Automotive finance product recommendation method and system based on artificial intelligence

Abstract

The artificial intelligence-based auto finance product recommendation method and system provided in this application relate to the field of artificial intelligence technology. In this application, first, a cluster of candidate auto finance products and a cluster of candidate financial product recommendation methods are determined; second, each candidate auto finance product in the candidate auto finance product cluster and each candidate financial product recommendation method in the candidate financial product recommendation method cluster are arbitrarily grouped to form multiple candidate product method groups; then, the product recommendation success rate corresponding to each candidate product method group is analyzed; finally, the target product method group is screened based on the product recommendation success rate, and the candidate auto finance products and candidate financial product recommendation methods in the target product method group are respectively used as the target auto finance products and target financial product recommendation methods. Based on the above content, the problem of relatively low reliability of auto finance product recommendations in the existing technology can be improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and more specifically, to an artificial intelligence-based automobile financial product recommendation method and system. Background Art

[0002] Auto finance products refer to automotive-related financial products such as auto loans, auto financing leases, auto insurance products, and auto value-preserving repurchase services. Different users generally choose different financial products based on their specific needs. Therefore, corresponding financial products are generally provided to users based on their actual needs. However, the inventors have discovered that in the prior art, financial product recommendations are generally made based on user preferences. However, due to a lack of relevant knowledge or experience, users struggle to select suitable financial products. Consequently, the reliability of recommendations based on user preferences is relatively low. Summary of the Invention

[0003] In view of this, the purpose of this application is to provide an artificial intelligence-based automobile financial product recommendation method and system to improve the problem of relatively low reliability of automobile financial product recommendations in the prior art.

[0004] To achieve the above objectives, this application adopts the following technical solutions:

[0005] An artificial intelligence-based auto finance product recommendation method, comprising:

[0006] Determining a candidate auto financial product cluster and a candidate financial product recommendation method cluster, wherein the candidate auto financial product cluster includes a plurality of candidate auto financial products, and the candidate financial product recommendation method cluster includes a plurality of candidate financial product recommendation methods;

[0007] performing arbitrary grouping processing on each candidate auto financial product in the candidate auto financial product cluster and each candidate financial product recommendation method in the candidate financial product recommendation method cluster to form a corresponding plurality of candidate product method groups, wherein each candidate product method group includes a candidate auto financial product and a candidate financial product recommendation method;

[0008] Loading user description data corresponding to each candidate product method group and the user to whom the auto finance product is recommended, analyzing the product recommendation success rate corresponding to each candidate product method group using a target product recommendation network, wherein the target product recommendation network is a pre-trained neural network, and the product recommendation success rate reflects the likelihood that the auto finance product recommendation user will accept the candidate auto finance product after the candidate auto finance product is recommended according to the corresponding candidate financial product recommendation method;

[0009] Based on the product recommendation success rate corresponding to each of the candidate product method groups, a target product method group is screened out from the multiple candidate product method groups, and the candidate auto finance products and candidate financial product recommendation methods in the target product method group are respectively used as the target auto finance products and target financial product recommendation methods corresponding to the auto finance product recommendation user, wherein the target auto finance product is recommended to the auto finance product recommendation user according to the target financial product recommendation method.

[0010] In a preferred embodiment of the present application, in the above-mentioned artificial intelligence-based auto finance product recommendation method, the step of loading the user description data corresponding to each candidate product mode group and the auto finance product recommendation user, and analyzing the product recommendation success rate corresponding to each candidate product mode group through the target product recommendation network, includes:

[0011] Loading user description data corresponding to each candidate product mode group and the user to whom the auto finance product is recommended, respectively, so that the target product recommendation network obtains user description data corresponding to each candidate product mode group and the user to whom the auto finance product is recommended, wherein the user description data includes at least user identity information and financial attribute information of the user to whom the auto finance product is recommended;

[0012] Using a first feature mining unit in the target product recommendation network, feature mining is performed on user description data corresponding to the user for whom the auto finance product is recommended, to obtain user semantic features corresponding to the user for whom the auto finance product is recommended;

[0013] For each candidate product mode group, using the second feature mining unit in the target product recommendation network, feature mining is performed on the financial product description data of the candidate auto financial products in the candidate product mode group to obtain the financial product semantic features corresponding to the candidate product mode group; and using the third feature mining unit in the target product recommendation network, feature mining is performed on the recommendation mode description data of the candidate financial product recommendation modes in the candidate product mode group to obtain the recommendation mode semantic features corresponding to the candidate product mode group;

[0014] For each of the candidate product mode groups, using a feature fusion unit in the target product recommendation network, feature fusion is performed on the financial product semantic features corresponding to the candidate product mode group, the recommendation mode semantic features corresponding to the candidate product mode group, and the user semantic features to obtain the financial product recommendation semantic features corresponding to the candidate product mode group;

[0015] For each of the candidate product mode groups, the analysis output unit in the target product recommendation network is used to process the financial product recommendation semantic features corresponding to the candidate product mode group and output the product recommendation success rate corresponding to the candidate product mode group.

[0016] In a preferred embodiment of the present application, in the above-mentioned artificial intelligence-based automobile financial product recommendation method, the step of, for each candidate product mode group, utilizing a feature fusion unit in the target product recommendation network to perform feature fusion on the financial product semantic features corresponding to the candidate product mode group, the recommendation mode semantic features corresponding to the candidate product mode group, and the user semantic features to obtain the financial product recommendation semantic features corresponding to the candidate product mode group includes:

[0017] Using a first feature fusion branch included in a feature fusion unit in the target product recommendation network, a first fusion is performed on the semantic features of financial products corresponding to the candidate product mode group and the semantic features of recommendation modes corresponding to the candidate product mode group to form a first fused semantic feature;

[0018] Using a second feature fusion branch included in a feature fusion unit in the target product recommendation network, performing a second fusion on the first fused semantic feature and the user semantic feature to form a second fused semantic feature;

[0019] Based at least on the second fused semantic feature, a financial product recommendation semantic feature corresponding to the candidate product mode group is determined.

[0020] In a preferred embodiment of the present application, in the above-mentioned artificial intelligence-based automobile financial product recommendation method, the step of utilizing the first feature fusion branch included in the feature fusion unit in the target product recommendation network to perform a first fusion on the financial product semantic features corresponding to the candidate product mode group and the recommendation mode semantic features corresponding to the candidate product mode group to form a first fused semantic feature includes:

[0021] Utilizing a first feature fusion branch including three feature mapping matrices included in a feature fusion unit in the target product recommendation network, the financial product semantic features corresponding to the candidate product mode group and the recommendation mode semantic features corresponding to the candidate product mode group are mapped respectively to form a first financial product mapping feature and a second financial product mapping feature corresponding to the financial product semantic features, and to form a recommendation mode mapping feature corresponding to the recommendation mode semantic features;

[0022] determining a feature correlation parameter between the recommendation method mapping feature and the first financial product mapping feature, and, based on the feature correlation parameter, updating the second financial product mapping feature to form an updated second financial product mapping feature;

[0023] A first fusion semantic feature is determined based on the updated second financial product mapping feature.

[0024] In a preferred embodiment of the present application, in the above-mentioned artificial intelligence-based auto finance product recommendation method, the step of utilizing the second feature fusion branch included in the feature fusion unit in the target product recommendation network to perform a second fusion on the first fused semantic feature and the user semantic feature to form a second fused semantic feature includes:

[0025] performing feature splitting processing on the user semantic feature to form at least a first local user semantic feature and a second local user semantic feature, wherein the first local user semantic feature is used to reflect the semantic information of the user identity information, and the second local user semantic feature is used to reflect the semantic information of the financial attribute information;

[0026] Using a first feature mapping matrix set included in a second feature fusion branch included in a feature fusion unit in the target product recommendation network, the first local user semantic feature and the second local user semantic feature are respectively mapped to form a first local user mapping feature corresponding to the first local user semantic feature, and to form a second local user mapping feature and a third local user mapping feature corresponding to the second local user semantic feature, wherein the first feature mapping matrix set includes three feature mapping matrices;

[0027] determining a feature correlation parameter between the first local user mapping feature and the second local user mapping feature, and, based on the feature correlation parameter, updating the third local user mapping feature to form an updated third local user mapping feature, and determining an associated user semantic feature based on the updated third local user mapping feature;

[0028] The first fused semantic feature and the associated user semantic feature are fused at multiple levels using the fusion network layer included in the second feature fusion branch to form a second fused semantic feature.

[0029] In a preferred embodiment of the present application, in the above-mentioned method for recommending automobile financial products based on artificial intelligence, the second feature fusion branch includes a plurality of cascaded fusion network layers, and the step of using the fusion network layers included in the second feature fusion branch to perform multi-level fusion of the first fused semantic feature and the associated user semantic feature to form a second fused semantic feature includes:

[0030] Using a first fusion network layer among the multiple fusion network layers, and based on the first fusion semantic feature, performing correlation fusion on the associated user semantic feature to form a correlation fusion semantic feature corresponding to the first fusion network layer;

[0031] For each fusion network layer other than the first fusion network layer in the multiple fusion network layers, using the current fusion network layer, respectively, performing pooling processing on the first fusion semantic feature and the correlation fusion semantic feature corresponding to the previous fusion network layer to form corresponding first pooling features and second pooling features, and, based on the first pooling feature, performing correlation fusion on the second pooling feature to form a correlation fusion semantic feature corresponding to the current fusion network layer, wherein the pooling processing method performed in each two adjacent fusion network layers is different;

[0032] The multiple correlation fusion semantic features corresponding to the multiple fusion network layers are superimposed or averaged to form corresponding second fusion semantic features.

[0033] In a preferred embodiment of the present application, in the above-mentioned method for recommending automobile financial products based on artificial intelligence, the step of utilizing the first fusion network layer among the multiple fusion network layers to perform correlation fusion on the associated user semantic features based on the first fusion semantic features to form the correlation fusion semantic features corresponding to the first fusion network layer includes:

[0034] Using a second feature mapping matrix set included in a first fusion network layer among the multiple fusion network layers, mapping the first fusion semantic feature and the associated user semantic feature respectively to form a first fusion mapping feature corresponding to the first fusion semantic feature, and forming a second fusion mapping feature and a third fusion mapping feature corresponding to the associated user semantic feature, wherein the second feature mapping matrix set includes three feature mapping matrices;

[0035] Determine the feature correlation parameter between the first fusion mapping feature and the second fusion mapping feature, and based on the feature correlation parameter, update the third fusion mapping feature to form an updated third fusion mapping feature, and based on the updated third fusion mapping feature, determine the correlation fusion semantic feature corresponding to the first fusion network layer.

[0036] In a preferred embodiment of the present application, in the above-mentioned method for recommending automobile financial products based on artificial intelligence, the method further comprises the step of training a target product recommendation network, which comprises:

[0037] Using the first feature mining unit in the candidate product recommendation network, feature mining is performed on the sample user description data to obtain the corresponding sample user semantic features;

[0038] Using the second feature mining unit in the candidate product recommendation network, perform feature mining on the sample financial product description data to obtain corresponding semantic features of the sample financial products;

[0039] Using the third feature mining unit in the candidate product recommendation network, feature mining is performed on the sample recommendation method description data to obtain corresponding sample recommendation method semantic features;

[0040] Using a feature fusion unit in the candidate product recommendation network, the semantic features of the sample financial products, the semantic features of the sample recommendation methods, and the semantic features of the sample users are fused to obtain corresponding semantic features for sample financial product recommendations;

[0041] Utilizing the analysis output unit in the candidate product recommendation network, processing the semantic features of the sample financial product recommendation and outputting the corresponding sample product recommendation success rate;

[0042] Based on the error between the sample product recommendation success rate and the product recommendation success rate label, the network parameters of the candidate product recommendation network are updated to form the target product recommendation network, wherein the product recommendation success rate label refers to the actual product recommendation success rate of recommending the auto financial product corresponding to the sample financial product description data to the user corresponding to the sample user description data according to the financial product recommendation method corresponding to the sample recommendation method description data.

[0043] In a preferred embodiment of the present application, in the above-mentioned artificial intelligence-based auto finance product recommendation method, the steps of screening out a target product method group from a plurality of candidate product method groups based on the product recommendation success rate corresponding to each candidate product method group, and using the candidate auto finance products and candidate financial product recommendation methods in the target product method group as the target auto finance products and target financial product recommendation methods corresponding to the auto finance product recommendation user, respectively, include:

[0044] Filtering out each candidate product mode group whose corresponding product recommendation success rate is greater than or equal to a predetermined recommendation success rate threshold from the plurality of candidate product mode groups, obtaining at least one candidate product mode group as the pending product mode group;

[0045] If the number of the pending product method groups is equal to 1, a corresponding pending product method group is determined as a target product method group; if the number of the pending product method groups is greater than 1, for each of the multiple pending product method groups, a first configuration weight of the candidate auto financial products in the pending product method group is determined, and a second configuration weight of the recommendation method of the candidate financial products in the pending product method group is determined, and based on the first configuration weight and the second configuration weight, a target configuration weight corresponding to the pending product method group is determined, and based on the target configuration weight, a product recommendation success rate corresponding to the pending product method group is updated to obtain an updated product recommendation success rate of the pending product method group, and, among the multiple pending product method groups, a pending product method group having the maximum updated product recommendation success rate is determined, and the pending product method group is determined as the target product method group;

[0046] The candidate automobile financial products and the candidate financial product recommendation methods in the target product method group are respectively used as the target automobile financial products and the target financial product recommendation method corresponding to the automobile financial product recommendation user.

[0047] Based on the above, this application also provides an artificial intelligence-based auto finance product recommendation method system, including:

[0048] memory for storing computer programs;

[0049] The processor connected to the memory is used to execute the computer program stored in the memory to implement the above-mentioned automobile financial product recommendation method based on artificial intelligence.

[0050] The artificial intelligence-based automobile financial product recommendation method and system provided in the present application firstly determines a candidate automobile financial product cluster and a candidate financial product recommendation method cluster; secondly, performs arbitrary grouping processing on each candidate automobile financial product in the candidate automobile financial product cluster and each candidate financial product recommendation method in the candidate financial product recommendation method cluster to form a plurality of candidate product method groups; then, analyzes the product recommendation success rate corresponding to each candidate product method group; finally, screens out a target product method group based on the product recommendation success rate, and uses the candidate automobile financial products and candidate financial product recommendation methods in the target product method group as target automobile financial products and target financial product recommendation methods, respectively. Based on the above content, since when analyzing the product recommendation success rate, not only the relevant information of the corresponding automobile financial products and automobile financial product recommendation users will be considered, but also the adopted financial product recommendation method will be considered, so that the basis for analysis can be more sufficient, thereby ensuring that the analyzed product recommendation success rate has a high reliability, so as to achieve reliable automobile financial product recommendations and improve the problem of relatively low reliability of automobile financial product recommendations in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings.

[0052] Figure 1 This is a structural block diagram of the artificial intelligence-based automobile financial product recommendation system provided in an embodiment of the present application.

[0053] Figure 2 A flowchart of an artificial intelligence-based auto finance product recommendation method provided in an embodiment of the present application.

[0054] Figure 3 A schematic diagram of multi-level fusion provided in an embodiment of the present application. DETAILED DESCRIPTION

[0055] To make the objectives, technical solutions, and advantages of the embodiments of the present application more clear, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Generally, the components of the embodiments of the present application described and shown in the drawings herein can be arranged and designed in various different configurations.

[0056] Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the present application for protection, but merely represents selected embodiments of the present application. All other embodiments obtained by persons of ordinary skill in the art based on the embodiments in the present application without creative work are within the scope of protection of the present application.

[0057] like Figure 1 As shown, an embodiment of the present application provides an artificial intelligence-based automobile financial product recommendation system. The artificial intelligence-based automobile financial product recommendation system may include a memory, a processor, and. In detail, the memory and the processor are electrically connected directly or indirectly to achieve data transmission or interaction. For example, the memory and the processor can be electrically connected through one or more communication buses or signal lines. The processor is used to execute an executable computer program stored in the memory to implement the artificial intelligence-based automobile financial product recommendation method provided in the embodiment of the present application.

[0058] Optionally, the memory may be, but is not limited to, a random access memory (RAM), a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), etc. Furthermore, the processor may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), a system on a chip (SoC), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0059] I understand. Figure 1 The structure shown is for illustration only. The automobile financial product recommendation system based on artificial intelligence may also include Figure 1 More or fewer components than shown, or with Figure 1 The different configurations shown, for example, may also include a communication unit for exchanging information with other devices. In addition, the artificial intelligence-based automobile financial product recommendation system may be a server.

[0060] Combine Figure 2The embodiment of the present application also provides an artificial intelligence-based automobile financial product recommendation method that can be applied to the above-mentioned artificial intelligence-based automobile financial product recommendation system. The method steps defined in the process related to the artificial intelligence-based automobile financial product recommendation method can be implemented by the artificial intelligence-based automobile financial product recommendation system (hereinafter referred to as the recommendation system). Figure 2 The specific process shown is explained in detail.

[0061] Step S110 : determining a candidate auto financial product cluster and a candidate financial product recommendation method cluster.

[0062] In an embodiment of the present application, the recommendation system can determine a cluster of candidate auto finance products and a cluster of recommendation methods for the candidate financial products. The cluster of candidate auto finance products includes multiple candidate auto finance products (e.g., multiple different types of auto insurance), and the cluster of recommendation methods includes multiple recommendation methods for the candidate financial products (e.g., online text recommendations, online graphic recommendations, online video recommendations, offline text recommendations, offline graphic recommendations, etc.).

[0063] Step S120 : performing arbitrary grouping processing on each of the candidate auto financial products in the candidate auto financial product cluster and each of the candidate financial product recommendation methods in the candidate financial product recommendation method cluster to form a corresponding plurality of candidate product method groups.

[0064] In this embodiment of the present application, after determining a cluster of candidate auto finance products and a cluster of candidate financial product recommendation methods, the recommendation system can arbitrarily group each candidate auto finance product in the cluster and each candidate financial product recommendation method in the cluster to form a corresponding plurality of candidate product method groups. Each candidate product method group includes a candidate auto finance product and a candidate financial product recommendation method, such as 1 million third-party liability insurance and an online text recommendation.

[0065] Step S130 , loading the user description data corresponding to each candidate product mode group and the auto finance product recommendation user respectively, so as to analyze the product recommendation success rate corresponding to each candidate product mode group through the target product recommendation network.

[0066] In this embodiment of the present application, after forming the plurality of candidate product method groups, the recommendation system may load user profile data corresponding to each candidate product method group and the user who recommended the auto finance product, and analyze the product recommendation success rate corresponding to each candidate product method group through a target product recommendation network. The target product recommendation network is a pre-trained neural network, and the product recommendation success rate reflects the likelihood (e.g., 0-1) that the auto finance product recommendation user will accept the candidate auto finance product after the candidate auto finance product is recommended according to the corresponding candidate financial product recommendation method.

[0067] Step S140, based on the product recommendation success rate corresponding to each of the candidate product method groups, a target product method group is screened out from the multiple candidate product method groups, and the candidate auto financial products and candidate financial product recommendation methods in the target product method group are respectively used as the target auto financial products and target financial product recommendation methods corresponding to the auto financial product recommendation user.

[0068] In this embodiment of the present application, after analyzing the corresponding product recommendation success rate, the recommendation system may filter out a target product method group from the multiple candidate product method groups based on the product recommendation success rate corresponding to each candidate product method group, and use the candidate auto finance products and candidate financial product recommendation methods in the target product method group as the target auto finance products and target financial product recommendation methods, respectively, corresponding to the auto finance product recommendation user. The target auto finance product is recommended to the auto finance product recommendation user according to the target financial product recommendation method.

[0069] Based on the above content, when analyzing the success rate of product recommendations, not only the relevant information of the corresponding auto finance products and the users who recommend the auto finance products will be considered, but also the financial product recommendation method adopted will be considered. This makes the basis for analysis more sufficient, thereby ensuring that the analyzed product recommendation success rate has a high reliability, so as to achieve reliable auto finance product recommendations and improve the problem of relatively low reliability of auto finance product recommendations in the existing technology.

[0070] It should be noted that for step S130 , the specific method of analyzing the product recommendation success rate corresponding to each candidate product mode group through the target product recommendation network is not limited and can be selected accordingly according to actual needs.

[0071] For example, in an alternative embodiment, in order to ensure that the product recommendation success rate corresponding to the candidate product mode group can be reliably determined, the above-mentioned step S130 can further include step S131, step S132, step S133, step S134 and step S135, the specific contents of which are as follows.

[0072] Step S131 : loading the user description data corresponding to each candidate product mode group and the user to whom the auto finance product is recommended, so that the target product recommendation network obtains the user description data corresponding to each candidate product mode group and the user to whom the auto finance product is recommended.

[0073] In this embodiment of the present application, user profile data corresponding to each candidate product method group and user for whom the auto finance product is recommended can be loaded separately, so that the target product recommendation network obtains user profile data corresponding to each candidate product method group and user for whom the auto finance product is recommended. The user profile data includes at least user identity information (e.g., gender, age, occupation, address, etc.) and financial attribute information (e.g., income, debt status, family situation, historical car usage, used car type, credit score, etc.) of the user for whom the auto finance product is recommended.

[0074] Step S132 : utilizing the first feature mining unit in the target product recommendation network to perform feature mining on the user description data corresponding to the user to whom the auto finance product is recommended, to obtain user semantic features corresponding to the user to whom the auto finance product is recommended.

[0075] In an embodiment of the present application, after obtaining the user description data corresponding to the user who recommends the auto finance product, the first feature mining unit in the target product recommendation network can be used to perform feature mining on the user description data corresponding to the user who recommends the auto finance product, and obtain the user semantic features corresponding to the user who recommends the auto finance product. Exemplarily, the first feature mining unit can be a word embedding model, so that the user description data can be embedded to obtain corresponding embedding features and serve as user semantic features. It should be noted that since the user description data can include user identity information and financial attribute information, they can be embedded separately to obtain corresponding local embedding features, and then the two local embedding features can be spliced ​​to obtain user semantic features. In addition, when performing the embedding process, the corresponding information can be segmented and embedded, and then the embedding vectors of each word can be averaged to obtain the corresponding local embedding features. For example, the "monthly income of 20,000 yuan" is embedded to obtain:

[0076] Month: [0.12,0.34,-0.25,0.06,-0.13,0.45,0.18,-0.02,0.11,0.23,-0.07,0.16, - 0.21,0.08,0.05,-0.13,...,0.31,-0.24,0.09,0.14];

[0077] Income: [-0.01, 0.22, 0.11, 0.44, -0.09, 0.37, -0.12, 0.28, -0.21, 0.19, 0.33, - 0.14,0.24,-0.08,0.21,0.17,...,-0.04,0.33,0.26,0.13];

[0078] 20,000: [0.11,0.24,-0.35,0.76,-0.83,0.45,0.98,-0.12,0.21,0.33,-0.47,0.56, - 0.27,0.06,0.05,-0.14,...,0.11,-0.24,0.19,0.15];

[0079] Yuan: [0.152,0.36,-0.27,0.08,-0.19,0.41,0.13,-0.04,0.16,0.27,-0.06,0.56, - 0.31,0.48,0.15,-0.23,...,0.81,-0.74,0.99,0.84].

[0080] Step S133: For each candidate product mode group, the second feature mining unit in the target product recommendation network is used to perform feature mining on the financial product description data of the candidate automobile financial products in the candidate product mode group to obtain the financial product semantic features corresponding to the candidate product mode group, and the third feature mining unit in the target product recommendation network is used to perform feature mining on the recommendation mode description data of the candidate financial product recommendation modes in the candidate product mode group to obtain the recommendation mode semantic features corresponding to the candidate product mode group.

[0081] In an embodiment of the present application, for each candidate product method group, the second feature mining unit in the target product recommendation network can be used to perform feature mining on the financial product description data of the candidate auto finance products in the candidate product method group to obtain the financial product semantic features corresponding to the candidate product method group, and the third feature mining unit in the target product recommendation network can be used to perform feature mining on the recommendation method description data of the candidate financial product recommendation methods in the candidate product method group to obtain the recommendation method semantic features corresponding to the candidate product method group. Exemplarily, the second feature mining unit and the third feature mining unit can be the same as the first feature mining unit, belonging to a word embedding model, that is, feature mining is completed through word embedding processing.

[0082] Step S134: For each of the candidate product mode groups, the feature fusion unit in the target product recommendation network is used to perform feature fusion on the financial product semantic features corresponding to the candidate product mode group, the recommendation mode semantic features corresponding to the candidate product mode group, and the user semantic features to obtain the financial product recommendation semantic features corresponding to the candidate product mode group.

[0083] In an embodiment of the present application, after determining the semantic features of the financial product, the semantic features of the recommendation method, and the semantic features of the user, for each candidate product method group, a feature fusion unit in the target product recommendation network can be used to perform feature fusion on the semantic features of the financial product corresponding to the candidate product method group, the semantic features of the recommendation method corresponding to the candidate product method group, and the semantic features of the user, to obtain the semantic features of the financial product recommendation corresponding to the candidate product method group. In other words, the semantic features of the financial product recommendation carry the semantic information of the semantic features of the financial product, the semantic information of the semantic features of the recommendation method, and the semantic information of the user, thereby enriching the semantic information in the semantic features of the financial product recommendation.

[0084] Step S135 : For each of the candidate product mode groups, the analysis output unit in the target product recommendation network is used to process the financial product recommendation semantic features corresponding to the candidate product mode group, and output the product recommendation success rate corresponding to the candidate product mode group.

[0085] In an embodiment of the present application, after obtaining the semantic features of financial product recommendation, for each of the candidate product mode groups, the analysis output unit in the target product recommendation network can be used to process the semantic features of financial product recommendation corresponding to the candidate product mode group, and output the product recommendation success rate corresponding to the candidate product mode group. Exemplarily, the analysis output unit may include at least a fully connected unit and an output function. After the fully connected unit processes the semantic features of financial product recommendation, the obtained fully connected semantic features may be processed by the output function to obtain the corresponding product recommendation success rate (i.e., the probability of success, which may be 0-1). In addition, the output function may be a function such as softmax, which is used to output the probability of success and the probability of failure; alternatively, a linear activation function may be used to output the probability of success.

[0086] It is understood that in step S134 above, the specific method for fusing the semantic features of the financial product, the semantic features of the recommendation method, and the semantic features of the user is not limited and can be selected based on actual needs. For example, in an alternative embodiment, to improve the reliability of feature fusion and achieve higher semantic representation accuracy of the semantic features of the financial product recommendation, step S134 above may further include steps S134a, S134b, and S134c, the specific contents of each step being described below.

[0087] Step S134a, utilizing the first feature fusion branch included in the feature fusion unit in the target product recommendation network, performs a first fusion on the financial product semantic features corresponding to the candidate product mode group and the recommendation mode semantic features corresponding to the candidate product mode group to form a first fused semantic feature.

[0088] In this embodiment of the present application, a first feature fusion branch included in a feature fusion unit in the target product recommendation network can be used to perform a first fusion of the financial product semantic features corresponding to the candidate product method group and the recommendation method semantic features corresponding to the candidate product method group to form a first fused semantic feature. In other words, the first fused semantic feature carries the semantic information of the corresponding financial product semantic features and the semantic information of the corresponding recommendation method semantic features, thereby achieving semantic fusion of the automotive financial product and the recommendation method.

[0089] Step S134b: Using the second feature fusion branch included in the feature fusion unit in the target product recommendation network, perform a second fusion on the first fused semantic feature and the user semantic feature to form a second fused semantic feature.

[0090] In this embodiment of the present application, after forming the first fused semantic feature, a second feature fusion branch included in the feature fusion unit in the target product recommendation network can be used to perform a second fusion on the first fused semantic feature and the user semantic feature to form a second fused semantic feature. In other words, the second fused semantic feature carries the semantic information in the corresponding first fused semantic feature and the semantic information in the user semantic feature, thus achieving the fusion of user data.

[0091] Step S134c: determining the financial product recommendation semantic features corresponding to the candidate product method group based at least on the second fused semantic features.

[0092] In an embodiment of the present application, after obtaining the second fused semantic feature, a financial product recommendation semantic feature corresponding to the candidate product method group can be determined based at least on the second fused semantic feature. For example, the second fused semantic feature can be directly used as the corresponding financial product recommendation semantic feature. Alternatively, the first fused semantic feature and the user semantic feature can be added together to obtain the corresponding financial product recommendation semantic feature.

[0093] It is understood that in the above step S134a, the specific method of performing the first fusion of the semantic features of the financial product and the semantic features of the recommendation method is not limited and can be selected according to actual needs. For example, in an alternative embodiment, in order to take into account both the efficiency and reliability of the fusion, the above step S134a may further include the following:

[0094] First, the first feature fusion branch of the feature fusion unit in the target product recommendation network includes three feature mapping matrices (which may be formed during the training process) to respectively map (e.g., multiply) the financial product semantic features corresponding to the candidate product mode group and the recommendation mode semantic features corresponding to the candidate product mode group to form a first financial product mapping feature and a second financial product mapping feature corresponding to the financial product semantic features, and to form a recommendation mode mapping feature corresponding to the recommendation mode semantic features. Exemplarily, the first feature mapping matrix and the second feature mapping matrix may be respectively multiplied by the financial product semantic features to obtain a first financial product mapping feature and a second financial product mapping feature, and the third feature mapping matrix may be multiplied by the recommendation mode semantic features to obtain a recommendation mode mapping feature.

[0095] Secondly, a feature correlation parameter between the recommendation method mapping feature and the first financial product mapping feature can be determined (for example, the recommendation method mapping feature can be multiplied by the transposed result of the first financial product mapping feature to obtain the corresponding feature correlation parameter), and based on the feature correlation parameter, the second financial product mapping feature can be updated to form an updated second financial product mapping feature (for example, the feature correlation parameter can be multiplied by the second financial product mapping feature to obtain the updated second financial product mapping feature. In this way, semantic information in the recommendation method semantic feature can be fused into the financial product semantic feature, or in other words, semantic information related to the recommendation method semantic feature can be mined from the financial product semantic feature to achieve associated fusion of semantic information).

[0096] Finally, the first fused semantic feature can be determined based on the updated second financial product mapping feature; for example, the updated second financial product mapping feature can be directly used as the first fused semantic feature, or the updated second financial product mapping feature can be added to the financial product semantic feature to obtain the corresponding first fused semantic feature.

[0097] It can be understood that in the above-mentioned step S134b, the specific method of performing the second fusion of the first fused semantic features and the user semantic features is not restricted and can be selected according to actual needs. For example, in an alternative embodiment, in order to ensure the reliability of the second fusion, the above-mentioned step S134b can further include step b1, step b2, step b3 and step b4, and the specific content of each step is described below.

[0098] Step b1: performing feature splitting processing on the user semantic feature to form at least a first local user semantic feature and a second local user semantic feature.

[0099] In an embodiment of the present application, the user semantic features may be subjected to feature splitting processing to generate at least a first local user semantic feature and a second local user semantic feature. The first local user semantic feature is used to reflect the semantic information of the user identity information, and the second local user semantic feature is used to reflect the semantic information of the financial attribute information. For example, when the user description data only includes user identity information and financial attribute information, after the feature splitting processing, only the first local user semantic feature and the second local user semantic feature may be obtained.

[0100] Step b2: Using the first feature mapping matrix set included in the second feature fusion branch included in the feature fusion unit in the target product recommendation network, the first local user semantic feature and the second local user semantic feature are respectively mapped to form a first local user mapping feature corresponding to the first local user semantic feature, and to form a second local user mapping feature and a third local user mapping feature corresponding to the second local user semantic feature, wherein the first feature mapping matrix set includes three feature mapping matrices.

[0101] In an embodiment of the present application, after obtaining the first local user semantic feature and the second local user semantic feature, the first local user semantic feature and the second local user semantic feature can be mapped using the first feature mapping matrix set included in the second feature fusion branch included in the feature fusion unit in the target product recommendation network (e.g., the first local user semantic feature is multiplied by the first feature mapping matrix, and the second local user semantic feature is multiplied by the second feature mapping matrix and the third feature mapping matrix, respectively). This forms a first local user mapping feature corresponding to the first local user semantic feature, and forms a second local user mapping feature and a third local user mapping feature corresponding to the second local user semantic feature. The first feature mapping matrix set includes three feature mapping matrices (which may be formed during training).

[0102] Step b3: Determine a feature correlation parameter between the first local user mapping feature and the second local user mapping feature, and based on the feature correlation parameter, update the third local user mapping feature to form an updated third local user mapping feature, and determine the associated user semantic feature based on the updated third local user mapping feature.

[0103] In an embodiment of the present application, after obtaining three local user mapping features, a feature correlation parameter between the first local user mapping feature and the second local user mapping feature can be determined (e.g., multiplying the first local user mapping feature by the transposed result of the second local user mapping feature), and, based on the feature correlation parameter, the third local user mapping feature is updated to form an updated third local user mapping feature (e.g., multiplying the feature correlation parameter by the third local user mapping feature), and based on the updated third local user mapping feature, an associated user semantic feature is determined (e.g., the updated third local user mapping feature can be directly determined as the associated user semantic feature; or, the updated third local user mapping feature can be added to the second local user semantic feature). Based on this, the semantic information in the first local user semantic feature can be fused into the second local user semantic feature, or, in other words, semantic information related to the first local user semantic feature can be mined from the second local user semantic feature to achieve associated fusion of semantic information.

[0104] Step b4: using the fusion network layer included in the second feature fusion branch, perform multi-level fusion on the first fused semantic feature and the associated user semantic feature to form a second fused semantic feature.

[0105] In an embodiment of the present application, after obtaining the associated user semantic feature, the fusion network layer included in the second feature fusion branch can be used to perform multi-level fusion of the first fused semantic feature and the associated user semantic feature to form a second fused semantic feature. In this way, deep information in the first fused semantic feature and the associated user semantic feature can be captured.

[0106] It is understandable that, in the above step b4, the specific method of multi-level fusion of the first fused semantic feature and the associated user semantic feature is not limited and can be selected according to actual needs. For example, in an alternative embodiment, in order to ensure that reliable semantic information is fully captured in the multi-level fusion process, the second feature fusion branch may include a plurality of cascaded fusion network layers. Based on this, the above step b4 may include the following content (combined with Figure 3 ):

[0107] First, a first fusion network layer among the multiple fusion network layers can be used to perform correlation fusion on the associated user semantic features based on the first fusion semantic features to form a correlation fusion semantic feature corresponding to the first fusion network layer; that is, the semantic information in the first fusion semantic feature can be fused into the associated user semantic feature;

[0108] Secondly, for each fusion network layer other than the first fusion network layer in the multiple fusion network layers, the current fusion network layer can be used to perform pooling processing on the first fusion semantic feature and the correlation fusion semantic feature corresponding to the previous fusion network layer, respectively, to form the corresponding first pooling feature and the second pooling feature, and, based on the first pooling feature, the second pooling feature is correlation fused to form the correlation fusion semantic feature corresponding to the current fusion network layer, wherein the pooling processing methods performed in each two adjacent fusion network layers are different. For example, the pooling processing method can be performed in each two fusion network layers, or the pooling processing method can be performed only in two adjacent fusion network layers, such as mean pooling in the second fusion network layer, maximum pooling in the third fusion network layer, mean pooling in the fourth fusion network layer, and maximum pooling in the fifth fusion network layer, alternating cycles; in addition, in the same fusion network layer, the method of pooling the first fusion semantic feature and the correlation fusion semantic feature corresponding to the previous fusion network layer is exactly the same;

[0109] Then, the multiple correlation fusion semantic features corresponding to the multiple fusion network layers may be superimposed or averaged to form a corresponding second fusion semantic feature.

[0110] It is understandable that the specific manner of performing correlation fusion in the above steps is not limited and can be selected according to actual needs. For example, in an alternative embodiment, in order to ensure that the associated semantic information can be fully mined when performing correlation fusion, the first fusion network layer of the multiple fusion network layers is used to perform correlation fusion on the associated user semantic features based on the first fusion semantic features to form the correlation fusion semantic features corresponding to the first fusion network layer. The step may further include the following:

[0111] First, the first fused semantic feature and the associated user semantic feature can be mapped separately using a second feature mapping matrix set included in a first fused network layer among the multiple fused network layers to form a first fused mapping feature corresponding to the first fused semantic feature, and to form a second fused mapping feature and a third fused mapping feature corresponding to the associated user semantic feature, wherein the second feature mapping matrix set includes three feature mapping matrices, as described above.

[0112] Secondly, determine the feature correlation parameter between the first fusion mapping feature and the second fusion mapping feature, and, based on the feature correlation parameter, update the third fusion mapping feature to form an updated third fusion mapping feature (i.e., fuse the semantic information in the first fusion semantic feature into the associated user semantic feature, or, in other words, mine the semantic information related to the first fusion semantic feature in the associated user semantic feature to achieve corresponding correlation fusion), and based on the updated third fusion mapping feature, determine the correlation fusion semantic feature corresponding to the first fusion network layer, as described above.

[0113] In addition, to ensure that the target product recommendation network can reliably analyze the product recommendation success rate corresponding to each candidate product mode group during step S130, the artificial intelligence-based auto finance product recommendation method may further include a step of training the target product recommendation network. This step may include the following specific implementation contents:

[0114] First, the first feature mining unit in the candidate product recommendation network can be used to perform feature mining on the sample user description data to obtain the corresponding sample user semantic features, as described above.

[0115] Secondly, the second feature mining unit in the candidate product recommendation network can be used to perform feature mining on the sample financial product description data to obtain the corresponding semantic features of the sample financial products, as described above.

[0116] Then, the third feature mining unit in the candidate product recommendation network can be used to perform feature mining on the sample recommendation method description data to obtain the corresponding sample recommendation method semantic features, as described above.

[0117] Afterwards, the feature fusion unit in the candidate product recommendation network may be used to perform feature fusion on the semantic features of the sample financial product, the semantic features of the sample recommendation method, and the semantic features of the sample user to obtain corresponding semantic features for sample financial product recommendations, as described above.

[0118] Furthermore, the analysis output unit in the candidate product recommendation network can be used to process the semantic features of the sample financial product recommendation and output the corresponding sample product recommendation success rate, as described above.

[0119] Finally, based on the error between the sample product recommendation success rate and the product recommendation success rate label (the specific error calculation method is not limited and can be selected according to actual needs), the network parameters of the candidate product recommendation network can be updated (for example, the network parameters can be updated and adjusted in the direction of reducing the error until the error converges) to form the target product recommendation network, wherein the product recommendation success rate label refers to the actual product recommendation success rate (such as 0 or 1, where 0 indicates unsuccessful recommendation and 1 indicates successful recommendation) of recommending the auto financial product corresponding to the sample financial product description data to the user corresponding to the sample user description data according to the financial product recommendation method corresponding to the sample recommendation method description data.

[0120] It should be noted that for step S140 , the specific method of determining the final target automobile financial product and the target financial product recommendation method is not limited and can be selected according to actual needs.

[0121] For example, in an alternative embodiment, to ensure that the final target auto finance product and the target finance product recommendation method determined are highly reliable and to make the recommendation more reliable, the above-mentioned step S140 may further include the following specific implementation contents:

[0122] First, each candidate product mode group whose corresponding product recommendation success rate is greater than or equal to a predetermined recommendation success rate threshold can be screened out from the plurality of candidate product mode groups to obtain at least one candidate product mode group as the pending product mode group; illustratively, the recommendation success rate threshold can be a value such as 0.5, 0.6, or 0.7, which can be configured according to needs;

[0123] Secondly, if the number of the pending product mode groups is equal to 1, a corresponding pending product mode group is determined as the target product mode group; if the number of the pending product mode groups is greater than 1, for each of the multiple pending product mode groups, a first configuration weight of the candidate auto finance products in the pending product mode group is determined (the first configuration weight may be generated in response to a configuration operation performed by a management user, and the first configuration weight of an important candidate auto finance product may be greater, and the first configuration weight of an unimportant candidate auto finance product may be smaller), and a second configuration weight of a recommended method for the candidate financial products in the pending product mode group is determined (the second configuration weight may also be generated in response to a configuration operation performed by a management user, and may be The first configuration weight and the second configuration weight are configured according to the importance of the candidate financial product recommendation method, and a target configuration weight corresponding to the pending product method group is determined based on the first configuration weight and the second configuration weight (for example, the first configuration weight and the second configuration weight can be averaged or weighted summed to obtain the target configuration weight), and the product recommendation success rate corresponding to the pending product method group is updated based on the target configuration weight (for example, the target configuration weight and the product recommendation success rate are multiplied together) to obtain the updated product recommendation success rate of the pending product method group, and, among the multiple pending product method groups, a pending product method group having the maximum updated product recommendation success rate is determined, and the pending product method group is determined as the target product method group;

[0124] Finally, the candidate automobile financial products and the candidate financial product recommendation methods in the target product method group may be used as the target automobile financial products and the target financial product recommendation method corresponding to the automobile financial product recommendation user, respectively.

[0125] In addition, in some embodiments, if there is no corresponding candidate product mode group whose product recommendation success rate is greater than or equal to a predetermined recommendation success rate threshold, then the candidate product mode group with the highest product recommendation success rate can be directly determined as the target product mode group.

[0126] In summary, the artificial intelligence-based automobile financial product recommendation method and system provided by the present application first determines a candidate automobile financial product cluster and a candidate financial product recommendation method cluster; secondly, each candidate automobile financial product in the candidate automobile financial product cluster and each candidate financial product recommendation method in the candidate financial product recommendation method cluster are arbitrarily grouped to form multiple candidate product method groups; then, the product recommendation success rate corresponding to each candidate product method group is analyzed; finally, the target product method group is screened out based on the product recommendation success rate, and the candidate automobile financial products and candidate financial product recommendation methods in the target product method group are respectively used as the target automobile financial products and target financial product recommendation methods. Based on the above content, since when analyzing the product recommendation success rate, not only the relevant information of the corresponding automobile financial products and automobile financial product recommendation users will be considered, but also the adopted financial product recommendation method will be considered, so that the basis for analysis can be more sufficient, thereby ensuring that the analyzed product recommendation success rate has a high reliability, so as to achieve reliable automobile financial product recommendations, thereby improving the problem of relatively low reliability of automobile financial product recommendations existing in the prior art.

[0127] In the several embodiments provided in the embodiments of the present application, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device and method embodiments described above are merely schematic. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architectures, functions and operations of the devices, methods and computer program products according to the multiple embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of the code, and the module, program segment or a part of the code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart, can be implemented with a dedicated hardware-based system that performs the specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0128] In addition, the functional modules in each embodiment of the present application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0129] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or partly contributed to the prior art or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, electronic device, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk. It should be noted that, in this article, the terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such a process, method, article or device. Without further constraints, an element defined by the phrase "comprises a..." does not preclude the existence of additional identical elements in the process, method, article or apparatus that includes the element.

[0130] The above description is merely a preferred embodiment of the present application and is not intended to limit the present application. Various modifications and variations are possible for those skilled in the art. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present application shall be included within the scope of protection of the present application.

Claims

1. An artificial intelligence-based automobile financial product recommendation method, characterized in that: include: Determining a candidate auto financial product cluster and a candidate financial product recommendation method cluster, wherein the candidate auto financial product cluster includes a plurality of candidate auto financial products, and the candidate financial product recommendation method cluster includes a plurality of candidate financial product recommendation methods; performing arbitrary grouping processing on each candidate auto financial product in the candidate auto financial product cluster and each candidate financial product recommendation method in the candidate financial product recommendation method cluster to form a corresponding plurality of candidate product method groups, wherein each candidate product method group includes a candidate auto financial product and a candidate financial product recommendation method; The user description data corresponding to each candidate product mode group and the user who recommends the auto financial product are loaded respectively, so that the target product recommendation network obtains the user description data corresponding to each candidate product mode group and the user who recommends the auto financial product, wherein the user description data at least includes the user identity information and financial attribute information of the user who recommends the auto financial product; the first feature mining unit in the target product recommendation network is used to perform feature mining on the user description data corresponding to the user who recommends the auto financial product to obtain the user semantic features corresponding to the user who recommends the auto financial product; for each candidate product mode group, the second feature mining unit in the target product recommendation network is used to perform feature mining on the user description data corresponding to the user who recommends the auto financial product to obtain the user semantic features corresponding to the user who recommends the auto financial product; The unit performs feature mining on the financial product description data of the candidate automobile financial products in the candidate product mode group to obtain the semantic features of the financial products corresponding to the candidate product mode group, and utilizes the third feature mining unit in the target product recommendation network to perform feature mining on the recommendation mode description data of the candidate financial product recommendation modes in the candidate product mode group to obtain the semantic features of the recommendation modes corresponding to the candidate product mode group; utilizes the first feature fusion branch of the feature fusion unit in the target product recommendation network to have three feature mapping matrices, and respectively maps the semantic features of the financial products corresponding to the candidate product mode group and the semantic features of the recommendation modes corresponding to the candidate product mode group to form the target product recommendation network. The first financial product mapping feature and the second financial product mapping feature corresponding to the financial product semantic feature are formed to form a recommendation method mapping feature corresponding to the recommendation method semantic feature; a feature correlation parameter between the recommendation method mapping feature and the first financial product mapping feature is determined, and based on the feature correlation parameter, the second financial product mapping feature is updated to form an updated second financial product mapping feature; based on the updated second financial product mapping feature, a first fused semantic feature is determined; and a second feature fusion branch included in the feature fusion unit in the target product recommendation network is used to perform a second fusion on the first fused semantic feature and the user semantic feature to form a second fusing semantic features; determining, based at least on the second fused semantic features, financial product recommendation semantic features corresponding to the candidate product method group; for each candidate product method group, processing the financial product recommendation semantic features corresponding to the candidate product method group using the analysis and output unit in the target product recommendation network, and outputting a product recommendation success rate corresponding to the candidate product method group, wherein the target product recommendation network is a pre-trained neural network, and the product recommendation success rate is used to reflect the likelihood that the user of the auto financial product recommendation will accept the candidate auto financial product after the candidate financial product is recommended to the corresponding candidate financial product in accordance with the corresponding candidate financial product recommendation method; Based on the product recommendation success rate corresponding to each of the candidate product method groups, a target product method group is screened out from the multiple candidate product method groups, and the candidate auto finance products and candidate financial product recommendation methods in the target product method group are respectively used as the target auto finance products and target financial product recommendation methods corresponding to the auto finance product recommendation user, wherein the target auto finance product is recommended to the auto finance product recommendation user according to the target financial product recommendation method.

2. The method for recommending automobile financial products based on artificial intelligence according to claim 1, characterized in that: The step of performing a second fusion on the first fused semantic feature and the user semantic feature by utilizing the second feature fusion branch included in the feature fusion unit in the target product recommendation network to form a second fused semantic feature includes: performing feature splitting processing on the user semantic feature to form at least a first local user semantic feature and a second local user semantic feature, wherein the first local user semantic feature is used to reflect the semantic information of the user identity information, and the second local user semantic feature is used to reflect the semantic information of the financial attribute information; Using a first feature mapping matrix set included in a second feature fusion branch included in a feature fusion unit in the target product recommendation network, the first local user semantic feature and the second local user semantic feature are respectively mapped to form a first local user mapping feature corresponding to the first local user semantic feature, and to form a second local user mapping feature and a third local user mapping feature corresponding to the second local user semantic feature, wherein the first feature mapping matrix set includes three feature mapping matrices; determining a feature correlation parameter between the first local user mapping feature and the second local user mapping feature, and, based on the feature correlation parameter, updating the third local user mapping feature to form an updated third local user mapping feature, and determining an associated user semantic feature based on the updated third local user mapping feature; The first fused semantic feature and the associated user semantic feature are fused at multiple levels using the fusion network layer included in the second feature fusion branch to form a second fused semantic feature.

3. The method for recommending automobile financial products based on artificial intelligence according to claim 2, characterized in that: The second feature fusion branch includes a plurality of cascaded fusion network layers, and the step of using the fusion network layers included in the second feature fusion branch to perform multi-level fusion on the first fused semantic feature and the associated user semantic feature to form a second fused semantic feature includes: Using a first fusion network layer among the multiple fusion network layers, and based on the first fusion semantic feature, performing correlation fusion on the associated user semantic feature to form a correlation fusion semantic feature corresponding to the first fusion network layer; For each fusion network layer other than the first fusion network layer in the multiple fusion network layers, using the current fusion network layer, respectively, performing pooling processing on the first fusion semantic feature and the correlation fusion semantic feature corresponding to the previous fusion network layer to form corresponding first pooling features and second pooling features, and, based on the first pooling feature, performing correlation fusion on the second pooling feature to form a correlation fusion semantic feature corresponding to the current fusion network layer, wherein the pooling processing method performed in each two adjacent fusion network layers is different; The multiple correlation fusion semantic features corresponding to the multiple fusion network layers are superimposed or averaged to form corresponding second fusion semantic features.

4. The method for recommending automobile financial products based on artificial intelligence according to claim 3, characterized in that: The step of using a first fusion network layer among the multiple fusion network layers to perform correlation fusion on the associated user semantic features based on the first fusion semantic features to form a correlation fusion semantic feature corresponding to the first fusion network layer includes: Using a second feature mapping matrix set included in a first fusion network layer among the multiple fusion network layers, mapping the first fusion semantic feature and the associated user semantic feature respectively to form a first fusion mapping feature corresponding to the first fusion semantic feature, and forming a second fusion mapping feature and a third fusion mapping feature corresponding to the associated user semantic feature, wherein the second feature mapping matrix set includes three feature mapping matrices; Determine the feature correlation parameter between the first fusion mapping feature and the second fusion mapping feature, and, based on the feature correlation parameter, update the third fusion mapping feature to form an updated third fusion mapping feature, and based on the updated third fusion mapping feature, determine the correlation fusion semantic feature corresponding to the first fusion network layer.

5. The method for recommending automobile financial products based on artificial intelligence according to claim 1, characterized in that: The artificial intelligence-based automobile financial product recommendation method further includes a step of training a target product recommendation network, which includes: Using the first feature mining unit in the candidate product recommendation network, feature mining is performed on the sample user description data to obtain the corresponding sample user semantic features; Using the second feature mining unit in the candidate product recommendation network, perform feature mining on the sample financial product description data to obtain corresponding semantic features of the sample financial products; Using the third feature mining unit in the candidate product recommendation network, feature mining is performed on the sample recommendation method description data to obtain corresponding sample recommendation method semantic features; Using a feature fusion unit in the candidate product recommendation network, the semantic features of the sample financial products, the semantic features of the sample recommendation methods, and the semantic features of the sample users are fused to obtain corresponding semantic features for sample financial product recommendations; Utilizing the analysis output unit in the candidate product recommendation network, processing the semantic features of the sample financial product recommendation and outputting the corresponding sample product recommendation success rate; Based on the error between the sample product recommendation success rate and the product recommendation success rate label, the network parameters of the candidate product recommendation network are updated to form the target product recommendation network, wherein the product recommendation success rate label refers to the actual product recommendation success rate of recommending the auto financial product corresponding to the sample financial product description data to the user corresponding to the sample user description data according to the financial product recommendation method corresponding to the sample recommendation method description data.

6. The method for recommending automobile financial products based on artificial intelligence according to any one of claims 1 to 5, characterized in that: The steps of screening a target product method group from a plurality of candidate product method groups based on the product recommendation success rate corresponding to each candidate product method group, and using the candidate auto finance products and candidate financial product recommendation methods in the target product method group as the target auto finance products and target financial product recommendation methods corresponding to the auto finance product recommendation user, respectively, include: Filtering out each candidate product mode group whose corresponding product recommendation success rate is greater than or equal to a predetermined recommendation success rate threshold from the plurality of candidate product mode groups, obtaining at least one candidate product mode group as the pending product mode group; If the number of the pending product method groups is equal to 1, a corresponding pending product method group is determined as a target product method group; if the number of the pending product method groups is greater than 1, for each of the multiple pending product method groups, a first configuration weight of the candidate auto financial products in the pending product method group is determined, and a second configuration weight of the recommendation method of the candidate financial products in the pending product method group is determined, and based on the first configuration weight and the second configuration weight, a target configuration weight corresponding to the pending product method group is determined, and based on the target configuration weight, a product recommendation success rate corresponding to the pending product method group is updated to obtain an updated product recommendation success rate of the pending product method group, and, among the multiple pending product method groups, a pending product method group having the maximum updated product recommendation success rate is determined, and the pending product method group is determined as the target product method group; The candidate automobile financial products and the candidate financial product recommendation methods in the target product method group are respectively used as the target automobile financial products and the target financial product recommendation method corresponding to the automobile financial product recommendation user.

7. An artificial intelligence-based automobile financial product recommendation system, characterized in that: include: memory for storing computer programs; A processor connected to the memory, configured to execute a computer program stored in the memory to implement the artificial intelligence-based automobile financial product recommendation method according to any one of claims 1 to 6.

Images