Financial product recommendation method and related device

Abstract

The invention discloses a financial product recommendation method and a related device, and the method comprises the steps: determining a target financial product from a plurality of candidate financial products according to the financial preference features of an object, the product features corresponding to the plurality of candidate financial products, and the historical income features corresponding to the plurality of candidate financial products in a historical period; the interest preference of the object is considered, whether the financial product accords with the interest preference is also considered, the income performance of the financial product is considered, the product characteristic of the correspondingly determined target financial product and the financial preference characteristic satisfy the preference screening condition, and the predicted income characteristic of the target financial product in the future period satisfies the income screening condition. And the method is more likely to have good income performance in the future time period. And finally, recommending the target financial product to the object. In the embodiment of the invention, the income performance of the financial product is considered, so that the problem that the financial product with poor income performance is recommended to the user and does not conform to the expectation of the user can be avoided, and the recommendation accuracy can be improved.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method and apparatus for recommending financial products. Background Technology

[0002] With the rapid development of internet technology, users can learn about financial products online, such as credit products and wealth management products. In practice, users can initiate credit applications and conduct wealth management through applications provided by banks and other institutions, or through financial websites.

[0003] Correspondingly, recommending suitable financial products to users helps them quickly find products that suit their needs or preferences, while also promoting the development of financial institutions (such as banks). Related technologies recommend financial products to users based on their interests and preferences and predefined recommendation rules. These recommendation rules can, for example, specify which financial products to recommend to user groups with particular interests and preferences.

[0004] However, the recommendation methods used in these technologies are difficult to recommend financial products that meet users' expectations, meaning there is a problem with poor recommendation accuracy. Summary of the Invention

[0005] To address the aforementioned technical problems, this application provides a method and related apparatus for recommending financial products, which can avoid the problem of recommending financial products with poor performance that do not meet user expectations, thereby improving the accuracy of recommendations.

[0006] The embodiments of this application disclose the following technical solutions:

[0007] On one hand, embodiments of this application provide a method for recommending financial products, the method comprising:

[0008] Obtain the financial preference characteristics corresponding to the object, as well as the product characteristics corresponding to multiple candidate financial products and the historical return characteristics of the multiple candidate financial products in historical periods.

[0009] Based on the financial preference characteristics, the product characteristics corresponding to the multiple candidate financial products, and the historical return characteristics corresponding to the multiple candidate financial products, a target financial product is determined from the multiple candidate financial products. The product characteristics of the target financial product satisfy the preference screening condition with the financial preference characteristics, and the predicted return characteristics of the target financial product in the future period satisfy the return screening condition. The predicted return characteristics corresponding to the target financial product are determined based on the historical return characteristics corresponding to the target financial product.

[0010] The target financial product is recommended to the target individual.

[0011] In another aspect, embodiments of this application provide a financial product recommendation device, the device comprising an acquisition unit, a determination unit, and a recommendation unit:

[0012] The acquisition unit is used to acquire the financial preference characteristics corresponding to the object, as well as the product characteristics corresponding to multiple candidate financial products and the historical return characteristics of the multiple candidate financial products in historical periods.

[0013] The determining unit is configured to determine a target financial product from multiple candidate financial products based on the financial preference features, the product features corresponding to the multiple candidate financial products respectively, and the historical return features corresponding to the multiple candidate financial products respectively. The product features of the target financial product satisfy the preference screening condition with the financial preference features, and the predicted return features of the target financial product in the future period satisfy the return screening condition. The predicted return features corresponding to the target financial product are determined based on the historical return features corresponding to the target financial product.

[0014] The recommendation unit is used to recommend the target financial product to the object.

[0015] In one possible implementation, the acquisition unit is further configured to:

[0016] Obtain the financial transaction data corresponding to the object and the interaction data of the object with respect to financial information;

[0017] The financial preference characteristics are determined based on the financial transaction data and the interaction data.

[0018] In one possible implementation, the acquisition unit is further configured to:

[0019] To obtain the market state characteristics corresponding to the financial market;

[0020] Based on the market state characteristics, a first weight corresponding to the financial transaction data and a second weight corresponding to the interaction data are determined. Different market state characteristics correspond to different weight combinations, and the weight combination includes the first weight and the second weight.

[0021] The financial preference features are determined based on the financial transaction data, the first weight, the interaction data, and the second weight.

[0022] In one possible implementation, the acquisition unit is further configured to:

[0023] Based on the market state characteristics, from the multiple sub-models included in the weighting model, the sub-model that matches the market state characteristics is determined as the target sub-model. Different sub-models match different market state characteristics.

[0024] Based on the financial transaction data and the interaction data, the first weight corresponding to the financial transaction data and the second weight corresponding to the interaction data are determined through the target sub-model.

[0025] The determining unit is further configured to:

[0026] Obtain a first training sample with a first sample label. The first training sample includes first sample financial transaction data and first sample interaction data. The first sample label is used to indicate the first sample weight corresponding to the first sample financial transaction data under the market state feature and the second sample weight corresponding to the first sample interaction data under the market state feature. Different market state features correspond to different combinations of sample weights. The combination of sample weights includes the first sample weight and the second sample weight.

[0027] Based on the first training sample, the first prediction weight corresponding to the first sample financial transaction data and the second prediction weight corresponding to the first sample interaction data are determined by the initial sub-model.

[0028] The initial sub-model is trained based on the difference between the first prediction weight, the second prediction weight, and the first sample label to obtain the target sub-model.

[0029] In one possible implementation, the acquisition unit is further configured to:

[0030] Obtain the market volatility parameters and market trend parameters corresponding to the financial market. The market volatility parameters are used to indicate the degree of volatility of the financial market in the historical period, and the market trend parameters are used to indicate the direction of the financial market in the future period.

[0031] The market state characteristics are determined based on the market fluctuation parameters and the market trend parameters.

[0032] In one possible implementation, the acquisition unit is further configured to:

[0033] Obtain multiple interaction data of the object regarding financial information;

[0034] The financial preference characteristics corresponding to the object are determined based on multiple interaction data.

[0035] In one possible implementation, if the multiple interaction data are multimodal data, the acquisition unit is further configured to:

[0036] Based on multiple interaction data, an interaction graph structure is constructed using a graph neural network. Nodes in the interaction graph structure are used to identify the interaction data. If the interaction data identified by the first node and the interaction data identified by the second node are data of different modalities, and there is an edge between the first node and the second node, then the edge is used to indicate the descriptive relationship between the interaction data identified by the first node and the interaction data identified by the second node.

[0037] Based on the interaction graph structure, the target interaction features are determined by the graph neural network;

[0038] The financial preference features are determined based on the target interaction features.

[0039] In one possible implementation, the acquisition unit is further configured to:

[0040] Multiple interaction features are obtained by extracting features from the multiple interaction data;

[0041] The fusion weight corresponding to the interaction feature is determined based on the occurrence time and interaction type of the interaction data.

[0042] The financial preference features are determined based on the multiple interaction features and the fusion weights corresponding to the multiple interaction features.

[0043] In one possible implementation, the acquisition unit is further configured to:

[0044] Multiple interaction data are clustered to obtain multiple interaction clusters. The interaction clusters are used to indicate the interaction behavior characteristics of the object in response to financial information of the target type. Different interaction clusters correspond to different target types.

[0045] The financial preference characteristics of the object are determined based on multiple interaction clusters.

[0046] In one possible implementation, if the candidate financial product belongs to the target financial type, the acquisition unit is further configured to:

[0047] Obtain the return status characteristics of the financial products of the target financial type in the financial market, and use them as the return status characteristics of the candidate financial products.

[0048] The determining unit is further configured to determine the target financial product from the multiple candidate financial products based on the financial preference characteristics, the product characteristics corresponding to the multiple candidate financial products respectively, the historical return characteristics corresponding to the multiple candidate financial products respectively, and the return status characteristics corresponding to the multiple candidate financial products respectively. The predicted return characteristics corresponding to the target financial product are determined based on the historical return characteristics and the return status characteristics corresponding to the target financial product.

[0049] In one possible implementation, the acquisition unit is further configured to:

[0050] Obtain the historical transaction data of the candidate financial product corresponding to the historical period and the product risk data of the candidate financial product;

[0051] The determining unit is further configured to determine the target financial product from the multiple candidate financial products based on the financial preference characteristics, the product characteristics corresponding to the multiple candidate financial products respectively, the historical return characteristics corresponding to the multiple candidate financial products respectively, the historical transaction data corresponding to the multiple candidate financial products respectively, and the product risk data corresponding to the multiple candidate financial products respectively. The predicted return characteristics corresponding to the target financial product are determined based on the historical return characteristics corresponding to the target financial product, the historical transaction data corresponding to the target financial product, and the product risk data corresponding to the target financial product.

[0052] In one possible implementation, the determining unit is further configured to:

[0053] Based on the financial preference characteristics, the product characteristics corresponding to the candidate financial products, and the historical return characteristics corresponding to the candidate financial products, a prediction matching degree is determined by a prediction model. The prediction matching degree is used to indicate whether the product characteristics of the candidate financial products and the financial preference characteristics meet the preference screening conditions, and whether the predicted return characteristics of the candidate financial products in the future period meet the return screening conditions.

[0054] Based on the predicted matching degree corresponding to each of the multiple candidate financial products, the candidate financial products with a predicted matching degree greater than the matching degree threshold are selected as the target financial products.

[0055] Obtain a second training sample with a second sample label. The second training sample includes the sample financial preference features of the sample object, the sample product features of the sample financial product, and the sample historical return features of the sample financial product. The second sample label is used to indicate the matching degree between the sample financial product and the sample object.

[0056] Based on the second training sample, the predicted matching degree corresponding to the sample financial product is determined by the initial model;

[0057] Based on the difference between the predicted matching degree corresponding to the sample financial product and the second sample label, the initial model is trained to obtain the prediction model.

[0058] In one possible implementation, the determining unit is further configured to:

[0059] Based on the financial preference characteristics and the product characteristics corresponding to the multiple candidate financial products, candidate financial products that satisfy the preference screening conditions between the product characteristics and the financial preference characteristics are selected as pending financial products.

[0060] Based on the historical return characteristics of the pending financial products, pending financial products whose predicted return characteristics meet the return screening conditions are selected as the target financial products.

[0061] On the other hand, embodiments of this application provide a computer device, the computer device including a processor and a memory:

[0062] The memory is used to store computer programs and to transfer the computer programs to the processor;

[0063] The processor is configured to execute the method described in any of the foregoing aspects according to instructions in the computer program.

[0064] On the other hand, embodiments of this application provide a computer-readable storage medium for storing a computer program, which, when run by a computer device, causes the computer device to perform the methods described in any of the foregoing aspects.

[0065] On the other hand, embodiments of this application provide a computer program product, including a computer program that, when run on a computer device, causes the computer device to perform the methods described in any of the foregoing aspects.

[0066] As can be seen from the above technical solution, a target financial product is determined from multiple candidate financial products based on the target's financial preference characteristics, the product characteristics of each candidate financial product, and the historical return characteristics of each candidate financial product. The product characteristics of the candidate financial products indicate their product status, while the historical return characteristics indicate their performance over historical periods. Therefore, determining the target financial product based on this method considers the target's individual interests and preferences, whether the financial product aligns with those preferences, and its return performance. Consequently, the product characteristics and financial preference characteristics of the determined target financial product satisfy the preference screening criteria, and the predicted return characteristics of the target financial product in the future satisfy the return screening criteria, meaning the target financial product is more likely to have good return performance in the future. Finally, the target financial product can be recommended to the target to meet their financial product needs and provide them with better financial returns. Compared to recommendation methods based on user interests and preferences combined with recommendation rules in related technologies, this application also considers the performance of financial products. Because it considers the performance of financial products, it can avoid the problem of recommending financial products with poor performance to users, which would not meet user expectations, thereby improving the accuracy of recommendations. Attached Figure Description

[0067] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0068] Figure 1 This is a schematic diagram illustrating an application scenario of a method for recommending financial products provided in an embodiment of this application.

[0069] Figure 2 A flowchart illustrating a method for recommending financial products provided in this application embodiment;

[0070] Figure 3 A schematic diagram illustrating the process of determining a first weight and a second weight, provided for an embodiment of this application;

[0071] Figure 4 A schematic diagram of the recommendation logic architecture of a financial product provided in this application embodiment;

[0072] Figure 5 A structural diagram of a financial product recommendation device provided in this application embodiment;

[0073] Figure 6A structural diagram of a terminal provided in an embodiment of this application;

[0074] Figure 7 This is a structural diagram of a server provided in an embodiment of this application. Detailed Implementation

[0075] The embodiments of this application will now be described with reference to the accompanying drawings.

[0076] Financial products can refer to products related to the financial sector (such as wealth management products), and can be provided by relevant institutions (such as banking institutions). Typically, users purchase financial products for purposes such as economic management; for example, they may purchase wealth management products to obtain additional returns.

[0077] In related technologies, financial products are recommended to users based on their interests and preferences, as well as predefined recommendation rules. These recommendation rules can, for example, specify which financial products to recommend to user groups with particular interests and preferences.

[0078] However, in the financial sector, the returns of financial products are highly volatile due to various factors (such as economic development trends and interest rate adjustments). Therefore, recommending financial products solely based on user interests and preferences, even if these recommendations align with user preferences, may result in poor subsequent returns, failing to meet user expectations. In other words, the recommendation methods employed by these technologies suffer from poor accuracy.

[0079] To address this, this application provides a method and related apparatus for recommending financial products. When recommending financial products, it considers not only the user's personal interests and preferences but also the historical return characteristics of the financial product over a historical period. The historical return characteristics of the financial product can indicate its performance over a historical period, thus reflecting its potential future performance. Therefore, financial products recommended by combining user interests and the product's performance are both compatible with the user's preferences and more likely to have good future performance, avoiding the problem of recommending financial products with poor performance that do not meet user expectations. Therefore, compared to the recommendation methods used in related technologies, this method has higher recommendation accuracy.

[0080] The method for recommending financial products provided in this application can be implemented using computer equipment, which can be a terminal or a server. The server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. Terminals include, but are not limited to, smartphones, computers, smart voice interaction devices, smart home appliances, and in-vehicle terminals. Terminals and servers can be directly or indirectly connected via wired or wireless communication, and this application does not impose any limitations on this connection.

[0081] The embodiments of this application can be specifically applied to scenarios involving the recommendation of financial products, such as recommending financial products to users through financial applications (such as mobile banking applications).

[0082] It should be noted that in the specific embodiments of this application, the process of recommending financial products may involve user information and other related data. When the above embodiments of this application are applied to specific products or technologies, separate consent or permission from the user is required, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0083] Figure 1 This illustration shows an application scenario of the financial product recommendation method provided in this application embodiment. Figure 1 In the scenario shown, server 100 is used as an example of the aforementioned computer device for illustration:

[0084] exist Figure 1 In the example, we take object A as an example, and multiple candidate financial products including candidate financial product 1, candidate financial product 2, ..., and candidate financial product n (n is a positive integer greater than 1).

[0085] First, server 100 can obtain the financial preference characteristics of object A, which can be used to indicate object A's interest in financial products within the financial field. Furthermore, server 100 can obtain the product characteristics corresponding to multiple candidate financial products and the historical return characteristics of multiple candidate financial products over historical periods. For example, it can obtain the product characteristics and historical return characteristics corresponding to candidate financial product 1, ..., and the product characteristics and historical return characteristics corresponding to candidate financial product n. The product characteristics can be used to indicate the product status of the candidate financial product, and the historical return characteristics can be used to indicate the return performance of the candidate financial product over historical periods.

[0086] Next, server 100 can determine the target financial product based on object A's financial preference characteristics, the product characteristics of multiple candidate financial products, and the historical return characteristics of each candidate financial product. It is evident that in determining the target financial product, not only are object A's personal interests and preferences considered, but also whether the candidate financial product itself aligns with object A's interests and preferences, and the return performance of the candidate financial product. Accordingly, the determined target financial product satisfies the preference screening criteria for its product characteristics and object A's financial preference characteristics, and its predicted return characteristics in the future period satisfy the return screening criteria. That is, the selected target financial product not only aligns with object A's personal interests and preferences but is also more likely to have good return performance in the future.

[0087] Finally, server 100 can recommend the aforementioned target financial products to object A in order to meet object A's personalized needs for financial products and bring object A more substantial financial returns.

[0088] Compared to recommendation methods based on user interests and preferences combined with recommendation rules in related technologies, this application also considers the performance of financial products. Because it considers the performance of financial products, it can avoid the problem of recommending financial products with poor performance to users, which would not meet user expectations, thereby improving the accuracy of recommendations.

[0089] Figure 2 A flowchart illustrating a method for recommending financial products provided in this application embodiment, using a server as an example of the aforementioned computer device, shows the method comprising S201-S203:

[0090] S201: Obtain the financial preference characteristics corresponding to the object, as well as the product characteristics corresponding to multiple candidate financial products and the historical return characteristics of multiple candidate financial products in historical periods.

[0091] The target audience can refer to users to whom financial products are recommended, and the candidate financial products can refer to financial products to be promoted. It should be noted that this application does not impose any limitations; the specifics can be determined based on the actual situation. For example, when applied to a banking institution, the target audience can refer to users with financial accounts at the banking institution, and the candidate financial products can be provided by the banking institution (such as wealth management products sold by the banking institution).

[0092] When recommending financial products to an object, the server can first obtain the object's corresponding financial preference characteristics. These characteristics can be used to indicate the object's interest in financial products within the financial field. For example, object A may be more interested in wealth management products, object B may be more interested in credit products, and object C may be more interested in fund products (a type of wealth management product).

[0093] Furthermore, the server can obtain the product characteristics corresponding to multiple candidate financial products and the historical return characteristics of multiple candidate financial products in different historical periods. Among them, the product characteristics can be used to indicate the product status of the candidate financial products, and the historical return characteristics can be used to indicate the return performance of the candidate financial products in historical periods.

[0094] It should be noted that this application does not impose any limitations on the methods used to determine product characteristics and historical revenue characteristics. For ease of understanding, the following examples are provided in the embodiments of this application:

[0095] As an example, the product description of a financial product (such as product details and features) can reflect the product's situation and specifically indicate what kind of financial product it is. Therefore, the corresponding product characteristics can be determined based on the product description of the candidate financial product. Furthermore, the historical yield and return trends of a financial product can reflect its performance over a historical period. Therefore, the corresponding historical return characteristics can be determined based on the historical yield and return trends of the candidate financial product.

[0096] The historical period can refer to a past timeframe of a preset duration. This application does not impose any limitations on this. In one example, the current moment can be the time when financial products need to be recommended, and the current moment can be used as the end point of the historical period, determined according to a preset duration. The preset duration can be set according to actual needs, such as a week or a month. Then, based on relevant data within this historical period, the aforementioned historical return characteristics are determined. Based on this, the latest data can be used to more accurately assess the situation of candidate financial products, thereby enabling more accurate recommendations.

[0097] S202: Based on financial preference characteristics, product characteristics corresponding to multiple candidate financial products, and historical return characteristics corresponding to multiple candidate financial products, determine the target financial product from multiple candidate financial products.

[0098] It is evident that in determining the target financial product, considerations were taken not only of the individual's interests and preferences, but also whether the candidate financial product itself aligned with those preferences, and its return performance. The historical return characteristics of the candidate financial product indicate its performance over a historical period, and thus can, to some extent, reflect its potential return performance in the future.

[0099] Thus, after comprehensively considering the above aspects, the target financial product selected from multiple candidate financial products meets the preference screening criteria in terms of its product characteristics and financial product features, that is, it aligns with the individual's interests and preferences. Furthermore, its predicted return characteristics in the future meet the return screening criteria, indicating a higher likelihood of good return performance in the future. The predicted return characteristics of the target financial product are determined based on its historical return characteristics and can be used to indicate its future return performance.

[0100] Based on this, by comprehensively considering the performance of financial products, it is helpful to select target financial products that not only match the individual's interests and preferences but are also more likely to have good performance in the future, so as to recommend them to the individual in the future.

[0101] The preference screening criteria can be used to determine whether candidate financial products match the interests and preferences of the target audience. Specifically, they can be determined based on the matching between product characteristics and financial preference characteristics. For example, if the product characteristics and financial preference characteristics match (such as a high degree of matching), then the preference screening criteria are considered to be met, and correspondingly, the candidate financial product matches the target audience's interests and preferences.

[0102] Furthermore, the return screening criteria can be used to determine whether a financial product's return performance is good in the future. Specifically, these criteria can be tailored to specific circumstances, such as setting specific return screening criteria based on predicted return characteristics. For example, predicted return characteristics can indicate the possible cumulative return value in the future; therefore, the return screening criteria can be set as the expected cumulative return value of a financial product in the future. If the cumulative return value indicated by the predicted return characteristics of a candidate financial product exceeds the expected cumulative return value, then the return screening criteria are considered met, and the candidate financial product is more likely to have good return performance in the future, thus it can be considered as a target financial product.

[0103] S203: Recommend target financial products to the target audience.

[0104] After identifying the target financial product, the server can recommend the target financial product to the object in order to meet the object's financial product needs and bring the object a better financial return.

[0105] It should be noted that this application does not impose any restrictions on how the target financial product is recommended to the recipient. In practical applications, users can learn about and purchase financial products through financial applications (such as mobile banking applications). Therefore, as an example, these applications can be used to recommend the target financial product to the recipient, making it easier for them to understand and purchase the target financial product.

[0106] As can be seen from the above technical solution, a target financial product is determined from multiple candidate financial products based on the target's financial preference characteristics, the product characteristics of each candidate financial product, and the historical return characteristics of each candidate financial product. The product characteristics of the candidate financial products indicate their product status, while the historical return characteristics indicate their performance over historical periods. Therefore, determining the target financial product based on this method considers the target's individual interests and preferences, whether the financial product aligns with those preferences, and its return performance. Consequently, the product characteristics and financial preference characteristics of the determined target financial product satisfy the preference screening criteria, and the predicted return characteristics of the target financial product in the future satisfy the return screening criteria, meaning the target financial product is more likely to have good return performance in the future. Finally, the target financial product can be recommended to the target to meet their financial product needs and provide them with better financial returns. Compared to recommendation methods based on user interests and preferences combined with recommendation rules in related technologies, this application also considers the performance of financial products. Because it considers the performance of financial products, it can avoid the problem of recommending financial products with poor performance to users, which would not meet user expectations, thereby improving the accuracy of recommendations.

[0107] The above embodiments illustrate the method for recommending financial products provided in this application. It should also be noted that this application does not limit the methods for obtaining the financial preference characteristics of the target audience or for determining the target financial product. For better understanding, this application will be illustrated through the following embodiments.

[0108] (i) Regarding the methods for obtaining the financial preference characteristics of an object, the embodiments of this application provide the following methods as examples:

[0109] In practical applications, financial transaction data refers to data generated during financial transactions (such as purchasing a financial product) by an individual in scenarios involving financial transactions (e.g., within a financial institution). This data reflects the individual's financial transaction history, economic situation, and thus their interest and preferences in financial products. For example, financial transaction data may include an individual's transaction records, consumption records, credit application records, and financial product purchase records.

[0110] Therefore, in one possible implementation, the server can obtain the financial transaction data corresponding to the object and determine the object's financial preference characteristics based on the financial transaction data. Since financial transaction data is data directly related to the financial field, it can directly reflect the object's interests and preferences in the financial field. Therefore, using financial transaction data is beneficial for determining more accurate financial preference characteristics.

[0111] For example, if the financial transaction data pertains to the financial products already held by an entity, such as entity A holding ten equity-type financial products and three bond-type financial products, then entity A's financial preference characteristics could indicate that entity A is more interested in equity-type financial products. Similarly, if entity B holds eight bond-type financial products, then entity B's financial preference characteristics could indicate that entity B is more interested in bond-type financial products.

[0112] Furthermore, interaction data between an object and financial information can refer to data generated during the interaction with financial information, reflecting the object's relevant interactive behaviors (such as forwarding, saving, browsing, commenting on, or sharing financial information). Financial information refers to information about financial types, typically used to introduce relevant situations within the financial field (such as the recent development of a certain type of financial product), thus also reflecting the object's interests and preferences in the financial field. For example, object A's interaction data could include browsing stock information 5 times or saving bond information. Similarly, object B's interaction data could include browsing history on a financial platform (such as a mobile banking application).

[0113] Therefore, in another possible implementation, the server can acquire multiple interaction data points of the object regarding financial information and determine the object's corresponding financial preference characteristics based on these data points. Since interaction data reflects the object's behavioral habits, it can indirectly reflect the object's interests and preferences in the financial field. Therefore, analyzing interaction data is beneficial for understanding the object's potential interests and preferences. Furthermore, as time progresses, the object's interests and preferences may change, and correspondingly, the object's interactive behaviors regarding financial information may also change. Therefore, using interaction data helps to determine more real-time financial preference characteristics that are closer to the object's current state.

[0114] To more accurately analyze an object's interests and preferences, another possible implementation involves the server acquiring the object's financial transaction data and interaction data related to financial information. Based on these data, financial preference characteristics can be determined. This approach, which comprehensively considers both dimensions of data, helps to identify more comprehensive financial preference characteristics, thus more accurately reflecting the object's interests and preferences in the financial field.

[0115] In practical applications, financial transaction data typically includes transaction values, and can therefore be considered a type of numerical data, also known as structured data. Interactive data, on the other hand, is much richer and more diverse in its data formats, including data generated by various interactive behaviors such as forwarding, browsing, saving, and commenting. Specifically, it can include text data, image data (such as images or emoticons entered during comments), and is a type of non-numerical data, also known as unstructured data. In this application, by integrating financial transaction data and interactive data, the integration of multi-source data—both structured and unstructured—is achieved, which is beneficial for a more comprehensive understanding of the financial preference characteristics of the target audience.

[0116] It should also be noted that this application does not impose any limitations on the specific methods for determining financial preference characteristics based on financial transaction data and / or interaction data. For ease of understanding, this application will provide examples and illustrations below:

[0117] (1) Regarding the method of determining financial preference characteristics based on interactive data, the embodiments of this application provide the following examples:

[0118] In practical applications, as recommendations are made over time, an individual's interests and preferences may change, and consequently, their behavior regarding financial information may also change. Therefore, one possible implementation is to assign weights to interaction data based on the time of occurrence, and then perform a weighted fusion of multiple interaction data points with their respective weights to determine financial preference characteristics. This approach helps to identify financial preference characteristics that better represent the individual's current interests and preferences.

[0119] Specifically, there is a positive correlation between weight and occurrence time; the longer the occurrence time, the closer it is to the present, and the better it represents the current interests and preferences of the target audience, thus it is given a greater weight. Conversely, the shorter the occurrence time, the further away it is from the present, thus it is given a smaller weight. For example, multiple interaction data points may have been generated within the aforementioned historical time period, in which case the aforementioned current moment can be used to represent the present.

[0120] It is understandable that different interactive behaviors performed by objects in response to financial information, i.e., different interaction types of interactive data, can reflect the degree of interest of the objects in that financial information. Therefore, interactive data of different interaction types can reflect the object's interest preferences to varying degrees. Thus, in another possible implementation, weights can be assigned to interactive data according to their interaction types, and then multiple interactive data and their respective weights can be weighted and fused to determine financial preference characteristics, which is beneficial for determining more accurate financial preference characteristics.

[0121] Specifically, preset weights can be assigned to each interaction type based on the degree to which different interaction types reflect the object's interests and preferences. The preset weight for each interaction type is positively correlated with the degree to which it reflects the object's interests and preferences. Then, the preset weight corresponding to the interaction type is assigned to the interaction data. For example, if an object saves financial information 1 compared to only browsing financial information 2, it indicates that the object has a greater interest in the financial products related to financial information 1. In other words, the save interaction type reflects the object's interests and preferences to a greater degree than the browse interaction type. Therefore, the preset weight for the save interaction type can be configured to be greater than that for the browse interaction type. The same approach can be used for other interaction types, such as forwarding, liking, commenting, and sharing comments.

[0122] To further improve the accuracy of financial preference characteristics, another possible implementation involves assigning weights based on both the occurrence time and interaction type. For better understanding, the embodiments of this application provide the following example:

[0123] In practice, the server can extract features from multiple interaction data points to obtain multiple interaction features. Then, the server can determine the fusion weights corresponding to the interaction features based on the occurrence time and interaction type of the interaction data. Finally, the server can determine the financial preference features based on the multiple interaction features and their respective fusion weights.

[0124] Among them, the fusion weight is determined by combining the two dimensions of occurrence time and interaction type, which can more accurately reflect the degree to which the interaction data can reflect the object's interests and preferences. Therefore, it is helpful to determine more accurate financial preference characteristics.

[0125] As an example, in practical implementation, a time weight can be determined for interactive data based on the time of its occurrence, and a preset weight corresponding to the interaction type can be used as the type weight for that interactive data. Then, the product of the time weight and the type weight is used as the fusion weight corresponding to that interaction feature. In this way, for any interactive data belonging to the same interaction type, their type weights are the same, but the time weight is also incorporated, thus distinguishing the importance of these interactive data belonging to the same interaction type. Similarly, for any interactive data belonging to the same occurrence time, their time weights are the same, but the type weight is also incorporated, thus distinguishing the importance of these interactive data belonging to the same occurrence time. Ultimately, this helps to determine more accurate financial preference characteristics.

[0126] Furthermore, this application does not impose any limitations on how to determine financial preference features based on multiple interaction features and their corresponding fusion weights. For ease of understanding, the following embodiments of this application are provided as examples:

[0127] In one possible implementation, multiple interaction features can be weighted and fused based on their respective fusion weights. The resulting feature (which can be denoted as the first fusion feature) can then be directly used as the financial preference feature, which helps to determine the financial preference feature more quickly.

[0128] This implementation method can be specifically implemented using the following formula:

[0129]

[0130] In the above formula, X can be used to represent the first fusion feature, m can be used to represent the number of interaction data (m is an integer greater than 1), i can be used to represent the i-th interaction data (i is an integer greater than or equal to 1 and less than or equal to m), x i w can be used to represent the i-th interaction feature corresponding to the i-th interaction data. i It can be used to represent the fusion weight corresponding to the i-th interaction feature.

[0131] In another possible implementation, multiple interactive features can be weighted and fused based on their respective fusion weights. Then, the feature obtained by weighted fusion (which can be denoted as the first fusion feature) and the feature corresponding to the aforementioned financial transaction data can be fused using multi-source features. The feature obtained by multi-source feature fusion (which can be denoted as the second fusion feature) can be used as the financial preference feature, which is beneficial for determining a more comprehensive financial preference feature.

[0132] It is understandable that interaction data can reflect an object's behavioral habits. Therefore, based on multiple interaction data points, it is possible to analyze an object's behavioral habits and thus its interests and preferences. Therefore, in another possible implementation, clustering can be used to analyze the distribution of an object's behavioral habits, thereby revealing its interests and preferences.

[0133] In practical implementation, the server can cluster multiple interaction data sets to obtain multiple interaction clusters. These interaction clusters indicate the behavioral characteristics of an object's interactions with target types of financial information. Different interaction clusters correspond to different target types. Therefore, multiple interaction clusters can reflect the distribution of an object's behavioral habits towards different types of financial information. Thus, the server can determine an object's financial preference characteristics based on multiple interaction clusters. Based on this, by analyzing the distribution of an object's behavioral habits through clustering, the server can uncover the object's interests and preferences in the financial field.

[0134] For example, the target type could be stock market type, dollar-cost averaging type, etc. If multiple interaction clusters reflect that the object has a lot of interaction data with financial information related to the stock market, it indicates that the object usually pays attention to stock market developments and may be more interested in stock-related financial products.

[0135] Correspondingly, in one possible implementation, the types of interactions that an object frequently focuses on can be determined based on the amount of interaction data included in each interaction cluster, thereby uncovering the object's stable interests and preferences. For example, interaction clusters containing more than a certain amount of interaction data can be designated as target interaction clusters, and then financial preference characteristics can be determined based on these target interaction clusters.

[0136] It should be noted that this application does not impose any restrictions on the choice of clustering method. As an example, the K-means clustering algorithm can be selected to cluster multiple interactive data to obtain K interactive clusters (K is a positive integer).

[0137] In practical applications, the interaction between objects and financial information may generate data in different modalities, thus making the aforementioned multiple interaction data multimodal data (such as text data, image data, audio and video data, etc.). Since different modalities of data may complement each other, they can better reflect the object's behavioral habits.

[0138] Therefore, in one possible implementation, financial preference characteristics can be determined by analyzing the correlations between data from different modalities. Graph Neural Networks (GNNs) or other deep models can model the correlations between multimodal data, thereby capturing the relationships between data from different modalities. Therefore, in practical implementation, graph neural networks or other deep models can be used to analyze multiple interactive data points that are multimodal data to better capture the correlations between data from different modalities.

[0139] To better understand, this application provides the following example, using graph neural networks for analysis:

[0140] In practical implementation, if multiple interaction data are multimodal data, the server can construct an interaction graph structure based on the multiple interaction data using a graph neural network. Nodes in the interaction graph structure can be used to identify interaction data. If the interaction data identified by the first node and the interaction data identified by the second node are different modalities, and there is an edge between the first node and the second node, then the edge can be used to indicate the descriptive relationship between the interaction data identified by the first node and the interaction data identified by the second node.

[0141] As can be seen, nodes are constructed using interactive data, and for data of different modalities, edges are constructed between nodes based on the descriptive relationships between the data of different modalities. Thus, edges can reflect the correlation between the two interactive data sets. Subsequently, the server determines the target interactive features using a graph neural network based on the interactive graph structure, and then determines the financial preference features based on the target interactive features. Based on this, by learning the nodes in the interactive graph structure through the graph neural network, as well as whether edges exist between nodes and the descriptive relationships indicated by the edges, it is possible to not only focus on the features of individual interactive data, but also understand the connections between interactive data of different modalities. This achieves the goal of cross-modal data fusion and is beneficial for determining more accurate financial preference features.

[0142] Correspondingly, the target interaction features determined by graph neural networks can refer to the overall features output after cross-modal data modeling of multiple interaction data belonging to multimodal data. These features can indicate the behavioral habits of an object in interacting with financial information, thus reflecting the object's interests and preferences in the financial field and can be used to determine the object's financial preference features. Moreover, because cross-modal data modeling is performed, it is beneficial to determine more accurate financial preference features.

[0143] In practical applications, multimodal data can include text data, image data, audio and video data, etc. For example, taking the comment interaction type as mentioned above as the interaction type of interactive data, the corresponding multiple interactive data can include at least two of the text data, image data, audio and video data, etc., entered by the object when commenting on financial information. As another example, taking the forwarding interaction type as mentioned above as the interaction type of interactive data, the corresponding multiple interactive data can include at least two of the text data, image data, audio and video data, etc., entered by the object when forwarding financial information.

[0144] The above embodiments provide a detailed explanation of how this application determines financial preference characteristics based on interactive data. It is understood that the above embodiments can also be used to describe methods for determining financial preferences based on financial transaction data. For example, weights can be assigned to financial transaction data based on transaction time to better analyze the object's recent financial transactions and thus analyze the object's current interests and preferences.

[0145] (2) Regarding the method of determining financial preference characteristics based on financial transaction data and interaction data, the embodiments of this application provide the following examples:

[0146] As mentioned earlier, financial transaction data and interaction data focus on different behaviors of the target audience. The former focuses on data generated when the target audience performs relevant actions in scenarios involving financial transactions (such as purchasing a financial product), while the latter focuses on data generated when interacting with financial information. Therefore, they are considered to be data from different sources. Because of their different sources, the degree to which they reflect the target audience's interests and preferences will also differ. In other words, the importance of different source data varies when determining financial preference characteristics. Therefore, one possible implementation is to assign weights to each source data and then perform a weighted fusion to determine the final financial preference characteristics.

[0147] It should be noted that this application does not impose any limitations on how to determine the weights for financial transaction data and interaction data. For ease of understanding, the following examples are provided in the embodiments of this application:

[0148] As an example, financial transaction data more directly reflects an individual's economic situation, thus directly reflecting their interests and preferences, while interaction data reflects an individual's behavioral habits, indirectly reflecting their interests and preferences. Therefore, it is simpler and more convenient to assign greater weight to financial transaction data and less weight to interaction data.

[0149] As another example, due to the volatility of financial markets, the generation of the aforementioned financial transaction data and interaction data will vary depending on different market conditions. For instance, when the financial market is relatively stable, individuals may be more likely to directly purchase financial products in scenarios involving financial transactions, spending less time focusing on financial-related information. Consequently, financial transaction data will better reflect the individual's interests and preferences. Conversely, when the financial market is highly volatile, individuals may be more cautious in directly purchasing financial products in scenarios involving financial transactions, and are more likely to spend more time fully understanding and learning about financial-related information. In this case, interaction data will better reflect the individual's interests and preferences.

[0150] Therefore, in another possible implementation, the weights of the aforementioned financial transaction data and interaction data can be determined based on the market conditions of the financial market. In this way, weights that are appropriate to the current situation of the financial market can be determined, which is conducive to more accurately analyzing the interests and preferences of the target and improving the accuracy of financial preference characteristics.

[0151] In practical implementation, the server can obtain the market state characteristics corresponding to the financial market. These market state characteristics can be used to indicate the current state of the financial market, specifically reflecting the current state of the financial market (such as the aforementioned stable state, volatile state, etc.). Next, based on the market state characteristics, the server can determine the first weight corresponding to the financial transaction data and the second weight corresponding to the interaction data. Then, based on the financial transaction data, the first weight, the interaction data, and the second weight, the server can determine the financial preference characteristics.

[0152] Different market state characteristics correspond to different weight combinations, including a first weight and a second weight. It is evident that both the first and second weights are adapted to the specific market conditions of the financial market; the weights differ depending on the market conditions. This allows for a more accurate analysis of an individual's interests and preferences in the current financial market, improving the accuracy of financial preference characteristics.

[0153] This application does not impose any limitations on the method of determining financial preference features based on financial transaction data, a first weight, interaction data, and a second weight. In practical applications, feature extraction can be performed on financial transaction data to obtain the first feature, and feature extraction can be performed on interaction data to obtain the second feature. Then, based on the respective weights of the two, a weighted fusion is performed, and the fused feature is used as the financial preference feature. Specifically, this can be implemented using the following formula:

[0154] h1 = f1(H1)

[0155] h2 = f2(H2)

[0156] h = r1 * h1 + r2 * h2

[0157] In the above formula, H1 can be used to represent financial transaction data, f1 can be used to represent the function for feature extraction from the financial transaction data, h1 can be used to represent the first feature, H2 can be used to represent interaction data, f2 can be used to represent the function for feature extraction from the interaction data, and h2 can be used to represent the second feature. Furthermore, r1 can be used to represent the first weight, r2 can be used to represent the second weight, and h can be used to represent the financial preference feature.

[0158] It is understood that h2 (i.e., the second feature) can be the first fused feature obtained by extracting features from multiple interaction behaviors separately as described above, and then weighting and fusing the multiple interaction features based on their respective fusion weights. For details, please refer to the aforementioned embodiments and their formula examples. Alternatively, h2 can also be the target interaction feature determined by the aforementioned graph neural network. For details, please refer to the relevant descriptions in the aforementioned embodiments. Similarly, after fusing the first feature corresponding to the financial transaction data and the second feature corresponding to the interaction data using multi-source features, the resulting h can be the aforementioned second fused feature. In this embodiment, h obtained from multi-source feature fusion is directly used as the financial preference feature. For details, please refer to the relevant descriptions in the aforementioned embodiments.

[0159] It should also be noted that this application does not impose any limitations on the methods for determining the first and second weights, or on the methods for determining the market state characteristics of the financial market. For ease of understanding, the embodiments of this application provide the following examples:

[0160] On the one hand, regarding how to determine the first weight and the second weight, this application provides the following method as an example:

[0161] In practical applications, due to the volatility of financial markets, the market conditions can vary significantly when recommending financial products to different individuals at different times. Therefore, to more quickly determine the weights appropriate for the current market situation when making recommendations, one possible implementation is to construct a weight model comprising multiple sub-models. This weight model can be used to determine the aforementioned first and second weights. Specifically, one sub-model can correspond to one market state characteristic, and different sub-models can correspond to different market state characteristics. Thus, when determining weights, the sub-model matching the current market state characteristics can be directly selected from among the multiple sub-models and directly invoked, thereby improving efficiency.

[0162] In practical implementation, the server can determine the target sub-model from among the multiple sub-models included in the weighted model, based on market state characteristics. Different market state characteristics match different sub-models; that is, different market state characteristics can determine different target sub-models. Next, the server can determine the first weight corresponding to the financial transaction data and the second weight corresponding to the interaction data based on the target sub-model, using the financial transaction data and interaction data. Therefore, when there is a recommendation requirement, the server can directly call the matching sub-model to determine the corresponding first and second weights based on the current market state characteristics, which improves efficiency.

[0163] To better understand, embodiments of this application also provide, as follows: Figure 3 The diagram illustrates the process of determining the first and second weights. Specifically:

[0164] exist Figure 3 In the example, the weight model can include sub-model 1, sub-model 2, ..., and sub-model q (q is an integer greater than 1), meaning there are q sub-models in total. Based on the market state characteristics corresponding to the current financial market, sub-model 2 is identified as the matching model; therefore, sub-model 2 is the target sub-model. Next, financial transaction data and interaction data can be input into sub-model 2, which outputs the first weight corresponding to the financial transaction data and the second weight corresponding to the interaction data. Thus, in practical applications, the required first and second weights can be quickly determined by directly calling the sub-model that matches the market state characteristics, improving efficiency.

[0165] To better understand the methods for determining each sub-model, this embodiment will use the determination of the target sub-model as an example. Specifically, the target sub-model can be determined in the following way:

[0166] First, the server can obtain a first training sample with a first sample label. This first training sample may include first sample financial transaction data and first sample interaction data. The first sample label indicates the first sample weight corresponding to the first sample financial transaction data under market state characteristics, and the second sample weight corresponding to the first sample interaction data under the same market state characteristics. Different market state characteristics correspond to different combinations of sample weights, including both the first and second sample weights. Next, based on the first training sample, the server can determine the first prediction weight corresponding to the first sample financial transaction data and the second prediction weight corresponding to the first sample interaction data through an initial sub-model. Then, based on the differences between the first prediction weight, the second prediction weight, and the first sample label, the server trains the initial sub-model to obtain the target sub-model.

[0167] Based on this, during model training, the initial sub-model can learn the correlation between different source data and their weights under the characteristics of this market state. It is understood that the training process for models with multiple sub-models can be explained in the examples above, and will not be repeated here.

[0168] It should also be noted that this application does not impose any restrictions on the settings of the weight model and the initial sub-model. In practical applications, a model with an attention mechanism can be used as the initial sub-model. Through the attention mechanism, the importance of different source data under the characteristics of this market state can be learned more quickly and effectively, that is, the relationship between different source data and their weights can be learned. Therefore, as an example, the aforementioned initial sub-model can be a model with an attention mechanism, and the weight model can also be called an attention model, which includes multiple sub-models.

[0169] On the other hand, regarding how to determine the market state characteristics of the financial market, the embodiments of this application provide the following methods as examples:

[0170] In one possible implementation, the server can obtain market volatility parameters and market trend parameters corresponding to the financial market, and determine the market state characteristics based on these parameters. The market volatility parameters can indicate the degree of volatility in the financial market over a historical period (e.g., volatility), while the market trend parameters can indicate the direction of the financial market's trend in the future (e.g., rising, falling, consolidation, etc.).

[0171] Based on this, by combining market volatility parameters and market trend parameters, we can consider both the fluctuations of the financial market in historical periods and the development direction of the financial market in future periods, thereby enabling a more accurate analysis of market conditions and the determination of more accurate market state characteristics.

[0172] It should be noted that this application does not impose any limitations on the methods for obtaining market fluctuation parameters and market trend parameters, nor on the methods for determining market state characteristics based on these parameters. For ease of understanding, the following examples are provided in the embodiments of this application:

[0173] In practical applications, the volatility of financial markets over historical periods can measure the degree of fluctuation in asset prices, and thus market volatility parameters can be determined based on this. For example, stock price data and trading volume in financial markets can be obtained to assess volatility over historical periods. Furthermore, in financial markets, comparing price highs and lows can identify trend indicators, reflecting the main direction of price development in the future, and thus market trend parameters can be determined based on this. For example, stock price highs and lows in financial markets over historical periods can be obtained to determine stock price trend indicators.

[0174] Understandably, in addition to these, a wider range of volatility indicators (such as Bollinger Bands) and trend indicators (such as moving averages) can be obtained to determine more accurate market volatility and trend parameters.

[0175] In practice, the aforementioned market data can be collected, cleaned, and then feature extracted to transform it into market state features (e.g., feature vectors) that can be understood and processed by computer equipment for subsequent use. Typically, suitable models (such as classification or regression models) can be built to analyze the obtained market state features (e.g., feature vectors) to determine the current market state. This allows for a more intuitive understanding of the current financial market situation through state categories, such as a stable state, a fluctuating state, a continuously declining state, or a state about to rebound.

[0176] (II) Regarding the method for determining the target financial product, the embodiments of this application provide the following methods as examples:

[0177] Understandably, target financial products can refer to candidate financial products that not only align with the target audience's interests and preferences but are also more likely to demonstrate strong returns in the future. Therefore, in another possible implementation, the screening process can be divided into two steps: first, screening candidate financial products that align with the target audience's interests and preferences, and then further screening from these candidate financial products that align with the target audience's interests and preferences for candidate financial products with strong return performance.

[0178] Therefore, as an example, in the specific implementation of the aforementioned S202, the server can first select candidate financial products as pending financial products based on financial preference characteristics and the product characteristics corresponding to multiple candidate financial products, where the product characteristics and financial preference characteristics satisfy the preference selection criteria. That is, the pending financial products conform to the object's interest preferences. Next, the server can select pending financial products as target financial products based on the historical return characteristics corresponding to the pending financial products, where the predicted return characteristics satisfy the return selection criteria.

[0179] Based on this, a two-step screening process can identify financial products that not only align with the target audience's interests and preferences but are also more likely to demonstrate strong returns in the future. Since the second step only requires evaluating the returns of the financial products identified in the first step, the number of potential financial products is typically less than the number of candidate financial products. Therefore, the two-step screening approach helps reduce computational load.

[0180] To improve the efficiency of identifying target financial products, one possible approach is to use a predictive model to quickly predict whether each candidate financial product is suitable as a target product to be recommended to the target audience, thereby improving efficiency.

[0181] Therefore, as another example, in the specific implementation of the aforementioned S202, the server can determine the prediction matching degree of the candidate financial product through the prediction model based on the financial preference characteristics, the product characteristics corresponding to the candidate financial products, and the historical return characteristics corresponding to the candidate financial products.

[0182] The predicted matching degree can be used to indicate whether the product characteristics of a candidate financial product meet the preference screening criteria, and whether the predicted return characteristics of the candidate financial product in the future meet the return screening criteria. Therefore, the server can select candidate financial products with predicted matching degrees greater than a matching degree threshold as target financial products based on the predicted matching degrees of multiple candidate financial products. The matching degree threshold can be preset according to actual circumstances. For example, a reasonable matching degree threshold can be set based on experience to assess whether it both matches the target's interests and preferences and is more likely to have good return performance.

[0183] Therefore, for multiple candidate financial products, predictive models can be used to determine their respective predictive matching degrees more quickly. Typically, the predictive matching degree can be a value between 0 and 1, thus providing a numerical and intuitive reflection of the degree of matching between the candidate financial product and the target. This matching degree comprehensively considers whether the candidate financial product aligns with the target's interests and preferences, as well as its future return performance, thus providing a numerical and intuitive representation of the overall matching outcome.

[0184] To better understand, embodiments of this application provide illustrative examples of how to determine the prediction model. Specifically, the aforementioned prediction model can be determined in the following manner:

[0185] First, the server can obtain second training samples with second sample labels. These second training samples may include the financial preference characteristics of the sample objects, the product characteristics of the sample financial products, and the historical return characteristics of the sample financial products. The second sample labels can be used to indicate the matching degree between the sample financial products and the sample objects. Next, the server can determine the predicted matching degree corresponding to the sample financial products using an initial model based on the second training samples. Then, based on the difference between the predicted matching degree corresponding to the sample financial products and the second sample labels, the initial model is trained to obtain the prediction model.

[0186] Based on this, during model training, the initial model can learn the correlation between three parts of data—sample financial preference features, sample product features, and sample historical return features—and the matching degree. Specifically, it can learn the correlation between comprehensively considering whether product features match interest preferences and how well the returns perform, and the matching degree. After model training is complete, these three parts of data can be directly input into the prediction model, and the predicted matching degree can be output to screen target financial products.

[0187] In practical applications, the model training process can employ online learning techniques to fully learn the correlation between the three parts of data and the matching degree. The goal of online learning can be to minimize the cumulative loss, thereby minimizing the difference between the predicted matching degree of the sample financial product and the second sample label. Specifically, the cumulative loss can be expressed by the following formula:

[0188]

[0189] In the above formula, t can be used to represent the number of second training samples used in one round of model training (t is an integer greater than or equal to 1), j can be used to represent the j-th second training sample (j is an integer greater than or equal to 1 and less than or equal to t), Y j It can be used to represent the specific content of the j-th second training sample (such as the sample financial preference characteristics of the sample object, the sample product characteristics of the sample financial product, and the sample historical return characteristics of the sample financial product mentioned above), y j θ can be used to represent the label of the j-th second training sample, θ can be used to represent the model parameters of the initial model, and f can be used to represent the data processing function of the initial model. j ;θ) can be used to represent the predicted matching degree corresponding to the j-th second training sample. L can be used to represent the difference between the predicted matching degree corresponding to the j-th second training sample and the label of the j-th second sample. t (θ) can be used for the cumulative loss corresponding to t second training samples.

[0190] In the actual model training process, t second training samples are used in one round of model training, and the cumulative loss is calculated. Then, the model parameters of the initial model are adjusted with the goal of minimizing the cumulative loss. After adjustment, the next round of model training is performed until the cumulative loss reaches the goal of minimization, at which point the model training can be terminated. Correspondingly, the initial model at the end of the training can be used as the aforementioned prediction model, and the model parameters at this time can be denoted as θ0.

[0191] Subsequently, in the application phase, the financial preference characteristics of the target, the product characteristics of the candidate financial products, and the historical return characteristics can be input into the prediction model. The prediction model then outputs the corresponding prediction matching degree. This can be implemented using the following formula:

[0192]

[0193] In the above formula, Y0 can be used to represent the input data, which can specifically include the financial preference characteristics of the aforementioned objects, the product characteristics of candidate financial products, and historical return characteristics; θ0 can be used to represent the model parameters of the prediction model; and f can be used to represent the data processing function of the prediction model. It can be used to represent the predicted matching degree determined based on the current input data.

[0194] In this way, the accurate predictive matching of the prediction model can be used to select target financial products from multiple candidate financial products for recommendation to the target individual. Understandably, to further improve recommendation accuracy, feedback information from the target financial product recipient can be collected after the recommendation. This feedback information can reflect whether the recipient meets the recommended target financial product or why they are dissatisfied. Then, based on the feedback information provided by the recipient, the prediction model can be adjusted using reinforcement learning, aiming to use the adjusted prediction model to identify target financial products that are more satisfactory to the recipient. Of course, the aforementioned feature extraction or feature fusion strategies can also be adjusted based on the feedback information to identify more accurate relevant features (such as more accurate financial preference features), thereby facilitating the identification of target financial products that are more satisfactory to the recipient.

[0195] It is evident that a mechanism is provided to optimize and adjust the recommendation strategy based on the feedback information provided by the object after the recommendation, in order to update the recommendation strategy according to the object's satisfaction level and improve the recommendation accuracy.

[0196] The above embodiments illustrate S202 from the perspective of specific determination methods. In the aforementioned embodiments, data from three aspects are integrated: the financial preference characteristics of the target, the product characteristics and historical return characteristics of the candidate financial products. It is understood that introducing more comprehensive data may improve prediction accuracy. Therefore, the following embodiments will further illustrate S202 from the perspective of introducing richer data:

[0197] In practical applications, the returns of individual financial products are also affected by overall market conditions. That is, due to the volatility of the financial market, the future returns of a particular financial product may be influenced. Therefore, to further improve recommendation accuracy, one possible approach is to combine the overall market performance of the financial product with the selection of a target financial product from multiple candidate products. Since the overall market performance of a financial product can reflect its potential future returns, incorporating this data helps improve the accuracy of the predicted return characteristics of candidate financial products, thereby increasing the accuracy of target product selection and ultimately improving recommendation accuracy.

[0198] In practical implementation, if the candidate financial product belongs to the target financial type, the server can also obtain the return status characteristics of financial products of the target financial type in the financial market, as the return status characteristics of the candidate financial product. These return status characteristics reflect the overall market performance of such financial products. Next, the server can determine the target financial product from multiple candidate financial products based on financial preference characteristics, product characteristics of each candidate financial product, historical return characteristics of each candidate financial product, and return status characteristics of each candidate financial product. Correspondingly, the predicted return characteristics of the target financial product can be determined based on its historical return characteristics and return status characteristics.

[0199] It is evident that by considering the historical return characteristics of candidate financial products and the overall market conditions for such products, it is possible to more accurately analyze their potential future performance and thus determine more accurate predictive return characteristics. Based on this, the impact of the overall market conditions for a class of financial products on the future performance of individual financial products is comprehensively considered. This allows for more accurate predictions of future financial product performance, enabling the selection of financial products with better future returns and their recommendation to the target audience. This approach better avoids the negative impact on satisfaction caused by recommending poorly performing financial products, thereby improving recommendation accuracy.

[0200] For example, if the target financial product type is an equity product, the historical return characteristics of candidate financial product A indicate that its performance over a historical period has been quite good. However, the return characteristics of equity products in the financial market indicate that the overall market for equity products is relatively sluggish. Based on this, it can be inferred that candidate financial product A will be affected by the overall sluggish market, resulting in a decline in its future returns compared to its historical returns.

[0201] Corresponding to this method of combining the overall market situation of a type of financial product, it can also be combined with the aforementioned implementation method based on prediction models. Accordingly, the formula for determining the prediction matching degree based on prediction models can be modified as follows:

[0202]

[0203] In the above formula, Y0 can be used to represent the input data, which can specifically include the financial preference characteristics of the aforementioned object, the product characteristics of the candidate financial products, and the historical return characteristics. V0 can be used to represent the return status characteristics corresponding to the candidate financial products. θ0 can be used to represent the model parameters of the prediction model, and f can be used to represent the data processing function of the prediction model. It can be used to represent the predictive matching degree determined based on the current input data and the characteristics of the profit status.

[0204] As can be seen, the return status features corresponding to candidate financial products have been added to the input of the prediction model to comprehensively consider the overall market conditions of such financial products, so as to output a more accurate prediction matching degree.

[0205] This application does not impose any limitations on the methods used to obtain the return characteristics of financial products of the target financial type in the financial market. In practical applications, the historical return characteristics of multiple financial products of the target financial type over historical periods can be obtained, and then the overall market trend of such financial products can be comprehensively analyzed. Of course, it is also possible to combine relevant financial information (such as media reports) related to the financial products of the target financial type to comprehensively analyze the overall market trend of such financial products.

[0206] In practical applications, financial products with higher risks are more prone to larger fluctuations in returns. Similarly, financial products with higher trading volumes indicate a larger user base and can, to some extent, reflect the product's return performance (e.g., a reasonably good return). Therefore, another possible approach is to combine these two data points to select a target financial product from multiple candidate products.

[0207] In practice, the server first obtains historical transaction data and product risk data for candidate financial products within a given historical period. The historical transaction data indicates the trading volume of the candidate financial product during that period (e.g., the number of trades, the total amount of assets traded), while the product risk data indicates the investment risk of the candidate financial product. Next, the server determines the target financial product from among the multiple candidate financial products based on financial preference characteristics, the product characteristics of each candidate financial product, the historical return characteristics of each candidate financial product, the historical transaction data of each candidate financial product, and the product risk data of each candidate financial product. Correspondingly, the predicted return characteristics of the target financial product can be determined based on its historical return characteristics, historical transaction data, and product risk data.

[0208] It is evident that comprehensively considering the historical trading volume and investment risk of candidate financial products allows for a more accurate analysis of their potential future returns, thereby determining more precise predictive return characteristics. Based on this, it is beneficial to identify financial products with more stable returns in the future and recommend them to potential clients, better avoiding the negative impact on satisfaction caused by recommending poorly performing products, and improving recommendation accuracy.

[0209] Corresponding to this implementation method, it can also be combined with the aforementioned implementation method based on the prediction model. Accordingly, the formula for determining the prediction matching degree based on the prediction model can be supplemented with historical transaction data and product risk data corresponding to the candidate financial products. For details, please refer to the aforementioned embodiments, which will not be repeated here.

[0210] It is understood that, in addition to the combined implementations shown in the above embodiments, this application can be flexibly combined with the implementations provided in the above aspects to provide more implementations in order to obtain higher recommendation accuracy.

[0211] The above embodiments have comprehensively illustrated the method for recommending financial products provided in this application. It should also be noted that, regarding the parts related to feature extraction (such as feature extraction from interaction data, feature extraction from financial transaction data, etc.), this application does not impose any limitations on the specific method of feature extraction. For ease of understanding, the embodiments of this application provide the following examples:

[0212] In practical applications, if the data to be processed (such as interactive data) is text-based, language models can be used to extract features from it, resulting in more accurate features, because language models have a better understanding of text-based data. For example, a pre-trained Bidirectional Encoder Representations from Transformers (BERT) model can be used. The text-based data to be processed (such as text-based interactive data, financial transaction data, or the aforementioned feedback information) can be input into the BERT model, which will then output the corresponding features, such as the interactive features for interactive data and the first feature for financial transaction data. In this way, text-based data can be converted into features (e.g., high-dimensional vector representations) for subsequent use.

[0213] The BERT model, in particular, can be a model pre-trained using Natural Language Processing (NLP) techniques, leveraging linguistic knowledge from a large-scale corpus. Specifically, it can be an encoder structure based on a multi-layered bidirectional Transformer, capable of capturing more comprehensive contextual information of the text. This ensures that the features of each word depend not only on itself but also on its surrounding words, thus providing better text understanding and improving the accuracy of the output features. NLP, on the other hand, is a technology involving the interaction between computers and natural language, specifically the ability of computer programs to process and analyze large amounts of natural language data.

[0214] To further improve the accuracy of BERT's output features, this application utilizes relevant data from the financial field (such as domain knowledge and financial transaction data) to fine-tune the BERT model. This enhances BERT's ability to understand textual data within the financial domain, specifically by improving its understanding of specific languages, symbols, and data structures. Based on this, more accurate features can be output for the aforementioned financial transaction data and interactive data.

[0215] In practical applications, the features output by BERT can be fused with other features to generate more comprehensive and richer features. For example, the first feature corresponding to financial transaction data can be output based on the BERT model, and the target interaction feature corresponding to the interaction data can be output based on the graph neural network. Then, multi-source feature fusion can be performed on these two parts to obtain the corresponding financial preference features, which can more comprehensively and accurately reflect the user's interests and preferences in the financial field. Subsequently, a better understanding of user needs and preferences can be achieved, leading to more accurate personalized recommendations, increasing user purchase rates and conversion rates for financial products, as well as improving user satisfaction and stickiness.

[0216] To better understand, this application also provides a recommendation system architecture, which may include a data collection module, a data preprocessing module, a model training module, an interest and preference analysis module, and a real-time recommendation module, specifically:

[0217] The data collection module can be used to collect and store relevant data required for the implementation process of this application (such as real-time collection to ensure the timeliness and accuracy of the data), such as the aforementioned financial transaction data and interaction data. In practical applications, distributed databases (such as Hadoop, Spark, etc.) can be used for data storage and management.

[0218] The data preprocessing module can preprocess the raw data stored in the data collection module. For example, it can perform data cleaning to remove noise and invalid information, and fill in missing values. It can perform data normalization to make the data meet the requirements for subsequent use (meeting the model's requirements for input data). It can perform data encoding for direct use later.

[0219] The model training module can be used to efficiently train models based on large amounts of data (such as the aforementioned first and second training samples) to obtain the desired models (such as the aforementioned weighted models and prediction models). In practical applications, deep learning frameworks (such as TensorFlow and PyTorch) can be used for model training, and graphics processing units (GPUs) can be used to accelerate computation and improve the efficiency of model training. Furthermore, to improve the model's data processing capabilities, a large model (LM) can be selected. For example, a large model can be chosen as the initial model, and then trained using the second training samples to obtain the prediction model, thereby improving the predictive ability of the prediction model. Here, a large model refers to a highly complex machine learning model trained on a large amount of data, typically containing hundreds of millions to tens of billions of parameters, thus possessing stronger data processing capabilities.

[0220] The interest and preference analysis module can be used to analyze the interests and preferences of an object. Specifically, it can be used to determine financial preference characteristics and then use these characteristics to indicate the object's interests and preferences.

[0221] The real-time recommendation module can be used to make recommendations to users in real time. If a user's interests and preferences change, the target financial products can be redefined based on the updated preferences for real-time recommendations. This allows for the immediate recommendation of relevant financial products to users as their needs change, combining timeliness and efficiency to improve user satisfaction. Typically, during the interaction between a user and the financial platform, the user's financial preference characteristics can be analyzed and updated in real time to promptly detect changes in their interests. In practical applications, streaming data processing technologies (such as Apache Kafka and Flink) can be used for real-time data processing and recommendations.

[0222] Furthermore, for a more comprehensive understanding, based on the foregoing embodiments, the embodiments of this application also provide, as follows: Figure 4 The diagram shown illustrates the recommendation logic architecture of a financial product, which may specifically include the following three parts:

[0223] The first part involves determining the financial preference characteristics of the target. Specifically, this involves acquiring the target's financial transaction data and interaction data, and extracting features from each, resulting in a first feature corresponding to the financial transaction data and a second feature corresponding to the interaction data. Next, multi-source feature fusion can be performed on the first and second features, and the fused feature can be used as the financial preference characteristic. For example, multi-source feature fusion can employ weighted fusion based on the first and second weights, as described in the preceding embodiments. Furthermore, the second feature can be obtained by extracting features from the interaction data. In one example, if the interaction data is multimodal, the multimodal data can be input into a graph neural network, which outputs a target interaction feature, which can be directly used as the second feature. In another example, features can be extracted from multiple interaction data separately to obtain multiple corresponding interaction features, and then these multiple interaction features can be weighted and fused (e.g., using the aforementioned weighted fusion based on their respective fusion weights). The fused feature can be denoted as the first fused feature, which can also be directly used as the second feature. It is understood that for implementation details of each part, please refer to the relevant descriptions in the preceding embodiments; they will not be repeated here.

[0224] The second part determines the predictive fit for each candidate financial product. Specifically, financial preference characteristics and relevant data of the candidate financial products (such as product characteristics, historical return characteristics, historical transaction data, product risk data, and return status characteristics) are input into the prediction model. The prediction model then outputs the predictive fit for each candidate financial product. For each candidate financial product, its corresponding predictive fit can be determined based on this.

[0225] The third part involves identifying the target financial product. For example, candidate financial product 1 corresponds to a predicted matching degree of 1, candidate financial product 2 corresponds to a predicted matching degree of 2, candidate financial product 3 corresponds to a predicted matching degree of 3, and so on, with candidate financial product n corresponding to a predicted matching degree of n. Next, based on whether the predicted matching degree is greater than a matching degree threshold, it is determined whether the candidate financial product can be used as the target financial product. For instance, the selected target financial products may include candidate financial product 2, candidate financial product 3, candidate financial product 5 (not shown in the figure), and candidate product n. Finally, the identified target financial products can be recommended to the target audience.

[0226] It should be noted that, based on the implementation methods provided in the above aspects, this application can be further combined to provide more implementation methods.

[0227] based on Figure 2Corresponding to the financial product recommendation method provided in the embodiments, this application also provides a financial product recommendation device 500, which includes an acquisition unit 501, a determination unit 502, and a recommendation unit 503.

[0228] The acquisition unit 501 is used to acquire the financial preference features corresponding to the object, as well as the product features corresponding to multiple candidate financial products and the historical return features corresponding to the multiple candidate financial products in historical periods.

[0229] The determining unit 502 is used to determine a target financial product from multiple candidate financial products based on the financial preference features, the product features corresponding to the multiple candidate financial products respectively, and the historical return features corresponding to the multiple candidate financial products respectively. The product features of the target financial product satisfy the preference screening condition with the financial preference features, and the predicted return features of the target financial product in the future period satisfy the return screening condition. The predicted return features corresponding to the target financial product are determined based on the historical return features corresponding to the target financial product.

[0230] The recommendation unit 503 is used to recommend the target financial product to the object.

[0231] In one possible implementation, the acquisition unit is further configured to:

[0232] Obtain the financial transaction data corresponding to the object and the interaction data of the object with respect to financial information;

[0233] The financial preference characteristics are determined based on the financial transaction data and the interaction data.

[0234] In one possible implementation, the acquisition unit is further configured to:

[0235] To obtain the market state characteristics corresponding to the financial market;

[0236] Based on the market state characteristics, a first weight corresponding to the financial transaction data and a second weight corresponding to the interaction data are determined. Different market state characteristics correspond to different weight combinations, and the weight combination includes the first weight and the second weight.

[0237] The financial preference features are determined based on the financial transaction data, the first weight, the interaction data, and the second weight.

[0238] In one possible implementation, the acquisition unit is further configured to:

[0239] Based on the market state characteristics, from the multiple sub-models included in the weighting model, the sub-model that matches the market state characteristics is determined as the target sub-model. Different sub-models match different market state characteristics.

[0240] Based on the financial transaction data and the interaction data, the first weight corresponding to the financial transaction data and the second weight corresponding to the interaction data are determined through the target sub-model.

[0241] The determining unit is further configured to:

[0242] Obtain a first training sample with a first sample label. The first training sample includes first sample financial transaction data and first sample interaction data. The first sample label is used to indicate the first sample weight corresponding to the first sample financial transaction data under the market state feature and the second sample weight corresponding to the first sample interaction data under the market state feature. Different market state features correspond to different combinations of sample weights. The combination of sample weights includes the first sample weight and the second sample weight.

[0243] Based on the first training sample, the first prediction weight corresponding to the first sample financial transaction data and the second prediction weight corresponding to the first sample interaction data are determined by the initial sub-model.

[0244] The initial sub-model is trained based on the difference between the first prediction weight, the second prediction weight, and the first sample label to obtain the target sub-model.

[0245] In one possible implementation, the acquisition unit is further configured to:

[0246] Obtain the market volatility parameters and market trend parameters corresponding to the financial market. The market volatility parameters are used to indicate the degree of volatility of the financial market in the historical period, and the market trend parameters are used to indicate the direction of the financial market in the future period.

[0247] The market state characteristics are determined based on the market fluctuation parameters and the market trend parameters.

[0248] In one possible implementation, the acquisition unit is further configured to:

[0249] Obtain multiple interaction data of the object regarding financial information;

[0250] The financial preference characteristics corresponding to the object are determined based on multiple interaction data.

[0251] In one possible implementation, if the multiple interaction data are multimodal data, the acquisition unit is further configured to:

[0252] Based on multiple interaction data, an interaction graph structure is constructed using a graph neural network. Nodes in the interaction graph structure are used to identify the interaction data. If the interaction data identified by the first node and the interaction data identified by the second node are data of different modalities, and there is an edge between the first node and the second node, then the edge is used to indicate the descriptive relationship between the interaction data identified by the first node and the interaction data identified by the second node.

[0253] Based on the interaction graph structure, the target interaction features are determined by the graph neural network;

[0254] The financial preference features are determined based on the target interaction features.

[0255] In one possible implementation, the acquisition unit is further configured to:

[0256] Multiple interaction features are obtained by extracting features from the multiple interaction data;

[0257] The fusion weight corresponding to the interaction feature is determined based on the occurrence time and interaction type of the interaction data.

[0258] The financial preference features are determined based on the multiple interaction features and the fusion weights corresponding to the multiple interaction features.

[0259] In one possible implementation, the acquisition unit is further configured to:

[0260] Multiple interaction data are clustered to obtain multiple interaction clusters. The interaction clusters are used to indicate the interaction behavior characteristics of the object in response to financial information of the target type. Different interaction clusters correspond to different target types.

[0261] The financial preference characteristics of the object are determined based on multiple interaction clusters.

[0262] In one possible implementation, if the candidate financial product belongs to the target financial type, the acquisition unit is further configured to:

[0263] Obtain the return status characteristics of the financial products of the target financial type in the financial market, and use them as the return status characteristics of the candidate financial products.

[0264] The determining unit is further configured to determine the target financial product from the multiple candidate financial products based on the financial preference characteristics, the product characteristics corresponding to the multiple candidate financial products respectively, the historical return characteristics corresponding to the multiple candidate financial products respectively, and the return status characteristics corresponding to the multiple candidate financial products respectively. The predicted return characteristics corresponding to the target financial product are determined based on the historical return characteristics and the return status characteristics corresponding to the target financial product.

[0265] In one possible implementation, the acquisition unit is further configured to:

[0266] Obtain the historical transaction data of the candidate financial product corresponding to the historical period and the product risk data of the candidate financial product;

[0267] The determining unit is further configured to determine the target financial product from the multiple candidate financial products based on the financial preference characteristics, the product characteristics corresponding to the multiple candidate financial products respectively, the historical return characteristics corresponding to the multiple candidate financial products respectively, the historical transaction data corresponding to the multiple candidate financial products respectively, and the product risk data corresponding to the multiple candidate financial products respectively. The predicted return characteristics corresponding to the target financial product are determined based on the historical return characteristics corresponding to the target financial product, the historical transaction data corresponding to the target financial product, and the product risk data corresponding to the target financial product.

[0268] In one possible implementation, the determining unit is further configured to:

[0269] Based on the financial preference characteristics, the product characteristics corresponding to the candidate financial products, and the historical return characteristics corresponding to the candidate financial products, a prediction matching degree is determined by a prediction model. The prediction matching degree is used to indicate whether the product characteristics of the candidate financial products and the financial preference characteristics meet the preference screening conditions, and whether the predicted return characteristics of the candidate financial products in the future period meet the return screening conditions.

[0270] Based on the predicted matching degree corresponding to each of the multiple candidate financial products, the candidate financial products with a predicted matching degree greater than the matching degree threshold are selected as the target financial products.

[0271] Obtain a second training sample with a second sample label. The second training sample includes the sample financial preference features of the sample object, the sample product features of the sample financial product, and the sample historical return features of the sample financial product. The second sample label is used to indicate the matching degree between the sample financial product and the sample object.

[0272] Based on the second training sample, the predicted matching degree corresponding to the sample financial product is determined by the initial model;

[0273] Based on the difference between the predicted matching degree corresponding to the sample financial product and the second sample label, the initial model is trained to obtain the prediction model.

[0274] In one possible implementation, the determining unit is further configured to:

[0275] Based on the financial preference characteristics and the product characteristics corresponding to the multiple candidate financial products, candidate financial products that satisfy the preference screening conditions between the product characteristics and the financial preference characteristics are selected as pending financial products.

[0276] Based on the historical return characteristics of the pending financial products, pending financial products whose predicted return characteristics meet the return screening conditions are selected as the target financial products.

[0277] As can be seen from the above technical solution, a target financial product is determined from multiple candidate financial products based on the target's financial preference characteristics, the product characteristics of each candidate financial product, and the historical return characteristics of each candidate financial product. The product characteristics of the candidate financial products indicate their product status, while the historical return characteristics indicate their performance over historical periods. Therefore, determining the target financial product based on this method considers the target's individual interests and preferences, whether the financial product aligns with those preferences, and its return performance. Consequently, the product characteristics and financial preference characteristics of the determined target financial product satisfy the preference screening criteria, and the predicted return characteristics of the target financial product in the future satisfy the return screening criteria, meaning the target financial product is more likely to have good return performance in the future. Finally, the target financial product can be recommended to the target to meet their financial product needs and provide them with better financial returns. Compared to recommendation methods based on user interests and preferences combined with recommendation rules in related technologies, this application also considers the performance of financial products. Because it considers the performance of financial products, it can avoid the problem of recommending financial products with poor performance to users, which would not meet user expectations, thereby improving the accuracy of recommendations.

[0278] This application also provides a computer device, which can be a terminal, taking a smartphone as an example:

[0279] Figure 6 The diagram shown is a block diagram of a portion of the structure of a smartphone provided in an embodiment of this application. (Reference) Figure 6The smartphone includes components such as a radio frequency (RF) circuit 610, a memory 620, an input unit 630, a display unit 640, a sensor 650, an audio circuit 660, a Wi-Fi module 670, a processor 680, and a power supply 690. The input unit 630 may include a touch panel 631 and other input devices 632, the display unit 640 may include a display panel 641, and the audio circuit 660 may include a speaker 661 and a microphone 662. Those skilled in the art will understand that... Figure 6 The smartphone structure shown does not constitute a limitation on smartphones and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0280] The memory 620 can be used to store software programs and modules. The processor 680 executes various functions and data processing of the smartphone by running the software programs and modules stored in the memory 620. The memory 620 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, applications required for at least one function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created based on the use of the smartphone (such as audio data, phonebook, etc.). In addition, the memory 620 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0281] The processor 680 is the control center of the smartphone, connecting various parts of the smartphone via various interfaces and lines. It performs various functions and processes data by running or executing software programs and / or modules stored in the memory 620, and by accessing data stored in the memory 620. Optionally, the processor 680 may include one or more processing units; preferably, the processor 680 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user interface, and applications, and the modem processor mainly handles wireless communication. It is understood that the modem processor may not be integrated into the processor 680.

[0282] In this embodiment, the steps performed by the processor 680 in the smartphone can be based on Figure 6 The structure shown is implemented.

[0283] The computer device provided in this application embodiment can also be a server. Please refer to [link / reference]. Figure 7 As shown, Figure 7This is a structural diagram of a server 700 provided in an embodiment of this application. The server 700 can vary significantly due to different configurations or performance. It may include one or more processors, such as a Central Processing Unit (CPU) 722, and a memory 732, and one or more storage media 730 (e.g., one or more mass storage devices) for storing application programs 742 or data 744. The memory 732 and storage media 730 can be temporary or persistent storage. The program stored in the storage media 730 may include one or more modules (not shown in the diagram), each module including a series of instruction operations on the server. Furthermore, the CPU 722 may be configured to communicate with the storage media 730 and execute the series of instruction operations stored in the storage media 730 on the server 700.

[0284] Server 700 may also include one or more power supplies 726, one or more wired or wireless network interfaces 750, one or more input / output interfaces 758, and / or one or more operating systems 741, such as Windows Server. TM Mac OS X TM Unix TM Linux TM FreeBSD TM etc.

[0285] In this embodiment, the central processing unit 722 in server 700 can perform the following steps:

[0286] Obtain the financial preference characteristics corresponding to the object, as well as the product characteristics corresponding to multiple candidate financial products and the historical return characteristics of the multiple candidate financial products in historical periods.

[0287] Based on the financial preference characteristics, the product characteristics corresponding to the multiple candidate financial products, and the historical return characteristics corresponding to the multiple candidate financial products, a target financial product is determined from the multiple candidate financial products. The product characteristics of the target financial product satisfy the preference screening condition with the financial preference characteristics, and the predicted return characteristics of the target financial product in the future period satisfy the return screening condition. The predicted return characteristics corresponding to the target financial product are determined based on the historical return characteristics corresponding to the target financial product.

[0288] The target financial product is recommended to the target individual.

[0289] According to one aspect of this application, a computer-readable storage medium is provided for storing a computer program that, when executed by a computer device, causes the computer device to perform the method for recommending financial products as described in the foregoing embodiments.

[0290] According to one aspect of this application, a computer program product is provided, comprising a computer program stored in a computer-readable storage medium. A processor of a computer device reads the computer program from the computer-readable storage medium and executes the computer program, causing the computer device to perform the methods provided in various optional implementations of the above embodiments.

[0291] The descriptions of the processes or structures corresponding to the above figures each have their own emphasis. For parts of a process or structure that are not described in detail, please refer to the relevant descriptions of other processes or structures.

[0292] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application 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 this application described herein can be implemented, for example, 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.

[0293] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between apparatuses or units through some interfaces, and may be electrical, mechanical, or other forms.

[0294] 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 network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0295] Furthermore, the functional units in the various embodiments of this application 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.

[0296] 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 this application, in essence, or the part that contributes to related technologies, 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 computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0297] In this application embodiment, the terms "module" or "unit" refer to a computer program or part of a computer program that has a predetermined function and works with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more modules or units. Furthermore, each module or unit can be part of an overall module or unit that includes the functionality of that module or unit.

[0298] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for recommending financial products, characterized in that, The method includes: Obtain the financial preference characteristics corresponding to the object, as well as the product characteristics corresponding to multiple candidate financial products and the historical return characteristics of the multiple candidate financial products in historical periods. Based on the financial preference characteristics, the product characteristics corresponding to the multiple candidate financial products, and the historical return characteristics corresponding to the multiple candidate financial products, a target financial product is determined from the multiple candidate financial products. The product characteristics of the target financial product satisfy the preference screening condition with the financial preference characteristics, and the predicted return characteristics of the target financial product in the future period satisfy the return screening condition. The predicted return characteristics corresponding to the target financial product are determined based on the historical return characteristics corresponding to the target financial product. The target financial product is recommended to the target individual.

2. The method according to claim 1, characterized in that, The financial preference characteristics corresponding to the acquired object include: Obtain the financial transaction data corresponding to the object and the interaction data of the object with respect to financial information; The financial preference characteristics are determined based on the financial transaction data and the interaction data.

3. The method according to claim 2, characterized in that, The method further includes: To obtain the market state characteristics corresponding to the financial market; Based on the market state characteristics, a first weight corresponding to the financial transaction data and a second weight corresponding to the interaction data are determined. Different market state characteristics correspond to different weight combinations, and the weight combination includes the first weight and the second weight. The step of determining the financial preference characteristics based on the financial transaction data and the interaction data includes: The financial preference features are determined based on the financial transaction data, the first weight, the interaction data, and the second weight.

4. The method according to claim 3, characterized in that, The step of determining the first weight corresponding to the financial transaction data and the second weight corresponding to the interaction data based on the market state characteristics includes: Based on the market state characteristics, from the multiple sub-models included in the weighting model, the sub-model that matches the market state characteristics is determined as the target sub-model. Different sub-models match different market state characteristics. Based on the financial transaction data and the interaction data, the first weight corresponding to the financial transaction data and the second weight corresponding to the interaction data are determined through the target sub-model. The target sub-model is determined in the following way: Obtain a first training sample with a first sample label. The first training sample includes first sample financial transaction data and first sample interaction data. The first sample label is used to indicate the first sample weight corresponding to the first sample financial transaction data under the market state feature and the second sample weight corresponding to the first sample interaction data under the market state feature. Different market state features correspond to different combinations of sample weights. The combination of sample weights includes the first sample weight and the second sample weight. Based on the first training sample, the first prediction weight corresponding to the first sample financial transaction data and the second prediction weight corresponding to the first sample interaction data are determined by the initial sub-model. The initial sub-model is trained based on the difference between the first prediction weight, the second prediction weight, and the first sample label to obtain the target sub-model.

5. The method according to claim 3, characterized in that, The acquisition of market state characteristics corresponding to the financial market includes: Obtain the market volatility parameters and market trend parameters corresponding to the financial market. The market volatility parameters are used to indicate the degree of volatility of the financial market in the historical period, and the market trend parameters are used to indicate the direction of the financial market in the future period. The market state characteristics are determined based on the market fluctuation parameters and the market trend parameters.

6. The method according to claim 1, characterized in that, The financial preference characteristics corresponding to the acquired object include: Obtain multiple interaction data of the object regarding financial information; The financial preference characteristics corresponding to the object are determined based on multiple interaction data.

7. The method according to claim 6, characterized in that, If the multiple interaction data are multimodal data, determining the financial preference features corresponding to the object based on the multiple interaction data includes: Based on multiple interaction data, an interaction graph structure is constructed using a graph neural network. Nodes in the interaction graph structure are used to identify the interaction data. If the interaction data identified by the first node and the interaction data identified by the second node are data of different modalities, and there is an edge between the first node and the second node, then the edge is used to indicate the descriptive relationship between the interaction data identified by the first node and the interaction data identified by the second node. Based on the interaction graph structure, the target interaction features are determined by the graph neural network; The financial preference features are determined based on the target interaction features.

8. The method according to claim 6, characterized in that, The step of determining the financial preference characteristics corresponding to the object based on multiple sets of interaction data includes: Multiple interaction features are obtained by extracting features from the multiple interaction data; The fusion weight corresponding to the interaction feature is determined based on the occurrence time and interaction type of the interaction data. The financial preference features are determined based on the multiple interaction features and the fusion weights corresponding to the multiple interaction features.

9. The method according to claim 6, characterized in that, The step of determining the financial preference characteristics corresponding to the object based on multiple sets of interaction data includes: Multiple interaction data are clustered to obtain multiple interaction clusters. The interaction clusters are used to indicate the interaction behavior characteristics of the object in response to financial information of the target type. Different interaction clusters correspond to different target types. The financial preference characteristics of the object are determined based on multiple interaction clusters.

10. The method according to any one of claims 1-9, characterized in that, If the candidate financial product belongs to the target financial type, the method further includes: Obtain the return status characteristics of the financial products of the target financial type in the financial market, and use them as the return status characteristics of the candidate financial products. The step of determining the target financial product from the multiple candidate financial products based on the financial preference characteristics, the product characteristics corresponding to the multiple candidate financial products, and the historical return characteristics corresponding to the multiple candidate financial products includes: Based on the financial preference characteristics, the product characteristics corresponding to the multiple candidate financial products, the historical return characteristics corresponding to the multiple candidate financial products, and the return status characteristics corresponding to the multiple candidate financial products, the target financial product is determined from the multiple candidate financial products. The predicted return characteristics of the target financial product are determined based on the historical return characteristics and the return status characteristics of the target financial product.

11. The method according to any one of claims 1-9, characterized in that, The method further includes: Obtain the historical transaction data of the candidate financial product corresponding to the historical period and the product risk data of the candidate financial product; The step of determining the target financial product from the multiple candidate financial products based on the financial preference characteristics, the product characteristics corresponding to the multiple candidate financial products, and the historical return characteristics corresponding to the multiple candidate financial products includes: Based on the financial preference characteristics, the product characteristics corresponding to the multiple candidate financial products, the historical return characteristics corresponding to the multiple candidate financial products, the historical transaction data corresponding to the multiple candidate financial products, and the product risk data corresponding to the multiple candidate financial products, the target financial product is determined from the multiple candidate financial products. The predicted return characteristics of the target financial product are determined based on the historical return characteristics, the historical transaction data, and the product risk data of the target financial product.

12. The method according to any one of claims 1-9, characterized in that, The step of determining the target financial product from the multiple candidate financial products based on the financial preference characteristics, the product characteristics corresponding to the multiple candidate financial products, and the historical return characteristics corresponding to the multiple candidate financial products includes: Based on the financial preference characteristics, the product characteristics corresponding to the candidate financial products, and the historical return characteristics corresponding to the candidate financial products, a prediction matching degree is determined by a prediction model. The prediction matching degree is used to indicate whether the product characteristics of the candidate financial products and the financial preference characteristics meet the preference screening conditions, and whether the predicted return characteristics of the candidate financial products in the future period meet the return screening conditions. Based on the predicted matching degree corresponding to each of the multiple candidate financial products, the candidate financial products with a predicted matching degree greater than the matching degree threshold are selected as the target financial products. The prediction model is determined in the following way: Obtain a second training sample with a second sample label. The second training sample includes the sample financial preference features of the sample object, the sample product features of the sample financial product, and the sample historical return features of the sample financial product. The second sample label is used to indicate the matching degree between the sample financial product and the sample object. Based on the second training sample, the predicted matching degree corresponding to the sample financial product is determined by the initial model; Based on the difference between the predicted matching degree corresponding to the sample financial product and the second sample label, the initial model is trained to obtain the prediction model.

13. The method according to any one of claims 1-9, characterized in that, The step of determining the target financial product from the multiple candidate financial products based on the financial preference characteristics, the product characteristics corresponding to the multiple candidate financial products, and the historical return characteristics corresponding to the multiple candidate financial products includes: Based on the financial preference characteristics and the product characteristics corresponding to the multiple candidate financial products, candidate financial products that satisfy the preference screening conditions between the product characteristics and the financial preference characteristics are selected as pending financial products. Based on the historical return characteristics of the pending financial products, pending financial products whose predicted return characteristics meet the return screening conditions are selected as the target financial products.

14. A device for recommending financial products, characterized in that, The device includes an acquisition unit, a determination unit, and a recommendation unit: The acquisition unit is used to acquire the financial preference characteristics corresponding to the object, as well as the product characteristics corresponding to multiple candidate financial products and the historical return characteristics of the multiple candidate financial products in historical periods. The determining unit is configured to determine a target financial product from multiple candidate financial products based on the financial preference features, the product features corresponding to the multiple candidate financial products respectively, and the historical return features corresponding to the multiple candidate financial products respectively. The product features of the target financial product satisfy the preference screening condition with the financial preference features, and the predicted return features of the target financial product in the future period satisfy the return screening condition. The predicted return features corresponding to the target financial product are determined based on the historical return features corresponding to the target financial product. The recommendation unit is used to recommend the target financial product to the object.

15. A computer device, characterized in that, The computer device includes a processor and memory: The memory is used to store computer programs and to transfer the computer programs to the processor; The processor is configured to execute the method according to any one of claims 1-13 according to instructions in the computer program.

16. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store a computer program, which, when executed by a computer device, causes the computer device to perform the method according to any one of claims 1-13.

17. A computer program product, comprising a computer program, characterized in that, When it is run on a computer device, it causes the computer device to perform the method according to any one of claims 1-13.

Images