the raw data to obtain multi-source data.

[0184] The processor can access the information and application programs stored in the memory via the transmission device to perform the following steps: receive the feature vectors of all financial products, aggregate the feature vectors of all financial products to construct an initial feature space; and standardize the initial feature space to obtain the feature space.

[0185] The processor can invoke information and application programs stored in the memory via a transmission device to perform the following steps: calculate the distance between any two product nodes in the feature space, and define the neighborhood of the product node based on the distance between the product nodes; calculate the neighborhood reachability distance between the product node and its neighboring nodes within the neighborhood, and calculate the local reachability density of the product node based on all neighborhood reachability distances of the product node; calculate the average local reachability density of all neighboring nodes within the neighborhood of the product node, and obtain the average local reachability density of the product node's neighborhood; calculate the ratio between the average local reachability density of the product node's neighborhood and the local reachability density of the product node, and obtain the local outlier factor value of the financial product.

[0186] The processor can call the information and application stored in the memory through the transmission device to perform the following steps: sorting the local outlier values ​​to obtain the sorting results; determining the local outlier threshold based on the sorting results, filtering out all financial products that are greater than the local outlier threshold, and marking the filtered financial products as abnormal financial products; and constructing an abnormal product candidate pool based on the abnormal financial products.

[0187] The processor can access the information and application programs stored in the memory via the transmission device to perform the following steps: inputting multi-source data of abnormal financial products into a classification model, wherein the classification model is a pre-built machine learning model for identifying the abnormal attributes of abnormal financial products; performing feature extraction and attribution analysis on the abnormal financial products through the classification model, and outputting the abnormal attributes of the abnormal financial products.

[0188] The processor can access information and applications stored in the memory via a transmission device to perform the following steps: calculate the user's risk preference value based on user profile data; classify abnormal financial products into positive and negative abnormal financial products based on abnormal attributes, and calculate the risk value of positive abnormal financial products; match positive abnormal financial products to the user by combining the user's risk preference value and the risk value of positive abnormal financial products, and push the successfully matched positive abnormal financial products as recommended investment financial products to the user's end; push all negative abnormal financial products as risky financial products to the user's end.

[0189] This invention provides a financial product recommendation scheme. For financial products, by calculating local outlier factor values, abnormal financial products are screened to construct an abnormal product candidate pool. This allows for the identification of financial products with unique advantages or risk characteristics based solely on their intrinsic features, even without behavioral data, effectively solving the cold-start problem in financial product recommendation. Subsequently, attribution analysis is performed on the financial products in the candidate pool, classifying abnormal financial products into positive and negative anomalies, and accurately matching them with users. Positive abnormal financial products with investment potential that meet the user's investment risk and needs are recommended, while negative abnormal financial products are flagged for warning. This enhances risk identification capabilities, avoids potential investment risks, and achieves accurate recommendations. This solves the technical problems of low accuracy and weak risk identification capabilities in related technologies based on collaborative filtering and content recommendation for financial product recommendation.

[0190] Those skilled in the art will understand that Figure 6 The structure shown is for illustrative purposes only. Electronic devices can also be smartphones, tablets, handheld computers, mobile internet devices (MIDs), PADs, and other terminal devices. Figure 6 This does not limit the structure of the aforementioned electronic device. For example, electronic devices may also include components that are more... Figure 6 The more or fewer components shown (such as network interfaces, display devices, etc.), or having the same Figure 6 The different configurations shown.

[0191] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing the hardware related to the terminal device. The program can be stored in a computer-readable storage medium, which may include: flash drive, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0192] The invention will now be described in conjunction with another alternative embodiment.

[0193] Example 4

[0194] This invention also provides a computer-readable storage medium. Optionally, in this invention, the computer-readable storage medium can be used to store the program code executed by the method for recommending financial products provided in Embodiment 1.

[0195] Optionally, in this embodiment of the invention, the storage medium may be located in any computer terminal in a group of computer terminals in a computer network, or in any mobile terminal in a group of mobile terminals.

[0196] This invention also provides a computer program product, which, when executed on a data processing device, is suitable for performing the steps of a financial product recommendation method: acquiring multi-source data and user profile data of financial products; aggregating the multi-source data to construct feature vectors for financial products; constructing a feature space based on the feature vectors of all financial products, calculating the local outlier factor value of each financial product in the feature space, and filtering financial products based on the local outlier factor value to construct an abnormal product candidate pool, wherein the local outlier factor value is used to characterize the local density deviation between product nodes and their neighboring nodes in the feature space, and the product node is the representation of the financial product in the feature space; performing attribution analysis on the abnormal financial products in the abnormal product candidate pool to determine the abnormal attributes of the abnormal financial products, and generating matching results for users by combining user profile data and abnormal attributes, wherein the matching results include at least one of the following: recommending investment financial products or risky financial products.

[0197] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0198] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0199] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0200] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0201] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0202] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0203] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for recommending financial products, characterized in that, include: Acquire multi-source data on financial products and user profile data; The multi-source data is aggregated to construct the feature vector of the financial product; A feature space is constructed based on the feature vectors of all the financial products. A local outlier factor value is calculated for each financial product in the feature space. The financial products are then screened based on the local outlier factor value to construct an abnormal product candidate pool. The local outlier factor value is used to characterize the local density deviation between the product node and its neighboring nodes in the feature space. The product node is the representation of the financial product in the feature space. Attribution analysis is performed on the abnormal financial products in the abnormal product candidate pool to determine the abnormal attributes of the abnormal financial products. The user profile data and the abnormal attributes are then combined to generate matching results for the user. The matching results include at least one of the following: recommended investment financial products and risky financial products.

2. The method according to claim 1, characterized in that, The steps to obtain multi-source data for financial products include: The raw data of the financial products are collected from multiple heterogeneous data sources. The original data is cleaned and integrated to obtain the multi-source data.

3. The method according to claim 1, characterized in that, The steps for constructing a feature space based on the feature vectors of all the aforementioned financial products include: Receive the feature vectors of all the financial products, and aggregate the feature vectors of all the financial products to construct an initial feature space; The initial feature space is standardized to obtain the feature space.

4. The method according to claim 1, characterized in that, The steps for calculating the local outlier factor value for each of the financial products in the feature space include: Calculate the distance between any two product nodes in the feature space, and define the neighborhood of the product node based on the distance between the product nodes; Calculate the neighborhood reachability distance between the product node and its neighboring nodes, and calculate the local reachability density of the product node based on all neighborhood reachability distances of the product node; Calculate the average local reachability density of all neighboring nodes within the neighborhood of the product node to obtain the average local reachability density of the product node's neighborhood. The local outlier factor value of the financial product is obtained by calculating the ratio between the average local reachability density of the neighborhood of the product node and the local reachability density of the product node.

5. The method according to claim 1, characterized in that, The steps of screening financial products based on the local outlier factor values ​​and constructing an abnormal product candidate pool include: The local outlier values ​​are sorted to obtain the sorting results; Based on the ranking results, a local outlier factor threshold is determined, and all financial products that exceed the local outlier factor threshold are selected and marked as abnormal financial products. The abnormal product candidate pool is constructed based on the abnormal financial products.

6. The method according to claim 1, characterized in that, The steps for performing attribution analysis on the abnormal financial products in the abnormal product candidate pool to determine the abnormal attributes of the abnormal financial products include: The multi-source data of the abnormal financial products are input into a classification model, wherein the classification model is a pre-built machine learning model for identifying the abnormal attributes of the abnormal financial products. The abnormal financial products are subjected to feature extraction and attribution analysis using the classification model, and the abnormal attributes of the abnormal financial products are output.

7. The method according to claim 1, characterized in that, The steps for generating matching results for users by combining the user profile data with the abnormal attributes include: Calculate the user's risk preference value based on the user profile data; Based on the aforementioned abnormal attributes, the abnormal financial products are classified into positive abnormal financial products and negative abnormal financial products, and the risk value of the positive abnormal financial products is calculated. The system combines the user's risk preference value and the risk value of the positively abnormal financial product to match the user with a positively abnormal financial product, and then pushes the successfully matched positively abnormal financial product as the recommended investment financial product to the user's end. All negative and abnormal financial products are pushed to the user terminal as risky financial products.

8. A device for recommending financial products, characterized in that, include: The acquisition unit is used to acquire multi-source data on financial products and user profile data. A construction unit is used to aggregate the multi-source data and construct the feature vector of the financial product; The computing unit is used to construct a feature space based on the feature vectors of all the financial products, calculate the local outlier factor value of each financial product in the feature space, and screen the financial products based on the local outlier factor value to construct an abnormal product candidate pool. The local outlier factor value is used to characterize the local density deviation between the product node and its neighboring nodes in the feature space. The product node is the representation of the financial product in the feature space. The analysis unit is used to perform attribution analysis on abnormal financial products in the abnormal product candidate pool, determine the abnormal attributes of the abnormal financial products, and generate matching results for users by combining the user profile data with the abnormal attributes. The matching results include at least one of the following: recommended investment financial products and risky financial products.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device on which the computer-readable storage medium is located to perform a method for recommending the financial product according to any one of claims 1 to 7.

10. An electronic device, characterized in that, It includes one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the method of recommending the financial product according to any one of claims 1 to 7.

Images