Non-vehicle product recommendation method and system

Abstract

The invention provides a non-vehicle product recommendation method and system, and the method comprises the steps: determining a first vector representation of a to-be-recommended user, the first vector representation representing first user data of the to-be-recommended user and first interaction behavior data of the to-be-recommended user for insurance types, the insurance types including non-vehicle products, and the first interaction behavior data of the to-be-recommended user for the non-vehicle products; the first vector representation is matched with second vector representations corresponding to the to-be-recommended non-vehicle products, target non-vehicle products recommended for the to-be-recommended users are obtained, and the second vector representation of any non-vehicle product represents product data of any non-vehicle product and interaction behavior data corresponding to any non-vehicle product. The individuation, effectiveness and reliability of recommendation can be improved.

Description

Technical Field

[0001] This specification relates to the field of artificial intelligence technology, and in particular to a recommendation method and system for non-vehicle products. Background Technology

[0002] In the auto insurance sector, bundling non-auto products (such as health insurance, home insurance, and medical insurance in addition to auto insurance) is a common business strategy.

[0003] In related technologies, non-motor products are mainly recommended to car insurance users through manual methods.

[0004] However, on the one hand, manual methods are easily affected by subjective human factors, resulting in low accuracy and reliability of recommendations; on the other hand, manual methods are also prone to inefficiency and lack of personalization.

[0005] It should be noted that the above-mentioned related technologies are only information known to the inventor personally, and do not mean that the above information had entered the public domain before the application date of this specification, nor do they mean that it can be considered prior art in this specification. Summary of the Invention

[0006] This specification provides a recommended method and system for non-vehicle products to avoid at least one of the above-mentioned technical problems.

[0007] Firstly, this specification provides a recommended method for non-vehicle products, including: A first vector representation of the user to be recommended is determined, wherein the first vector representation represents the first user data of the user to be recommended and the first interaction behavior data of the user to be recommended with the insurance type, and the insurance type includes non-vehicle products; The first vector representation is matched with the second vector representation corresponding to each non-vehicle product to be recommended to obtain the target non-vehicle product recommended to the user to be recommended. The second vector of any non-vehicle product represents the product data of the non-vehicle product and the interaction behavior data corresponding to the non-vehicle product.

[0008] Secondly, this specification provides a recommendation system for non-vehicle products, including: At least one storage medium stores at least one instruction set for recommending non-vehicle products; At least one processor is communicatively connected to the at least one storage medium, wherein when the at least one processor is running, it reads the at least one instruction set and executes the method as described in the first aspect according to the instructions of the at least one instruction set.

[0009] Thirdly, this specification provides a computer-readable non-transitory storage medium, wherein the computer-readable non-transitory storage medium stores at least one instruction set, which is executed by at least one processor to implement the method as described in the first aspect.

[0010] As can be seen from the above technical solutions, the non-vehicle product recommendation method and system provided in this specification recommend non-vehicle products to users by matching the vector representations of the users to be recommended with the vector representations of the non-vehicle products to be recommended. On one hand, the recommendation system uses vector representation technology to abstract complex user and product information into computable objects, and then uses matching to recommend non-vehicle products, which can improve the efficiency and reliability of the recommendation. On the other hand, since the first vector representation utilizes both user attribute data and user interaction data with non-vehicle products, the first vector strongly represents the personalized characteristics of the users to be recommended and their preference characteristics under different non-vehicle products, thus improving the targeting and personalization of the recommendation. Furthermore, since the second vector representation utilizes both non-vehicle product attribute data and data on different users' interactions with non-vehicle products, the second vector representation strongly represents the characteristics of non-vehicle products and their characteristics under different user preferences, thus improving the accuracy and reliability of the recommendation through matching.

[0011] Other functions of the recommended methods and systems for non-vehicle products provided in this specification will be partially listed in the following description. The inventive aspects of the recommended methods and systems for non-vehicle products provided in this specification can be fully explained by practice or use of the methods, devices, and combinations described in the detailed examples below. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of this specification, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a schematic diagram illustrating the application scenario of the recommendation method for non-vehicle products provided in the embodiments of this specification; Figure 2 This is a schematic diagram of the structure of the recommendation system for non-vehicle products provided in the embodiments of this specification; Figure 3 A flowchart illustrating a method for recommending non-vehicle products according to one embodiment of this specification; Figure 4 A flowchart illustrating a method for recommending non-vehicle products according to another embodiment of this specification; Figure 5 This is a schematic diagram of the layer structure of the recommendation system for non-vehicle products provided in the embodiments of this specification; Figure 6 This is a schematic diagram illustrating the principle of a recommended method for non-vehicle products provided in one embodiment of this specification. Detailed Implementation

[0014] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this specification. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this specification as detailed in the appended claims.

[0015] It should be understood that the terms “comprising” and “having”, and any variations thereof, in the embodiments of this specification are intended to cover but not exclude inclusion. For example, a product or device that includes a series of components is not necessarily limited to those components that are explicitly listed, but may include other components that are not explicitly listed or that are inherent to such product or device.

[0016] The term "and / or" in the embodiments of this specification describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. The character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0017] In the embodiments of this specification, the term "multiple" refers to two or more, and other quantifiers are similar.

[0018] The terms “first,” “second,” “third,” “target,” etc., used in this specification are used to distinguish similar or related objects or entities and do not necessarily imply a specific order or sequence, unless otherwise indicated. It should be understood that such terms can be used interchangeably where appropriate, for example, in situations where implementation is possible in a sequence other than those given in the embodiments illustrated or described in this specification.

[0019] As used in this specification, the term "unit / module" means any known or subsequently developed hardware, software, firmware, artificial intelligence, fuzzy logic, or combination of hardware and / or software code capable of performing the functions associated with that element.

[0020] To avoid at least one of the technical problems mentioned in the background section above, this specification proposes a technical concept developed through inventive effort: obtaining a vector representation of the user to be recommended in the user dimension and the insurance type interaction dimension, and matching this vector representation with the vector representations of the product dimension and the interaction dimension corresponding to each different insurance type, so as to determine the non-vehicle products to be recommended to the user to be recommended.

[0021] Taking recommending non-vehicle products to user X as an example: First, the recommendation system can obtain a vector representation of user X to represent two dimensions of content. One dimension includes data about user X themselves (data such as user identity and the vehicle they own). The other dimension includes data about user X's interaction with insurance products (such as data related to user X's access to, clicks on, and order placement for non-vehicle products).

[0022] Then, the recommendation system can match the vector representation of user X with the corresponding vector representation of each non-vehicle product to be recommended, so as to determine and recommend the corresponding non-vehicle products to user X from the products to be recommended.

[0023] Similarly, the vector representation of a non-vehicle product can also characterize its characteristics from two dimensions. For example, the vector representation of non-vehicle product Y can represent two dimensions. One dimension includes product data of non-vehicle product Y itself (such as cost, duration, etc.). The other dimension includes interactive behavior data corresponding to non-vehicle product Y (such as data corresponding to user interactions with non-vehicle product Y, such as access, clicks, and order placement).

[0024] The technical solution provided in this specification is implemented based on the above-described technical concept. As can be seen from the description of the above technical concept, in the technical solution provided in this specification, the vector representation of the user to be recommended is matched with the corresponding vector representations of each non-vehicle product to be recommended, thereby obtaining the corresponding non-vehicle products recommended to the user. This avoids the problem of low recommendation accuracy and reliability caused by subjective human factors, thus improving the accuracy and reliability of the recommendation. Furthermore, it also avoids the problems of low recommendation efficiency and lack of personalization, thereby improving recommendation efficiency and enabling personalized recommendations.

[0025] To facilitate readers' understanding of this manual, the application scenarios of this manual are introduced below.

[0026] The technical solutions provided in this manual are applicable to scenarios where it is necessary to recommend non-vehicle products, such as recommending non-vehicle products to users.

[0027] Figure 1This diagram illustrates an application scenario of the recommendation method (hereinafter referred to as the recommendation method) for non-vehicle products according to embodiments of this specification. The recommendation method can be applied to, for example... Figure 1 Scenario 100 is shown. (e.g.) Figure 1 As shown, scenario 100 may include target user 101, client 102, server 103, and network 104.

[0028] Target user 101 can be the user who triggers the recommendation of non-vehicle products. For example, target user 101 can perform a targeted action on client 102 to trigger the recommendation of non-vehicle products.

[0029] Client 102 may be an electronic device that provides interactive functionality to target user 101. For example, client 102 may provide an interactive interface to target user 101, where target user 101 can perform target operations. In some embodiments, client 102 executes the recommended method described herein in response to detecting a target operation triggered by target user 101. In this case, client 102 may store data or instructions for executing the recommended method described herein, and may execute or be used to execute the data or instructions. In some embodiments, client 102 may include a hardware device with data processing capabilities and the necessary programs required to drive the hardware device to execute the recommended method described herein.

[0030] In some embodiments, client 102 may include a mobile device, tablet, laptop, built-in device in a motor vehicle, or similar content, or any combination thereof. In some embodiments, the mobile device may include a smart home device, a smart mobile device, a virtual reality device, an augmented reality device, or similar device, or any combination thereof. In some embodiments, smart home devices may include a smart TV, a desktop computer, etc., or any combination thereof. In some embodiments, smart mobile devices may include a smartphone, a personal digital assistant, a gaming device, a navigation device, etc., or any combination thereof. In some embodiments, built-in devices in a motor vehicle may include an in-vehicle computer, an in-vehicle television, etc.

[0031] In some embodiments, client 102 may have one or more applications (APPs) installed. APPs provide target user 101 with the ability and interface to interact with the outside world via network 104. APPs include, but are not limited to: web browser APPs, search APPs, chat APPs, shopping APPs, video APPs, financial management APPs, instant messaging tools, email clients, social media platform software, etc.

[0032] like Figure 1As shown, client 102 can establish a communication connection with server 103. Server 103 can communicate with one client 102 or multiple clients 102. In some embodiments, client 102 can interact with server 103 via network 104 to receive or send messages, etc.

[0033] Server 103 can be a server that provides various services. For example, server 103 can be a cloud server or a local server. Server 103 can communicate with one client 102 and receive data sent by that client 102, or it can communicate with multiple clients 102 and receive data sent by each client 102.

[0034] In some embodiments, the recommended methods described herein can be executed on server 103. In this case, server 103 may store data or instructions for executing the recommended methods described herein, and may execute or be used to execute the data or instructions. Server 103 may include hardware devices with data processing capabilities and the necessary programs required to drive the hardware devices.

[0035] Network 104 is a medium used to provide a communication connection between client 102 and server 103. Network 104 can facilitate the exchange of information or data. Figure 1 As shown, client 102 and server 103 can connect to network 104 respectively and transmit information or data to each other through network 104.

[0036] In some embodiments, network 104 can be any type of wired or wireless network, or a combination thereof. For example, network 104 may include a cable network, a wired network, a fiber optic network, a telecommunications network, an intranet, the Internet, a local area network (LAN), a wide area network (WAN), a wireless local area network (WLAN), a metropolitan area network (MAN), a public switched telephone network (PSTN), a Bluetooth network™, a ZigBee™ short-range wireless network, a near field communication (NFC) network, or a similar network.

[0037] In some embodiments, network 104 may include one or more network access points. For example, network 104 may include wired or wireless network access points, such as base stations or internet switching points, through which one or more components of client 102 and server 103 can connect to network 104 to exchange data or information.

[0038] It is worth noting that, Figure 1 The number of clients 102, servers 103, and networks 104 shown is merely illustrative. Depending on implementation needs, there can be any number of clients 102, servers 103, and networks 104. Furthermore, the recommended method provided in this specification can be implemented entirely on client 102, entirely on server 103, or partially on client 102 and partially on server 103.

[0039] In other words, Figure 1 and targeting Figure 1 The above description is only used to illustrate the application scenarios in which the recommended methods in this specification may be applicable, and should not be construed as limiting the application scenarios.

[0040] Figure 2 A hardware structure diagram of a non-vehicle product recommendation system (hereinafter referred to as the recommendation system) 200 provided according to an embodiment of this specification is shown. The recommendation system 200 can execute the recommendation methods described in this specification. The recommendation methods are described in other parts of this specification. When the recommendation methods are executed on client 102, the recommendation system 200 can be client 102. When the recommendation methods are executed on server 103, the recommendation system 200 can be server 103. When the recommendation methods are executed partly on client 102 and partly on server 103, the recommendation system 200 can be a system including client 102 and server 103.

[0041] like Figure 2 As shown, the recommendation system 200 may include at least one storage medium 203 and at least one processor 202. In some embodiments, the recommendation system 200 may also include a communication port 204 and an internal communication bus 201. The recommendation system 200 may also include I / O components 205.

[0042] The internal communication bus 201 can connect to different system components. For example, the internal communication bus 201 can connect to storage medium 203, processor 202, communication port 204, and I / O component 205.

[0043] I / O component 205 supports input / output between recommendation system 200 and other components.

[0044] Communication port 204 is used for data communication between the recommendation system 200 and the outside world. For example, communication port 204 can be used for data communication between the recommendation system 200 and network 104. Communication port 204 can be a wired communication port or a wireless communication port.

[0045] Storage medium 203 may include a data storage device. The data storage device may be a non-transitory storage medium or a temporary storage medium. For example, the data storage device may include one or more of a disk 2031, a read-only storage medium (ROM) 2032, or a random access storage medium (RAM) 2033. Storage medium 203 also includes at least one instruction set stored in the data storage device. The instruction set includes computer program code, which may include programs, routines, objects, components, data structures, procedures, modules, etc., that perform the recommended methods provided in this specification.

[0046] At least one processor 202 may be communicatively connected to at least one storage medium 203. At least one processor 202 is used to execute at least one instruction set described above. When the recommendation system 200 is running, at least one processor 202 reads the at least one instruction set and, according to the instructions of the at least one instruction set, executes the recommended method provided in this specification. Processor 202 may execute all steps included in the recommended method. Processor 202 may be in the form of one or more processors. In some embodiments, processor 202 may include one or more hardware processors, such as microcontrollers, microprocessors, reduced instruction set computers (RISC), application-specific integrated circuits (ASICs), application-specific instruction set processors (ASIPs), central processing units (CPUs), graphics processing units (GPUs), physical processing units (PPUs), microcontroller units, digital signal processors (DSPs), field-programmable gate arrays (FPGAs), advanced RISC machines (ARMs), programmable logic devices (PLDs), any circuit or processor capable of performing one or more functions, or any combination thereof.

[0047] For illustrative purposes only, the accompanying drawings show only one processor 202 in the recommendation system 200. However, it should be noted that the recommendation system 200 may also include multiple processors. Therefore, the operation and / or method steps disclosed herein may be executed by one processor or by multiple processors in combination. For example, if the processor 202 of the recommendation system 200 described in this specification executes steps A and B, it should be understood that steps A and B may also be executed jointly or separately by two different processors 202 (e.g., the first processor executes step A, the second processor executes step B, or the first and second processors jointly execute steps A and B).

[0048] Please see Figure 3 , Figure 3 This is a flowchart illustrating a method for recommending non-vehicle products according to one embodiment of this specification. Wherein, Figure 3 The entity executing the recommendation method shown can be a recommendation system. For a description of recommendation systems, please refer to the example above; it will not be repeated here.

[0049] like Figure 3 As shown, the method includes the following steps S301 and S302: S301: Determine the first vector representation of the user to be recommended, wherein the first vector representation represents the first user data of the user to be recommended and the first interaction behavior data of the user to be recommended on the insurance type through vector representation, and the insurance type includes non-vehicle products.

[0050] Users to be recommended can be understood as non-vehicle products that the recommendation system recommends.

[0051] This embodiment does not limit the triggering of recommending non-vehicle products to the user to be recommended. For example, in response to the user to be recommended logging into an insurance (or insurance-related) platform, the recommendation system can recommend non-vehicle products to the user to be recommended. As another example, in response to the user to be recommended browsing non-vehicle products, the recommendation system can recommend non-vehicle products to the user to be recommended.

[0052] Vector representation (embedding) can be understood as transforming discrete, unstructured raw data (such as text, categories, etc.) into continuous, low-dimensional real number vectors, giving them semantic and mathematical computation capabilities.

[0053] The first vector representation can be understood as a vector representation determined based on the static attributes of the user to be recommended (such as the user's first user data in the user dimension) + behavioral data (the user's first interaction behavior data with the insurance product in the interaction dimension). The first vector representation can be used to characterize the comprehensive characteristics and insurance product preferences of the user to be recommended.

[0054] The first user data may include static profile data such as age, gender, city, occupation, income, and vehicle-related information (such as vehicle model, vehicle age, vehicle price, etc.).

[0055] The first interaction behavior data may include: the historical clicks, browsing, adding to cart, and purchasing data of the user to be recommended to non-vehicle products.

[0056] S302: Match the first vector representation with the second vector representation corresponding to each non-vehicle product to be recommended to obtain the target non-vehicle product recommended to the user. The second vector representation of any non-vehicle product represents the product data and the interaction behavior data corresponding to any non-vehicle product through vector representation.

[0057] The second vector representation can be understood as a vector representation determined based on the attributes of each non-vehicle product and market feedback data (such as user interaction data with non-vehicle products). The second vector representation can be used to characterize the comprehensive characteristics and market characteristics (or user preference characteristics) of the corresponding non-vehicle product.

[0058] Product data may include: product name, coverage, price, target audience, and key terms and conditions.

[0059] Interaction data for a non-vehicle product may include metrics such as which users clicked / purchased the product, conversion rate, popularity, and click-through rate (CTR).

[0060] For example, there are multiple non-vehicle products to be recommended.

[0061] For each non-vehicle product to be recommended, the recommendation system first matches (or compares) its first vector representation with the second vector representation of that product to obtain the corresponding matching result. This allows the system to obtain the matching result for each non-vehicle product to be recommended.

[0062] Then, the recommendation system can determine the target non-vehicle product from the non-vehicle products to be recommended based on the matching results, and recommend the target non-vehicle product to the user to be recommended.

[0063] Based on the above analysis of S301 and S302, in this embodiment, the recommendation system recommends non-vehicle products to the user by matching the vector representation of the user to be recommended with the vector representation of the non-vehicle product to be recommended. On one hand, the recommendation system uses vector representation technology to abstract complex user and product information into computable objects, and then uses matching to recommend non-vehicle products, which can improve the efficiency and reliability of the recommendation. On the other hand, since the first vector representation utilizes both user attribute data and user interaction data with non-vehicle products, it strongly represents the personalized characteristics of the user to be recommended and the user's preference characteristics under different non-vehicle products, thus improving the targeting and personalization of the recommendation. Furthermore, since the second vector representation utilizes both non-vehicle product attribute data and data on different users' interactions with non-vehicle products, it strongly represents the characteristics of non-vehicle products and the characteristics of non-vehicle products under different user preferences. Therefore, recommending through matching can improve the accuracy and reliability of the recommendation.

[0064] To facilitate a deeper understanding of the technical solutions provided in this manual, the following is combined with... Figure 4 The recommended methods provided in this manual will be described in more detail.

[0065] in, Figure 4 This is a flowchart illustrating a recommended method for non-vehicle products provided in another embodiment of this specification. (Example:) Figure 4 As shown, the recommended method includes the following steps S401 to S403: S401: Obtain the first vector representation of the user to be recommended, and the second vector representations of M (M is an integer greater than 1) non-vehicle products to be recommended. The first vector representation represents the first user data and the first interaction behavior data of the user with respect to insurance products, including non-vehicle products. The second vector representation of any non-vehicle product represents the product data and the corresponding interaction behavior data of that product.

[0066] It is understood that, to avoid tedious explanations, this embodiment will not repeat the same or similar content as the examples above. For example, regarding the understanding of the first vector representation and the second vector representation, please refer to the examples above, and they will not be repeated here.

[0067] Based on the above analysis, the triggering conditions for recommending non-vehicle products to users may differ. Accordingly, when recommending non-vehicle products to users, the recommendation system can obtain a first vector representation and M second vector representations. Each non-vehicle product to be recommended corresponds to one second vector representation.

[0068] In some embodiments, the recommender system includes a data layer that stores vector representations of user dimensions and vector representations of non-vehicle product dimensions. The user-dimensional vector representation includes a first vector representation, and the non-vehicle product-dimensional vector representation includes M second vector representations.

[0069] For example, such as Figure 5 As shown, the recommendation system may include a data layer. The data layer stores the vector representation of each user and the vector representation of each non-vehicle product to be recommended.

[0070] Correspondingly, when there is a recommendation requirement, such as recommending non-vehicle products to the user to be recommended, the recommendation system can obtain the first vector representation of the user to be recommended and M second vector representations from the data layer.

[0071] For example, the recommendation system could be an insurance agency platform that sells various types of insurance (e.g., M non-motor insurance products to be recommended). The insurance agency platform includes a data layer. This data layer stores the vector representations of each registered user on the insurance agency platform, as well as the vector representations of the M non-motor insurance products to be recommended.

[0072] In this embodiment, by storing the vector representations corresponding to the user and non-vehicle product dimensions in the data layer, such as determining and storing the vector representations corresponding to the user and non-vehicle product dimensions offline, the corresponding vector representations can be obtained when there is a recommendation need. The recommendation of non-vehicle products can be improved by matching the obtained vector representations.

[0073] In some embodiments, the first vector representation and the second vector representation are obtained by feature extraction of the input data to be recommended using a network model, wherein the input data to be recommended is determined based on the first user data, the first interaction behavior data, and the product data and interaction behavior data corresponding to each non-vehicle product to be recommended.

[0074] The network model is trained to determine the corresponding vector representation based on the input data.

[0075] A network model can be understood as a trained machine learning model (such as DNN, Transformer, DeepFM, BERT, etc.). A network model is used to map its inputs to vector representations. For example, the input to a network model is the data to be mapped, and the output is the mapped vector representation. In other words, the network model transforms the input data into a vector representation through features extraction and other processing.

[0076] In other words, the input data to be recommended can be understood as the "raw materials" of the network model. The network model uses these "raw materials" to construct the corresponding vector representation.

[0077] For example, network models can extract, abstract, and combine useful information from the input data to be recommended layer by layer, and finally output a meaningful vector representation.

[0078] In this embodiment, by combining a network model to determine the vector representations, manual intervention in the recommendation process can be reduced, and the content of the input data to be recommended can be better captured, thereby improving the accuracy and reliability of the determined vector representations.

[0079] In some embodiments, the network model includes a general model, a first vector representation, and a second vector representation: obtained by feature extraction of the input data to be recommended using the parameterized knowledge of the general model; and / or The network model includes a vertical model for the auto insurance field. The first vector representation and the second vector representation are obtained by extracting features from the input data to be recommended using the domain knowledge of the vertical model.

[0080] A general-purpose model can also be called a general-purpose large-scale model, a general-purpose language model, a large-scale language model, a basic model, etc. A general-purpose model can be understood as a model pre-trained on a large-scale general-purpose corpus, which has a wide range of language understanding and generation capabilities.

[0081] Through training, a general-purpose model can acquire parametric knowledge, also known as world knowledge. For example, parametric knowledge can be understood as the world knowledge learned by the general-purpose model, and this world knowledge can be implicitly stored in the model parameters. World knowledge can be understood as not being specialized knowledge in a specific field (such as medicine or law), but rather as common-sense knowledge that universally exists in human society.

[0082] Vertical models can also be called vertical mini-models, domain-specific models, task-specific models, etc. In this embodiment, a vertical model can be understood as a model trained in the application scenario of this embodiment to capture knowledge (referred to as domain knowledge) of user behavior (such as clicks, orders, etc.) towards non-vehicle products in that scenario.

[0083] The domain knowledge of a vertical model can be called professional knowledge. Domain knowledge can be understood as the knowledge, rules, terminology, patterns, and contextual understanding capabilities that the vertical model possesses in the domain corresponding to the technical solutions provided in this specification.

[0084] In other words, general-purpose models and vertical-specific models are relative probabilistic models. Relatively speaking, general-purpose models possess knowledge from various fields, while vertical-specific models possess knowledge from their respective fields. That is, general-purpose models have universality, while vertical-specific models have domain-specific specialization.

[0085] For example, there are three ways to determine the first vector representation and the second vector representation: Method 1: Determine the first vector representation and the second vector representation based on the general model.

[0086] Method 2: Determine the first vector representation and the second vector representation based on the vertical category model.

[0087] Method 3: Determine the first vector representation based on the general model and the vertical model; determine the second vector representation based on the general model and the vertical model.

[0088] Based on the above analysis, it can be seen that in this embodiment, the recommender system determines the corresponding vector representation through a general model and / or a vertical model, which can achieve diversity and flexibility in determining the vector representation. Especially when the recommender system combines a general model and a vertical model to jointly determine the vector representation, it can leverage the semantic advantages of the general model while retaining the domain-specific expertise, thereby improving the accuracy and reliability of the determined vector representation.

[0089] In some embodiments, for method 3 above, when the network model includes a general model and a vertical model, the recommendation system determines the first vector representation and the second vector representation based on the general model and the vertical model, which includes: inputting the input data to be recommended into the general model, outputting a first sub-vector representation of the user dimension and a third sub-vector representation of the non-vehicle product dimension; inputting the input data to be recommended into the vertical model, outputting a second sub-vector representation of the user dimension and a fourth sub-vector representation of the non-vehicle product dimension; fusing the first sub-vector representation and the second sub-vector representation to obtain the first vector representation; and fusing the third sub-vector representation and the fourth sub-vector representation to obtain the second vector representation.

[0090] The first subvector representation can be understood as a semantic vector extracted by the general model from the input data to be recommended, mainly from the first user data and the first interaction behavior data, to reflect the intention, interest semantics and other information of the user to be recommended from the user dimension.

[0091] The second sub-vector representation can be understood as a behavioral vector extracted by the vertical model from the input data to be recommended, mainly from the first user data and the first interaction behavior data, to reflect information such as the historical preferences of the user to be recommended from the user dimension.

[0092] The third sub-vector representation can be understood as a semantic vector extracted by the general model from the input data to be recommended, mainly from the product data and interaction behavior data corresponding to each non-vehicle product to be recommended, in order to reflect the textual meaning and other information of the non-vehicle products to be recommended from the product dimension.

[0093] The fourth sub-vector representation can be understood as a behavioral vector extracted by the vertical model from the input data to be recommended, mainly from the product data and interaction behavior data corresponding to each non-vehicle product to be recommended, in order to reflect information such as the popularity, audience profile, and conversion characteristics of the non-vehicle products to be recommended in the market from the product dimension.

[0094] Fusion can be understood as the operation of merging two sub-vector representations into a unified vector representation.

[0095] This embodiment does not limit the fusion method, which can be determined by the recommendation system based on needs, historical records, experiments, etc. Fusion methods may include concat, weighted summation, attention fusion, etc.

[0096] In this embodiment, the recommendation system first obtains sub-vector representations using a general model and a vertical model, and then merges the sub-vector representations into a final vector representation to determine the first vector representation and the second vector representation. This effectively combines semantic and behavioral information, thereby improving the accuracy and reliability of the determined first and second vector representations.

[0097] Based on the above analysis, it can be seen that network models can include general models and vertical models, and recommendation systems can determine vector representations (such as first vector representation and second vector representation) based on general models and vertical models.

[0098] In some embodiments, the general model and the vertical model can be trained in different ways. For example, the general model can be trained in an unsupervised manner, while the vertical model can be trained in a supervised manner.

[0099] Furthermore, the general model and the vertical model obtained through training can be used for recommendation systems or other systems (such as training systems), and this embodiment does not impose any limitations. This embodiment mainly uses a recommendation system as an example for illustration.

[0100] Taking the training of a general model as an example: the base model of the general model can be a large language model, and the large language model can be trained using sample data from various domains through prompts, thereby obtaining the general model.

[0101] Taking the training of a general model as an example: it can be trained using a supervised learning framework, with the label being whether a certain type of insurance is issued to a certain user. The training samples are determined based on user data and interaction behavior data corresponding to the sample users, as well as product data corresponding to the insurance types.

[0102] Supervised learning frameworks can be understood as a machine learning paradigm that uses labeled datasets for training, with the goal of teaching the model to predict output labels from input features.

[0103] Whether a policy has been issued can be understood as whether a user has successfully purchased insurance for a non-vehicle product and generated a policy.

[0104] Training samples can be understood as the data used to train the vertical category model. Training samples may include user data (such as user profile data) for each user, data obtained from each user's interaction behavior with each non-vehicle product, and product data for each non-vehicle product.

[0105] In addition, the training samples may also include data for each non-vehicle product under each user interaction behavior. Since the data for each non-vehicle product under each user interaction behavior can be determined through the aforementioned data obtained from user interactions with each non-vehicle product, this data is not emphasized in the training samples.

[0106] In some embodiments, training samples are obtained by processing user data and interaction behavior data corresponding to sample users, as well as product data corresponding to insurance types. The processing includes statistics and / or determining interaction preferences.

[0107] Combining the above examples and Figure 5 It can be seen that a recommendation system may include a data layer, which includes user data corresponding to each user, interaction behavior data (such as interaction behavior data of each user on different non-vehicle products, and interaction behavior data of different users on each non-vehicle product), and product data corresponding to each non-vehicle product.

[0108] The data in this data layer can be encrypted. This allows for the consumption of various data in a privacy-preserving environment, thereby preventing privacy leaks and improving data security.

[0109] Recommendation systems can process data in the data layer, such as performing statistical analysis and determining interaction preferences, to obtain processed data. For example, by processing data in the data layer, a recommendation system can obtain processed data such as: the frequency of user purchases of car insurance, accident insurance, and medical insurance; user preferences for insurance company purchases / clicks; and the coverage amount / premium / profit margin of accident insurance and medical insurance.

[0110] Correspondingly, recommendation systems can train vertical and general models based on the processed data.

[0111] In this embodiment, the recommendation system can train the vertical category model with "predicting a single probability" as the objective. However, in specific non-vehicle product recommendation applications, the recommendation system can focus more on the latent vector output of its intermediate layers to obtain the corresponding vector representation.

[0112] Based on the above analysis, it can be seen that in this embodiment, by employing a supervised learning framework and using whether a corresponding insurance product is sold to a corresponding user as a label, the vertical model trained can achieve accurate conversion prediction. For example, it can recommend products not only based on dimensions of interest but also on conversion rates, such as the dimension of the most likely transaction, thereby improving the effectiveness, reliability, and accuracy of the recommendations. This allows the vector representation determined based on the vertical model to also contain information about conversion logic, further enhancing the practicality and accuracy of recommendations based on the corresponding vector representation.

[0113] S402: Match the first vector representation with each of the M second vector representations to obtain the candidate non-vehicle products corresponding to the first N matched second vectors, where M≥N≥1.

[0114] like Figure 6As shown, the M second vector representations include second vector representation 1, second vector representation 2, and so on up to second vector representation M. The recommendation system can match the first vector representation with each of the second vector representations 1, 2, and M to obtain M matching results. It then selects the top N most closely matched second vectors and identifies the N non-vehicle products corresponding to these top N second vectors as candidate non-vehicle products. For example, as shown... Figure 6 The candidate non-vehicle product 1, candidate non-vehicle product 2, up to candidate non-vehicle product N are shown.

[0115] In some embodiments, S402 may include: calculating the similarity between the first vector representation and M second vector representations respectively, and determining the top N non-vehicle products with the highest similarity as candidate non-vehicle products.

[0116] Continuing with the example above, the recommendation system can calculate the similarity between the first vector representation and the second vector representation 1, obtaining similarity 1; calculate the similarity between the first vector representation and the second vector representation 2, obtaining similarity 2, and so on, until the similarity between the first vector representation and the second vector representation M is calculated, obtaining similarity M.

[0117] Recommendation systems can compare similarity scores from 1 to M to obtain the top N similarities. For example, a recommendation system can sort the M similarities in descending (or ascending) order to determine the top N (or last N) similarities as the highest similarity scores.

[0118] Based on this, the recommendation system can identify the top N non-vehicle products with the highest similarity as candidate non-vehicle products.

[0119] Similarly, this embodiment does not limit the method for calculating similarity, which can be determined by the recommendation system based on needs, historical records, experiments, etc. For example, the recommendation system can use cosine similarity to calculate it.

[0120] In some embodiments, combined with Figure 5 It can be seen that the recommendation system can also include an application layer, which can include a recall sub-layer, and the recall sub-layer can include a non-vehicle recall model.

[0121] Based on the above analysis, the input of the non-vehicle recall model can include a first vector representation and M second vector representations, and the output of the non-vehicle recall model is N candidate non-vehicle products.

[0122] Based on the above analysis, it can be seen that in this embodiment, by calculating similarity, N candidate non-vehicle products are determined from M non-vehicle products to be recommended, which can make the determined candidate non-vehicle products have high accuracy and reliability.

[0123] S403: Based on the first user data and the product data corresponding to each of the N candidate non-vehicle products, determine the target non-vehicle product from the N candidate non-vehicle products.

[0124] Continuing with the examples above and Figure 6 Given N candidate non-vehicle products, the recommendation system combines the user dimension data of the user to be recommended (i.e., the first user data) and the product dimension data (i.e., the product data) of each of the N candidate non-vehicle products to determine the target non-vehicle product and recommend the determined target non-vehicle product to the user to be recommended.

[0125] The number of target non-vehicle products can be one or more. For example, using the recommendation method described above, the recommendation system can identify candidate non-vehicle product 1 and candidate non-vehicle product 2 as target non-vehicle products, and recommend candidate non-vehicle product 1 and candidate non-vehicle product 2 to the user to be recommended.

[0126] Based on the above analysis of S402 and S403, it can be seen that in this embodiment, the recommendation system first obtains N candidate non-vehicle products through matching, and then determines the final target non-vehicle product to be recommended to the user by combining user-dimensional data and product-dimensional data. This can ensure that the target non-vehicle product meets the recommendation requirements from both user and product dimensions, thereby improving the accuracy and reliability of the recommendation.

[0127] In some embodiments, S403 may include: inputting the first user data and the product data corresponding to each of the N candidate non-vehicle products into a pre-trained deep learning model to obtain the rating information of each candidate non-vehicle product applicable to the user to be recommended, wherein the deep learning model is trained to predict the rating information of the corresponding non-vehicle product applicable to the corresponding user based on the user data and product data, and to determine the target non-vehicle product from the N candidate non-vehicle products based on the N rating information.

[0128] Similarly, the deep learning model trained can be used for recommendation systems or other systems (such as training systems), and this embodiment is not limited to any particular system. Furthermore, this specification primarily uses a recommendation system as an example for illustration.

[0129] For example, a recommender system can be trained to generate a deep learning model that predicts the rating information of a non-vehicle product for a given user based on user data and product data. For instance, the input to the deep learning model includes user data and product data, enabling the model to learn how to determine the rating information of a non-vehicle product for a given user based on the input.

[0130] Specifically, the input to the deep learning model includes user data from multiple users and product data from multiple non-vehicle products. The deep learning model learns to determine the rating information applicable to any non-vehicle product for any user.

[0131] Similarly, this embodiment does not limit the framework, structure, type, or training method of the deep model, which can be determined by the recommendation system based on requirements, historical records, experiments, etc.

[0132] Continuing with the example above, after obtaining N candidate non-vehicle products, the recommendation system can determine the corresponding rating information for each candidate non-vehicle product. For example, rating information 1 for candidate non-vehicle product 1, rating information 2 for candidate non-vehicle product 2, and so on up to rating information N for candidate non-vehicle product N.

[0133] Among them, rating information 1 can be understood as the rating of candidate non-vehicle product 1 for the user to be recommended.

[0134] Correspondingly, the recommendation system can determine the target non-vehicle product from candidate non-vehicle product 1 to candidate non-vehicle product N based on N rating information.

[0135] Based on the above analysis and Figure 5 It can be seen that the recall sub-layer can also include a fine-ranking model, which can be a deep learning model. The input of the fine-ranking model includes the first user data and the product data corresponding to each of the N candidate non-vehicle products. The output is the rating information, and the recommendation system can determine the target non-vehicle product based on the rating information.

[0136] In this embodiment, the recommendation system determines the scoring information through a deep learning model to identify target non-vehicle products from the perspective of their suitability for users. This can improve the automation and intelligence of identifying target non-vehicle products, as well as enhance the effectiveness and reliability of the identified target non-vehicle products.

[0137] In some embodiments, the N rating information includes score pairs corresponding to the user to be recommended and the N candidate non-vehicle products, and the score pairs include the scores of the corresponding candidate non-vehicle products applicable to the user to be recommended.

[0138] And, based on N scoring information, determine the target non-vehicle product from N candidate non-vehicle products, including: determining the candidate non-vehicle product with the highest score from N candidate non-vehicle products as the target non-vehicle product.

[0139] For example, each candidate non-vehicle product corresponds to a scoring pair. For instance, candidate non-vehicle product 1 corresponds to a scoring pair, which includes a score representing the suitability of the candidate non-vehicle product for the user to be recommended to. Relatively speaking, the higher the score, the more suitable candidate non-vehicle product 1 is to be recommended to the user. Conversely, the lower the score, the less suitable candidate non-vehicle product 1 is to be recommended to the user.

[0140] For example, if N=3, meaning the recommendation system identifies 3 candidate non-vehicle products: Accident Insurance A, Accident Insurance B, and Health Insurance A, then after the above method, the recommendation system can obtain 3 score pairs: <User to be recommended, Accident Insurance A, 0.9>, <User to be recommended, Accident Insurance B, 0.94>, and <User to be recommended, Health Insurance A, 0.88>.

[0141] In other words, the score for candidate non-motor accident insurance A is 0.9 for the user to be recommended; the score for candidate non-motor accident insurance B is 0.94 for the user to be recommended; and the score for candidate non-motor health insurance A is 0.88 for the user to be recommended.

[0142] Accordingly, the recommendation system can identify the candidate non-motor accident insurance A with the highest score (0.94) as the target non-motor product and recommend candidate non-motor accident insurance A to the user to be recommended.

[0143] Based on the above analysis, it can be seen that in this embodiment, the recommendation system determines the target non-vehicle product by combining the scores in the scoring pairs, so as to recommend the candidate non-vehicle product that is most suitable for the user to be recommended, thereby improving the accuracy and reliability of the recommendation.

[0144] In some embodiments, a scoring pair includes scores corresponding to multiple indicator dimensions, including conversion rate, profit margin, and total transaction volume (GMV).

[0145] And, from N candidate non-vehicle products, the candidate non-vehicle product with the highest score is determined as the target non-vehicle product, including: determining the total score corresponding to each candidate non-vehicle product, and determining the candidate non-vehicle product with the highest total score as the target non-vehicle product.

[0146] For example, a scoring pair may include multiple scores. A scoring pair could include three scores: a score for conversion rate, a score for profit margin, and a score for total transaction amount.

[0147] The conversion rate score can be understood as the score by which non-vehicle products are deemed suitable for users, determined from the perspective of conversion rate. Relatively speaking, all other things being equal, non-vehicle products with higher conversion rates have relatively higher scores.

[0148] The profit margin score can be understood as the score by which non-vehicle products are deemed suitable for users, determined from the perspective of profit margin. Relatively speaking, all other things being equal, non-vehicle products with higher profit margins will have higher scores.

[0149] The score corresponding to the total transaction amount can be understood as the score for non-vehicle products applicable to users, determined from the perspective of the total transaction amount. Relatively speaking, all other things being equal, non-vehicle products with higher total transaction amounts have higher scores.

[0150] For any candidate non-vehicle product, the recommendation system can determine a total score based on the scores of each candidate non-vehicle product. If N total scores are obtained, the recommendation system can compare these N total scores to determine the highest total score, and then recommend the candidate non-vehicle product corresponding to the highest total score to the user.

[0151] This embodiment does not limit the method of determining the total score based on the scores of each dimension. For example, in Implementation 1, the recommendation system can determine the total score by summing and averaging; in Implementation 2, the recommendation system can determine the total score by weighting the scores of each dimension according to their respective weights.

[0152] Implementation method 1: The score pair 1 of candidate non-vehicle product 1 includes score 11, score 12, and score 13. The recommendation system can calculate the total score of candidate non-vehicle product 1 by (score 11 + score 12 + score 13) / 3.

[0153] Implementation Method 2: The score pair 1 of candidate non-vehicle product 1 includes score 11, score 12, and score 13. The recommendation system can calculate the total score of candidate non-vehicle product 1 using (a*score 11 + b*score 12 + c*score 13). a, b, and c are preset weight coefficients.

[0154] Based on the above analysis, it can be seen that in this embodiment, by determining the scores of candidate non-vehicle products applicable to the user to be recommended from different dimensions, and recommending the corresponding non-vehicle products to the user to be recommended based on the total score, it is possible to consider factors from different dimensions when recommending non-vehicle products, thereby further improving the effectiveness and reliability of the recommendation.

[0155] In some embodiments, for implementation method 2 described above, the weighting coefficients can be adjusted based on requirements, etc.

[0156] For example, if the profit margin needs to be increased this month, that is, to recommend corresponding non-vehicle products to users based on the increased profit margin, the recommendation system can adjust the weight coefficient corresponding to the profit margin to a larger value than before.

[0157] For example, continuing to combine the above examples and Figure 5 The application layer can also include a sub-layer for dynamically changing metrics. This sub-layer includes dynamic adjustments to metrics such as conversion rate, profit margin, and total transaction volume.

[0158] It is worth noting that the above examples and Figure 5 This is merely an illustrative example of possible layered structures for a recommendation system and should not be construed as a limitation on the layered structure of a recommendation system. For example, the data layer can be further divided into a user data layer and a product data layer, or it can be further divided into a data layer and a vector representation layer; similarly, the application layer can be divided into an application layer and a metric layer; and so on, which will not be listed here.

[0159] In this specification, the Large Language Model (LLM) may also be referred to simply as the Large Model. A Large Language Model is a natural language processing model based on deep learning techniques, typically with billions to hundreds of billions or even more parameters, possessing powerful language understanding and generation capabilities. Large Language Models can employ the Transformer architecture or its variants (such as GPT, BERT, etc.), which utilizes an attention mechanism to globally model sequential data, efficiently handling long-distance dependencies and thus performing exceptionally well in natural language tasks. Large Language Models learn the statistical features and semantic relationships of language through pre-training on large-scale corpora, giving them outstanding generalization capabilities. The core capabilities of Large Language Models include, but are not limited to: understanding contextual semantics, generating coherent and grammatically correct text, performing logical reasoning, and handling multi-task scenarios. Its usage typically includes two modes: direct inference and fine-tuning. In direct inference mode, the user guides the Large Language Model to generate specific outputs by designing prompts. Cue words can be task descriptions or instructions in text form, used to stimulate the semantic understanding and generation capabilities of large language models. In fine-tuning mode, large language models are further trained on small-scale datasets in specific domains to optimize their performance on specific tasks. The powerful generalization ability and flexibility of large language models make them an important tool in the field of artificial intelligence, providing efficient and accurate solutions for automated text generation and understanding.

[0160] In some embodiments, large language models can also understand and generate data from other modalities (such as visual and audio data). In this case, large language models can also be called multimodal large language models (MLLMs). MLLMs provide a richer and more natural interactive experience by integrating multiple types of input and output, such as text, images, and sound. The core advantage of MLLMs lies in their ability to process and understand information from different modalities and fuse this information to complete complex tasks. For example, MLLMs can analyze an image and generate descriptive text, or generate a corresponding image based on a text description. This cross-modal understanding and generation capability makes MLLMs widely applicable across multiple fields.

[0161] It should be noted that the key technologies of large language models can be found in the detailed description in the paper "A Survey of Large Language Models" (paper number: arXiv:2303.18223v16, published on March 11, 2025, public link: https: / / doi.org / 10.48550 / arXiv.2303.18223), and will not be repeated here.

[0162] It is worth noting that the above examples are merely illustrative of possible implementations of the recommended methods in this specification, and should not be construed as limiting the implementation of the recommended methods in this specification. For example, based on the above technical concepts, some of the technical features described above can be combined to obtain new embodiments; new technical features can be added to the above examples to obtain new embodiments; some technical features can be removed from the above examples to obtain new embodiments; some technical features in the above examples can be replaced with other technical features; some technical features and their order in the above examples can be adjusted to obtain new embodiments, and so on, which will not be listed here.

[0163] Based on the above-described technical concept, this specification also provides a computer-readable non-transitory storage medium storing at least one instruction set, which, when executed by a processor, performs the steps of the recommended method described in this specification.

[0164] In some possible implementations, various aspects of this specification can also be implemented as a program product comprising program code. When the program product is run on the recommendation system 200, the program code causes the recommendation system 200 to perform the steps of the recommendation method described in this specification. The program product for implementing the above method may employ a portable compact disc read-only memory (CD-ROM) containing program code and may run on the recommendation system 200. However, the program product of this specification is not limited thereto. In this specification, a readable storage medium may be any tangible medium containing or storing a program that may be used by or in conjunction with an instruction execution system. The program product may employ any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of readable storage media include: electrical connections having one or more wires, portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. The computer-readable storage medium may include data signals propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable storage medium may also be any readable medium other than a readable storage medium that can send, propagate, or transmit programs for use by or in connection with an instruction execution system, apparatus, or device. Program code contained on a readable storage medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof. Program code for performing the operations described herein can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the recommendation system 200, partially on the recommendation system 200, as a standalone software package, partially on the recommendation system 200 and partially on a remote recommendation system, or entirely on the remote recommendation system 200.

[0165] It should be noted that the collection, storage, use, processing, transmission, provision, and disclosure of user-related information (such as user data and interaction behavior data) involved in the technical solutions of this specification all comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0166] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0167] In summary, after reading this detailed disclosure, those skilled in the art will understand that the foregoing detailed disclosure is presented by way of example only and is not restrictive. Although not explicitly stated herein, those skilled in the art will understand that this specification requires various reasonable changes, improvements, and modifications to the embodiments. These changes, improvements, and modifications are intended to be made by this specification and are within the spirit and scope of the exemplary embodiments described herein.

[0168] Furthermore, certain terms in this specification have been used to describe embodiments of this specification. For example, "an embodiment," "an embodiment," and / or "some embodiments" mean that a particular feature, structure, or characteristic described in connection with that embodiment may be included in at least one embodiment of this specification. Therefore, it is to be emphasized and understood that two or more references to "an embodiment" or "an embodiment" or "alternative embodiment" in various parts of this specification do not necessarily refer to the same embodiment. Moreover, specific features, structures, or characteristics may be suitably combined in one or more embodiments of this specification.

[0169] It should be understood that in the foregoing description of the embodiments in this specification, various features are combined in a single embodiment, drawing, or description for the purpose of simplifying the description and to aid in understanding a feature. However, this does not mean that the combination of these features is necessary, and those skilled in the art, upon reading this specification, may readily identify some of the devices as separate embodiments. That is, the embodiments in this specification can also be understood as an integration of multiple secondary embodiments. And the content of each secondary embodiment is valid even if it contains fewer than all the features of a single foregoing disclosed embodiment.

[0170] Every patent, patent application, publication of a patent application, and other material cited herein, such as articles, books, specifications, publications, documents, and literature (excluding any related historical examination documents), is referenced for all purposes relevant to this document, including in the specification and claims herein. However, in the event of any inconsistency or conflict between the descriptions, definitions, and / or terms used in the foregoing and those used herein, the descriptions, definitions, and / or terms used herein shall prevail.

[0171] Finally, it should be understood that the embodiments disclosed herein are illustrative of the principles of the embodiments described in this specification. Other modified embodiments are also within the scope of this specification. Therefore, the embodiments disclosed in this specification are merely examples and not limitations. Those skilled in the art can implement the applications described in this specification using alternative configurations based on the embodiments in this specification. Therefore, the embodiments in this specification are not limited to the embodiments precisely described in the applications.

Claims

1. A method for recommending non-vehicle products, comprising: A first vector representation of the user to be recommended is determined, wherein the first vector representation represents the first user data of the user to be recommended and the first interaction behavior data of the user to be recommended with the insurance type, and the insurance type includes non-vehicle products; The first vector representation is matched with the second vector representation corresponding to each non-vehicle product to be recommended to obtain the target non-vehicle product recommended to the user to be recommended. The second vector representation of any non-vehicle product represents the product data of the non-vehicle product and the interaction behavior data corresponding to the non-vehicle product.

2. The method according to claim 1, wherein, The number of non-vehicle products to be recommended is M; the first vector representation is matched with the second vector representation corresponding to each non-vehicle product to be recommended, to obtain the target non-vehicle products recommended to the user, including: The first vector representation is matched with each of the M second vector representations to obtain the candidate non-vehicle products corresponding to the top N matching second vectors, where M ≥ N ≥ 1; and Based on the first user data and the product data corresponding to each of the N candidate non-vehicle products, the target non-vehicle product is determined from the N candidate non-vehicle products.

3. The method according to claim 2, wherein, Based on the first user characteristics and the product data corresponding to each of the N candidate non-vehicle products, the target non-vehicle product is determined from the N candidate non-vehicle products, including: The first user data and the product data corresponding to each of the N candidate non-vehicle products are input into a pre-trained deep learning model to obtain the rating information of each candidate non-vehicle product applicable to the user to be recommended. The deep learning model is trained to predict the rating information of the corresponding non-vehicle product applicable to the corresponding user based on the user data and product data. Based on N scoring information, the target non-vehicle product is determined from the N candidate non-vehicle products.

4. The method according to claim 3, wherein, The N rating information includes the score pairs corresponding to the user to be recommended and the N candidate non-vehicle products, and the score pairs include the scores of the corresponding candidate non-vehicle products applicable to the user to be recommended. And, based on N scoring information, determining the target non-vehicle product from the N candidate non-vehicle products includes: determining the candidate non-vehicle product with the highest score from the N candidate non-vehicle products as the target non-vehicle product.

5. The method according to claim 4, wherein, A scoring pair includes scores for multiple indicator dimensions, including conversion rate, profit margin, and total transaction amount. And, from the N candidate non-vehicle products, the candidate non-vehicle product with the highest score is determined as the target non-vehicle product, including: determining the total score corresponding to each candidate non-vehicle product, and determining the candidate non-vehicle product with the highest total score as the target non-vehicle product.

6. The method according to claim 1, wherein, The first vector representation and the second vector representation are obtained by feature extraction of the input data to be recommended using a network model. The input data to be recommended is determined based on the first user data, the first interaction behavior data, and the product data and interaction behavior data corresponding to each of the non-vehicle products to be recommended. The network model is trained to determine the corresponding vector representation based on the input data.

7. The method according to claim 6, wherein, The network model includes a general model, and the first vector representation and the second vector representation are obtained by extracting features from the input data to be recommended using the parameterized knowledge of the general model; and / or The network model includes a vertical model for the auto insurance field. The first vector representation and the second vector representation are obtained by extracting features from the input data to be recommended using the domain knowledge of the vertical model.

8. The method according to claim 7, wherein, The network model includes the general model and the vertical model; determining the first vector representation and the second vector representation using the network model includes: The input data to be recommended is input into the general model, and the first sub-vector representation of the user dimension and the third sub-vector representation of the non-vehicle product dimension are output. The input data to be recommended is input into the vertical category model, which outputs a second sub-vector representation of the user dimension and a fourth sub-vector representation of the non-vehicle product dimension; and The first sub-vector representation and the second sub-vector representation are fused to obtain the first vector representation, and the third sub-vector representation and the fourth sub-vector representation are fused to obtain the second vector representation.

9. The method according to claim 7, wherein, The vertical category model is obtained by training the training samples using a supervised learning framework and labeling whether a corresponding insurance product is issued to a corresponding user. The training samples are determined based on user data and interaction behavior data corresponding to the sample users, as well as product data corresponding to the insurance products.

10. The method according to claim 9, wherein, The training samples are obtained by processing user data and interaction behavior data corresponding to sample users, as well as product data corresponding to insurance types. The processing includes statistics and / or determining interaction preferences.

11. The method according to claim 2, wherein, The first vector representation is matched with M second vector representations to obtain the candidate non-vehicle products corresponding to the top N matching second vector representations, including: Calculate the similarity between the first vector representation and the M second vector representations respectively; and The top N non-vehicle products with the highest similarity were identified as candidate non-vehicle products.

12. The method according to claim 1, wherein, The method is applied to a recommendation system for non-vehicle products. The recommendation system includes a data layer, which stores vector representations of user dimensions and vector representations of non-vehicle product dimensions. The vector representation of the user dimension includes the first vector representation, and the vector representation of the non-vehicle product dimension includes the second vector representation corresponding to each non-vehicle product to be recommended.

13. A recommendation system for non-vehicle products, comprising: At least one storage medium stores at least one instruction set for recommending non-vehicle products; At least one processor is communicatively connected to the at least one storage medium, wherein when the at least one processor is running, it reads the at least one instruction set and executes the method as described in any one of claims 1 to 12 according to the instructions of the at least one instruction set.

Images