Financial product push method and device for atypical information overload

Abstract

The present invention provides a method and device for pushing financial products with atypical information overload, which can be used in the field of artificial intelligence technology. The method includes: extracting features from user information to obtain user demand features; predicting the pushed financial products by a deep learning model formed based on machine learning; expanding and predicting the pushed financial products according to a directed relationship graph of all preset financial products to obtain expanded financial products, and obtaining an expanded product portfolio based on the pushed financial products and the expanded financial products to push to users. The present invention proposes a method for pushing financial products with a small number of products, no information overload, and weak homogeneity between products, improving the accuracy of financial product pushing.

Description

Technical Field

[0001] The present invention relates to the technical field of financial product push, in particular to the field of artificial intelligence technology, and more particularly to a method and device for pushing financial products with atypical information overload. Background Art

[0002] Existing e-commerce platforms usually provide products to consumers through the recommendation algorithms of the recommendation system. However, the current recommendation algorithms cannot be well applied to the recommendation of financial products for corporate users of financial institutions. Specifically, first, one of the important assumptions of the existing recommendation algorithms is obvious information overload, that is, the number of products is huge. Users cannot directly select products and need to rely on the recommendation algorithm because they cannot exhaustively browse the products, rather than because they do not understand the product content. That is, assuming that the number of products retrieved by a user for a certain type of product on the e-commerce platform is very limited, then the user does not need to rely on the existing recommendation system and can select by himself. The number of financial products provided by a financial institution for corporate users is limited, and the order of magnitude is far lower than that of the commodities on an e-commerce platform. Therefore, there is no typical information overload situation when users make selections. Second, one of the important assumptions of the existing recommendation system algorithms is that the homogeneity among similar products is strong. However, the financial products provided by a financial institution for corporate users have weak homogeneity, and most of them are point-to-point coverage, that is, a certain function is only provided by a certain product. Third, although the number of products on the e-commerce platform is large, most of the products are for ordinary consumers, and the information on product functions, risks, etc. has no barriers and is easy to obtain and understand. However, financial products have a certain degree of information barriers, and non-professionals have obstacles in information acquisition and understanding. Therefore, corporate users can usually only put forward the direction of the required financial products, but can hardly clarify specific financial products. Summary of the Invention

[0003] An object of the present invention is to provide a method for pushing financial products with atypical information overload, and to propose a method for pushing financial products with few products, no information overload and weak homogeneity among products, so as to improve the accuracy of financial product push. Another object of the present invention is to provide a device for pushing financial products with atypical information overload. Still another object of the present invention is to provide a computer device. Yet another object of the present invention is to provide a readable medium.

[0004] To achieve the above objects, on the one hand, the present invention discloses a method for pushing financial products with atypical information overload, including:

[0005] Extracting feature of user information to obtain user demand feature;

[0006] Predicting the pushed financial products by using a deep learning model formed based on machine learning for the user demand feature;

[0007] Based on the directed relationship graph of all preset financial products, the pushed financial products are expanded and predicted to obtain expanded financial products, and an expanded product portfolio is obtained based on the pushed financial products and the expanded financial products to be pushed to users.

[0008] Preferably, the feature extraction of user information to obtain user demand features specifically includes:

[0009] Feature extraction is performed on the basic information and behavioral information in the user information to obtain user demand features.

[0010] Preferably, it further includes the step of obtaining the deep learning model based on the principle of machine learning in advance:

[0011] Historical user demand features are obtained based on the basic information and behavioral information of user information in historical data;

[0012] The pre-constructed machine learning model is trained according to the user's product subscription information, industry and macro information in historical data and the historical user demand features to obtain a deep learning model.

[0013] Preferably, the training of the pre-constructed machine learning model according to the user's product subscription information, industry and macro information in historical data and the historical user demand features to obtain a deep learning model specifically includes:

[0014] An undirected relationship graph of financial products is obtained according to the user's product subscription information in the historical data;

[0015] The pre-constructed machine learning model is trained according to the undirected relationship graph, the industry and macro information, and the historical user demand features to obtain a deep learning model.

[0016] Preferably, the obtaining of the undirected relationship graph of financial products according to the user's product subscription information in the historical data specifically includes:

[0017] A financial product graph is formed according to all financial product information;

[0018] The relationship strength factor between every two financial products in the financial product graph is determined according to the subscription information of users in historical data;

[0019] The relationship strength factor is used as the edge of the corresponding two financial product nodes to obtain an undirected relationship graph.

[0020] Preferably, it further includes the step of forming the directed relationship graph in advance:

[0021] A financial product graph is formed according to all financial product information;

[0022] Determine the directed strength factor between every two financial products in the financial product graph according to the subscription information of users in historical data;

[0023] Use the directed strength factor as the edge between the corresponding two financial product nodes to obtain a directed relationship graph.

[0024] Preferably, the forming of the financial product graph according to all financial product information specifically includes:

[0025] Determine the product feature information of each financial product according to all financial product information;

[0026] Establish the financial product graph with each financial product as a node and the corresponding product feature information as the node attribute.

[0027] The present invention also discloses a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor,

[0028] When the processor executes the program, the above-mentioned method is implemented.

[0029] The present invention also discloses a computer-readable medium, on which a computer program is stored,

[0030] When the program is executed by the processor, the above-mentioned method is implemented.

[0031] The method for pushing financial products with non-atypical information overload of the present invention extracts the feature of user information to obtain user demand features, predicts the pushed financial products through a deep learning model formed based on machine learning for the user demand features, expands and predicts the pushed financial products according to the preset directed relationship graph of all financial products to obtain expanded financial products, and obtains an expanded product portfolio according to the pushed financial products and the expanded financial products to push to users. Thus, the present invention predicts the user demand features extracted from the user information features through a preset deep learning model formed based on machine learning to obtain the pushed financial products that can be recommended to users, and then expands and predicts the pushed financial products through the preset directed relationship graph of financial products to obtain expanded financial products to form an expanded product portfolio to push to users, thereby completing the matching mapping between financial products and users. Thus, the present invention proposes a method for pushing financial products with few product numbers, no information overload, and weak homogeneity between products, improving the accuracy of financial product pushing. Description of the Drawings

[0032] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0033] Figure 1 Flowchart showing a specific embodiment of the method for pushing financial products with atypical information overload of the present invention;

[0034] Figure 2 Flowchart showing the process of obtaining the deep learning model based on the principle of machine learning in advance in a specific embodiment of the method for pushing financial products with atypical information overload of the present invention;

[0035] Figure 3 Flowchart showing a specific embodiment S020 of the method for pushing financial products with atypical information overload of the present invention;

[0036] Figure 4 Flowchart showing a specific embodiment S021 of the method for pushing financial products with atypical information overload of the present invention;

[0037] Figure 5 Flowchart showing a specific embodiment S030 of the method for pushing financial products with atypical information overload of the present invention;

[0038] Figure 6 Flowchart showing a specific embodiment S031 of the method for pushing financial products with atypical information overload of the present invention;

[0039] Figure 7 Structure diagram showing a specific embodiment of the device for pushing financial products with atypical information overload of the present invention;

[0040] Figure 8 Structure diagram showing a computer device suitable for implementing the embodiments of the present invention. Detailed implementation manners

[0041] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0042] It should be noted that the method and device for pushing financial products with atypical information overload disclosed in this application can be used in the field of artificial intelligence technology, and can also be used in any field other than the field of artificial intelligence technology. The application field of the method and device for pushing financial products with atypical information overload disclosed in this application is not limited.

[0043] According to one aspect of the present invention, an embodiment discloses a method for pushing financial products with atypical information overload. As Figure 1 shown, in this embodiment, the method includes:

[0044] S100: Extract features from user information to obtain user demand features.

[0045] S200: Predict the pushed financial products by using a deep learning model formed based on machine learning for the user demand features.

[0046] S300: Expand and predict the pushed financial products according to the directed relationship graph of all preset financial products to obtain expanded financial products, and obtain an expanded product portfolio based on the pushed financial products and the expanded financial products to push to users.

[0047] The method for pushing financial products with atypical information overload of the present invention extracts features from user information to obtain user demand features, predicts the pushed financial products by using a deep learning model formed based on machine learning for the user demand features, expands and predicts the pushed financial products according to the directed relationship graph of all preset financial products to obtain expanded financial products, and obtains an expanded product portfolio based on the pushed financial products and the expanded financial products to push to users. Thus, the present invention predicts the user demand features extracted from the user information characteristics through a preset deep learning model formed based on machine learning to obtain the pushed financial products that can be recommended to users, and then expands and predicts the pushed financial products through the preset directed relationship graph of financial products to obtain expanded financial products to form an expanded product portfolio to push to users, thereby completing the matching mapping between financial products and users. Therefore, the present invention proposes a method for pushing financial products with a small number of products, no information overload, and weak homogeneity between products, improving the accuracy of financial product pushing.

[0048] In a preferred embodiment, the specific steps of S100 for extracting features from user information to obtain user demand features may include:

[0049] S110: Extract features from the basic information and behavioral information in the user information to obtain user demand features.

[0050] Specifically, the basic information and behavioral information in the user information can be analyzed to obtain the user information, and then feature extraction is performed on the basic information and behavioral information and converted into feature vectors to obtain user demand features. Thus, the user demand features can be directly input into a deep learning model to predict and push financial products. Among them, the basic information in the user information can include at least one of the information such as the registration location, the industry to which it belongs, the nature of the enterprise, the registered capital, and the registration time. The behavioral information in the user information can include transactions and the economic relationships established by the transactions. The transaction information can include at least one of the information such as the total settlement amount, the average daily balance for the year, various transaction volumes, and various credit limits.

[0051] In a preferred embodiment, as Figure 2 shown, the method further includes the step of obtaining the deep learning model based on the principle of machine learning in advance:

[0052] S010: Obtain historical user demand features according to the basic information and behavioral information of the user information in the historical data;

[0053] S020: Train a pre-constructed machine learning model according to the product subscription information, industry and macro information in the historical data and the historical user demand features to obtain a deep learning model.

[0054] Specifically, in this preferred embodiment, in order to improve the prediction accuracy, considering the influence of multi-dimensional factors on financial product recommendation, historical user demand features of feature vectors are obtained by performing feature extraction on the basic information and behavioral information of the user information in the historical data. Further considering the industry and macro information, a training sample set is formed according to the historical user demand features and the industry and macro information, and a machine learning model formed based on the principle of machine learning is trained to obtain a deep learning model.

[0055] For example, in a specific example, define the full user set C = {Ci}. For each user Ci, construct a user basic feature vector (historical user demand feature) CiF = (Cif1, Cif2,...). The set of all user features is obtained as CF = {CiF}, where the full user set includes three categories: user basic information, user behavioral information, industry, and macroeconomic information. Among them, the basic information in the user information can include at least one of the information such as the registration location, the industry to which it belongs, the nature of the enterprise, the registered capital, and the registration time. The behavioral information in the user information takes the information registered and filed by corporate users in the industrial and commercial bureau as the core, and can include transactions and the economic relationships established by the transactions. The transaction information can include at least one of the information such as the total settlement amount, the average daily balance for the year, various transaction volumes, and various credit limits.

[0056] For industry and macro information, define the full set of industries I = {Iu}. The characteristics of a certain industry Iu are represented by the vector IuF = (Iuf1, Iuf2,...), and the characteristics include information such as industry output value, production capacity, common upstream and downstream, etc. According to the industry to which it belongs in the basic information of Ci, it is associated with the corresponding Iu, and its IuF is taken as a part of Ci. Macro data may include at least one of the money supply of various calibers (M0 / M1 / M2), various monetary policy tools (SLO / SLF / MLF / TMLF / PSL / CRA / TLF) quantification and short- and long-term deposit and loan interest rates, various liquidity regulatory indicators (LCR / NSFR), GDP at the national and provincial and municipal levels, output value and production capacity distribution of the primary, secondary, and tertiary industries, population, number of registered enterprises, and the change trends of the above-mentioned values, etc.

[0057] Finally, according to the full set of users C = {Ci} and the user characteristics CiF = (Cif1, Cif2,...), the set of users (regarded as having relatively sufficient service opening, and n can be customized according to the actual situation) with the number of financial products / services (hereinafter referred to as "services") opened or used greater than or equal to n (n is an integer greater than 1) is taken as the basic sample set {CSbn} for establishing user information, and it is classified according to the service opening situation, and the characteristics CiF of each type of user and other means are used to aggregate and extract user needs.

[0058] In a preferred embodiment, as Figure 3 shown, the S020 trains a pre-constructed machine learning model according to the product subscription information, industry and macro information, and the historical user demand characteristics of users in the historical data to obtain a deep learning model, which specifically may include:

[0059] S021: Obtain an undirected relationship graph of financial products according to the product subscription information of users in the historical data.

[0060] S022: Train a pre-constructed machine learning model according to the undirected relationship graph, the industry and macro information, and the historical user demand characteristics to obtain a deep learning model.

[0061] Specifically, in this preferred embodiment, the product subscription information of users in the historical data is further analyzed to obtain an undirected relationship graph of financial products. The deep learning model trained on the machine learning model according to the undirected relationship graph of this financial product, the historical user demand characteristics obtained by feature extraction of user information, and the industry and macro information can predict financial products with high matching degrees for users.

[0062] In a preferred embodiment, as Figure 4As shown, the specific process of the S021 obtaining the undirected relationship graph of financial products based on the product subscription information of users in the historical data includes:

[0063] S0211: Form a financial product graph based on all financial product information.

[0064] S0212: Determine the relationship strength factor between every two financial products in the financial product graph according to the subscription information of users in the historical data.

[0065] S0213: Use the relationship strength factor as the edge between the corresponding two financial product nodes to obtain an undirected relationship graph.

[0066] Specifically, an undirected relationship graph of all financial products can be obtained according to the relevant information of users' subscribed financial products in the historical data. For example, in a specific example, the situation of users opening financial products is expressed in the form of a graph. Define the full set of corporate financial products S = {Sj}, where the granularity of a financial product can be defined by the recommendation system according to the actual situation, and each Sj is a node of the graph. For a certain financial product Sj, describe the feature vector SjF = (Sjf1, Sjf2,...) of the financial product in the specified dimension according to the historical data, including the occurrence frequency, total transaction volume (only service items involving amounts, otherwise recorded as null), core function size categories (such as payment and settlement, information management, liquidity, financing, etc.) and main user groups (such as small and micro enterprises, manufacturing enterprises, group users, etc.) and other product feature information.

[0067] The relationship strength factor between every two financial products in the financial product graph can be calculated in the following way: The financial products opened by user Ci are expressed as a vector CiS = (Si1, Si2,..., Sij,...) in the order of opening time, where {Si1, Si2,..., Sij,...} is a subset of {Sj}. Obtain the vectors of all users opening financial products, and calculate the following probabilities for any two financial products Sm and Sn (m < n, that is, Sm is opened before Sn) one by one: The probability Psmn that Sm and Sn appear in the same vector, which has nothing to do with the opening order of m and n. Theoretically, when the financial product Sm is a prerequisite for the financial product Sn, Ps(n|m) should be equal to 1; when the functions of Sm and Sn are mutually exclusive, Ps(mn), Ps(m|n), and Ps(m|n) are all 0; there is no inevitable mathematical relationship between Ps(m|n) and Ps(n|m). The undirected relationship graph uses the financial products {Sj} as the node set, and Psmn as the relationship strength factor between Sm and Sn. This graph reflects the degree of closeness between products through the probability that several financial products are opened by users at the same time, and the relationship has strength but no direction.

[0068] The object of machine learning is the mapping relationship between the requirements in the user information module (generated by user characteristics CF) and the service combination for starting a business, using the undirected relationship graph in the financial product network module. The construction of the training sample set is based on the {CSbn} set defined above, and the node attributes, connection strength, and direction of the {Sj} nodes involved in this set on the unordered financial product graph. The user to be recommended is used as the sample set to be predicted, and the deep learning result {CSf0} is obtained, where CiSf0 = (CiS1, CiS2,..., CiSj), that is, a total of j products are recommended for user Ci.

[0069] In a preferred embodiment, as Figure 5 shown, the method further includes the step S030 of pre-forming the directed relationship graph:

[0070] S031: Form a financial product graph according to all financial product information;

[0071] S032: Determine the directed strength factor between every two financial products in the financial product graph according to the subscription information of users in historical data;

[0072] S033: Use the directed strength factor as the edge of the corresponding two financial product nodes to obtain a directed relationship graph.

[0073] Specifically, the relationship strength factor between every two financial products in the financial product graph can be calculated as follows: The financial products started by user Ci are expressed as a vector CiS = (Si1, Si2,..., Sij,...) in the order of starting time, where {Si1, Si2,..., Sij,...} is a subset of {Sj}. Obtain the vectors of all users starting financial products, and calculate the following probabilities for any two financial products Sm and Sn one by one (m < n, that is, Sm is started before Sn). When Sm has been started, the probability of starting Sn later, Ps(m|n), that is, the probability of Sn appearing after Sm in the CiS vector, where the prior probability Ps(m|n) is calculated using the likelihood function method. Theoretically, when the financial product Sm is a prerequisite for the financial product Sn, Ps(n|m) should be equal to 1; when the functions of the Sm and Sn products are mutually exclusive, Ps(mn), Ps(m|n), and Ps(m|n) are all 0; there is no necessary mathematical relationship between Ps(m|n) and Ps(n|m). The directed relationship graph takes the financial products {Sj} as the node set, and Ps(m|n) as the relationship strength from node Sm to node Sn. This graph reflects the probability that the users of the current financial product node will have the adjacent node financial product, and the relationship has direction and strength.

[0074] In the actual application process, due to the unique product information barriers of corporate financial products, it is impossible to ensure that all customers in the training sample set have opened all the required financial products. Therefore, this paper believes that the correspondence between customers and financial products included in {CSbn} is incomplete. Then, {CSf} learned based on {CSbn} is also incomplete. Based on the ordered financial product map in the previous module, this incompleteness can be corrected.

[0075] For the deep learning prediction result CiSf = (CiS1, CiS2,..., CiSj) of customer Ci, take the complete set of services Sf0 = {S1, S2,..., Sj} involved in the service, and obtain the expanded all services Sfex0 by taking the adjacent nodes and connection strengths pointed to by each node Sj on the ordered financial product map. Depending on the number of newly added nodes, limit the number of nodes expanded outward by each node or the lower limit of the connection strength to reduce its quantity to a level suitable for business practice, and obtain the expanded product portfolio Sfex = (Sex1, Sex2,..., Sexk) based on the map. Use it as a supplement to Sf0 to obtain the final recommendation result CiSf = (CiS1, CiS2,..., CiSj, CiSex1, CiSex2,..., CiSexk) for customer Ci.

[0076] In a preferred embodiment, as Figure 6 shown, the specific process of forming the financial product map by S031 according to all financial product information includes:

[0077] S0311: Determine the product feature information of each financial product according to all financial product information.

[0078] S0312: Establish the financial product map with each financial product as a node and the corresponding product feature information as the node attribute.

[0079] It can be understood that when expressing the situation of users opening financial products in the form of a graph, each financial product can be used as a node of the graph, the product feature information corresponding to each financial product can be used as the attribute of the node, the relationship between every two financial products can be used as an edge, and the intensity factor can be used as the specific value of the edge to form a financial product graph of all financial products. For example, in a specific example, a full set of corporate financial products S = {Sj} is defined. Here, the granularity of a financial product can be defined by the recommendation system according to the actual situation, and each Sj is a node of the graph. For a certain financial product Sj, the feature vector SjF = (Sjf1, Sjf2,...) of the financial product in the specified dimension is described according to historical data, including the occurrence frequency, total transaction volume (only service items involving amounts, otherwise recorded as null), core function size categories (such as payment and settlement, information management, liquidity, financing, etc.), and main user groups (such as small and micro enterprises, manufacturing enterprises, group users, etc.) and other product feature information.

[0080] It should be noted that the present invention supports user cold start and product cold start. Among them, user cold start refers to a new user, and there is no user behavior information in the user information module. The user information module retrieves other information except behavior, that is, basic information, industry and macro information, and other processing is consistent with the above method. In this case, the recommendation result will rely more on some basic information (such as scale) and industry information of the user, and the recommendation result can be updated regularly to incorporate the subsequent business behavior data of the user into the user information module as soon as possible. Product cold start refers to a new product without user opening or usage records. In the product features, the transaction volume is discarded, the size category information is taken, and the relationship with the previous and subsequent products, that is, the intensity, is supplemented by business rules. Other processing is consistent with the above method. In this case, the recommendation result will rely more on whether the delineation of the product size category is reasonable and whether the preset of the relationship intensity with possible subsequent products is reasonable. The recommendation result can be updated regularly to incorporate the product business volume and user opening data into the financial product network module as soon as possible.

[0081] Based on the same principle, this embodiment also discloses a financial product push device for non - typical information overload. As Figure 7 shown, in this embodiment, the device includes a feature extraction module 11, a product prediction module 12, and a supplementary prediction module 13.

[0082] Among them, the feature extraction module 11 is used to extract features from user information to obtain user demand features.

[0083] The product prediction module 12 is used to predict the push financial products by using a deep learning model formed based on machine learning for the user demand features.

[0084] The supplementary prediction module 13 is used to perform an expansion prediction on the pushed financial products according to the preset directed relationship graph of all financial products to obtain expanded financial products, and obtain an expanded product portfolio based on the pushed financial products and the expanded financial products to be pushed to users.

[0085] The financial product pushing method with atypical information overload of the present invention extracts the user information features to obtain user demand features, predicts the pushed financial products through a deep learning model formed based on machine learning, performs an expansion prediction on the pushed financial products according to the preset directed relationship graph of all financial products to obtain expanded financial products, and obtains an expanded product portfolio based on the pushed financial products and the expanded financial products to be pushed to users. Thus, the present invention predicts the user demand features extracted from the user information features through a preset deep learning model formed based on machine learning to obtain the pushed financial products that can be recommended to users, and then performs an expansion prediction on the pushed financial products through the preset directed relationship graph of financial products to obtain expanded financial products to form an expanded product portfolio to be pushed to users, thereby completing the matching mapping between financial products and users. Therefore, the present invention proposes a financial product pushing method with a small number of products, no information overload, and weak product homogeneity, improving the accuracy of financial product pushing.

[0086] In a preferred embodiment, the feature extraction module 11 is specifically configured to extract features from the basic information and behavior information in the user information to obtain user demand features.

[0087] Specifically, the user information can be analyzed to obtain the basic information and behavior information in the user information, and then the basic information and behavior information are subjected to feature extraction and converted into feature vectors to obtain user demand features, so that the user demand features can be directly input into the deep learning model to predict the pushed financial products. Among them, the basic information in the user information may include at least one of information such as the registration location, the industry to which it belongs, the nature of the enterprise, the registered capital, and the registration time. The behavior information in the user information may include transactions and economic relationships established by transactions, and the transaction information may include at least one of information such as the total settlement amount, the average daily balance per year, various trading volumes, and various credit limits.

[0088] In a preferred embodiment, the feature extraction module 11 is further configured to obtain historical user demand features according to the basic information and behavior information of the user information in the historical data; train a pre-constructed machine learning model according to the product subscription information, industry and macro information of the user in the historical data and the historical user demand features to obtain a deep learning model.

[0089] Specifically, in this preferred embodiment, to improve the prediction accuracy, the impact of multi-dimensional factors on financial product recommendations is considered. The historical user demand characteristics of feature vectors are obtained by extracting features from the basic information and behavioral information of user information in historical data. Further considering industry and macro information, a training sample set is formed based on the historical user demand characteristics and industry and macro information to train a machine learning model based on the principle of machine learning to obtain a deep learning model.

[0090] For example, in a specific example, a full set of users C = {Ci} is defined. For each user Ci, a user-based feature vector (historical user demand characteristics) CiF = (Cif1, Cif2,...) is constructed. The set of all user characteristics is CF = {CiF}. The full set of users includes three categories: user basic information, user behavioral information, industry, and macroeconomic information. Among them, the basic information in user information may include at least one of the information such as the registration location, the industry to which it belongs, the nature of the enterprise, the registered capital, and the registration time. The behavioral information in user information is centered around the information recorded and filed by corporate users in the industrial and commercial bureau, and may include transactions and economic relationships established by transactions. The transaction information may include at least one of the information such as the total settlement amount, the average daily balance for the year, various transaction volumes, and various credit limits.

[0091] For industry and macro information, a full set of industries I = {Iu} is defined. The characteristics of a certain industry Iu are represented by a vector IuF = (Iuf1, Iuf2,...). The characteristics include information such as industry output value, production capacity, and common upstream and downstream. According to the industry to which Ci belongs in the basic information, the corresponding Iu is associated, and its IuF is taken as part of Ci. The macro data may include at least one of the information such as the money supply of each caliber (M0 / M1 / M2), various monetary policy tools (SLO / SLF / MLF / TMLF / PSL / CRA / TLF) quantification and long-term and short-term deposit and loan interest rates, various liquidity regulatory indicators (LCR / NSFR), GDP at the national and provincial and municipal levels to which it belongs, the output value and production capacity distribution of the primary, secondary, and tertiary industries, the population number, the number of registered enterprises, and the change trends of the above-mentioned various values.

[0092] Finally, according to the full set of users C = {Ci} and the user characteristics CiF = (Cif1, Cif2,...), the set of users (regarded as having relatively sufficient service opening, and n can be customized according to the actual situation) with the number of financial products / services (hereinafter referred to as "services") opened or used greater than or equal to n (n is an integer greater than 1) is taken as the basic sample set {CSbn} for establishing user information, and classified according to the service opening situation, and user needs are aggregated and extracted by means of the characteristics CiF of each type of user.

[0093] In a preferred embodiment, the feature extraction module 11 is specifically configured to obtain an undirected relationship graph of financial products according to the product subscription information of users in the historical data; and train a pre-constructed machine learning model according to the undirected relationship graph, the industry and macro information, and the historical user demand characteristics to obtain a deep learning model.

[0094] Specifically, in this preferred embodiment, the product subscription information of users in the historical data is further analyzed to obtain an undirected relationship graph of financial products. The deep learning model obtained by training the machine learning model according to the undirected relationship graph of financial products, the historical user demand characteristics obtained by feature extraction of user information, and the industry and macro information can be used to predict financial products with a high degree of matching with users.

[0095] In a preferred embodiment, the feature extraction module 11 is specifically configured to form a financial product graph according to all financial product information. Determine the relationship strength factor between every two financial products in the financial product graph according to the subscription information of users in the historical data. Use the relationship strength factor as the edge of the corresponding two financial product nodes to obtain an undirected relationship graph.

[0096] Specifically, an undirected relationship graph of all financial products can be obtained according to the relevant information of users' subscribed financial products in the historical data. For example, in a specific example, the situation of users opening financial products is expressed in the form of a graph. Define the full set of corporate financial products S = {Sj}, where the granularity of a financial product can be defined by the recommendation system according to the actual situation, and each Sj is a node of the graph. For a certain financial product Sj, describe the feature vector SjF=(Sjf1,Sjf2,...) of the financial product in the specified dimension according to the historical data, including the occurrence frequency, total transaction volume (only service items involving amounts, otherwise recorded as null), core function size category (such as payment and settlement, information management, liquidity, financing, etc.) and main user groups (such as small and micro enterprises, manufacturing enterprises, group users, etc.) and other product feature information.

[0097] The relationship strength factor between every two financial products in the financial product map can be calculated as follows: The financial products opened by user Ci are expressed as a vector CiS = (Si1, Si2,..., Sij,...) in the order of opening time, where {Si1, Si2,..., Sij,...} is a subset of {Sj}. Obtain the vectors of financial products opened by all users, and calculate the following probabilities for any two financial products Sm and Sn (m < n, that is, Sm is opened before Sn) one by one: The probability Psmn that Sm and Sn appear in the same vector, which has nothing to do with the opening order of m and n. Theoretically, when financial product Sm is a prerequisite for financial product Sn, Ps(n|m) should be equal to 1; when the functions of Sm and Sn are mutually exclusive, Ps(mn), Ps(m|n), and Ps(n|m) are all 0; there is no necessary mathematical relationship between Ps(m|n) and Ps(n|m). The undirected relationship map uses the financial products {Sj} as the node set, and Psmn as the relationship strength factor between Sm and Sn. This map reflects the degree of closeness between products through the probability that several financial products are opened by users at the same time. The relationship has strength but no direction.

[0098] The object of machine learning is the mapping relationship between the requirements in the user information module (generated by user characteristics CF) and the combination of opened services, using the undirected relationship map in the financial product network module. The construction of the training sample set is based on the {CSbn} set defined above, and the node attributes, connection strength, and direction of the {Sj} nodes involved in this set on the unordered financial product map. The user to be recommended is used as the sample set to be predicted, and the deep learning result {CSf0} is obtained, where CiSf0 = (CiS1, CiS2,..., CiSj), that is, j products are recommended for user Ci.

[0099] In a preferred embodiment, the feature extraction module 11 is further configured to form a financial product map according to all financial product information; determine the directed strength factor between every two financial products in the financial product map according to the subscription information of users in historical data; and obtain a directed relationship map by using the directed strength factor as the edge of the corresponding two financial product nodes.

[0100] Specifically, the relationship strength factor between every two financial products in the financial product map can be calculated as follows: The financial products opened by user Ci are expressed as a vector CiS = (Si1, Si2,..., Sij,...) in the order of opening time, where {Si1, Si2,..., Sij,...} is a subset of {Sj}. Obtain the vectors of financial products opened by all users, and calculate the following probabilities for any two financial products Sm and Sn one by one (m < n, that is, Sm is opened before Sn). When Sm has been opened, the probability of opening Sn later, Ps(m|n), that is, the probability of Sn appearing after Sm in the CiS vector, where the prior probability Ps(m|n) is calculated using the likelihood function method. Theoretically, when the financial product Sm is a necessary prerequisite for the financial product Sn, Ps(n|m) should be equal to 1; when the functions of the financial products Sm and Sn are mutually exclusive, Ps(mn), Ps(m|n), and Ps(m|n) are all 0; there is no necessary mathematical relationship between Ps(m|n) and Ps(n|m). The directed relationship map takes the financial products {Sj} as the node set and Ps(m|n) as the relationship strength from the node Sm to the node Sn. This map reflects the probability that the users of the current financial product node will have the financial products of the adjacent nodes, and the relationship has a direction and strength.

[0101] In the actual application process, due to the unique product information barriers of corporate financial products, it is impossible to ensure that all customers in the training sample set have opened all the required financial products. Therefore, this paper believes that the correspondence between the customers included in {CSbn} and the financial products is incomplete. Then, {CSf} obtained by learning based on {CSbn} is also incomplete. Based on the ordered financial product map in the previous module, this kind of incompleteness can be corrected.

[0102] For the deep learning prediction result CiSf = (CiS1, CiS2,..., CiSj) of customer Ci, take the complete set Sf0 of services involved = {S1, S2,..., Sj}, and obtain the adjacent nodes and connection strengths pointed to by each node Sj on the ordered financial product map to obtain the expanded complete set of services Sfex0. Then, depending on the number of newly added nodes, reduce its quantity to a level suitable for business practice by limiting the number of nodes expanded outward by each node or the lower limit of the connection strength, and obtain the expanded product portfolio Sfex = (Sex1, Sex2,..., Sexk) based on the map. Use it as a supplement to Sf0 to obtain the final recommendation result CiSf = (CiS1, CiS2,..., CiSj, CiSex1, CiSex2,..., CiSexk) for customer Ci.

[0103] In a preferred embodiment, the feature extraction module 11 is specifically configured to determine the product feature information of each financial product according to all financial product information. A financial product graph is established with each financial product as a node and the corresponding product feature information as the node attribute.

[0104] It can be understood that expressing the situation of users opening financial products in the form of a graph, each financial product can be used as a node of the graph, the product feature information corresponding to each financial product can be used as the attribute of the node, the relationship between every two financial products can be used as an edge, and the intensity factor can be used as the specific value of the edge to form a financial product graph of all financial products. For example, in a specific example, a full set of corporate financial products S = {Sj} is defined, where the granularity of a financial product can be defined by the recommendation system according to the actual situation, and each Sj is a node of the graph. For a certain financial product Sj, the feature vector SjF = (Sjf1, Sjf2,...) of the financial product in the specified dimension is described according to historical data, including occurrence frequency, total transaction volume (only for service items involving amounts, otherwise recorded as null), core function size categories (such as payment and settlement, information management, liquidity, financing, etc.) and main user groups (such as small and micro enterprises, manufacturing enterprises, group users, etc.) and other product feature information.

[0105] It should be noted that the present invention supports user cold start and product cold start. Among them, user cold start refers to new users, and there is no user behavior information in the user information module. The user information module obtains other information except behavior, that is, basic information, industry and macro information, and other processing is consistent with the above method. In this case, the recommendation result will depend more on some basic information (such as scale) and industry information of the user, and the recommendation result can be updated regularly to include the subsequent business behavior data of the user into the user information module as soon as possible. Product cold start refers to new products without user opening or usage records. In the product features, the transaction volume is discarded, the size category information is taken, and the relationship with the previous and subsequent products, that is, the intensity, is supplemented by business rules, and other processing is consistent with the above method. In this case, the recommendation result will depend more on whether the delineation of the product size category is reasonable and whether the preset of the relationship intensity with possible subsequent products is reasonable. The recommendation result can be updated regularly to include the product business volume and user opening data into the financial product network module as soon as possible.

[0106] Since the principle of this device for solving problems is similar to the above method, the implementation of this device can refer to the implementation of the method and will not be elaborated here.

[0107] The systems, devices, modules or units illustrated in the above embodiments can be specifically implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer device. Specifically, the computer device can be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or any combination of these devices.

[0108] In a typical example, the computer device specifically includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements the method executed by the user terminal as described above, or when the processor executes the program, it implements the method executed by the server as described above.

[0109] The following refers to Figure 8 , which shows a schematic structural diagram of a computer device 600 suitable for implementing the embodiments of the present application.

[0110] As Figure 8 shown, the computer device 600 includes a central processing unit (CPU) 601, which can perform various appropriate tasks and processes according to the program stored in the read-only memory (ROM) 602 or the program loaded from the storage section 608 into the random access memory (RAM) 603. In the RAM 603, various programs and data required for the operation of the system 600 are also stored. The CPU 601, ROM 602, and RAM 603 are connected to each other via a bus 604. The input / output (I / O) interface 605 is also connected to the bus 604.

[0111] The following components are connected to the I / O interface 605: an input section 606 including a keyboard, a mouse, etc.; an output section 607 including, for example, a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 608 including a hard disk, etc.; and a communication section 609 including a network interface card such as a LAN card, a modem, etc. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to the I / O interface 605 as required. A removable medium 611, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 610 as required, so that the computer program read from it can be installed into the storage section 608 as required.

[0112] In particular, according to an embodiment of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, an embodiment of the present invention includes a computer program product that includes a computer program tangibly embodied on a machine-readable medium, the computer program including program code for performing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 609, and / or installed from the removable medium 611.

[0113] Computer-readable media includes both permanent and non-permanent, removable and non-removable media implemented by any method or technology for storage of information such as computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read only memory (ROM), electrically erasable programmable read only memory (EEPROM), flash memory or other memory technologies, compact disc read only memory (CD-ROM), digital versatile discs (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transitory medium that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media such as modulated data signals and carrier waves.

[0114] For convenience of description, the above-described apparatus is described by dividing it into various units according to functions. Of course, when implementing the present application, the functions of the various units can be implemented in one or more pieces of software and / or hardware.

[0115] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the present invention. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and combinations of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded processor or other programmable data processing device to produce a machine such that the instructions executed by the processor of the computer or other programmable data processing device produce means for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0116] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to operate in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means that implement the functions specified in one or more of the blocks and / or processes. Figure 1 one or more of the processes and / or blocks Figure 1 specified in one or more of the blocks and / or processes.

[0117] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, so that the instructions executed on the computer or other programmable apparatus provide steps for implementing the functions specified in one or more of the blocks and / or processes. Figure 1 one or more of the processes and / or blocks Figure 1 specified in one or more of the blocks and / or processes.

[0118] It should also be noted that the term "comprising", "including" or any other variation thereof is intended to cover a non-exclusive inclusion, such that a process, method, article or apparatus comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or apparatus. Without further limitation, an element defined by the phrase "comprising a..." does not exclude the presence of additional identical elements in the process, method, article or apparatus comprising the element.

[0119] Those skilled in the art will appreciate that the embodiments of the present application may be provided as a method, system or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Moreover, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0120] The present application may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types. The present application may also be practiced in a distributed computing environment where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in both local and remote computer storage media including storage devices.

[0121] Each embodiment in this specification is described in a progressive manner. For the same or similar parts among the embodiments, reference can be made to each other. Each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple. For the relevant parts, reference can be made to the corresponding description in the method embodiment.

[0122] The above description is only for the embodiments of the present application and is not intended to limit the present application. For those skilled in the art, various modifications and changes can be made to the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included within the scope of the claims of the present application.

Claims

1. A method for pushing financial products with atypical information overload, characterized in that, Including: Performing feature extraction on user information to obtain user demand features; Predicting the pushed financial products by using a deep learning model formed based on machine learning for the user demand features; Performing extended prediction on the pushed financial products according to a directed relationship graph of all preset financial products to obtain extended financial products, and obtaining an extended product portfolio based on the pushed financial products and the extended financial products to push to users; The steps of pre-forming the directed relationship graph include: Forming a financial product graph according to all financial product information; Determining a directed intensity factor between every two financial products in the financial product graph according to the subscription information of users in historical data; Taking the directed intensity factor as the edge of the corresponding two financial product nodes to obtain a directed relationship graph; The determining the directed intensity factor between every two financial products in the financial product graph according to the subscription information of users in historical data includes: The financial products opened by user Ci are expressed as a vector CiS=(Si1, Si2,..., Sij,...) in the order of opening time. Among them, the full set of users is defined as C={Ci}, the full set of corporate financial products is S={Sj}, and {Si1, Si2,..., Sij,...} is a subset of {Sj}; obtaining vectors of financial products opened by all users, and calculating the probability between any two financial products Sm and Sn one by one; where m < n, that is, Sm is opened before Sn; When Sm has been opened, the probability of opening Sn later is Ps(m|n), that is, the probability of Sn appearing after Sm in the CiS vector; where the prior probability Ps(m|n) is calculated by using the likelihood function method; The directed relationship graph takes the financial product set {Sj} as the node set, and takes Ps(m|n) as the relationship intensity from node Sm to node Sn. This graph reflects the probability that the users of the current financial product node have adjacent node financial products, where the relationship has direction and intensity.

2. The method for pushing financial products with atypical information overload according to claim 1, characterized in that, The performing feature extraction on user information to obtain user demand features specifically includes: Performing feature extraction on the basic information and behavior information in user information to obtain user demand features.

3. The method for pushing financial products with atypical information overload according to claim 1, characterized in that, Further including the steps of pre-obtaining the deep learning model based on the principle of machine learning: Obtaining historical user demand features according to the basic information and behavior information of users in historical data; Training a pre-constructed machine learning model according to the product subscription information, industry and macro information of users in historical data and the historical user demand features to obtain a deep learning model.

4. The method for pushing financial products with atypical information overload according to claim 3, characterized in that, The training a pre-constructed machine learning model according to the product subscription information, industry and macro information of users in historical data and the historical user demand features to obtain a deep learning model specifically includes: Obtaining an undirected relationship graph of financial products according to the product subscription information of users in the historical data; Training a pre-constructed machine learning model according to the undirected relationship graph, the industry and macro information, and the historical user demand features to obtain a deep learning model.

5. The method for pushing financial products with atypical information overload according to claim 4, characterized in that, The obtaining an undirected relationship graph of financial products according to the product subscription information of users in the historical data specifically includes: Form a financial product map based on all financial product information; Determine the relationship strength factor between every two financial products in the financial product map according to the subscription information of users in historical data; Use the relationship strength factor as the edge of the corresponding two financial product nodes to obtain an undirected relationship map.

6. The method for pushing financial products with atypical information overload according to claim 5, characterized in that, The forming of the financial product map according to all financial product information specifically includes: Determine the product feature information of each financial product according to all financial product information; Establish the financial product map with each financial product as a node and the corresponding product feature information as the node attribute.

7. A device for pushing financial products with atypical information overload, characterized in that, Include: A feature extraction module for extracting features from user information to obtain user demand features; A product prediction module for predicting the user demand features through a deep learning model formed based on machine learning to obtain push financial products; A supplementary prediction module for performing an extended prediction on the push financial products according to the directed relationship map of all preset financial products to obtain extended financial products, and obtaining an extended product portfolio based on the push financial products and the extended financial products to be pushed to users; The feature extraction module is further configured to: form a financial product map according to all financial product information; determine the directed strength factor between every two financial products in the financial product map according to the subscription information of users in historical data; use the directed strength factor as the edge of the corresponding two financial product nodes to obtain a directed relationship map; Among them, the directed strength factor between every two financial products is calculated in the following way: Express the financial products opened by user Ci in chronological order of opening as a vector CiS=(Si1, Si2,..., Sij,...), where the full set of users is defined as C={Ci}, the full set of corporate financial products is S={Sj}, and {Si1, Si2,..., Sij,...} is a subset of {Sj}; Obtain the vectors of financial products opened by all users, and calculate the probability between any two financial products Sm and Sn one by one; where, m < n, that is, Sm is opened before Sn; when Sm has been opened, the probability of opening Sn later is Ps(m|n), that is, the probability of Sn appearing after Sm in the CiS vector; among them, the prior probability Ps(m|n) is calculated using the likelihood function method; The directed relationship map uses the financial product set {Sj} as the node set, and Ps(m|n) as the relationship strength from node Sm to node Sn. This map reflects the probability that the users of the current financial product node will have the adjacent node financial product, where the relationship has direction and strength.

8. A computer device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method described in any one of claims 1-6.

9. A computer-readable medium, having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method described in any one of claims 1-6.

10. A computer program product, characterized in that, The computer program product includes a computer program, and when the computer program is executed by the processor, it implements the method described in any one of claims 1-6.

Images